Spark 3.5.x on Java 21?

[Stephen Coy]

Hi everyone,

We’re about to upgrade our Spark clusters from Java 11 and Spark 3.2.1 to Spark 
3.5.1.

I know that 3.5.1 is supposed to be fine on Java 17, but will it run OK on Java 
21?

Thanks,

Steve C


This email contains confidential information of and is the copyright of 
Infomedia. It must not be forwarded, amended or disclosed without consent of 
the sender. If you received this message by mistake, please advise the sender 
and delete all copies. Security of transmission on the internet cannot be 
guaranteed, could be infected, intercepted, or corrupted and you should ensure 
you have suitable antivirus protection in place. By sending us your or any 
third party personal details, you consent to (or confirm you have obtained 
consent from such third parties) to Infomedia’s privacy policy. 
http://www.infomedia.com.au/privacy-policy/

Re: [Spark Streaming]: Save the records that are dropped by watermarking in spark structured streaming

[Mich Talebzadeh]

you may consider

- Increase Watermark Retention: Consider increasing the watermark retention
duration. This allows keeping records for a longer period before dropping
them. However, this might increase processing latency and violate
at-least-once semantics if the watermark lags behind real-time.

OR

- Use a separate stream for dropped records: Create a separate streaming
pipeline to process the dropped records. Try:


   - Filter: Filter out records older than the watermark in the main
   pipeline.  say

   resultC = streamingDataFrame.select( \
 col("parsed_value.rowkey").alias("rowkey") \
   , col("parsed_value.timestamp").alias("timestamp") \
   , col("parsed_value.temperature").alias("temperature"))

"""
We work out the window and the AVG(temperature) in the window's
timeframe below
This should return back the following Dataframe as struct

 root
 |-- window: struct (nullable = false)
 ||-- start: timestamp (nullable = true)
 ||-- end: timestamp (nullable = true)
 |-- avg(temperature): double (nullable = true)

"""
resultM = resultC. \
 *withWatermark("timestamp", "5 minutes").* \
 groupBy(window(resultC.timestamp, "5 minutes", "5
minutes")). \
 avg('temperature')

   - Write to Sink: Write the filtered records (dropped records) to a
   separate Kafka topic.
   - Consume and Store: Consume the dropped records topic with another
   streaming job and store them in a Postgres table or S3 using lib


HTH

Mich Talebzadeh,
Technologist | Architect | Data Engineer  | Generative AI | FinCrime
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Wed, 8 May 2024 at 05:13, Nandha Kumar  wrote:

> Hi Team,
>We are trying to use *spark structured streaming *for our use
> case. We will be joining 2 streaming sources(from kafka topic) with
> watermarks. As time progresses, the records that are prior to the watermark
> timestamp are removed from the state. For our use case, we want to *store
> these dropped records* in some postgres table or s3.
>
> When searching, we found a similar question
> in
> StackOverflow which is unanswered.
> *We would like to know how to store these dropped records due to the
> watermark.*
>

Spark not creating staging dir for insertInto partitioned table

[Sanskar Modi]

Hi Folks,

I wanted to check why spark doesn't create staging dir while doing an
insertInto on partitioned tables. I'm running below example code –
```
spark.sql("set hive.exec.dynamic.partition.mode=nonstrict")

val rdd = sc.parallelize(Seq((1, 5, 1), (2, 1, 2), (4, 4, 3)))
val df = spark.createDataFrame(rdd)
df.write.insertInto("testing_table") // testing table is partitioned on "_1"
```
In this scenario FileOutputCommitter considers table path as output path
and creates temporary folders like
`/testing_table/_temporary/0` and then moves them to the
partition location when the job commit happens.

But in-case if multiple parallel apps are inserting into the same
partition, this can cause race condition issues while deleting the
`_temporary` dir. Ideally for each app there should be a unique staging dir
where the job should write its output.

Is there any specific reason for this? or am i missing something here?
Thanks for your time and assistance regarding this!

Kind regards
Sanskar

[Spark Streaming]: Save the records that are dropped by watermarking in spark structured streaming

[Nandha Kumar]

Hi Team,
   We are trying to use *spark structured streaming *for our use case.
We will be joining 2 streaming sources(from kafka topic) with watermarks.
As time progresses, the records that are prior to the watermark timestamp
are removed from the state. For our use case, we want to *store these
dropped records* in some postgres table or s3.

When searching, we found a similar question
in
StackOverflow which is unanswered.
*We would like to know how to store these dropped records due to the
watermark.*

unsubscribe

[Wojciech Bombik]

unsubscribe

unsubscribe

[Moise]

unsubscribe

Re: ********Spark streaming issue to Elastic data**********

[Mich Talebzadeh]

Hi Kartrick,

Unfortunately Materialised views are not available in Spark as yet. I
raised Jira [SPARK-48117] Spark Materialized Views: Improve Query
Performance and Data Management - ASF JIRA (apache.org)
 as a feature request.

Let me think about another way and revert

HTH

Mich Talebzadeh,
Technologist | Architect | Data Engineer  | Generative AI | FinCrime
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Mon, 6 May 2024 at 07:54, Karthick Nk  wrote:

> Thanks Mich,
>
> can you please confirm me is my understanding correct?
>
> First, we have to create the materialized view based on the mapping
> details we have by using multiple tables as source(since we have multiple
> join condition from different tables). From the materialised view we can
> stream the view data into elastic index by using cdc?
>
> Thanks in advance.
>
> On Fri, May 3, 2024 at 3:39 PM Mich Talebzadeh 
> wrote:
>
>> My recommendation! is using materialized views (MVs) created in Hive with
>> Spark Structured Streaming and Change Data Capture (CDC) is a good
>> combination for efficiently streaming view data updates in your scenario.
>>
>> HTH
>>
>> Mich Talebzadeh,
>> Technologist | Architect | Data Engineer  | Generative AI | FinCrime
>> London
>> United Kingdom
>>
>>
>>view my Linkedin profile
>> 
>>
>>
>>  https://en.everybodywiki.com/Mich_Talebzadeh
>>
>>
>>
>> *Disclaimer:* The information provided is correct to the best of my
>> knowledge but of course cannot be guaranteed . It is essential to note
>> that, as with any advice, quote "one test result is worth one-thousand
>> expert opinions (Werner
>> Von Braun
>> )".
>>
>>
>> On Thu, 2 May 2024 at 21:25, Karthick Nk  wrote:
>>
>>> Hi All,
>>>
>>> Requirements:
>>> I am working on the data flow, which will use the view definition(view
>>> definition already defined in schema), there are multiple tables used in
>>> the view definition. Here we want to stream the view data into elastic
>>> index based on if any of the table(used in the view definition) data got
>>> changed.
>>>
>>>
>>> Current flow:
>>> 1. we are inserting id's from the table(which used in the view
>>> definition) into the common table.
>>> 2. From the common table by using the id, we will be streaming the view
>>> data (by using if any of the incomming id is present in the collective id
>>> of all tables used from view definition) by using spark structured
>>> streaming.
>>>
>>>
>>> Issue:
>>> 1. Here we are facing issue - For each incomming id here we running view
>>> definition(so it will read all the data from all the data) and check if any
>>> of the incomming id is present in the collective id's of view result, Due
>>> to which it is taking more memory in the cluster driver and taking more
>>> time to process.
>>>
>>>
>>> I am epxpecting an alternate solution, if we can avoid full scan of view
>>> definition every time, If you have any alternate deisgn flow how we can
>>> achieve the result, please suggest for the same.
>>>
>>>
>>> Note: Also, it will be helpfull, if you can share the details like
>>> community forum or platform to discuss this kind of deisgn related topics,
>>> it will be more helpfull.
>>>
>>

Re: ********Spark streaming issue to Elastic data**********

[Karthick Nk]

Thanks Mich,

can you please confirm me is my understanding correct?

First, we have to create the materialized view based on the mapping details
we have by using multiple tables as source(since we have multiple
join condition from different tables). From the materialised view we can
stream the view data into elastic index by using cdc?

Thanks in advance.

On Fri, May 3, 2024 at 3:39 PM Mich Talebzadeh 
wrote:

> My recommendation! is using materialized views (MVs) created in Hive with
> Spark Structured Streaming and Change Data Capture (CDC) is a good
> combination for efficiently streaming view data updates in your scenario.
>
> HTH
>
> Mich Talebzadeh,
> Technologist | Architect | Data Engineer  | Generative AI | FinCrime
> London
> United Kingdom
>
>
>view my Linkedin profile
> 
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> *Disclaimer:* The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner  Von
> Braun )".
>
>
> On Thu, 2 May 2024 at 21:25, Karthick Nk  wrote:
>
>> Hi All,
>>
>> Requirements:
>> I am working on the data flow, which will use the view definition(view
>> definition already defined in schema), there are multiple tables used in
>> the view definition. Here we want to stream the view data into elastic
>> index based on if any of the table(used in the view definition) data got
>> changed.
>>
>>
>> Current flow:
>> 1. we are inserting id's from the table(which used in the view
>> definition) into the common table.
>> 2. From the common table by using the id, we will be streaming the view
>> data (by using if any of the incomming id is present in the collective id
>> of all tables used from view definition) by using spark structured
>> streaming.
>>
>>
>> Issue:
>> 1. Here we are facing issue - For each incomming id here we running view
>> definition(so it will read all the data from all the data) and check if any
>> of the incomming id is present in the collective id's of view result, Due
>> to which it is taking more memory in the cluster driver and taking more
>> time to process.
>>
>>
>> I am epxpecting an alternate solution, if we can avoid full scan of view
>> definition every time, If you have any alternate deisgn flow how we can
>> achieve the result, please suggest for the same.
>>
>>
>> Note: Also, it will be helpfull, if you can share the details like
>> community forum or platform to discuss this kind of deisgn related topics,
>> it will be more helpfull.
>>
>

unsubscribe

[chen...@birdiexx.com]

unsubscribe

unsubscribe

[Bing]

 Replied Message 
| From | Wood Super |
| Date | 05/01/2024 07:49 |
| To | user  |
| Subject | unsubscribe |
unsubscribe

Spark Materialized Views: Improve Query Performance and Data Management

[Mich Talebzadeh]

Hi,

I have raised a ticket SPARK-48117
 for enhancing Spark
capabilities with Materialised Views (MV). Currently both Hive and
Databricks support this. I have added these potential benefits  to the
ticket

-* Improved Query Performance (especially for Streaming Data):*
Materialized Views can significantly improve query performance,
particularly for use cases involving Spark Structured Streaming. When
dealing with continuous data streams, materialized views can pre-compute
and store frequently accessed aggregations or transformations. Subsequent
queries on the materialized view can retrieve the results much faster
compared to continuously processing the entire streaming data. This is
crucial for real-time analytics where low latency is essential.
*Enhancing Data Management:* They offer a way to pre-aggregate or transform
data, making complex queries more efficient.
- *Reduced Data Movement*: Materialized Views can be materialized on
specific clusters or storage locations closer to where the data will be
consumed. This minimizes data movement across the network, further
improving query performance and reducing overall processing time.
- *Simplified Workflows:* Developers and analysts can leverage pre-defined
Materialized Views that represent specific business logic or data subsets.
This simplifies data access, reduces development time for queries that rely
on these views, and fosters code reuse.

Please have a look at the ticket and add your comments.

Thanks

Mich Talebzadeh,

Technologist | Architect | Data Engineer  | Generative AI | FinCrime

London
United Kingdom


   view my Linkedin profile


 https://en.everybodywiki.com/Mich_Talebzadeh



Disclaimer: The information provided is correct to the best of my knowledge
but of course cannot be guaranteed . It is essential to note that, as with
any advice, quote "one test result is worth one-thousand expert opinions
(Werner Von Braun)".

Re: Issue with Materialized Views in Spark SQL

[Mich Talebzadeh]

Sadly Apache Spark sounds like it has nothing to do within materialised
views. I was hoping it could read it!

>>> *spark.sql("SELECT * FROM test.mv ").show()*
Traceback (most recent call last):
  File "", line 1, in 
  File "/opt/spark/python/pyspark/sql/session.py", line 1440, in sql
return DataFrame(self._jsparkSession.sql(sqlQuery, litArgs), self)
  File
"/usr/src/Python-3.9.16/venv/venv3.9/lib/python3.9/site-packages/py4j/java_gateway.py",
line 1321, in __call__
return_value = get_return_value(
  File "/opt/spark/python/pyspark/errors/exceptions/captured.py", line 175,
in deco
raise converted from None
*Pyspark.errors.exceptions.captured.AnalysisException: Hive materialized
view is not supported.*


HTH

Mch Talebzadeh,
Technologist | Architect | Data Engineer  | Generative AI | FinCrime
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Fri, 3 May 2024 at 11:03, Mich Talebzadeh 
wrote:

> Thanks for the comments I received.
>
> So in summary, Apache Spark itself doesn't directly manage materialized
> views,(MV)  but it can work with them through integration with the
> underlying data storage systems like Hive or through iceberg. I believe
> databricks through unity catalog support MVs as well.
>
> Moreover, there is a case for supporting MVs. However, Spark can utilize
> materialized views even though it doesn't directly manage them.. This came
> about because someone in the Spark user forum enquired about "Spark
> streaming issue to Elastic data*". One option I thought of was that uUsing
> materialized views with Spark Structured Streaming and Change Data Capture
> (CDC) is a potential solution for efficiently streaming view data updates
> in this scenario. .
>
>
> Mich Talebzadeh,
> Technologist | Architect | Data Engineer  | Generative AI | FinCrime
> London
> United Kingdom
>
>
>view my Linkedin profile
> 
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> *Disclaimer:* The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner  Von
> Braun )".
>
>
> On Fri, 3 May 2024 at 00:54, Mich Talebzadeh 
> wrote:
>
>> An issue I encountered while working with Materialized Views in Spark
>> SQL. It appears that there is an inconsistency between the behavior of
>> Materialized Views in Spark SQL and Hive.
>>
>> When attempting to execute a statement like DROP MATERIALIZED VIEW IF
>> EXISTS test.mv in Spark SQL, I encountered a syntax error indicating
>> that the keyword MATERIALIZED is not recognized. However, the same
>> statement executes successfully in Hive without any errors.
>>
>> pyspark.errors.exceptions.captured.ParseException:
>> [PARSE_SYNTAX_ERROR] Syntax error at or near 'MATERIALIZED'.(line 1, pos
>> 5)
>>
>> == SQL ==
>> DROP MATERIALIZED VIEW IF EXISTS test.mv
>> -^^^
>>
>> Here are the versions I am using:
>>
>>
>>
>> *Hive: 3.1.1Spark: 3.4*
>> my Spark session:
>>
>> spark = SparkSession.builder \
>>   .appName("test") \
>>   .enableHiveSupport() \
>>   .getOrCreate()
>>
>> Has anyone seen this behaviour or encountered a similar issue or if there
>> are any insights into why this discrepancy exists between Spark SQL and
>> Hive.
>>
>> Thanks
>>
>> Mich Talebzadeh,
>>
>> Technologist | Architect | Data Engineer  | Generative AI | FinCrime
>>
>> London
>> United Kingdom
>>
>>
>>view my Linkedin profile
>>
>>
>>  https://en.everybodywiki.com/Mich_Talebzadeh
>>
>>
>>
>> Disclaimer: The information provided is correct to the best of my
>> knowledge but of course cannot be guaranteed . It is essential to note
>> that, as with any advice, quote "one test result is worth one-thousand
>> expert opinions (Werner Von Braun)".
>>
>

Help needed optimize spark history server performance

[Vikas Tharyani]

Dear Spark Community,

I'm writing to seek your expertise in optimizing the performance of our
Spark History Server (SHS) deployed on Amazon EKS. We're encountering
timeouts (HTTP 504) when loading large event logs exceeding 5 GB.

*Our Setup:*

   - Deployment: SHS on EKS with Nginx ingress (idle connection timeout: 60
   seconds)
   - Instance: Memory-optimized with sufficient RAM and CPU
   - Spark Daemon Memory: 30 GB
   - Spark History Server Options:
   - K8s Namespace has a limit of 128Gb
   - The backend S3 has a lifecycle policy to delete objects that are older
   than *7 days*.

sparkHistoryOpts:
"-Dspark.history.fs.logDirectory=s3a:///eks-infra-use1/
-Dspark.history.retainedApplications=1
-Dspark.history.ui.maxApplications=20
-Dspark.history.store.serializer=PROTOBUF
-Dspark.hadoop.fs.s3a.threads.max=25
-Dspark.hadoop.fs.s3a.connection.maximum=650
-Dspark.hadoop.fs.s3a.readahead.range=512K
-Dspark.history.fs.endEventReparseChunkSize=2m
-Dspark.history.store.maxDiskUsage=30g"

*Problem:*

   - SHS times out when loading large event logs (8 GB or more).

*Request:*

We would greatly appreciate any insights or suggestions you may have to
improve the performance of our SHS and prevent these timeouts. Here are
some areas we're particularly interested in exploring:

   - Are there additional configuration options we should consider for
   handling large event logs?
   - Could Nginx configuration adjustments help with timeouts?
   - Are there best practices for optimizing SHS performance on EKS?

We appreciate any assistance you can provide.

Thank you for your time and support.

Sincerely,
-- 

Vikas Tharyani

Associate Manager, DevOps

Nielsen

www.nielsen.com 

Re: ********Spark streaming issue to Elastic data**********

[Mich Talebzadeh]

My recommendation! is using materialized views (MVs) created in Hive with
Spark Structured Streaming and Change Data Capture (CDC) is a good
combination for efficiently streaming view data updates in your scenario.

HTH

Mich Talebzadeh,
Technologist | Architect | Data Engineer  | Generative AI | FinCrime
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Thu, 2 May 2024 at 21:25, Karthick Nk  wrote:

> Hi All,
>
> Requirements:
> I am working on the data flow, which will use the view definition(view
> definition already defined in schema), there are multiple tables used in
> the view definition. Here we want to stream the view data into elastic
> index based on if any of the table(used in the view definition) data got
> changed.
>
>
> Current flow:
> 1. we are inserting id's from the table(which used in the view definition)
> into the common table.
> 2. From the common table by using the id, we will be streaming the view
> data (by using if any of the incomming id is present in the collective id
> of all tables used from view definition) by using spark structured
> streaming.
>
>
> Issue:
> 1. Here we are facing issue - For each incomming id here we running view
> definition(so it will read all the data from all the data) and check if any
> of the incomming id is present in the collective id's of view result, Due
> to which it is taking more memory in the cluster driver and taking more
> time to process.
>
>
> I am epxpecting an alternate solution, if we can avoid full scan of view
> definition every time, If you have any alternate deisgn flow how we can
> achieve the result, please suggest for the same.
>
>
> Note: Also, it will be helpfull, if you can share the details like
> community forum or platform to discuss this kind of deisgn related topics,
> it will be more helpfull.
>

Re: Issue with Materialized Views in Spark SQL

[Mich Talebzadeh]

Thanks for the comments I received.

So in summary, Apache Spark itself doesn't directly manage materialized
views,(MV)  but it can work with them through integration with the
underlying data storage systems like Hive or through iceberg. I believe
databricks through unity catalog support MVs as well.

Moreover, there is a case for supporting MVs. However, Spark can utilize
materialized views even though it doesn't directly manage them.. This came
about because someone in the Spark user forum enquired about "Spark
streaming issue to Elastic data*". One option I thought of was that uUsing
materialized views with Spark Structured Streaming and Change Data Capture
(CDC) is a potential solution for efficiently streaming view data updates
in this scenario. .


Mich Talebzadeh,
Technologist | Architect | Data Engineer  | Generative AI | FinCrime
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Fri, 3 May 2024 at 00:54, Mich Talebzadeh 
wrote:

> An issue I encountered while working with Materialized Views in Spark SQL.
> It appears that there is an inconsistency between the behavior of
> Materialized Views in Spark SQL and Hive.
>
> When attempting to execute a statement like DROP MATERIALIZED VIEW IF
> EXISTS test.mv in Spark SQL, I encountered a syntax error indicating that
> the keyword MATERIALIZED is not recognized. However, the same statement
> executes successfully in Hive without any errors.
>
> pyspark.errors.exceptions.captured.ParseException:
> [PARSE_SYNTAX_ERROR] Syntax error at or near 'MATERIALIZED'.(line 1, pos 5)
>
> == SQL ==
> DROP MATERIALIZED VIEW IF EXISTS test.mv
> -^^^
>
> Here are the versions I am using:
>
>
>
> *Hive: 3.1.1Spark: 3.4*
> my Spark session:
>
> spark = SparkSession.builder \
>   .appName("test") \
>   .enableHiveSupport() \
>   .getOrCreate()
>
> Has anyone seen this behaviour or encountered a similar issue or if there
> are any insights into why this discrepancy exists between Spark SQL and
> Hive.
>
> Thanks
>
> Mich Talebzadeh,
>
> Technologist | Architect | Data Engineer  | Generative AI | FinCrime
>
> London
> United Kingdom
>
>
>view my Linkedin profile
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> Disclaimer: The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner Von Braun)".
>

Re: Issue with Materialized Views in Spark SQL

[Jungtaek Lim]

(removing dev@ as I don't think this is dev@ related thread but more about
"question")

My understanding is that Apache Spark does not support Materialized View.
That's all. IMHO it's not a proper expectation that all operations in
Apache Hive will be supported in Apache Spark. They are different projects
and Apache Spark does not aim to be 100% compatible with Apache Hive. There
was a time the community tried to provide some sort of compatibility, but
both projects are 10+ years old, and mature enough to have their own
roadmap to drive.

That said, that's not a bug or an issue. You can initiate a feature request
and wish the community to include that into the roadmap.

On Fri, May 3, 2024 at 12:01 PM Mich Talebzadeh 
wrote:

> An issue I encountered while working with Materialized Views in Spark SQL.
> It appears that there is an inconsistency between the behavior of
> Materialized Views in Spark SQL and Hive.
>
> When attempting to execute a statement like DROP MATERIALIZED VIEW IF
> EXISTS test.mv in Spark SQL, I encountered a syntax error indicating that
> the keyword MATERIALIZED is not recognized. However, the same statement
> executes successfully in Hive without any errors.
>
> pyspark.errors.exceptions.captured.ParseException:
> [PARSE_SYNTAX_ERROR] Syntax error at or near 'MATERIALIZED'.(line 1, pos 5)
>
> == SQL ==
> DROP MATERIALIZED VIEW IF EXISTS test.mv
> -^^^
>
> Here are the versions I am using:
>
>
>
> *Hive: 3.1.1Spark: 3.4*
> my Spark session:
>
> spark = SparkSession.builder \
>   .appName("test") \
>   .enableHiveSupport() \
>   .getOrCreate()
>
> Has anyone seen this behaviour or encountered a similar issue or if there
> are any insights into why this discrepancy exists between Spark SQL and
> Hive.
>
> Thanks
>
> Mich Talebzadeh,
>
> Technologist | Architect | Data Engineer  | Generative AI | FinCrime
>
> London
> United Kingdom
>
>
>view my Linkedin profile
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> Disclaimer: The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner Von Braun)".
>

Re: Issue with Materialized Views in Spark SQL

[Walaa Eldin Moustafa]

I do not think the issue is with DROP MATERIALIZED VIEW only, but also with
CREATE MATERIALIZED VIEW, because neither is supported in Spark. I guess
you must have created the view from Hive and are trying to drop it from
Spark and that is why you are running to the issue with DROP first.

There is some work in the Iceberg community to add the support to Spark
through SQL extensions, and Iceberg support for views and
materialization tables. Some recent discussions can be found here [1] along
with a WIP Iceberg-Spark PR.

[1] https://lists.apache.org/thread/rotmqzmwk5jrcsyxhzjhrvcjs5v3yjcc

Thanks,
Walaa.

On Thu, May 2, 2024 at 4:55 PM Mich Talebzadeh 
wrote:

> An issue I encountered while working with Materialized Views in Spark SQL.
> It appears that there is an inconsistency between the behavior of
> Materialized Views in Spark SQL and Hive.
>
> When attempting to execute a statement like DROP MATERIALIZED VIEW IF
> EXISTS test.mv in Spark SQL, I encountered a syntax error indicating that
> the keyword MATERIALIZED is not recognized. However, the same statement
> executes successfully in Hive without any errors.
>
> pyspark.errors.exceptions.captured.ParseException:
> [PARSE_SYNTAX_ERROR] Syntax error at or near 'MATERIALIZED'.(line 1, pos 5)
>
> == SQL ==
> DROP MATERIALIZED VIEW IF EXISTS test.mv
> -^^^
>
> Here are the versions I am using:
>
>
>
> *Hive: 3.1.1Spark: 3.4*
> my Spark session:
>
> spark = SparkSession.builder \
>   .appName("test") \
>   .enableHiveSupport() \
>   .getOrCreate()
>
> Has anyone seen this behaviour or encountered a similar issue or if there
> are any insights into why this discrepancy exists between Spark SQL and
> Hive.
>
> Thanks
>
> Mich Talebzadeh,
>
> Technologist | Architect | Data Engineer  | Generative AI | FinCrime
>
> London
> United Kingdom
>
>
>view my Linkedin profile
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> Disclaimer: The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner Von Braun)".
>

Issue with Materialized Views in Spark SQL

[Mich Talebzadeh]

An issue I encountered while working with Materialized Views in Spark SQL.
It appears that there is an inconsistency between the behavior of
Materialized Views in Spark SQL and Hive.

When attempting to execute a statement like DROP MATERIALIZED VIEW IF
EXISTS test.mv in Spark SQL, I encountered a syntax error indicating that
the keyword MATERIALIZED is not recognized. However, the same statement
executes successfully in Hive without any errors.

pyspark.errors.exceptions.captured.ParseException:
[PARSE_SYNTAX_ERROR] Syntax error at or near 'MATERIALIZED'.(line 1, pos 5)

== SQL ==
DROP MATERIALIZED VIEW IF EXISTS test.mv
-^^^

Here are the versions I am using:



*Hive: 3.1.1Spark: 3.4*
my Spark session:

spark = SparkSession.builder \
  .appName("test") \
  .enableHiveSupport() \
  .getOrCreate()

Has anyone seen this behaviour or encountered a similar issue or if there
are any insights into why this discrepancy exists between Spark SQL and
Hive.

Thanks

Mich Talebzadeh,

Technologist | Architect | Data Engineer  | Generative AI | FinCrime

London
United Kingdom


   view my Linkedin profile


 https://en.everybodywiki.com/Mich_Talebzadeh



Disclaimer: The information provided is correct to the best of my knowledge
but of course cannot be guaranteed . It is essential to note that, as with
any advice, quote "one test result is worth one-thousand expert opinions
(Werner Von Braun)".

********Spark streaming issue to Elastic data**********

[Karthick Nk]

Hi All,

Requirements:
I am working on the data flow, which will use the view definition(view
definition already defined in schema), there are multiple tables used in
the view definition. Here we want to stream the view data into elastic
index based on if any of the table(used in the view definition) data got
changed.


Current flow:
1. we are inserting id's from the table(which used in the view definition)
into the common table.
2. From the common table by using the id, we will be streaming the view
data (by using if any of the incomming id is present in the collective id
of all tables used from view definition) by using spark structured
streaming.


Issue:
1. Here we are facing issue - For each incomming id here we running view
definition(so it will read all the data from all the data) and check if any
of the incomming id is present in the collective id's of view result, Due
to which it is taking more memory in the cluster driver and taking more
time to process.


I am epxpecting an alternate solution, if we can avoid full scan of view
definition every time, If you have any alternate deisgn flow how we can
achieve the result, please suggest for the same.


Note: Also, it will be helpfull, if you can share the details like
community forum or platform to discuss this kind of deisgn related topics,
it will be more helpfull.

unsubscribe

[Nebi Aydin]

unsubscribe

unsubscribe

[Atakala Selam]

unsubscribe

unsubscribe

[Yoel Benharrous]

unsubscribe

Traceback is missing content in pyspark when invoked with UDF

[Indivar Mishra]

Hi

*Tl;Dr:* I have a scenario where I generate code string on fly and execute
that code, now for me if an error occurs I need the traceback but for
executable code I just get partial traceback i.e. the line which caused the
error is missing.

Consider below MRC:
def fun():
from pyspark.sql import SparkSession
from pyspark.sql.functions import col, udf
from pyspark.sql.types import StringType

spark = SparkSession.builder.appName("some_name").getOrCreate()

columns = ["Seqno", "Name"]
data = [("1", "john jones"), ("2", "tracey smith"), ("3", "amy sanders"
)]

df = spark.createDataFrame(data=data, schema=columns)

def errror_func(str):
def internal_error_method():
raise RuntimeError

return internal_error_method()

# Converting function to UDF
errror_func_udf = udf(lambda z: errror_func(z), StringType())

df.select(col("Seqno"), errror_func_udf(col("Name")).alias("Name")).show
(truncate=False)

fun()


This gives below shown Traceback, (Notice we are also getting the line
content that caused error

> Traceback (most recent call last):
>
>   File "temp.py", line 28, in 
>
> fun()
>
>   File "temp.py", line 25, in fun
>
> df.select(col("Seqno"),
>> errror_func_udf(col("Name")).alias("Name")).show(truncate=False)
>
>   File
>> "/home/indivar/corridor/code/corridor-platforms/venv/lib/python3.8/site-packages/pyspark/sql/dataframe.py",
>> line 502, in show
>
> print(self._jdf.showString(n, int_truncate, vertical))
>
>   File
>> "/home/indivar/corridor/code/corridor-platforms/venv/lib/python3.8/site-packages/py4j/java_gateway.py",
>> line 1321, in __call__
>
> return_value = get_return_value(
>
>   File
>> "/home/indivar/corridor/code/corridor-platforms/venv/lib/python3.8/site-packages/pyspark/sql/utils.py",
>> line 117, in deco
>
> raise converted from None
>
> pyspark.sql.utils.PythonException:
>
>   An exception was thrown from the Python worker. Please see the stack
>> trace below.
>
> Traceback (most recent call last):
>
>   File "temp.py", line 23, in 
>
> errror_func_udf = udf(lambda z: errror_func(z), StringType())
>
>   File "temp.py", line 20, in errror_func
>
> return internal_error_method()
>
>   File "temp.py", line 18, in internal_error_method
>
> raise RuntimeError
>
> RuntimeError
>
>
>
But now if i run the same code by doing an exec i loose the traceback line
content although line number is there
import linecache

code = """
def fun():
from pyspark.sql import SparkSession
from pyspark.sql.functions import col, udf
from pyspark.sql.types import StringType

spark = SparkSession.builder.appName("some_name").getOrCreate()

columns = ["Seqno", "Name"]
data = [("1", "john jones"), ("2", "tracey smith"), ("3", "amy
sanders")]

df = spark.createDataFrame(data=data, schema=columns)

def errror_func(str):
def internal_error_method():
raise RuntimeError

return internal_error_method()

# Converting function to UDF
errror_func_udf = udf(lambda z: errror_func(z), StringType())

df.select(col("Seqno"),
errror_func_udf(col("Name")).alias("Name")).show(truncate=False)
"""


scope = {}
filename = ""
compiled_code = compile(code, filename, "exec")
if filename not in linecache.cache:
linecache.cache[filename] = (
len(scope),
None,
code.splitlines(keepends=True),
filename,
)
exec(compiled_code, scope, scope)
fun = scope["fun"]

fun()


Traceback of this code is

> Traceback (most recent call last):
>
>   File "temp.py", line 74, in 
>
> fun()
>
>   File "", line 23, in fun
>
>   File
>> "/home/indivar/corridor/code/corridor-platforms/venv/lib/python3.8/site-packages/pyspark/sql/dataframe.py",
>> line 502, in show
>
> print(self._jdf.showString(n, int_truncate, vertical))
>
>   File
>> "/home/indivar/corridor/code/corridor-platforms/venv/lib/python3.8/site-packages/py4j/java_gateway.py",
>> line 1321, in __call__
>
> return_value = get_return_value(
>
>   File
>> "/home/indivar/corridor/code/corridor-platforms/venv/lib/python3.8/site-packages/pyspark/sql/utils.py",
>> line 117, in deco
>
> raise converted from None
>
> pyspark.sql.utils.PythonException:
>
>   An exception was thrown from the Python worker. Please see the stack
>> trace below.
>
> Traceback (most recent call last):
>
>   File "", line 21, in 
>
>   File "", line 18, in errror_func
>
>   File "", line 16, in internal_error_method
>
> RuntimeError
>
>
> As you can see this has missing line content.

initially i thought this was a python issue, so i tried to do some reading,
python internally seems to be using linecache module to get content of
line, now when doing exec uptill python 3.12 python also had same issue
which they have fixed in python 3.13 [issue ref for details]: Support
multi-line error locations in traceback and other related improvements
(PEP-657, 3.11) · Issue #106922 · python/cpython (github.com)

Re: [spark-graphframes]: Generating incorrect edges

[Mich Talebzadeh]

Hi Steve,

Thanks for your statement. I tend to use uuid myself to avoid
collisions. This built-in function generates random IDs that are highly
likely to be unique across systems. My concerns are on edge so to speak. If
the Spark application runs for a very long time or encounters restarts, the
monotonically_increasing_id() sequence might restart from the beginning.
This could again cause duplicate IDs if other Spark applications are
running concurrently or if data is processed across multiple runs of the
same application..

HTH

Mich Talebzadeh,
Technologist | Architect | Data Engineer  | Generative AI | FinCrime
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Wed, 1 May 2024 at 01:22, Stephen Coy  wrote:

> Hi Mich,
>
> I was just reading random questions on the user list when I noticed that
> you said:
>
> On 25 Apr 2024, at 2:12 AM, Mich Talebzadeh 
> wrote:
>
> 1) You are using monotonically_increasing_id(), which is not
> collision-resistant in distributed environments like Spark. Multiple hosts
>can generate the same ID. I suggest switching to UUIDs (e.g.,
> uuid.uuid4()) for guaranteed uniqueness.
>
>
> It’s my understanding that the *Spark* `monotonically_increasing_id()`
> function exists for the exact purpose of generating a collision-resistant
> unique id across nodes on different hosts.
> We use it extensively for this purpose and have never encountered an issue.
>
> Are we wrong or are you thinking of a different (not Spark) function?
>
> Cheers,
>
> Steve C
>
>
>
>
> This email contains confidential information of and is the copyright of
> Infomedia. It must not be forwarded, amended or disclosed without consent
> of the sender. If you received this message by mistake, please advise the
> sender and delete all copies. Security of transmission on the internet
> cannot be guaranteed, could be infected, intercepted, or corrupted and you
> should ensure you have suitable antivirus protection in place. By sending
> us your or any third party personal details, you consent to (or confirm you
> have obtained consent from such third parties) to Infomedia’s privacy
> policy. http://www.infomedia.com.au/privacy-policy/
>

Re: [spark-graphframes]: Generating incorrect edges

[Stephen Coy]

Hi Mich,

I was just reading random questions on the user list when I noticed that you 
said:

On 25 Apr 2024, at 2:12 AM, Mich Talebzadeh  wrote:

1) You are using monotonically_increasing_id(), which is not 
collision-resistant in distributed environments like Spark. Multiple hosts
   can generate the same ID. I suggest switching to UUIDs (e.g., uuid.uuid4()) 
for guaranteed uniqueness.


It’s my understanding that the *Spark* `monotonically_increasing_id()` function 
exists for the exact purpose of generating a collision-resistant unique id 
across nodes on different hosts.
We use it extensively for this purpose and have never encountered an issue.

Are we wrong or are you thinking of a different (not Spark) function?

Cheers,

Steve C




This email contains confidential information of and is the copyright of 
Infomedia. It must not be forwarded, amended or disclosed without consent of 
the sender. If you received this message by mistake, please advise the sender 
and delete all copies. Security of transmission on the internet cannot be 
guaranteed, could be infected, intercepted, or corrupted and you should ensure 
you have suitable antivirus protection in place. By sending us your or any 
third party personal details, you consent to (or confirm you have obtained 
consent from such third parties) to Infomedia’s privacy policy. 
http://www.infomedia.com.au/privacy-policy/

unsubscribe

[Wood Super]

unsubscribe

unsubscribe

[junhua . xie]

unsubscribe

unsubscribe

[Yoel Benharrous]

Re: spark.sql.shuffle.partitions=auto

[Mich Talebzadeh]

spark.sql.shuffle.partitions=auto

Because Apache Spark does not build clusters. This configuration option is
specific to Databricks, with their managed Spark offering. It allows
Databricks to automatically determine an optimal number of shuffle
partitions for your workload.

HTH

Mich Talebzadeh,
Technologist | Architect | Data Engineer  | Generative AI | FinCrime
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Tue, 30 Apr 2024 at 11:51, second_co...@yahoo.com.INVALID
 wrote:

> May i know is
>
> spark.sql.shuffle.partitions=auto
>
> only available on Databricks? what about on vanilla Spark ? When i set
> this, it gives error need to put int.  Any open source library that auto
> find the best partition , block size for dataframe?
>
>
>

Re: Spark on Kubernetes

[Mich Talebzadeh]

Hi,
In k8s the driver is responsible for executor creation. The likelihood of
your problem is that Insufficient memory allocated for executors in the K8s
cluster. Even with dynamic allocation, k8s won't  schedule executor pods if
there is not enough free memory to fulfill their resource requests.

My suggestions

   - Increase Executor Memory: Allocate more memory per executor (e.g., 2GB
   or 3GB) to allow for multiple executors within available cluster memory.
   - Adjust Driver Pod Resources: Ensure the driver pod has enough memory
   to run Spark and manage executors.
   - Optimize Resource Management: Explore on-demand allocation or
   adjusting allocation granularity for better resource utilization. For
   example look at documents for Executor On-Demand Allocation
   (spark.executor.cores=0): and spark.dynamicAllocation.minExecutors &
   spark.dynamicAllocation.maxExecutors

HTH

Mich Talebzadeh,
Technologist | Architect | Data Engineer  | Generative AI | FinCrime
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Tue, 30 Apr 2024 at 04:29, Tarun raghav 
wrote:

> Respected Sir/Madam,
> I am Tarunraghav. I have a query regarding spark on kubernetes.
>
> We have an eks cluster, within which we have spark installed in the pods.
> We set the executor memory as 1GB and set the executor instances as 2, I
> have also set dynamic allocation as true. So when I try to read a 3 GB CSV
> file or parquet file, it is supposed to increase the number of pods by 2.
> But the number of executor pods is zero.
> I don't know why executor pods aren't being created, even though I set
> executor instance as 2. Please suggest a solution for this.
>
> Thanks & Regards,
> Tarunraghav
>
>

spark.sql.shuffle.partitions=auto

[second_co...@yahoo.com.INVALID]

May i know is spark.sql.shuffle.partitions=auto only available on Databricks? 
what about on vanilla Spark ? When i set this, it gives error need to put int.  
Any open source library that auto find the best partition , block size for 
dataframe?

Spark on Kubernetes

[Tarun raghav]

Respected Sir/Madam,
I am Tarunraghav. I have a query regarding spark on kubernetes.

We have an eks cluster, within which we have spark installed in the pods.
We set the executor memory as 1GB and set the executor instances as 2, I
have also set dynamic allocation as true. So when I try to read a 3 GB CSV
file or parquet file, it is supposed to increase the number of pods by 2.
But the number of executor pods is zero.
I don't know why executor pods aren't being created, even though I set
executor instance as 2. Please suggest a solution for this.

Thanks & Regards,
Tarunraghav

Re: Python for the kids and now PySpark

[Meena Rajani]

Mitch, you are right these days the attention span is getting shorter.
Christian could work on a completely new thing for 3 hours and is proud to
explain. It is amazing.

Thanks for sharing.



On Sat, Apr 27, 2024 at 9:40 PM Farshid Ashouri 
wrote:

> Mich, this is absolutely amazing.
>
> Thanks for sharing.
>
> On Sat, 27 Apr 2024, 22:26 Mich Talebzadeh, 
> wrote:
>
>> Python for the kids. Slightly off-topic but worthwhile sharing.
>>
>> One of the things that may benefit kids is starting to learn something
>> new. Basically anything that can focus their attention away from games for
>> a few hours. Around 2020, my son Christian (now nearly 15) decided to
>> learn a programming language. So somehow he picked Python to start with.
>> The kids are good when they focus. However, they live in a virtual reality
>> world and they cannot focus for long hours. I let him explore Python on his
>> Windows 10 laptop and download it himself. In the following video Christian
>> explains to his mother what he started to do just before going to bed. BTW,
>> when he says 32M he means 32-bit. I leave it to you to judge :) Now the
>> idea is to start learning PySpark. So I will let him do it himself and
>> learn from his mistakes.  For those who have kids, I would be interested to
>> know their opinion.
>>
>> Cheers
>>
>>
>> Mich Talebzadeh,
>> Technologist | Architect | Data Engineer  | Generative AI | FinCrime
>> London
>> United Kingdom
>>
>>
>>view my Linkedin profile
>> 
>>
>>
>>  https://en.everybodywiki.com/Mich_Talebzadeh
>>
>>
>>
>> *Disclaimer:* The information provided is correct to the best of my
>> knowledge but of course cannot be guaranteed . It is essential to note
>> that, as with any advice, quote "one test result is worth one-thousand
>> expert opinions (Werner
>> Von Braun
>> )".
>>
>> -
>> To unsubscribe e-mail: user-unsubscr...@spark.apache.org
>
>

Re: Python for the kids and now PySpark

[Farshid Ashouri]

Mich, this is absolutely amazing.

Thanks for sharing.

On Sat, 27 Apr 2024, 22:26 Mich Talebzadeh, 
wrote:

> Python for the kids. Slightly off-topic but worthwhile sharing.
>
> One of the things that may benefit kids is starting to learn something
> new. Basically anything that can focus their attention away from games for
> a few hours. Around 2020, my son Christian (now nearly 15) decided to
> learn a programming language. So somehow he picked Python to start with.
> The kids are good when they focus. However, they live in a virtual reality
> world and they cannot focus for long hours. I let him explore Python on his
> Windows 10 laptop and download it himself. In the following video Christian
> explains to his mother what he started to do just before going to bed. BTW,
> when he says 32M he means 32-bit. I leave it to you to judge :) Now the
> idea is to start learning PySpark. So I will let him do it himself and
> learn from his mistakes.  For those who have kids, I would be interested to
> know their opinion.
>
> Cheers
>
>
> Mich Talebzadeh,
> Technologist | Architect | Data Engineer  | Generative AI | FinCrime
> London
> United Kingdom
>
>
>view my Linkedin profile
> 
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> *Disclaimer:* The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner  Von
> Braun )".
>
> -
> To unsubscribe e-mail: user-unsubscr...@spark.apache.org

[Release Question]: Estimate on 3.5.2 release?

[Paul Gerver]

Hello,

I'm curious if there is an estimate when 3.5.2 for Spark Core will be released.
There are several bug and security vulnerability fixes in the dependencies we 
are excited to receive!

If anyone has any insights, that would be greatly appreciated. Thanks!
- ​Paul





[cid:8a2e80d5-1a98-4eca-b993-46937e35b2e9]



Paul Gerver

Streams Software Engineer

[cid:0412bb89-89ee-4329-a946-00e56d95d85c][cid:8543c1a1-eba3-4153-9c05-88c2e1d9bc13]


This e-mail (including any attachments) may contain privileged, confidential, 
proprietary, private, copyrighted, or other legally protected information. The 
information is intended to be for the use of the individual or entity 
designated above. If you are not the intended recipient (even if the e-mail 
address above is yours), please notify us by return e-mail immediately, and 
delete the message and any attachments. Any disclosure, reproduction, 
distribution or other use of this message or any attachments by an individual 
or entity other than the intended recipient is prohibited.

[SparkListener] Accessing classes loaded via the '--packages' option

[Damien Hawes]

Hi folks,

I'm contributing to the OpenLineage project, specifically the Apache Spark
integration. My current focus is on extending the project to support data
lineage extraction for Spark Streaming, beginning with Apache Kafka sources
and sinks.

I've encountered an obstacle when attempting to access information
essential for lineage extraction from Apache Kafka-related classes within
the OpenLineage Spark code base. Specifically, I need to access details
like Kafka topic names and bootstrap servers from objects like
StreamingDataSourceV2Relation.

While I can successfully access these details if the Kafka JARs are placed
directly in the 'spark/jars' directory, I'm unable to do so when using the
`--packages` option for dependency management. This creates a significant
obstacle for users who rely on `--packages` for their Spark applications.

I've taken initial steps to investigate (viewable in this GitHub PR
, the class in
question is *StreamingDataSourceV2RelationVisitor*), but I'd greatly
appreciate any insights or guidance on the following:

*1. Understanding the Issue:* Are there known reasons within Spark that
could explain this difference in behavior when loading dependencies via
`--packages` versus placing JARs directly?
*2. Alternative Approaches:*  Are there recommended techniques or patterns
to access the necessary Kafka class information within a SparkListener
extension, especially when dependencies are managed via `--packages`?

I'm eager to find a solution that avoids heavy reliance on reflection.

Thank you for your time and assistance!

Kind regards,
Damien

unsubscribe

[Code Tutelage]

Re:RE: How to add MaxDOP option in spark mssql JDBC

[Elite]

Thank you.

My main purpose is pass "MaxDop 1" to MSSQL to control the CPU usage. From the 
offical doc, I guess the problem of my codes is spark wrap the query to 




select * from (SELECT TOP 10 * FROM dbo.Demo with (nolock) WHERE Id = 1 option 
(maxdop 1)) spark_gen_alias




Apparently, this valilate MSSQL syntax, because "option (maxdop 1)" is not 
placed at the end.

May I know,  how spark wrap the query if I use prepareQuery?

I do not have spark 3.4+ env now, so did not got a chance to try this option.







At 2024-04-24 20:51:45, "Appel, Kevin"  wrote:

You might be able to leverage the prepareQuery option, that is at 
https://spark.apache.org/docs/3.5.1/sql-data-sources-jdbc.html#data-source-option
 … this was introduced in Spark 3.4.0 to handle temp table query and CTE query 
against MSSQL server since what you send in is not actually what gets sent, 
there is some items that get wrapped.

 

There is more of the technical info in 
https://issues.apache.org/jira/browse/SPARK-37259 with the PR’s linked that had 
the fix done for this

 

 

From: Elite 
Sent: Tuesday, April 23, 2024 10:28 PM
To: user@spark.apache.org
Subject: How to add MaxDOP option in spark mssql JDBC

 

[QUESTION] How to pass MAXDOP option · Issue #2395 · microsoft/mssql-jdbc 
(github.com)

 

Hi team, 

 

I am suggested to require help form spark community.

 

We suspect spark rewerite the query before pass to ms sql, and it lead to 
syntax error.

Is there any work around to let make my codes work? 

 

spark.read()
.format("jdbc")
.option("driver","com.microsoft.sqlserver.jdbc.SQLServerDriver")
.option("url", "jdbc:sqlserver://xxx.database.windows.net;databaseName=")
.option("query", "SELECT TOP 10 * FROM dbo.Demo with (nolock) WHERE Id = 1 
option (maxdop 1)")
.load()
.show();

com.microsoft.sqlserver.jdbc.SQLServerException: Incorrect syntax near the 
keyword 'option'.
at 
com.microsoft.sqlserver.jdbc.SQLServerException.makeFromDatabaseError(SQLServerException.java:270)
at 
com.microsoft.sqlserver.jdbc.SQLServerStatement.getNextResult(SQLServerStatement.java:1778)
at 
com.microsoft.sqlserver.jdbc.SQLServerPreparedStatement.doExecutePreparedStatement(SQLServerPreparedStatement.java:697)
at 
com.microsoft.sqlserver.jdbc.SQLServerPreparedStatement$PrepStmtExecCmd.doExecute(SQLServerPreparedStatement.java:616)
at com.microsoft.sqlserver.jdbc.TDSCommand.execute(IOBuffer.java:7775)
at 
com.microsoft.sqlserver.jdbc.SQLServerConnection.executeCommand(SQLServerConnection.java:4397)
at 
com.microsoft.sqlserver.jdbc.SQLServerStatement.executeCommand(SQLServerStatement.java:293)
at 
com.microsoft.sqlserver.jdbc.SQLServerStatement.executeStatement(SQLServerStatement.java:263)
at 
com.microsoft.sqlserver.jdbc.SQLServerPreparedStatement.executeQuery(SQLServerPreparedStatement.java:531)
at 
org.apache.spark.sql.execution.datasources.jdbc.JDBCRDD$.resolveTable(JDBCRDD.scala:61)
at 
org.apache.spark.sql.execution.datasources.jdbc.JDBCRelation$.getSchema(JDBCRelation.scala:226)
at 
org.apache.spark.sql.execution.datasources.jdbc.JdbcRelationProvider.createRelation(JdbcRelationProvider.scala:35)
at 
org.apache.spark.sql.execution.datasources.DataSource.resolveRelation(DataSource.scala:344)
at org.apache.spark.sql.DataFrameReader.loadV1Source(DataFrameReader.scala:297)
at 
org.apache.spark.sql.DataFrameReader.$anonfun$load$2(DataFrameReader.scala:286)
at scala.Option.getOrElse(Option.scala:189)
at org.apache.spark.sql.DataFrameReader.load(DataFrameReader.scala:286)
at org.apache.spark.sql.DataFrameReader.load(DataFrameReader.scala:221)

This message, and any attachment(s), is for the intended recipient(s) only, may 
contain information that is privileged, confidential and/or proprietary and 
subject to important terms and conditions available at 
http://www.bankofamerica.com/electronic-disclaimer. If you are not the intended 
recipient, please delete this message. For more information about how Bank of 
America protects your privacy, including specific rights that may apply, please 
visit the following pages: 
https://business.bofa.com/en-us/content/global-privacy-notices.html (which 
includes global privacy notices) and 
https://www.bankofamerica.com/security-center/privacy-overview/ (which includes 
US State specific privacy notices such as the 
http://www.bankofamerica.com/ccpa-notice).

Re: [spark-graphframes]: Generating incorrect edges

[Nijland, J.G.W. (Jelle, Student M-CS)]

Hi Mich,

Thanks for your suggestions.
1) It currently runs on one server with plenty of resources assigned. But I 
will keep it in mind to replace monotonically_increasing_id() with uuid() once 
we scale up.
2) I have replaced the null values in origin with a string 
{prefix}-{mnt_by}-{organisation}

replacement_string = psf.concat_ws("-", psf.col("prefix"), psf.col("mnt_by"), 
psf.col("descr"))
df = df.withColumn("origin", psf.coalesce(psf.col("origin"), 
replacement_string))

I have verified my other columns have no Null values.

3) This is my logic how i generate IDs

mnt_by_id = df.select(MNT_BY).distinct().withColumn(MAINTAINER_ID, 
psf.concat(psf.lit('m_'), psf.monotonically_increasing_id()))
prefix_id = df.select(PREFIX).distinct().withColumn(PREFIX_ID, 
psf.concat(psf.lit('p_'), psf.monotonically_increasing_id()))
origin_id = df.select(ORIGIN).distinct().withColumn(ORIGIN_ID, 
psf.concat(psf.lit('o_'), psf.monotonically_increasing_id()))
organisation_id = df.select(DESCR).distinct().withColumn(ORGANISATION_ID, 
psf.concat(psf.lit('org_'), psf.monotonically_increasing_id()))

df = df.join(mnt_by_id, on=MNT_BY, how="left").join(prefix_id, on=PREFIX, 
how="left").join(origin_id, on=ORIGIN, how="left").join(organisation_id, 
on=DESCR, how="left")

I create the ID using the distinct values in the columns "mnt_by", "prefix", 
"origin" and "descr". The same columns I join "on".

4) This is my current resource allocation, I run it on the server of my 
university.
It has 112 cores and 1.48T ram, I can request more resources but in my eyes 
this sound be plenty.
If you think more resource would help, I will ask them.

spark_conf = SparkConf().setAppName(f"pyspark-{APP_NAME}-{int(time())}").set(
"spark.submit.deployMode", "client"
).set("spark.sql.parquet.binaryAsString", "true"
).set("spark.driver.bindAddress", "localhost"
).set("spark.driver.host", "127.0.0.1"
# ).set("spark.driver.port", "0"
).set("spark.ui.port", "4041"
).set("spark.executor.instances", "1"
).set("spark.executor.cores", "50"
).set("spark.executor.memory", "128G"
).set("spark.executor.memoryOverhead", "32G"
).set("spark.driver.cores", "16"
).set("spark.driver.memory", "64G"
)

I dont think b) applies as its a single machine.

Kind regards,
Jelle


From: Mich Talebzadeh 
Sent: Wednesday, April 24, 2024 6:12 PM
To: Nijland, J.G.W. (Jelle, Student M-CS) 
Cc: user@spark.apache.org 
Subject: Re: [spark-graphframes]: Generating incorrect edges

OK let us have a look at these

1) You are using monotonically_increasing_id(), which is not 
collision-resistant in distributed environments like Spark. Multiple hosts
   can generate the same ID. I suggest switching to UUIDs (e.g., uuid.uuid4()) 
for guaranteed uniqueness.

2) Missing values in the Origin column lead to null IDs, potentially causing 
problems downstream. You can handle missing values appropriately, say
   a) Filter out rows with missing origins or b) impute missing values with a 
strategy that preserves relationships (if applicable).

3) With join code, you mentioned left joining on the same column used for ID 
creation, not very clear!

4) Edge Issue, it appears to me the issue seems to occur with larger datasets 
(>100K records). Possible causes could be
   a) Resource Constraints as data size increases, PySpark might struggle with 
joins or computations if resources are limited (memory, CPU).
   b) Data Skew: Uneven distribution of values in certain columns could lead to 
imbalanced processing across machines.  Check Spark UI (4040) on staging and 
execution tabs

HTH

Mich Talebzadeh,
Technologist | Architect | Data Engineer  | Generative AI | FinCrime
London
United Kingdom


 
[https://ci3.googleusercontent.com/mail-sig/AIorK4zholKucR2Q9yMrKbHNn-o1TuS4mYXyi2KO6Xmx6ikHPySa9MLaLZ8t2hrA6AUcxSxDgHIwmKE]
   view my Linkedin 
profile


 https://en.everybodywiki.com/Mich_Talebzadeh



Disclaimer: The information provided is correct to the best of my knowledge but 
of course cannot be guaranteed . It is essential to note that, as with any 
advice, quote "one test result is worth one-thousand expert opinions (Werner 
 Von 
Braun)".


On Wed, 24 Apr 2024 at 16:44, Nijland, J.G.W. (Jelle, Student M-CS) 
mailto:j.g.w.nijl...@student.utwente.nl>> 
wrote:
Hi Mich,

Thanks for your reply,
1) ID generation is done using 
monotonically_increasing_id()
 this is then prefixed with "p_", "m_", "o_" or "org_" depending on the type of 
the value it identifies.
2) There are some missing values in the Origin column, these will result in a 
Null ID
3) The join code is present in [1], I join "left" on the same column I create 
the ID on
4) I dont think the issue is in ID or edge 

DataFrameReader: timestampFormat default value

[keen]

Is anyone familiar with [Datetime patterns](
https://spark.apache.org/docs/latest/sql-ref-datetime-pattern.html) and
`TimestampType` parsing in PySpark?
When reading CSV or JSON files, timestamp columns need to be parsed. via
datasource property `timestampFormat`.
[According to documentation](
https://spark.apache.org/docs/3.3.1/sql-data-sources-json.html#data-source-option:~:text=read/write-,timestampFormat,-%2DMM%2Ddd%27T%27HH)
default value is `-MM-dd'T'HH:mm:ss[.SSS][XXX]`.

However, I noticed some weird behavior:
```python
from pyspark.sql import types as T

json_lines =[
"{'label': 'no tz'  , 'value':
'2023-12-24T20:00:00'  }",
"{'label': 'UTC', 'value':
'2023-12-24T20:00:00Z' }",
"{'label': 'tz offset hour' , 'value':
'2023-12-24T20:00:00+01'   }",
"{'label': 'tz offset minute no colon'  , 'value':
'2023-12-24T20:00:00+0100' }",
"{'label': 'tz offset minute with colon', 'value':
'2023-12-24T20:00:00+01:00'}",
"{'label': 'tz offset second no colon'  , 'value':
'2023-12-24T20:00:00+01'   }",
"{'label': 'tz offset second with colon', 'value':
'2023-12-24T20:00:00+01:00:00' }",
]

schema = T.StructType([
T.StructField("label", T.StringType()),
T.StructField("value", T.TimestampType()),
T.StructField("t_corrupt_record", T.StringType()),
])

df = (spark.read
.schema(schema)
.option("timestampFormat", "-MM-dd'T'HH:mm:ss[.SSS][XXX]") # <--
using the "default" from doc
.option("mode", "PERMISSIVE")
.option("columnNameOfCorruptRecord", "t_corrupt_record")
.json(sc.parallelize(json_lines))
)

df.show(truncate=False)
+---+---+--+
|label  |value  |t_corrupt_record
   |
+---+---+--+
|no tz  |2023-12-24 20:00:00|null
   |
|UTC|2023-12-24 20:00:00|null
   |
|tz offset hour |null   |{'label': 'tz offset hour'
, 'value': '2023-12-24T20:00:00+01'   }|
|tz offset minute no colon  |null   |{'label': 'tz offset
minute no colon'  , 'value': '2023-12-24T20:00:00+0100' }|
|tz offset minute with colon|2023-12-24 19:00:00|null
   |
|tz offset second no colon  |null   |{'label': 'tz offset
second no colon'  , 'value': '2023-12-24T20:00:00+01'   }|
|tz offset second with colon|null   |{'label': 'tz offset
second with colon', 'value': '2023-12-24T20:00:00+01:00:00' }|
+---+---+--+
```

however, when omitting timestampFormat , the values are parsed just fine
```python
df = (spark.read
.schema(schema)
.option("mode", "PERMISSIVE")
.option("columnNameOfCorruptRecord", "t_corrupt_record")
.json(sc.parallelize(json_lines))
)

df.show(truncate=False)
+---+---++
|label  |value  |t_corrupt_record|
+---+---++
|no tz  |2023-12-24 20:00:00|null|
|UTC|2023-12-24 20:00:00|null|
|tz offset hour |2023-12-24 19:00:00|null|
|tz offset minute no colon  |2023-12-24 19:00:00|null|
|tz offset minute with colon|2023-12-24 19:00:00|null|
|tz offset second no colon  |2023-12-24 19:00:00|null|
|tz offset second with colon|2023-12-24 19:00:00|null|
+---+---++
```

This is not plausible to me.
Using the default value explicitly should lead to the same results as
omitting it.


Thanks and regards
Martin

Re: [spark-graphframes]: Generating incorrect edges

[Mich Talebzadeh]

OK let us have a look at these

1) You are using monotonically_increasing_id(), which is not
collision-resistant in distributed environments like Spark. Multiple hosts
   can generate the same ID. I suggest switching to UUIDs (e.g.,
uuid.uuid4()) for guaranteed uniqueness.

2) Missing values in the Origin column lead to null IDs, potentially
causing problems downstream. You can handle missing values appropriately,
say
   a) Filter out rows with missing origins or b) impute missing values with
a strategy that preserves relationships (if applicable).

3) With join code, you mentioned left joining on the same column used for
ID creation, not very clear!

4) Edge Issue, it appears to me the issue seems to occur with larger
datasets (>100K records). Possible causes could be
   a) Resource Constraints as data size increases, PySpark might struggle
with joins or computations if resources are limited (memory, CPU).
   b) Data Skew: Uneven distribution of values in certain columns could
lead to imbalanced processing across machines.  Check Spark UI (4040) on
staging and execution tabs

HTH

Mich Talebzadeh,
Technologist | Architect | Data Engineer  | Generative AI | FinCrime
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Wed, 24 Apr 2024 at 16:44, Nijland, J.G.W. (Jelle, Student M-CS) <
j.g.w.nijl...@student.utwente.nl> wrote:

> Hi Mich,
>
> Thanks for your reply,
> 1) ID generation is done using monotonically_increasing_id()
> 
>  this
> is then prefixed with "p_", "m_", "o_" or "org_" depending on the type of
> the value it identifies.
> 2) There are some missing values in the Origin column, these will result
> in a Null ID
> 3) The join code is present in [1], I join "left" on the same column
> I create the ID on
> 4) I dont think the issue is in ID or edge generation, if i limit my input
> dataframe and union it with my Utwente data row, I can verify those edges
> are created correctly up to 100K records.
> Once I go past that amount of records the results become inconsistent and
> incorrect.
>
> Kind regards,
> Jelle Nijland
>
>
> --
> *From:* Mich Talebzadeh 
> *Sent:* Wednesday, April 24, 2024 4:40 PM
> *To:* Nijland, J.G.W. (Jelle, Student M-CS) <
> j.g.w.nijl...@student.utwente.nl>
> *Cc:* user@spark.apache.org 
> *Subject:* Re: [spark-graphframes]: Generating incorrect edges
>
> OK few observations
>
> 1) ID Generation Method: How are you generating unique IDs (UUIDs,
> sequential numbers, etc.)?
> 2) Data Inconsistencies: Have you checked for missing values impacting ID
> generation?
> 3) Join Verification: If relevant, can you share the code for joining data
> points during ID creation? Are joins matching columns correctly?
> 4) Specific Edge Issues: Can you share examples of vertex IDs with
> incorrect connections? Is this related to ID generation or edge creation
> logic?
>
> HTH
> Mich Talebzadeh,
> Technologist | Architect | Data Engineer  | Generative AI, FinCrime
> London
> United Kingdom
>
>
>view my Linkedin profile
> 
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> *Disclaimer:* The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner Von
> Braun )".
>
>
> On Wed, 24 Apr 2024 at 12:24, Nijland, J.G.W. (Jelle, Student M-CS) <
> j.g.w.nijl...@student.utwente.nl> wrote:
>
> tags: pyspark,spark-graphframes
>
> Hello,
>
> I am running pyspark in a podman container and I have issues with
> incorrect edges when I build my graph.
> I start with loading a source dataframe from a parquet directory on my
> server. The source dataframe has the following columns:
>
> +-+---+-+-+--+-+--+---+
> |created |descr |last_modified|mnt_by |origin|start_address|prefix
> |external_origin|
>
> +-+---+-+-+--+-+--+---+
>
> I aim to build a graph connecting prefix, mnt_by, origin and descr with
> edges storing the created and last_modified values.
> I start with generating IDs for the prefix, mnt_by, origin and descr using

Re: [spark-graphframes]: Generating incorrect edges

[Nijland, J.G.W. (Jelle, Student M-CS)]

Hi Mich,

Thanks for your reply,
1) ID generation is done using 
monotonically_increasing_id()
 this is then prefixed with "p_", "m_", "o_" or "org_" depending on the type of 
the value it identifies.
2) There are some missing values in the Origin column, these will result in a 
Null ID
3) The join code is present in [1], I join "left" on the same column I create 
the ID on
4) I dont think the issue is in ID or edge generation, if i limit my input 
dataframe and union it with my Utwente data row, I can verify those edges are 
created correctly up to 100K records.
Once I go past that amount of records the results become inconsistent and 
incorrect.

Kind regards,
Jelle Nijland



From: Mich Talebzadeh 
Sent: Wednesday, April 24, 2024 4:40 PM
To: Nijland, J.G.W. (Jelle, Student M-CS) 
Cc: user@spark.apache.org 
Subject: Re: [spark-graphframes]: Generating incorrect edges

OK few observations

1) ID Generation Method: How are you generating unique IDs (UUIDs, sequential 
numbers, etc.)?
2) Data Inconsistencies: Have you checked for missing values impacting ID 
generation?
3) Join Verification: If relevant, can you share the code for joining data 
points during ID creation? Are joins matching columns correctly?
4) Specific Edge Issues: Can you share examples of vertex IDs with incorrect 
connections? Is this related to ID generation or edge creation logic?

HTH
Mich Talebzadeh,
Technologist | Architect | Data Engineer  | Generative AI, FinCrime
London
United Kingdom


 
[https://ci3.googleusercontent.com/mail-sig/AIorK4zholKucR2Q9yMrKbHNn-o1TuS4mYXyi2KO6Xmx6ikHPySa9MLaLZ8t2hrA6AUcxSxDgHIwmKE]
   view my Linkedin 
profile


 https://en.everybodywiki.com/Mich_Talebzadeh



Disclaimer: The information provided is correct to the best of my knowledge but 
of course cannot be guaranteed . It is essential to note that, as with any 
advice, quote "one test result is worth one-thousand expert opinions (Werner 
 Von 
Braun)".


On Wed, 24 Apr 2024 at 12:24, Nijland, J.G.W. (Jelle, Student M-CS) 
mailto:j.g.w.nijl...@student.utwente.nl>> 
wrote:
tags: pyspark,spark-graphframes

Hello,

I am running pyspark in a podman container and I have issues with incorrect 
edges when I build my graph.
I start with loading a source dataframe from a parquet directory on my server. 
The source dataframe has the following columns:
+-+---+-+-+--+-+--+---+
|created |descr |last_modified|mnt_by |origin|start_address|prefix 
|external_origin|
+-+---+-+-+--+-+--+---+

I aim to build a graph connecting prefix, mnt_by, origin and descr with edges 
storing the created and last_modified values.
I start with generating IDs for the prefix, mnt_by, origin and descr using 
monotonically_increasing_id() [1]
These IDs are prefixed with "m_", "p_", "o_" or "org_" to ensure they are 
unique IDs across the dataframe.

Then I construct the vertices dataframe by collecting the ID, value and whether 
they are external for each vertex. [2]
These vertices are then unioned together.
Following the vertices, I construct the edges dataframe by selecting the IDs 
that I want to be the src and the dst and union those together. [3]
These edges store the created and last_modified.

Now I am ready to construct the graph. Here is where I run into my issue.

When verifying my graph, I looked at a couple of vertices to see if they have 
the correct edges.
I looked at the Utwente prefix, origin, descr and mnt_by and found that it 
generates incorrect edges.

I saw edges going out to vertices that are not associated with the utwente 
values at all.
The methods to find the vertices, edges and the output can be found in [4]
We can already observe inconsistencies by viewing the prefix->maintainer and 
origin -> prefix edges. [5]
Depending on what column I filter on the results are inconsistent.
To make matters worse some edges contain IDs that are not connected to the 
original values in the source dataframe at all.

What I have tried to resolve my issue:

  *
Write a checker that verifies edges created against the source dataframe. [6]
The aim of this checker was to determine where the inconsistency comes fro, to 
locate the bug and resolve it.
I ran this checker a limited graphs from n=10 upwards to n=100 (or 1m).
This felt close enough as there are only ~6.5m records in my source dataframe.
This ran correctly, near the 1m it did experience significant slowdown at the 
full dataframe it errors/times out.
I blamed this on the large joins that it performs on the source dataframe.
  *
I found a github issue of 

Re: [spark-graphframes]: Generating incorrect edges

[Mich Talebzadeh]

OK few observations

1) ID Generation Method: How are you generating unique IDs (UUIDs,
sequential numbers, etc.)?
2) Data Inconsistencies: Have you checked for missing values impacting ID
generation?
3) Join Verification: If relevant, can you share the code for joining data
points during ID creation? Are joins matching columns correctly?
4) Specific Edge Issues: Can you share examples of vertex IDs with
incorrect connections? Is this related to ID generation or edge creation
logic?

HTH
Mich Talebzadeh,
Technologist | Architect | Data Engineer  | Generative AI, FinCrime
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Wed, 24 Apr 2024 at 12:24, Nijland, J.G.W. (Jelle, Student M-CS) <
j.g.w.nijl...@student.utwente.nl> wrote:

> tags: pyspark,spark-graphframes
>
> Hello,
>
> I am running pyspark in a podman container and I have issues with
> incorrect edges when I build my graph.
> I start with loading a source dataframe from a parquet directory on my
> server. The source dataframe has the following columns:
>
> +-+---+-+-+--+-+--+---+
> |created |descr |last_modified|mnt_by |origin|start_address|prefix
> |external_origin|
>
> +-+---+-+-+--+-+--+---+
>
> I aim to build a graph connecting prefix, mnt_by, origin and descr with
> edges storing the created and last_modified values.
> I start with generating IDs for the prefix, mnt_by, origin and descr using
> monotonically_increasing_id() [1]
> These IDs are prefixed with "m_", "p_", "o_" or "org_" to ensure they are
> unique IDs across the dataframe.
>
> Then I construct the vertices dataframe by collecting the ID, value and
> whether they are external for each vertex. [2]
> These vertices are then unioned together.
> Following the vertices, I construct the edges dataframe by selecting the
> IDs that I want to be the src and the dst and union those together. [3]
> These edges store the created and last_modified.
>
> Now I am ready to construct the graph. Here is where I run into my issue.
>
> When verifying my graph, I looked at a couple of vertices to see if they
> have the correct edges.
> I looked at the Utwente prefix, origin, descr and mnt_by and found that it
> generates incorrect edges.
>
> I saw edges going out to vertices that are not associated with the utwente
> values at all.
> The methods to find the vertices, edges and the output can be found in [4]
> We can already observe inconsistencies by viewing the prefix->maintainer
> and origin -> prefix edges. [5]
> Depending on what column I filter on the results are inconsistent.
> To make matters worse some edges contain IDs that are not connected to the
> original values in the source dataframe at all.
>
> What I have tried to resolve my issue:
>
>- Write a checker that verifies edges created against the source
>dataframe. [6]
>The aim of this checker was to determine where the inconsistency comes
>fro, to locate the bug and resolve it.
>I ran this checker a limited graphs from n=10 upwards to n=100 (or
>1m).
>This felt close enough as there are only ~6.5m records in my source
>dataframe.
>This ran correctly, near the 1m it did experience significant slowdown
>at the full dataframe it errors/times out.
>I blamed this on the large joins that it performs on the source
>dataframe.
>- I found a github issue of someone with significantly larger graphs
>have similar issues.
>One suggestion there blamed indexing using strings rather than ints or
>longs.
>I rewrote my system to use int for IDs but I ran into the same issue.
>The amount of incorrect edges was the same, the link to which
>incorrects vertices it links to was the same too.
>- I re-ordered my source dataframe to see what the impact was.
>This results in considerably more incorrect edges using the checker in
>[4]
>If helpful I can post the output of this checker as well.
>
>
> Can you give me any pointers in what I can try or what I can do to clarify
> my situation better?
> Thanks in advance for your time.
>
> Kind regards,
> Jelle Nijland
>
>
>
>
> [1]
> import pyspark.sql.functions as psf
>
> # ID labels
> PREFIX_ID = "prefix_id"
> MAINTAINER_ID = "mnt_by_id"
> ORIGIN_ID = "origin_id"
> ORGANISATION_ID = "organisation_id"
>
> # Source dataframe column names
> MNT_BY = "mnt_by"
> PREFIX = "prefix"
> ORIGIN = "origin"
> DESCR = "descr"
> 

RE: How to add MaxDOP option in spark mssql JDBC

[Appel, Kevin]

You might be able to leverage the prepareQuery option, that is at 
https://spark.apache.org/docs/3.5.1/sql-data-sources-jdbc.html#data-source-option
 ... this was introduced in Spark 3.4.0 to handle temp table query and CTE 
query against MSSQL server since what you send in is not actually what gets 
sent, there is some items that get wrapped.

There is more of the technical info in 
https://issues.apache.org/jira/browse/SPARK-37259 with the PR's linked that had 
the fix done for this


From: Elite 
Sent: Tuesday, April 23, 2024 10:28 PM
To: user@spark.apache.org
Subject: How to add MaxDOP option in spark mssql JDBC

[QUESTION] How to pass MAXDOP option * Issue #2395 * microsoft/mssql-jdbc 
(github.com)

Hi team,

I am suggested to require help form spark community.

We suspect spark rewerite the query before pass to ms sql, and it lead to 
syntax error.
Is there any work around to let make my codes work?


spark.read()
.format("jdbc")
.option("driver","com.microsoft.sqlserver.jdbc.SQLServerDriver")
.option("url", "jdbc:sqlserver://xxx.database.windows.net;databaseName=")
.option("query", "SELECT TOP 10 * FROM dbo.Demo with (nolock) WHERE Id = 1 
option (maxdop 1)")
.load()
.show();

com.microsoft.sqlserver.jdbc.SQLServerException: Incorrect syntax near the 
keyword 'option'.
at 
com.microsoft.sqlserver.jdbc.SQLServerException.makeFromDatabaseError(SQLServerException.java:270)
at 
com.microsoft.sqlserver.jdbc.SQLServerStatement.getNextResult(SQLServerStatement.java:1778)
at 
com.microsoft.sqlserver.jdbc.SQLServerPreparedStatement.doExecutePreparedStatement(SQLServerPreparedStatement.java:697)
at 
com.microsoft.sqlserver.jdbc.SQLServerPreparedStatement$PrepStmtExecCmd.doExecute(SQLServerPreparedStatement.java:616)
at com.microsoft.sqlserver.jdbc.TDSCommand.execute(IOBuffer.java:7775)
at 
com.microsoft.sqlserver.jdbc.SQLServerConnection.executeCommand(SQLServerConnection.java:4397)
at 
com.microsoft.sqlserver.jdbc.SQLServerStatement.executeCommand(SQLServerStatement.java:293)
at 
com.microsoft.sqlserver.jdbc.SQLServerStatement.executeStatement(SQLServerStatement.java:263)
at 
com.microsoft.sqlserver.jdbc.SQLServerPreparedStatement.executeQuery(SQLServerPreparedStatement.java:531)
at 
org.apache.spark.sql.execution.datasources.jdbc.JDBCRDD$.resolveTable(JDBCRDD.scala:61)
at 
org.apache.spark.sql.execution.datasources.jdbc.JDBCRelation$.getSchema(JDBCRelation.scala:226)
at 
org.apache.spark.sql.execution.datasources.jdbc.JdbcRelationProvider.createRelation(JdbcRelationProvider.scala:35)
at 
org.apache.spark.sql.execution.datasources.DataSource.resolveRelation(DataSource.scala:344)
at org.apache.spark.sql.DataFrameReader.loadV1Source(DataFrameReader.scala:297)
at 
org.apache.spark.sql.DataFrameReader.$anonfun$load$2(DataFrameReader.scala:286)
at scala.Option.getOrElse(Option.scala:189)
at org.apache.spark.sql.DataFrameReader.load(DataFrameReader.scala:286)
at org.apache.spark.sql.DataFrameReader.load(DataFrameReader.scala:221)

--
This message, and any attachment(s), is for the intended recipient(s) only, may 
contain information that is privileged, confidential and/or proprietary and 
subject to important terms and conditions available at 
http://www.bankofamerica.com/electronic-disclaimer. If you are not the intended 
recipient, please delete this message. For more information about how Bank of 
America protects your privacy, including specific rights that may apply, please 
visit the following pages: 
https://business.bofa.com/en-us/content/global-privacy-notices.html (which 
includes global privacy notices) and 
https://www.bankofamerica.com/security-center/privacy-overview/ (which includes 
US State specific privacy notices such as the 
http://www.bankofamerica.com/ccpa-notice).

[spark-graphframes]: Generating incorrect edges

[Nijland, J.G.W. (Jelle, Student M-CS)]

tags: pyspark,spark-graphframes

Hello,

I am running pyspark in a podman container and I have issues with incorrect 
edges when I build my graph.
I start with loading a source dataframe from a parquet directory on my server. 
The source dataframe has the following columns:
+-+---+-+-+--+-+--+---+
|created |descr |last_modified|mnt_by |origin|start_address|prefix 
|external_origin|
+-+---+-+-+--+-+--+---+

I aim to build a graph connecting prefix, mnt_by, origin and descr with edges 
storing the created and last_modified values.
I start with generating IDs for the prefix, mnt_by, origin and descr using 
monotonically_increasing_id() [1]
These IDs are prefixed with "m_", "p_", "o_" or "org_" to ensure they are 
unique IDs across the dataframe.

Then I construct the vertices dataframe by collecting the ID, value and whether 
they are external for each vertex. [2]
These vertices are then unioned together.
Following the vertices, I construct the edges dataframe by selecting the IDs 
that I want to be the src and the dst and union those together. [3]
These edges store the created and last_modified.

Now I am ready to construct the graph. Here is where I run into my issue.

When verifying my graph, I looked at a couple of vertices to see if they have 
the correct edges.
I looked at the Utwente prefix, origin, descr and mnt_by and found that it 
generates incorrect edges.

I saw edges going out to vertices that are not associated with the utwente 
values at all.
The methods to find the vertices, edges and the output can be found in [4]
We can already observe inconsistencies by viewing the prefix->maintainer and 
origin -> prefix edges. [5]
Depending on what column I filter on the results are inconsistent.
To make matters worse some edges contain IDs that are not connected to the 
original values in the source dataframe at all.

What I have tried to resolve my issue:

  *
Write a checker that verifies edges created against the source dataframe. [6]
The aim of this checker was to determine where the inconsistency comes fro, to 
locate the bug and resolve it.
I ran this checker a limited graphs from n=10 upwards to n=100 (or 1m).
This felt close enough as there are only ~6.5m records in my source dataframe.
This ran correctly, near the 1m it did experience significant slowdown at the 
full dataframe it errors/times out.
I blamed this on the large joins that it performs on the source dataframe.
  *
I found a github issue of someone with significantly larger graphs have similar 
issues.
One suggestion there blamed indexing using strings rather than ints or longs.
I rewrote my system to use int for IDs but I ran into the same issue.
The amount of incorrect edges was the same, the link to which incorrects 
vertices it links to was the same too.
  *
I re-ordered my source dataframe to see what the impact was.
This results in considerably more incorrect edges using the checker in [4]
If helpful I can post the output of this checker as well.

Can you give me any pointers in what I can try or what I can do to clarify my 
situation better?
Thanks in advance for your time.

Kind regards,
Jelle Nijland




[1]
import pyspark.sql.functions as psf

# ID labels
PREFIX_ID = "prefix_id"
MAINTAINER_ID = "mnt_by_id"
ORIGIN_ID = "origin_id"
ORGANISATION_ID = "organisation_id"

# Source dataframe column names
MNT_BY = "mnt_by"
PREFIX = "prefix"
ORIGIN = "origin"
DESCR = "descr"
EXTERNAL_O = "external_origin"


def generate_ids(df: DataFrame) -> DataFrame:
"""
Generates a unique ID for each distinct maintainer, prefix, origin and 
organisation

Parameters
--
df : DataFrame
DataFrame to generate IDs for
"""
mnt_by_id = df.select(MNT_BY).distinct().withColumn(MAINTAINER_ID, 
psf.concat(psf.lit('m_'), psf.monotonically_increasing_id()))
prefix_id = df.select(PREFIX).distinct().withColumn(PREFIX_ID, 
psf.concat(psf.lit('p_'), psf.monotonically_increasing_id()))
origin_id = df.select(ORIGIN).distinct().withColumn(ORIGIN_ID, 
psf.concat(psf.lit('o_'), psf.monotonically_increasing_id()))
organisation_id = df.select(DESCR).distinct().withColumn(ORGANISATION_ID, 
psf.concat(psf.lit('org_'), psf.monotonically_increasing_id()))

df = df.join(mnt_by_id, on=MNT_BY, how="left").join(prefix_id, on=PREFIX, 
how="left").join(origin_id, on=ORIGIN, how="left").join(organisation_id, 
on=DESCR, how="left")
return df

def create_vertices(df: DataFrame) -> DataFrame:
"""
Creates vertices from a DataFrame with IDs
Vertices have the format:
ID (str) | VALUE (str) | EXTERNAL (bool)

ID follows the format [p_|o_|m_|org_][0-9]+

Parameters
--
df : DataFrame
DataFrame to generate vertices for
"""
prefixes = df.select(PREFIX_ID, PREFIX, psf.lit(False))
maintainers = df.select(MAINTAINER_ID, 

How to add MaxDOP option in spark mssql JDBC

[Elite]

[QUESTION] How to pass MAXDOP option · Issue #2395 · microsoft/mssql-jdbc 
(github.com)


Hi team, 


I am suggested to require help form spark community.


We suspect spark rewerite the query before pass to ms sql, and it lead to 
syntax error.
Is there any work around to let make my codes work? 



spark.read()
.format("jdbc")
.option("driver","com.microsoft.sqlserver.jdbc.SQLServerDriver")
.option("url", "jdbc:sqlserver://xxx.database.windows.net;databaseName=")
.option("query", "SELECT TOP 10 * FROM dbo.Demo with (nolock) WHERE Id = 1 
option (maxdop 1)")
.load()
.show();

com.microsoft.sqlserver.jdbc.SQLServerException: Incorrect syntax near the 
keyword 'option'.
at 
com.microsoft.sqlserver.jdbc.SQLServerException.makeFromDatabaseError(SQLServerException.java:270)
at 
com.microsoft.sqlserver.jdbc.SQLServerStatement.getNextResult(SQLServerStatement.java:1778)
at 
com.microsoft.sqlserver.jdbc.SQLServerPreparedStatement.doExecutePreparedStatement(SQLServerPreparedStatement.java:697)
at 
com.microsoft.sqlserver.jdbc.SQLServerPreparedStatement$PrepStmtExecCmd.doExecute(SQLServerPreparedStatement.java:616)
at com.microsoft.sqlserver.jdbc.TDSCommand.execute(IOBuffer.java:7775)
at 
com.microsoft.sqlserver.jdbc.SQLServerConnection.executeCommand(SQLServerConnection.java:4397)
at 
com.microsoft.sqlserver.jdbc.SQLServerStatement.executeCommand(SQLServerStatement.java:293)
at 
com.microsoft.sqlserver.jdbc.SQLServerStatement.executeStatement(SQLServerStatement.java:263)
at 
com.microsoft.sqlserver.jdbc.SQLServerPreparedStatement.executeQuery(SQLServerPreparedStatement.java:531)
at 
org.apache.spark.sql.execution.datasources.jdbc.JDBCRDD$.resolveTable(JDBCRDD.scala:61)
at 
org.apache.spark.sql.execution.datasources.jdbc.JDBCRelation$.getSchema(JDBCRelation.scala:226)
at 
org.apache.spark.sql.execution.datasources.jdbc.JdbcRelationProvider.createRelation(JdbcRelationProvider.scala:35)
at 
org.apache.spark.sql.execution.datasources.DataSource.resolveRelation(DataSource.scala:344)
at org.apache.spark.sql.DataFrameReader.loadV1Source(DataFrameReader.scala:297)
at 
org.apache.spark.sql.DataFrameReader.$anonfun$load$2(DataFrameReader.scala:286)
at scala.Option.getOrElse(Option.scala:189)
at org.apache.spark.sql.DataFrameReader.load(DataFrameReader.scala:286)
at org.apache.spark.sql.DataFrameReader.load(DataFrameReader.scala:221)

How to use Structured Streaming in Spark SQL

[????]

In Flink, you can create flow calculation tables using Flink SQL, and directly 
connect with SQL through CDC and Kafka. How to use SQL for flow calculation in 
Spark



308027...@qq.com

How to access the internal hidden columns of table by spark jdbc

[casel.chen]

I want to use spark jdbc to access alibaba cloud hologres 
(https://www.alibabacloud.com/product/hologres)  internal hidden column 
`hg_binlog_timestamp_us ` but met the following error:


Exception in thread "main" org.apache.spark.sql.AnalysisException: cannot 
resolve 'hg_binlog_timestamp_us' given input columns: 
[top_trans_order.acc_split_bunch, top_trans_order.acct_code, ...]; line 4 pos 
30;


So, my question is how to access the internal hidden columns of table by spark 
jdbc (sql and DataFrame) ? Thanks!

Accounting the impact of failures in spark jobs

[Faiz Halde]

Hello,

In my organization, we have an accounting system for spark jobs that uses
the task execution time to determine how much time a spark job uses the
executors for and we use it as a way to segregate cost. We sum all the task
times per job and apply proportions. Our clusters follow a 1 task per core
model & this works well.

A job goes through several failures during its run, due to executor
failure, node failure ( spot interruptions ), and spark retries tasks &
sometimes entire stages.

We now want to account for this failure and determine what % of a job's
total task time is due to these retries. Basically, if a job with failures
& retries has a total task time of X, there is a X' representing the
goodput of this job – i.e. a hypothetical run of the job with 0 failures &
retries. In this case, ( X-X' ) / X quantifies the cost of failures.

This form of accounting requires tracking execution history of each task
i.e. tasks that compute the same logical partition of some RDD. This was
quite easy with AQE disabled as stage ids never changed, but with AQE
enabled that's no longer the case.

Do you have any suggestions on how I can use the Spark event system?

Thanks
Faiz

StreamingQueryListener integration with Spark native metric sink (JmxSink)

[Mason Chen]

Hi all,

Is it possible to integrate StreamingQueryListener with Spark metrics so
that metrics can be reported through Spark's internal metric system?
Ideally, I would like to report some custom metrics through
StreamingQueryListener and export them to Spark's JmxSink.

Best,
Mason

[ANNOUNCE] Apache Spark 3.4.3 released

[Dongjoon Hyun]

We are happy to announce the availability of Apache Spark 3.4.3!

Spark 3.4.3 is a maintenance release containing many fixes including
security and correctness domains. This release is based on the
branch-3.4 maintenance branch of Spark. We strongly
recommend all 3.4 users to upgrade to this stable release.

To download Spark 3.4.3, head over to the download page:
https://spark.apache.org/downloads.html

To view the release notes:
https://spark.apache.org/releases/spark-release-3-4-3.html

We would like to acknowledge all community members for contributing to this
release. This release would not have been possible without you.

Dongjoon Hyun

[Spark SQL][How-To] Remove builtin function support from Spark

[Matthew McMillian]

Hello,

I'm very new to the Spark ecosystem, apologies if this question is a bit
simple.

I want to modify a custom fork of Spark to remove function support. For
example, I want to remove the query runners ability to call reflect and
java_method. I saw that there exists a data structure in spark-sql called
FunctionRegistry that seems to act as an allowlist on what functions Spark
can execute. If I remove a function of the registry, is that enough
guarantee that that function can "never" be invoked in Spark, or are there
other areas that would need to be changed as well?

Thanks,
Matthew McMillian

[Spark SQL][How-To] Remove builtin function support from Spark

[Matthew McMillian]

Hello,

I'm very new to the Spark ecosystem, apologies if this question is a bit
simple.

I want to modify a custom fork of Spark to remove function support. For
example, I want to remove the query runners ability to call reflect and
java_method. I saw that there exists a data structure in spark-sql called
FunctionRegistry that seems to act as an allowlist on what functions Spark
can execute. If I remove a function of the registry, is that enough
guarantee that that function can "never" be invoked in Spark, or are there
other areas that would need to be changed as well?

Thanks,
Matthew McMillian

should OutputCommitCoordinator fail stages for authorized committer failures when using s3a optimized committers?

[Dylan McClelland]

In https://issues.apache.org/jira/browse/SPARK-39195,
OutputCommitCoordinator was modified to fail a stage if an authorized
committer task fails.

We run our spark jobs on a k8s cluster managed by karpenter and mostly
built from spot instances. As a result, our executors are frequently
killed. With the above change, that leads to expensive stage failures at
the final write stage.

I think I understand why the above is needed when using
FileOutputCommitter, but it seems like we can handle things like the magic
s3a committer differently. For those, we could instead abort the task
attempt, which will the data files that are awaiting the final PUT
operation, and remove them from the list of files to be completed during
the job commit phase

Does this seem reasonable? I think the change could go in
OutputCommitCoordinator (as a case in the taskCompleted block), but there
are other options as well

Any other ideas on how stop individual failures of authorized committer
tasks from failing the whole job?

[Spark SQL] xxhash64 default seed of 42 confusion

[Igor Calabria]

Hi all,

I've noticed that spark's xxhas64 output doesn't match other tool's due to
using seed=42 as a default. I've looked at a few libraries and they use 0
as a default seed:

- python https://github.com/ifduyue/python-xxhash
- java https://github.com/OpenHFT/Zero-Allocation-Hashing/
- java (slice library, used by trino)
https://github.com/airlift/slice/blob/master/src/main/java/io/airlift/slice/XxHash64.java

Was there a special motivation behind this? or is 42 just used for the sake
of the hitchhiker's guide reference? It's very common for spark to interact
with other tools (either via data or direct connection) and this just seems
like a unnecessary footgun.

auto create event log directory if not exist

[second_co...@yahoo.com.INVALID]

Spark history server is set to use s3a, like below
spark.eventLog.enabled true
spark.eventLog.dir s3a://bucket-test/test-directory-log

any configuration option i can set on the Spark config such that if the 
directory 'test-directory-log' does not exist auto create it before start Spark 
history server?

Thank you

Re: Spark streaming job for kafka transaction does not consume read_committed messages correctly.

[Kidong Lee]

Thanks, Mich for your reply.

I agree, it is not so scalable and efficient. But it works correctly for
kafka transaction, and there is no problem with committing offset to kafka
async for now.

I try to tell you some more details about my streaming job.
CustomReceiver does not receive anything from outside and just forward
notice message to run an executor in which kafka consumer will be run.
See my CustomReceiver.

private static class CustomReceiver extends Receiver {

public CustomReceiver() {
super(StorageLevel.MEMORY_AND_DISK_2());
}

@Override
public void onStart() {
new Thread(this::receive).start();
}

private void receive() {
String input = "receiver input " + UUID.randomUUID().toString();
store(input);
}

@Override
public void onStop() {

}
}


Actually, just one Kafka consumer will be run which consumes committed
messages from kafka directly(, which is not so scalable, I think.).
But the main point of this approach which I need is that spark
session needs to be used to save rdd(parallelized consumed messages) to
iceberg table.
Consumed messages will be converted to spark rdd which will be saved to
iceberg table using spark session.

I have tested this spark streaming job with transactional producers which
send several millions of messages. Correctly consumed and saved to iceberg
tables correctly.

- Kidong.



2024년 4월 14일 (일) 오후 11:05, Mich Talebzadeh 님이 작성:

> Interesting
>
> My concern is infinite Loop in* foreachRDD*: The *while(true)* loop
> within foreachRDD creates an infinite loop within each Spark executor. This
> might not be the most efficient approach, especially since offsets are
> committed asynchronously.?
>
> HTH
>
> Mich Talebzadeh,
> Technologist | Solutions Architect | Data Engineer  | Generative AI
> London
> United Kingdom
>
>
>view my Linkedin profile
> 
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> *Disclaimer:* The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner  Von
> Braun )".
>
>
> On Sun, 14 Apr 2024 at 13:40, Kidong Lee  wrote:
>
>>
>> Because spark streaming for kafk transaction does not work correctly to
>> suit my need, I moved to another approach using raw kafka consumer which
>> handles read_committed messages from kafka correctly.
>>
>> My codes look like the following.
>>
>> JavaDStream stream = ssc.receiverStream(new CustomReceiver()); // 
>> CustomReceiver does nothing special except awaking foreach task.
>>
>> stream.foreachRDD(rdd -> {
>>
>>   KafkaConsumer consumer = new 
>> KafkaConsumer<>(consumerProperties);
>>
>>   consumer.subscribe(Arrays.asList(topic));
>>
>>   while(true){
>>
>> ConsumerRecords records =
>> consumer.poll(java.time.Duration.ofMillis(intervalMs));
>>
>> Map offsetMap = new HashMap<>();
>>
>> List someList = new ArrayList<>();
>>
>> for (ConsumerRecord consumerRecord : records) {
>>
>>   // add something to list.
>>
>>   // put offset to offsetMap.
>>
>> }
>>
>> // process someList.
>>
>> // commit offset.
>>
>> consumer.commitAsync(offsetMap, null);
>>
>>   }
>>
>> });
>>
>>
>> In addition, I increased max.poll.records to 10.
>>
>> Even if this raw kafka consumer approach is not so scalable, it consumes
>> read_committed messages from kafka correctly and is enough for me at the
>> moment.
>>
>> - Kidong.
>>
>>
>>
>> 2024년 4월 12일 (금) 오후 9:19, Kidong Lee 님이 작성:
>>
>>> Hi,
>>>
>>> I have a kafka producer which sends messages transactionally to kafka
>>> and spark streaming job which should consume read_committed messages from
>>> kafka.
>>> But there is a problem for spark streaming to consume read_committed
>>> messages.
>>> The count of messages sent by kafka producer transactionally is not the
>>> same to the count of the read_committed messages consumed by spark
>>> streaming.
>>>
>>> Some consumer properties of my spark streaming job are as follows.
>>>
>>> auto.offset.reset=earliest
>>> enable.auto.commit=false
>>> isolation.level=read_committed
>>>
>>>
>>> I also added the following spark streaming configuration.
>>>
>>> sparkConf.set("spark.streaming.kafka.allowNonConsecutiveOffsets", "true");
>>> sparkConf.set("spark.streaming.kafka.consumer.poll.ms", String.valueOf(2 * 
>>> 60 * 1000));
>>>
>>>
>>> My spark streaming is using DirectStream like this.
>>>
>>> JavaInputDStream> stream =
>>> KafkaUtils.createDirectStream(
>>> ssc,
>>> LocationStrategies.PreferConsistent(),
>>> ConsumerStrategies.>> GenericRecord>Subscribe(topics, kafkaParams)
>>> );
>>>
>>>
>>> stream.foreachRDD(rdd -> O
>>>
>>> 

Re: Spark streaming job for kafka transaction does not consume read_committed messages correctly.

[Mich Talebzadeh]

Interesting

My concern is infinite Loop in* foreachRDD*: The *while(true)* loop within
foreachRDD creates an infinite loop within each Spark executor. This might
not be the most efficient approach, especially since offsets are committed
asynchronously.?

HTH

Mich Talebzadeh,
Technologist | Solutions Architect | Data Engineer  | Generative AI
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Sun, 14 Apr 2024 at 13:40, Kidong Lee  wrote:

>
> Because spark streaming for kafk transaction does not work correctly to
> suit my need, I moved to another approach using raw kafka consumer which
> handles read_committed messages from kafka correctly.
>
> My codes look like the following.
>
> JavaDStream stream = ssc.receiverStream(new CustomReceiver()); // 
> CustomReceiver does nothing special except awaking foreach task.
>
> stream.foreachRDD(rdd -> {
>
>   KafkaConsumer consumer = new 
> KafkaConsumer<>(consumerProperties);
>
>   consumer.subscribe(Arrays.asList(topic));
>
>   while(true){
>
> ConsumerRecords records =
> consumer.poll(java.time.Duration.ofMillis(intervalMs));
>
> Map offsetMap = new HashMap<>();
>
> List someList = new ArrayList<>();
>
> for (ConsumerRecord consumerRecord : records) {
>
>   // add something to list.
>
>   // put offset to offsetMap.
>
> }
>
> // process someList.
>
> // commit offset.
>
> consumer.commitAsync(offsetMap, null);
>
>   }
>
> });
>
>
> In addition, I increased max.poll.records to 10.
>
> Even if this raw kafka consumer approach is not so scalable, it consumes
> read_committed messages from kafka correctly and is enough for me at the
> moment.
>
> - Kidong.
>
>
>
> 2024년 4월 12일 (금) 오후 9:19, Kidong Lee 님이 작성:
>
>> Hi,
>>
>> I have a kafka producer which sends messages transactionally to kafka and
>> spark streaming job which should consume read_committed messages from kafka.
>> But there is a problem for spark streaming to consume read_committed
>> messages.
>> The count of messages sent by kafka producer transactionally is not the
>> same to the count of the read_committed messages consumed by spark
>> streaming.
>>
>> Some consumer properties of my spark streaming job are as follows.
>>
>> auto.offset.reset=earliest
>> enable.auto.commit=false
>> isolation.level=read_committed
>>
>>
>> I also added the following spark streaming configuration.
>>
>> sparkConf.set("spark.streaming.kafka.allowNonConsecutiveOffsets", "true");
>> sparkConf.set("spark.streaming.kafka.consumer.poll.ms", String.valueOf(2 * 
>> 60 * 1000));
>>
>>
>> My spark streaming is using DirectStream like this.
>>
>> JavaInputDStream> stream =
>> KafkaUtils.createDirectStream(
>> ssc,
>> LocationStrategies.PreferConsistent(),
>> ConsumerStrategies.Subscribe(topics, 
>> kafkaParams)
>> );
>>
>>
>> stream.foreachRDD(rdd -> O
>>
>>// get offset ranges.
>>
>>OffsetRange[] offsetRanges = ((HasOffsetRanges) 
>> rdd.rdd()).offsetRanges();
>>
>>// process something.
>>
>>
>>// commit offset.
>>((CanCommitOffsets) stream.inputDStream()).commitAsync(offsetRanges);
>>
>> }
>> );
>>
>>
>>
>> I have tested with a kafka consumer written with raw kafka-clients jar
>> library without problem that it consumes read_committed messages correctly,
>> and the count of consumed read_committed messages is equal to the count of
>> messages sent by kafka producer.
>>
>>
>> And sometimes, I got the following exception.
>>
>> Job aborted due to stage failure: Task 0 in stage 324.0 failed 1 times,
>> most recent failure: Lost task 0.0 in stage 324.0 (TID 1674)
>> (chango-private-1.chango.private executor driver):
>> java.lang.IllegalArgumentException: requirement failed: Failed to get
>> records for compacted spark-executor-school-student-group school-student-7
>> after polling for 12
>>
>> at scala.Predef$.require(Predef.scala:281)
>>
>> at
>> org.apache.spark.streaming.kafka010.InternalKafkaConsumer.compactedNext(KafkaDataConsumer.scala:186)
>>
>> at
>> org.apache.spark.streaming.kafka010.KafkaDataConsumer.compactedNext(KafkaDataConsumer.scala:60)
>>
>> at
>> org.apache.spark.streaming.kafka010.KafkaDataConsumer.compactedNext$(KafkaDataConsumer.scala:59)
>>
>> at
>> org.apache.spark.streaming.kafka010.KafkaDataConsumer$CachedKafkaDataConsumer.compactedNext(KafkaDataConsumer.scala:219)
>>
>>
>>
>> I have experienced spark streaming job which works fine with kafka
>> messages which are non-transactional, and I 

Re: Spark streaming job for kafka transaction does not consume read_committed messages correctly.

[Kidong Lee]

Because spark streaming for kafk transaction does not work correctly to
suit my need, I moved to another approach using raw kafka consumer which
handles read_committed messages from kafka correctly.

My codes look like the following.

JavaDStream stream = ssc.receiverStream(new CustomReceiver());
// CustomReceiver does nothing special except awaking foreach task.

stream.foreachRDD(rdd -> {

  KafkaConsumer consumer = new
KafkaConsumer<>(consumerProperties);

  consumer.subscribe(Arrays.asList(topic));

  while(true){

ConsumerRecords records =
consumer.poll(java.time.Duration.ofMillis(intervalMs));

Map offsetMap = new HashMap<>();

List someList = new ArrayList<>();

for (ConsumerRecord consumerRecord : records) {

  // add something to list.

  // put offset to offsetMap.

}

// process someList.

// commit offset.

consumer.commitAsync(offsetMap, null);

  }

});


In addition, I increased max.poll.records to 10.

Even if this raw kafka consumer approach is not so scalable, it consumes
read_committed messages from kafka correctly and is enough for me at the
moment.

- Kidong.



2024년 4월 12일 (금) 오후 9:19, Kidong Lee 님이 작성:

> Hi,
>
> I have a kafka producer which sends messages transactionally to kafka and
> spark streaming job which should consume read_committed messages from kafka.
> But there is a problem for spark streaming to consume read_committed
> messages.
> The count of messages sent by kafka producer transactionally is not the
> same to the count of the read_committed messages consumed by spark
> streaming.
>
> Some consumer properties of my spark streaming job are as follows.
>
> auto.offset.reset=earliest
> enable.auto.commit=false
> isolation.level=read_committed
>
>
> I also added the following spark streaming configuration.
>
> sparkConf.set("spark.streaming.kafka.allowNonConsecutiveOffsets", "true");
> sparkConf.set("spark.streaming.kafka.consumer.poll.ms", String.valueOf(2 * 60 
> * 1000));
>
>
> My spark streaming is using DirectStream like this.
>
> JavaInputDStream> stream =
> KafkaUtils.createDirectStream(
> ssc,
> LocationStrategies.PreferConsistent(),
> ConsumerStrategies.Subscribe(topics, 
> kafkaParams)
> );
>
>
> stream.foreachRDD(rdd -> O
>
>// get offset ranges.
>
>OffsetRange[] offsetRanges = ((HasOffsetRanges) 
> rdd.rdd()).offsetRanges();
>
>// process something.
>
>
>// commit offset.
>((CanCommitOffsets) stream.inputDStream()).commitAsync(offsetRanges);
>
> }
> );
>
>
>
> I have tested with a kafka consumer written with raw kafka-clients jar
> library without problem that it consumes read_committed messages correctly,
> and the count of consumed read_committed messages is equal to the count of
> messages sent by kafka producer.
>
>
> And sometimes, I got the following exception.
>
> Job aborted due to stage failure: Task 0 in stage 324.0 failed 1 times,
> most recent failure: Lost task 0.0 in stage 324.0 (TID 1674)
> (chango-private-1.chango.private executor driver):
> java.lang.IllegalArgumentException: requirement failed: Failed to get
> records for compacted spark-executor-school-student-group school-student-7
> after polling for 12
>
> at scala.Predef$.require(Predef.scala:281)
>
> at
> org.apache.spark.streaming.kafka010.InternalKafkaConsumer.compactedNext(KafkaDataConsumer.scala:186)
>
> at
> org.apache.spark.streaming.kafka010.KafkaDataConsumer.compactedNext(KafkaDataConsumer.scala:60)
>
> at
> org.apache.spark.streaming.kafka010.KafkaDataConsumer.compactedNext$(KafkaDataConsumer.scala:59)
>
> at
> org.apache.spark.streaming.kafka010.KafkaDataConsumer$CachedKafkaDataConsumer.compactedNext(KafkaDataConsumer.scala:219)
>
>
>
> I have experienced spark streaming job which works fine with kafka
> messages which are non-transactional, and I never encountered the
> exceptions like above.
> It seems that spark streaming for kafka transaction does not handle such
> as kafka consumer properties like isolation.level=read_committed and
> enable.auto.commit=false correctly.
>
> Any help appreciated.
>
> - Kidong.
>
>
> --
> *이기동 *
> *Kidong Lee*
>
> Email: mykid...@gmail.com
> Chango: https://cloudcheflabs.github.io/chango-private-docs
> Web Site: http://www.cloudchef-labs.com/
> Mobile: +82 10 4981 7297
> 
>


-- 
*이기동 *
*Kidong Lee*

Email: mykid...@gmail.com
Chango: https://cloudcheflabs.github.io/chango-private-docs
Web Site: http://www.cloudchef-labs.com/
Mobile: +82 10 4981 7297

Re: Spark streaming job for kafka transaction does not consume read_committed messages correctly.

[Kidong Lee]

Thank you Mich for your reply.

Actually, I tried to do most of your advice.

When spark.streaming.kafka.allowNonConsecutiveOffsets=false, I got the
following error.

Job aborted due to stage failure: Task 0 in stage 1.0 failed 1 times, most
recent failure: Lost task 0.0 in stage 1.0 (TID 3)
(chango-private-1.chango.private executor driver):
java.lang.IllegalArgumentException: requirement failed: Got wrong record
for spark-executor-school-student-group school-student-7 even after seeking
to offset 11206961 got offset 11206962 instead. If this is a compacted
topic, consider enabling spark.streaming.kafka.allowNonConsecutiveOffsets

at scala.Predef$.require(Predef.scala:281)

at
org.apache.spark.streaming.kafka010.InternalKafkaConsumer.get(KafkaDataConsumer.scala:155)

at
org.apache.spark.streaming.kafka010.KafkaDataConsumer.get(KafkaDataConsumer.scala:40)

at
org.apache.spark.streaming.kafka010.KafkaDataConsumer.get$(KafkaDataConsumer.scala:39)

at
org.apache.spark.streaming.kafka010.KafkaDataConsumer$CachedKafkaDataConsumer.get(KafkaDataConsumer.scala:219)


And I tried to increase spark.streaming.kafka.consumer.poll.ms to avoid the
exceptions, but it did not help.


- Kidong.




2024년 4월 14일 (일) 오전 4:25, Mich Talebzadeh 님이 작성:

> Hi Kidong,
>
> There may be few potential reasons why the message counts from your Kafka
> producer and Spark Streaming consumer might not match, especially with
> transactional messages and read_committed isolation level.
>
> 1) Just ensure that both your Spark Streaming job and the Kafka consumer
> written with raw kafka-clients use the same consumer group. Messages are
> delivered to specific consumer groups, and if they differ, Spark Streaming
> might miss messages consumed by the raw consumer.
> 2) Your Spark Streaming configuration sets *enable.auto.commit=false* and
> uses *commitAsync manually*. However, I noted
> *spark.streaming.kafka.allowNonConsecutiveOffsets=true* which may be
> causing the problem. This setting allows Spark Streaming to read offsets
> that are not strictly increasing, which can happen with transactional
> reads. Generally recommended to set this to* false *for transactional
> reads to ensure Spark Streaming only reads committed messages.
> 3) Missed messages, in transactional messages, Kafka guarantees *delivery
> only after the transaction commits successfully. *There could be a slight
> delay between the producer sending the message and it becoming visible to
> consumers under read_committed isolation level. Spark Streaming could
> potentially miss messages during this window.
> 4) The exception Lost task 0.0 in stage 324.0, suggests a problem fetching
> records for a specific topic partition. Review your code handling of
> potential exceptions during rdd.foreachRDD processing. Ensure retries or
> appropriate error handling if encountering issues with specific partitions.
> 5) Try different configurations for *spark.streaming.kafka.consumer.poll.ms
> * to adjust polling
> frequency and potentially improve visibility into committed messages.
>
> HTH
>
> Mich Talebzadeh,
> Technologist | Solutions Architect | Data Engineer  | Generative AI
> London
> United Kingdom
>
>
>view my Linkedin profile
> 
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> *Disclaimer:* The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner  Von
> Braun )".
>
>
> On Fri, 12 Apr 2024 at 21:38, Kidong Lee  wrote:
>
>> Hi,
>>
>> I have a kafka producer which sends messages transactionally to kafka and
>> spark streaming job which should consume read_committed messages from kafka.
>> But there is a problem for spark streaming to consume read_committed
>> messages.
>> The count of messages sent by kafka producer transactionally is not the
>> same to the count of the read_committed messages consumed by spark
>> streaming.
>>
>> Some consumer properties of my spark streaming job are as follows.
>>
>> auto.offset.reset=earliest
>> enable.auto.commit=false
>> isolation.level=read_committed
>>
>>
>> I also added the following spark streaming configuration.
>>
>> sparkConf.set("spark.streaming.kafka.allowNonConsecutiveOffsets", "true");
>> sparkConf.set("spark.streaming.kafka.consumer.poll.ms", String.valueOf(2 * 
>> 60 * 1000));
>>
>>
>> My spark streaming is using DirectStream like this.
>>
>> JavaInputDStream> stream =
>> KafkaUtils.createDirectStream(
>> ssc,
>> LocationStrategies.PreferConsistent(),
>> ConsumerStrategies.Subscribe(topics, 
>> kafkaParams)
>> );
>>
>>
>> stream.foreachRDD(rdd -> O
>>
>>// get offset ranges.
>>
>> 

Re: Spark streaming job for kafka transaction does not consume read_committed messages correctly.

[Mich Talebzadeh]

Hi Kidong,

There may be few potential reasons why the message counts from your Kafka
producer and Spark Streaming consumer might not match, especially with
transactional messages and read_committed isolation level.

1) Just ensure that both your Spark Streaming job and the Kafka consumer
written with raw kafka-clients use the same consumer group. Messages are
delivered to specific consumer groups, and if they differ, Spark Streaming
might miss messages consumed by the raw consumer.
2) Your Spark Streaming configuration sets *enable.auto.commit=false* and
uses *commitAsync manually*. However, I noted
*spark.streaming.kafka.allowNonConsecutiveOffsets=true* which may be
causing the problem. This setting allows Spark Streaming to read offsets
that are not strictly increasing, which can happen with transactional
reads. Generally recommended to set this to* false *for transactional reads
to ensure Spark Streaming only reads committed messages.
3) Missed messages, in transactional messages, Kafka guarantees *delivery
only after the transaction commits successfully. *There could be a slight
delay between the producer sending the message and it becoming visible to
consumers under read_committed isolation level. Spark Streaming could
potentially miss messages during this window.
4) The exception Lost task 0.0 in stage 324.0, suggests a problem fetching
records for a specific topic partition. Review your code handling of
potential exceptions during rdd.foreachRDD processing. Ensure retries or
appropriate error handling if encountering issues with specific partitions.
5) Try different configurations for *spark.streaming.kafka.consumer.poll.ms
* to adjust polling
frequency and potentially improve visibility into committed messages.

HTH

Mich Talebzadeh,
Technologist | Solutions Architect | Data Engineer  | Generative AI
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Fri, 12 Apr 2024 at 21:38, Kidong Lee  wrote:

> Hi,
>
> I have a kafka producer which sends messages transactionally to kafka and
> spark streaming job which should consume read_committed messages from kafka.
> But there is a problem for spark streaming to consume read_committed
> messages.
> The count of messages sent by kafka producer transactionally is not the
> same to the count of the read_committed messages consumed by spark
> streaming.
>
> Some consumer properties of my spark streaming job are as follows.
>
> auto.offset.reset=earliest
> enable.auto.commit=false
> isolation.level=read_committed
>
>
> I also added the following spark streaming configuration.
>
> sparkConf.set("spark.streaming.kafka.allowNonConsecutiveOffsets", "true");
> sparkConf.set("spark.streaming.kafka.consumer.poll.ms", String.valueOf(2 * 60 
> * 1000));
>
>
> My spark streaming is using DirectStream like this.
>
> JavaInputDStream> stream =
> KafkaUtils.createDirectStream(
> ssc,
> LocationStrategies.PreferConsistent(),
> ConsumerStrategies.Subscribe(topics, 
> kafkaParams)
> );
>
>
> stream.foreachRDD(rdd -> O
>
>// get offset ranges.
>
>OffsetRange[] offsetRanges = ((HasOffsetRanges) 
> rdd.rdd()).offsetRanges();
>
>// process something.
>
>
>// commit offset.
>((CanCommitOffsets) stream.inputDStream()).commitAsync(offsetRanges);
>
> }
> );
>
>
>
> I have tested with a kafka consumer written with raw kafka-clients jar
> library without problem that it consumes read_committed messages correctly,
> and the count of consumed read_committed messages is equal to the count of
> messages sent by kafka producer.
>
>
> And sometimes, I got the following exception.
>
> Job aborted due to stage failure: Task 0 in stage 324.0 failed 1 times,
> most recent failure: Lost task 0.0 in stage 324.0 (TID 1674)
> (chango-private-1.chango.private executor driver):
> java.lang.IllegalArgumentException: requirement failed: Failed to get
> records for compacted spark-executor-school-student-group school-student-7
> after polling for 12
>
> at scala.Predef$.require(Predef.scala:281)
>
> at
> org.apache.spark.streaming.kafka010.InternalKafkaConsumer.compactedNext(KafkaDataConsumer.scala:186)
>
> at
> org.apache.spark.streaming.kafka010.KafkaDataConsumer.compactedNext(KafkaDataConsumer.scala:60)
>
> at
> org.apache.spark.streaming.kafka010.KafkaDataConsumer.compactedNext$(KafkaDataConsumer.scala:59)
>
> at
> 

Spark streaming job for kafka transaction does not consume read_committed messages correctly.

[Kidong Lee]

Hi,

I have a kafka producer which sends messages transactionally to kafka and
spark streaming job which should consume read_committed messages from kafka.
But there is a problem for spark streaming to consume read_committed
messages.
The count of messages sent by kafka producer transactionally is not the
same to the count of the read_committed messages consumed by spark
streaming.

Some consumer properties of my spark streaming job are as follows.

auto.offset.reset=earliest
enable.auto.commit=false
isolation.level=read_committed


I also added the following spark streaming configuration.

sparkConf.set("spark.streaming.kafka.allowNonConsecutiveOffsets", "true");
sparkConf.set("spark.streaming.kafka.consumer.poll.ms",
String.valueOf(2 * 60 * 1000));


My spark streaming is using DirectStream like this.

JavaInputDStream> stream =
KafkaUtils.createDirectStream(
ssc,
LocationStrategies.PreferConsistent(),
ConsumerStrategies.Subscribe(topics, kafkaParams)
);


stream.foreachRDD(rdd -> O

   // get offset ranges.

   OffsetRange[] offsetRanges = ((HasOffsetRanges)
rdd.rdd()).offsetRanges();

   // process something.


   // commit offset.
   ((CanCommitOffsets) stream.inputDStream()).commitAsync(offsetRanges);

}
);



I have tested with a kafka consumer written with raw kafka-clients jar
library without problem that it consumes read_committed messages correctly,
and the count of consumed read_committed messages is equal to the count of
messages sent by kafka producer.


And sometimes, I got the following exception.

Job aborted due to stage failure: Task 0 in stage 324.0 failed 1 times,
most recent failure: Lost task 0.0 in stage 324.0 (TID 1674)
(chango-private-1.chango.private executor driver):
java.lang.IllegalArgumentException: requirement failed: Failed to get
records for compacted spark-executor-school-student-group school-student-7
after polling for 12

at scala.Predef$.require(Predef.scala:281)

at
org.apache.spark.streaming.kafka010.InternalKafkaConsumer.compactedNext(KafkaDataConsumer.scala:186)

at
org.apache.spark.streaming.kafka010.KafkaDataConsumer.compactedNext(KafkaDataConsumer.scala:60)

at
org.apache.spark.streaming.kafka010.KafkaDataConsumer.compactedNext$(KafkaDataConsumer.scala:59)

at
org.apache.spark.streaming.kafka010.KafkaDataConsumer$CachedKafkaDataConsumer.compactedNext(KafkaDataConsumer.scala:219)



I have experienced spark streaming job which works fine with kafka messages
which are non-transactional, and I never encountered the exceptions like
above.
It seems that spark streaming for kafka transaction does not handle such as
kafka consumer properties like isolation.level=read_committed and
enable.auto.commit=false correctly.

Any help appreciated.

- Kidong.


-- 
*이기동 *
*Kidong Lee*

Email: mykid...@gmail.com
Chango: https://cloudcheflabs.github.io/chango-private-docs
Web Site: http://www.cloudchef-labs.com/
Mobile: +82 10 4981 7297

Spark column headings, camelCase or snake case?

[Mich Talebzadeh]

I know this is a bit of a silly question. But what is the norm for
 Sparkcolumn headings? Is it camelCase or snakec_ase. For example here "
someone suggested and I quote
SumTotalInMillionGBP" accurately conveys the meaning but is a bit long and
uses camelCase, which is not the standard convention for Spark DataFrames
(usually snake_case). Use snake_case for better readability like:
"total_price_in_millions_gbp"

So this is the gist

+--+-+---+
|district  |NumberOfOffshoreOwned|total_price_in_millions_gbp|
+--+-+---+
|CITY OF WESTMINSTER   |4452 |21472.5|
|KENSINGTON AND CHELSEA|2403 |6544.8 |
|CAMDEN|1023 |4275.9 |
|SOUTHWARK |1080 |3938.0 |
|ISLINGTON |627  |3062.0 |
|TOWER HAMLETS |1715 |3008.0 |
|HAMMERSMITH AND FULHAM|765  |2137.2 |

Now I recently saw a note (if i recall correctly) that Spark should be
using camelCase in new spark related documents. What are the accepted views
or does it matter?

Thanks
Mich Talebzadeh,

Technologist | Solutions Architect | Data Engineer  | Generative AI

London
United Kingdom


   view my Linkedin profile


 https://en.everybodywiki.com/Mich_Talebzadeh



Disclaimer: The information provided is correct to the best of my knowledge
but of course cannot be guaranteed . It is essential to note that, as with
any advice, quote "one test result is worth one-thousand expert opinions
(Werner Von Braun)".

Re: External Spark shuffle service for k8s

[Bjørn Jørgensen]

I think this answers your question about what to do if you need more space
on nodes.

https://spark.apache.org/docs/latest/running-on-kubernetes.html#local-storage

Local Storage


Spark supports using volumes to spill data during shuffles and other
operations. To use a volume as local storage, the volume’s name should
starts with spark-local-dir-, for example:

--conf 
spark.kubernetes.driver.volumes.[VolumeType].spark-local-dir-[VolumeName].mount.path=
--conf 
spark.kubernetes.driver.volumes.[VolumeType].spark-local-dir-[VolumeName].mount.readOnly=false

Specifically, you can use persistent volume claims if the jobs require
large shuffle and sorting operations in executors.

spark.kubernetes.executor.volumes.persistentVolumeClaim.spark-local-dir-1.options.claimName=OnDemand
spark.kubernetes.executor.volumes.persistentVolumeClaim.spark-local-dir-1.options.storageClass=gp
spark.kubernetes.executor.volumes.persistentVolumeClaim.spark-local-dir-1.options.sizeLimit=500Gi
spark.kubernetes.executor.volumes.persistentVolumeClaim.spark-local-dir-1.mount.path=/data
spark.kubernetes.executor.volumes.persistentVolumeClaim.spark-local-dir-1.mount.readOnly=false

To enable shuffle data recovery feature via the built-in
KubernetesLocalDiskShuffleDataIO plugin, we need to have the followings.
You may want to enable
spark.kubernetes.driver.waitToReusePersistentVolumeClaim additionally.

spark.kubernetes.executor.volumes.persistentVolumeClaim.spark-local-dir-1.mount.path=/data/spark-x/executor-x
spark.shuffle.sort.io.plugin.class=org.apache.spark.shuffle.KubernetesLocalDiskShuffleDataIO

If no volume is set as local storage, Spark uses temporary scratch space to
spill data to disk during shuffles and other operations. When using
Kubernetes as the resource manager the pods will be created with an emptyDir
 volume
mounted for each directory listed in spark.local.dir or the environment
variable SPARK_LOCAL_DIRS . If no directories are explicitly specified then
a default directory is created and configured appropriately.

emptyDir volumes use the ephemeral storage feature of Kubernetes and do not
persist beyond the life of the pod.

tor. 11. apr. 2024 kl. 10:29 skrev Bjørn Jørgensen :

> " In the end for my usecase I started using pvcs and pvc aware scheduling
> along with decommissioning. So far performance is good with this choice."
> How did you do this?
>
>
> tor. 11. apr. 2024 kl. 04:13 skrev Arun Ravi :
>
>> Hi Everyone,
>>
>> I had to explored IBM's and AWS's S3 shuffle plugins (some time back), I
>> had also explored AWS FSX lustre in few of my production jobs which has
>> ~20TB of shuffle operations with 200-300 executors. What I have observed is
>> S3 and fax behaviour was fine during the write phase, however I faced iops
>> throttling during the read phase(read taking forever to complete). I think
>> this might be contributed by the heavy use of shuffle index file (I didn't
>> perform any extensive research on this), so I believe the shuffle manager
>> logic have to be intelligent enough to reduce the fetching of files from
>> object store. In the end for my usecase I started using pvcs and pvc aware
>> scheduling along with decommissioning. So far performance is good with this
>> choice.
>>
>> Thank you
>>
>> On Tue, 9 Apr 2024, 15:17 Mich Talebzadeh, 
>> wrote:
>>
>>> Hi,
>>>
>>> First thanks everyone for their contributions
>>>
>>> I was going to reply to @Enrico Minack   but
>>> noticed additional info. As I understand for example,  Apache Uniffle is an
>>> incubating project aimed at providing a pluggable shuffle service for
>>> Spark. So basically, all these "external shuffle services" have in common
>>> is to offload shuffle data management to external services, thus reducing
>>> the memory and CPU overhead on Spark executors. That is great.  While
>>> Uniffle and others enhance shuffle performance and scalability, it would be
>>> great to integrate them with Spark UI. This may require additional
>>> development efforts. I suppose  the interest would be to have these
>>> external matrices incorporated into Spark with one look and feel. This may
>>> require customizing the UI to fetch and display metrics or statistics from
>>> the external shuffle services. Has any project done this?
>>>
>>> Thanks
>>>
>>> Mich Talebzadeh,
>>> Technologist | Solutions Architect | Data Engineer  | Generative AI
>>> London
>>> United Kingdom
>>>
>>>
>>>view my Linkedin profile
>>> 
>>>
>>>
>>>  https://en.everybodywiki.com/Mich_Talebzadeh
>>>
>>>
>>>
>>> *Disclaimer:* The information provided is correct to the best of my
>>> knowledge but of course cannot be guaranteed . It is essential to note
>>> that, as with any advice, quote "one test result is worth one-thousand
>>> expert opinions (Werner
>>> 

Re: [Spark SQL]: Source code for PartitionedFile

[Ashley McManamon]

Hi Mich,

Thanks for the reply.

I did come across that file but it didn't align with the appearance of
`PartitionedFile`:
https://github.com/apache/spark/blob/master/sql/core/src/main/scala/org/apache/spark/sql/execution/PartitionedFileUtil.scala

In fact, the code snippet you shared also references the type
`PartitionedFile`.

There's actually this javadoc.io page for a `PartitionedFile`
at org.apache.spark.sql.execution.datasources for spark-sql_2.12:3.0.2:
https://javadoc.io/doc/org.apache.spark/spark-sql_2.12/3.0.2/org/apache/spark/sql/execution/datasources/PartitionedFile.html.
I double checked the source code for version 3.0.2 and doesn't seem to
exist there either:
https://github.com/apache/spark/tree/v3.0.2/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources

Ashley


On Mon, 8 Apr 2024 at 22:41, Mich Talebzadeh 
wrote:

> Hi,
>
> I believe this is the package
>
>
> https://raw.githubusercontent.com/apache/spark/master/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FilePartition.scala
>
> And the code
>
> case class FilePartition(index: Int, files: Array[PartitionedFile])
>   extends Partition with InputPartition {
>   override def preferredLocations(): Array[String] = {
> // Computes total number of bytes that can be retrieved from each host.
> val hostToNumBytes = mutable.HashMap.empty[String, Long]
> files.foreach { file =>
>   file.locations.filter(_ != "localhost").foreach { host =>
> hostToNumBytes(host) = hostToNumBytes.getOrElse(host, 0L) +
> file.length
>   }
> }
>
> // Selects the first 3 hosts with the most data to be retrieved.
> hostToNumBytes.toSeq.sortBy {
>   case (host, numBytes) => numBytes
> }.reverse.take(3).map {
>   case (host, numBytes) => host
> }.toArray
>   }
> }
>
> HTH
>
> Mich Talebzadeh,
> Technologist | Solutions Architect | Data Engineer  | Generative AI
> London
> United Kingdom
>
>
>view my Linkedin profile
> 
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> *Disclaimer:* The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner  Von
> Braun )".
>
>
> On Mon, 8 Apr 2024 at 20:31, Ashley McManamon <
> ashley.mcmana...@quantcast.com> wrote:
>
>> Hi All,
>>
>> I've been diving into the source code to get a better understanding of
>> how file splitting works from a user perspective. I've hit a deadend at
>> `PartitionedFile`, for which I cannot seem to find a definition? It appears
>> though it should be found at
>> org.apache.spark.sql.execution.datasources but I find no definition in
>> the entire source code. Am I missing something?
>>
>> I appreciate there may be an obvious answer here, apologies if I'm being
>> naive.
>>
>> Thanks,
>> Ashley McManamon
>>
>>

Re: External Spark shuffle service for k8s

[Bjørn Jørgensen]

" In the end for my usecase I started using pvcs and pvc aware scheduling
along with decommissioning. So far performance is good with this choice."
How did you do this?


tor. 11. apr. 2024 kl. 04:13 skrev Arun Ravi :

> Hi Everyone,
>
> I had to explored IBM's and AWS's S3 shuffle plugins (some time back), I
> had also explored AWS FSX lustre in few of my production jobs which has
> ~20TB of shuffle operations with 200-300 executors. What I have observed is
> S3 and fax behaviour was fine during the write phase, however I faced iops
> throttling during the read phase(read taking forever to complete). I think
> this might be contributed by the heavy use of shuffle index file (I didn't
> perform any extensive research on this), so I believe the shuffle manager
> logic have to be intelligent enough to reduce the fetching of files from
> object store. In the end for my usecase I started using pvcs and pvc aware
> scheduling along with decommissioning. So far performance is good with this
> choice.
>
> Thank you
>
> On Tue, 9 Apr 2024, 15:17 Mich Talebzadeh, 
> wrote:
>
>> Hi,
>>
>> First thanks everyone for their contributions
>>
>> I was going to reply to @Enrico Minack   but
>> noticed additional info. As I understand for example,  Apache Uniffle is an
>> incubating project aimed at providing a pluggable shuffle service for
>> Spark. So basically, all these "external shuffle services" have in common
>> is to offload shuffle data management to external services, thus reducing
>> the memory and CPU overhead on Spark executors. That is great.  While
>> Uniffle and others enhance shuffle performance and scalability, it would be
>> great to integrate them with Spark UI. This may require additional
>> development efforts. I suppose  the interest would be to have these
>> external matrices incorporated into Spark with one look and feel. This may
>> require customizing the UI to fetch and display metrics or statistics from
>> the external shuffle services. Has any project done this?
>>
>> Thanks
>>
>> Mich Talebzadeh,
>> Technologist | Solutions Architect | Data Engineer  | Generative AI
>> London
>> United Kingdom
>>
>>
>>view my Linkedin profile
>> 
>>
>>
>>  https://en.everybodywiki.com/Mich_Talebzadeh
>>
>>
>>
>> *Disclaimer:* The information provided is correct to the best of my
>> knowledge but of course cannot be guaranteed . It is essential to note
>> that, as with any advice, quote "one test result is worth one-thousand
>> expert opinions (Werner
>> Von Braun
>> )".
>>
>>
>> On Mon, 8 Apr 2024 at 14:19, Vakaris Baškirov <
>> vakaris.bashki...@gmail.com> wrote:
>>
>>> I see that both Uniffle and Celebron support S3/HDFS backends which is
>>> great.
>>> In the case someone is using S3/HDFS, I wonder what would be the
>>> advantages of using Celebron or Uniffle vs IBM shuffle service plugin
>>>  or Cloud Shuffle Storage
>>> Plugin from AWS
>>> 
>>> ?
>>>
>>> These plugins do not require deploying a separate service. Are there any
>>> advantages to using Uniffle/Celebron in the case of using S3 backend, which
>>> would require deploying a separate service?
>>>
>>> Thanks
>>> Vakaris
>>>
>>> On Mon, Apr 8, 2024 at 10:03 AM roryqi  wrote:
>>>
 Apache Uniffle (incubating) may be another solution.
 You can see
 https://github.com/apache/incubator-uniffle

 https://uniffle.apache.org/blog/2023/07/21/Uniffle%20-%20New%20chapter%20for%20the%20shuffle%20in%20the%20cloud%20native%20era

 Mich Talebzadeh  于2024年4月8日周一 07:15写道：

> Splendid
>
> The configurations below can be used with k8s deployments of Spark.
> Spark applications running on k8s can utilize these configurations to
> seamlessly access data stored in Google Cloud Storage (GCS) and Amazon S3.
>
> For Google GCS we may have
>
> spark_config_gcs = {
> "spark.kubernetes.authenticate.driver.serviceAccountName":
> "service_account_name",
> "spark.hadoop.fs.gs.impl":
> "com.google.cloud.hadoop.fs.gcs.GoogleHadoopFileSystem",
> "spark.hadoop.google.cloud.auth.service.account.enable": "true",
> "spark.hadoop.google.cloud.auth.service.account.json.keyfile":
> "/path/to/keyfile.json",
> }
>
> For Amazon S3 similar
>
> spark_config_s3 = {
> "spark.kubernetes.authenticate.driver.serviceAccountName":
> "service_account_name",
> "spark.hadoop.fs.s3a.impl":
> "org.apache.hadoop.fs.s3a.S3AFileSystem",
> "spark.hadoop.fs.s3a.access.key": "s3_access_key",
> "spark.hadoop.fs.s3a.secret.key": "secret_key",
> }
>
>
> To implement these configurations and enable Spark applications to
> interact with GCS and S3, I 

Re: External Spark shuffle service for k8s

[Arun Ravi]

Hi Everyone,

I had to explored IBM's and AWS's S3 shuffle plugins (some time back), I
had also explored AWS FSX lustre in few of my production jobs which has
~20TB of shuffle operations with 200-300 executors. What I have observed is
S3 and fax behaviour was fine during the write phase, however I faced iops
throttling during the read phase(read taking forever to complete). I think
this might be contributed by the heavy use of shuffle index file (I didn't
perform any extensive research on this), so I believe the shuffle manager
logic have to be intelligent enough to reduce the fetching of files from
object store. In the end for my usecase I started using pvcs and pvc aware
scheduling along with decommissioning. So far performance is good with this
choice.

Thank you

On Tue, 9 Apr 2024, 15:17 Mich Talebzadeh, 
wrote:

> Hi,
>
> First thanks everyone for their contributions
>
> I was going to reply to @Enrico Minack   but
> noticed additional info. As I understand for example,  Apache Uniffle is an
> incubating project aimed at providing a pluggable shuffle service for
> Spark. So basically, all these "external shuffle services" have in common
> is to offload shuffle data management to external services, thus reducing
> the memory and CPU overhead on Spark executors. That is great.  While
> Uniffle and others enhance shuffle performance and scalability, it would be
> great to integrate them with Spark UI. This may require additional
> development efforts. I suppose  the interest would be to have these
> external matrices incorporated into Spark with one look and feel. This may
> require customizing the UI to fetch and display metrics or statistics from
> the external shuffle services. Has any project done this?
>
> Thanks
>
> Mich Talebzadeh,
> Technologist | Solutions Architect | Data Engineer  | Generative AI
> London
> United Kingdom
>
>
>view my Linkedin profile
> 
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> *Disclaimer:* The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner  Von
> Braun )".
>
>
> On Mon, 8 Apr 2024 at 14:19, Vakaris Baškirov 
> wrote:
>
>> I see that both Uniffle and Celebron support S3/HDFS backends which is
>> great.
>> In the case someone is using S3/HDFS, I wonder what would be the
>> advantages of using Celebron or Uniffle vs IBM shuffle service plugin
>>  or Cloud Shuffle Storage
>> Plugin from AWS
>> 
>> ?
>>
>> These plugins do not require deploying a separate service. Are there any
>> advantages to using Uniffle/Celebron in the case of using S3 backend, which
>> would require deploying a separate service?
>>
>> Thanks
>> Vakaris
>>
>> On Mon, Apr 8, 2024 at 10:03 AM roryqi  wrote:
>>
>>> Apache Uniffle (incubating) may be another solution.
>>> You can see
>>> https://github.com/apache/incubator-uniffle
>>>
>>> https://uniffle.apache.org/blog/2023/07/21/Uniffle%20-%20New%20chapter%20for%20the%20shuffle%20in%20the%20cloud%20native%20era
>>>
>>> Mich Talebzadeh  于2024年4月8日周一 07:15写道：
>>>
 Splendid

 The configurations below can be used with k8s deployments of Spark.
 Spark applications running on k8s can utilize these configurations to
 seamlessly access data stored in Google Cloud Storage (GCS) and Amazon S3.

 For Google GCS we may have

 spark_config_gcs = {
 "spark.kubernetes.authenticate.driver.serviceAccountName":
 "service_account_name",
 "spark.hadoop.fs.gs.impl":
 "com.google.cloud.hadoop.fs.gcs.GoogleHadoopFileSystem",
 "spark.hadoop.google.cloud.auth.service.account.enable": "true",
 "spark.hadoop.google.cloud.auth.service.account.json.keyfile":
 "/path/to/keyfile.json",
 }

 For Amazon S3 similar

 spark_config_s3 = {
 "spark.kubernetes.authenticate.driver.serviceAccountName":
 "service_account_name",
 "spark.hadoop.fs.s3a.impl":
 "org.apache.hadoop.fs.s3a.S3AFileSystem",
 "spark.hadoop.fs.s3a.access.key": "s3_access_key",
 "spark.hadoop.fs.s3a.secret.key": "secret_key",
 }


 To implement these configurations and enable Spark applications to
 interact with GCS and S3, I guess we can approach it this way

 1) Spark Repository Integration: These configurations need to be added
 to the Spark repository as part of the supported configuration options for
 k8s deployments.

 2) Configuration Settings: Users need to specify these configurations
 when submitting Spark applications to a Kubernetes cluster. They can
 include 

Re: Re: [Spark SQL] How can I use .sql() in conjunction with watermarks?

[Mich Talebzadeh]

interesting. So below should be the corrected code with the suggestion in
the [SPARK-47718] .sql() does not recognize watermark defined upstream -
ASF JIRA (apache.org) 

# Define schema for parsing Kafka messages
schema = StructType([
StructField('createTime', TimestampType(), True),
StructField('orderId', LongType(), True),
StructField('payAmount', DoubleType(), True),
StructField('payPlatform', IntegerType(), True),
StructField('provinceId', IntegerType(), True),
])

# Read streaming data from Kafka source
streaming_df = session.readStream \
.format("kafka") \
.option("kafka.bootstrap.servers", "localhost:9092") \
.option("subscribe", "payment_msg") \
.option("startingOffsets", "earliest") \
.load() \
.select(from_json(col("value").cast("string"),
schema).alias("parsed_value")) \
.select("parsed_value.*") \
.withWatermark("createTime", "10 seconds")

# Create temporary view for SQL queries
*streaming_df.createOrReplaceTempView("streaming_df")*
# Define SQL query with correct window function usage
query = """
*SELECT*
*window(start, '1 hour', '30 minutes') as window,*
provinceId,
sum(payAmount) as totalPayAmount
FROM streaming_df
GROUP BY provinceId, window(start, '1 hour', '30 minutes')
ORDER BY window.start
"""

# Write the aggregated results to Kafka sink
stream = session.sql(query) \
.writeStream \
.format("kafka") \
.option("checkpointLocation", "checkpoint") \
.option("kafka.bootstrap.servers", "localhost:9092") \
.option("topic", "sink") \
.start()


Mich Talebzadeh,
Technologist | Solutions Architect | Data Engineer  | Generative AI
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Tue, 9 Apr 2024 at 21:45, 刘唯  wrote:

> Sorry this is not a bug but essentially a user error. Spark throws a
> really confusing error and I'm also confused. Please see the reply in the
> ticket for how to make things correct.
> https://issues.apache.org/jira/browse/SPARK-47718
>
> 刘唯  于2024年4月6日周六 11:41写道：
>
>> This indeed looks like a bug. I will take some time to look into it.
>>
>> Mich Talebzadeh  于2024年4月3日周三 01:55写道：
>>
>>>
>>> hm. you are getting below
>>>
>>> AnalysisException: Append output mode not supported when there are
>>> streaming aggregations on streaming DataFrames/DataSets without watermark;
>>>
>>> The problem seems to be that you are using the append output mode when
>>> writing the streaming query results to Kafka. This mode is designed for
>>> scenarios where you want to append new data to an existing dataset at the
>>> sink (in this case, the "sink" topic in Kafka). However, your query
>>> involves a streaming aggregation: group by provinceId, window('createTime',
>>> '1 hour', '30 minutes'). The problem is that Spark Structured Streaming
>>> requires a watermark to ensure exactly-once processing when using
>>> aggregations with append mode. Your code already defines a watermark on the
>>> "createTime" column with a delay of 10 seconds (withWatermark("createTime",
>>> "10 seconds")). However, the error message indicates it is missing on the
>>> start column. Try adding watermark to "start" Column: Modify your code as
>>> below  to include a watermark on the "start" column generated by the
>>> window function:
>>>
>>> from pyspark.sql.functions import col, from_json, explode, window, sum,
>>> watermark
>>>
>>> streaming_df = session.readStream \
>>>   .format("kafka") \
>>>   .option("kafka.bootstrap.servers", "localhost:9092") \
>>>   .option("subscribe", "payment_msg") \
>>>   .option("startingOffsets", "earliest") \
>>>   .load() \
>>>   .select(from_json(col("value").cast("string"),
>>> schema).alias("parsed_value")) \
>>>   .select("parsed_value.*") \
>>>   .withWatermark("createTime", "10 seconds")  # Existing watermark on
>>> createTime
>>>
>>> *# Modified section with watermark on 'start' column*
>>> streaming_df = streaming_df.groupBy(
>>>   col("provinceId"),
>>>   window(col("createTime"), "1 hour", "30 minutes")
>>> ).agg(
>>>   sum(col("payAmount")).alias("totalPayAmount")
>>> ).withWatermark(expr("start"), "10 seconds")  # Watermark on
>>> window-generated 'start'
>>>
>>> # Rest of the code remains the same
>>> streaming_df.createOrReplaceTempView("streaming_df")
>>>
>>> spark.sql("""
>>> SELECT
>>>   window.start, window.end, provinceId, totalPayAmount
>>> FROM streaming_df
>>> ORDER BY window.start
>>> """) \
>>> .writeStream \
>>> .format("kafka") \
>>> .option("checkpointLocation", 

Re: Re: [Spark SQL] How can I use .sql() in conjunction with watermarks?

[刘唯]

Sorry this is not a bug but essentially a user error. Spark throws a really
confusing error and I'm also confused. Please see the reply in the ticket
for how to make things correct.
https://issues.apache.org/jira/browse/SPARK-47718

刘唯  于2024年4月6日周六 11:41写道：

> This indeed looks like a bug. I will take some time to look into it.
>
> Mich Talebzadeh  于2024年4月3日周三 01:55写道：
>
>>
>> hm. you are getting below
>>
>> AnalysisException: Append output mode not supported when there are
>> streaming aggregations on streaming DataFrames/DataSets without watermark;
>>
>> The problem seems to be that you are using the append output mode when
>> writing the streaming query results to Kafka. This mode is designed for
>> scenarios where you want to append new data to an existing dataset at the
>> sink (in this case, the "sink" topic in Kafka). However, your query
>> involves a streaming aggregation: group by provinceId, window('createTime',
>> '1 hour', '30 minutes'). The problem is that Spark Structured Streaming
>> requires a watermark to ensure exactly-once processing when using
>> aggregations with append mode. Your code already defines a watermark on the
>> "createTime" column with a delay of 10 seconds (withWatermark("createTime",
>> "10 seconds")). However, the error message indicates it is missing on the
>> start column. Try adding watermark to "start" Column: Modify your code as
>> below  to include a watermark on the "start" column generated by the
>> window function:
>>
>> from pyspark.sql.functions import col, from_json, explode, window, sum,
>> watermark
>>
>> streaming_df = session.readStream \
>>   .format("kafka") \
>>   .option("kafka.bootstrap.servers", "localhost:9092") \
>>   .option("subscribe", "payment_msg") \
>>   .option("startingOffsets", "earliest") \
>>   .load() \
>>   .select(from_json(col("value").cast("string"),
>> schema).alias("parsed_value")) \
>>   .select("parsed_value.*") \
>>   .withWatermark("createTime", "10 seconds")  # Existing watermark on
>> createTime
>>
>> *# Modified section with watermark on 'start' column*
>> streaming_df = streaming_df.groupBy(
>>   col("provinceId"),
>>   window(col("createTime"), "1 hour", "30 minutes")
>> ).agg(
>>   sum(col("payAmount")).alias("totalPayAmount")
>> ).withWatermark(expr("start"), "10 seconds")  # Watermark on
>> window-generated 'start'
>>
>> # Rest of the code remains the same
>> streaming_df.createOrReplaceTempView("streaming_df")
>>
>> spark.sql("""
>> SELECT
>>   window.start, window.end, provinceId, totalPayAmount
>> FROM streaming_df
>> ORDER BY window.start
>> """) \
>> .writeStream \
>> .format("kafka") \
>> .option("checkpointLocation", "checkpoint") \
>> .option("kafka.bootstrap.servers", "localhost:9092") \
>> .option("topic", "sink") \
>> .start()
>>
>> Try and see how it goes
>>
>> HTH
>>
>> Mich Talebzadeh,
>>
>> Technologist | Solutions Architect | Data Engineer  | Generative AI
>>
>> London
>> United Kingdom
>>
>>
>>view my Linkedin profile
>>
>>
>>  https://en.everybodywiki.com/Mich_Talebzadeh
>>
>>
>>
>> Disclaimer: The information provided is correct to the best of my
>> knowledge but of course cannot be guaranteed . It is essential to note
>> that, as with any advice, quote "one test result is worth one-thousand
>> expert opinions (Werner Von Braun)".
>>
>> Mich Talebzadeh,
>> Technologist | Solutions Architect | Data Engineer  | Generative AI
>> London
>> United Kingdom
>>
>>
>>view my Linkedin profile
>> 
>>
>>
>>  https://en.everybodywiki.com/Mich_Talebzadeh
>>
>>
>>
>> *Disclaimer:* The information provided is correct to the best of my
>> knowledge but of course cannot be guaranteed . It is essential to note
>> that, as with any advice, quote "one test result is worth one-thousand
>> expert opinions (Werner
>> Von Braun
>> )".
>>
>>
>> On Tue, 2 Apr 2024 at 22:43, Chloe He 
>> wrote:
>>
>>> Hi Mich,
>>>
>>> Thank you so much for your response. I really appreciate your help!
>>>
>>> You mentioned "defining the watermark using the withWatermark function
>>> on the streaming_df before creating the temporary view” - I believe this is
>>> what I’m doing and it’s not working for me. Here is the exact code snippet
>>> that I’m running:
>>>
>>> ```
>>> >>> streaming_df = session.readStream\
>>> .format("kafka")\
>>> .option("kafka.bootstrap.servers", "localhost:9092")\
>>> .option("subscribe", "payment_msg")\
>>> .option("startingOffsets","earliest")\
>>> .load()\
>>> .select(from_json(col("value").cast("string"),
>>> schema).alias("parsed_value"))\
>>> .select("parsed_value.*")\
>>> .withWatermark("createTime", "10 seconds")
>>>
>>> >>> streaming_df.createOrReplaceTempView("streaming_df”)
>>>
>>> >>> spark.sql("""
>>> SELECT
>>> window.start, window.end, provinceId, sum(payAmount) as
>>> totalPayAmount
>>> FROM 

Re: How to get db related metrics when use spark jdbc to read db table?

[Femi Anthony]

If you're using just Spark you could try turning on the history server
 and try to glean
statistics from there.  But there is no one location or log file which
stores them all.
Databricks, which is a managed Spark solution, provides such features in an
Enterprise setting.
I am unsure whether AWS EMR or Google Data Proc does the same.

Femi



On Mon, Apr 8, 2024 at 5:34 AM casel.chen  wrote:

> Hello, I have a spark application with jdbc source and do some
> calculation.
> To monitor application healthy, I need db related metrics per database
> like number of connections, sql execution time and sql fired time
> distribution etc.
> Does anybody know how to get them? Thanks!
>
>

-- 
http://dataphantik.com

"Great spirits have always encountered violent opposition from mediocre
minds." - Albert Einstein.

Re: [Spark SQL]: Source code for PartitionedFile

[Mich Talebzadeh]

Hi,

I believe this is the package

https://raw.githubusercontent.com/apache/spark/master/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/FilePartition.scala

And the code

case class FilePartition(index: Int, files: Array[PartitionedFile])
  extends Partition with InputPartition {
  override def preferredLocations(): Array[String] = {
// Computes total number of bytes that can be retrieved from each host.
val hostToNumBytes = mutable.HashMap.empty[String, Long]
files.foreach { file =>
  file.locations.filter(_ != "localhost").foreach { host =>
hostToNumBytes(host) = hostToNumBytes.getOrElse(host, 0L) +
file.length
  }
}

// Selects the first 3 hosts with the most data to be retrieved.
hostToNumBytes.toSeq.sortBy {
  case (host, numBytes) => numBytes
}.reverse.take(3).map {
  case (host, numBytes) => host
}.toArray
  }
}

HTH

Mich Talebzadeh,
Technologist | Solutions Architect | Data Engineer  | Generative AI
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Mon, 8 Apr 2024 at 20:31, Ashley McManamon <
ashley.mcmana...@quantcast.com> wrote:

> Hi All,
>
> I've been diving into the source code to get a better understanding of how
> file splitting works from a user perspective. I've hit a deadend at
> `PartitionedFile`, for which I cannot seem to find a definition? It appears
> though it should be found at
> org.apache.spark.sql.execution.datasources but I find no definition in the
> entire source code. Am I missing something?
>
> I appreciate there may be an obvious answer here, apologies if I'm being
> naive.
>
> Thanks,
> Ashley McManamon
>
>

Re: How to get db related metrics when use spark jdbc to read db table?

[Mich Talebzadeh]

Well you can do a fair bit with the available tools

The Spark UI, particularly the Staging and Executors tabs, do provide some
valuable insights related to database health metrics for applications using
a JDBC source.

Stage Overview:

This section provides a summary of all the stages executed during the
application's lifetime. It includes details such as the stage ID,
description, submission time, duration, and number of tasks.
Each Stage represents a set of tasks that perform the same computation,
typically applied to a partition of the input data. The Stages tab offers
insights into how these stages are executed and their associated metrics.
This tab may include a directed acyclic graph (DAG) visualization,
illustrating the logical and physical execution plan of the Spark
application.

Executors Tab:

The Executors tab provides detailed information about the executors running
in the Spark application. Executors are responsible for executing tasks on
behalf of the Spark application. The "Executors" tab offers insights into
the current state and resource usage of each executor.

In addition, the underlying database will have some instrumentation to
assist you with your work. say with Oracle (as an example), utilise tools
like OEM, VM StatPack, SQL*Plus scripts etc or third-party monitoring tools
to collect detailed database health metrics directly from the Oracle
database server.

HTH

Mich Talebzadeh,
Technologist | Solutions Architect | Data Engineer  | Generative AI
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Mon, 8 Apr 2024 at 19:35, casel.chen  wrote:

> Hello, I have a spark application with jdbc source and do some
> calculation.
> To monitor application healthy, I need db related metrics per database
> like number of connections, sql execution time and sql fired time
> distribution etc.
> Does anybody know how to get them? Thanks!
>
>

[Spark SQL]: Source code for PartitionedFile

[Ashley McManamon]

Hi All,

I've been diving into the source code to get a better understanding of how
file splitting works from a user perspective. I've hit a deadend at
`PartitionedFile`, for which I cannot seem to find a definition? It appears
though it should be found at
org.apache.spark.sql.execution.datasources but I find no definition in the
entire source code. Am I missing something?

I appreciate there may be an obvious answer here, apologies if I'm being
naive.

Thanks,
Ashley McManamon

Re: External Spark shuffle service for k8s

[Mich Talebzadeh]

Hi,

First thanks everyone for their contributions

I was going to reply to @Enrico Minack   but
noticed additional info. As I understand for example,  Apache Uniffle is an
incubating project aimed at providing a pluggable shuffle service for
Spark. So basically, all these "external shuffle services" have in common
is to offload shuffle data management to external services, thus reducing
the memory and CPU overhead on Spark executors. That is great.  While
Uniffle and others enhance shuffle performance and scalability, it would be
great to integrate them with Spark UI. This may require additional
development efforts. I suppose  the interest would be to have these
external matrices incorporated into Spark with one look and feel. This may
require customizing the UI to fetch and display metrics or statistics from
the external shuffle services. Has any project done this?

Thanks

Mich Talebzadeh,
Technologist | Solutions Architect | Data Engineer  | Generative AI
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Mon, 8 Apr 2024 at 14:19, Vakaris Baškirov 
wrote:

> I see that both Uniffle and Celebron support S3/HDFS backends which is
> great.
> In the case someone is using S3/HDFS, I wonder what would be the
> advantages of using Celebron or Uniffle vs IBM shuffle service plugin
>  or Cloud Shuffle Storage Plugin
> from AWS
> 
> ?
>
> These plugins do not require deploying a separate service. Are there any
> advantages to using Uniffle/Celebron in the case of using S3 backend, which
> would require deploying a separate service?
>
> Thanks
> Vakaris
>
> On Mon, Apr 8, 2024 at 10:03 AM roryqi  wrote:
>
>> Apache Uniffle (incubating) may be another solution.
>> You can see
>> https://github.com/apache/incubator-uniffle
>>
>> https://uniffle.apache.org/blog/2023/07/21/Uniffle%20-%20New%20chapter%20for%20the%20shuffle%20in%20the%20cloud%20native%20era
>>
>> Mich Talebzadeh  于2024年4月8日周一 07:15写道：
>>
>>> Splendid
>>>
>>> The configurations below can be used with k8s deployments of Spark.
>>> Spark applications running on k8s can utilize these configurations to
>>> seamlessly access data stored in Google Cloud Storage (GCS) and Amazon S3.
>>>
>>> For Google GCS we may have
>>>
>>> spark_config_gcs = {
>>> "spark.kubernetes.authenticate.driver.serviceAccountName":
>>> "service_account_name",
>>> "spark.hadoop.fs.gs.impl":
>>> "com.google.cloud.hadoop.fs.gcs.GoogleHadoopFileSystem",
>>> "spark.hadoop.google.cloud.auth.service.account.enable": "true",
>>> "spark.hadoop.google.cloud.auth.service.account.json.keyfile":
>>> "/path/to/keyfile.json",
>>> }
>>>
>>> For Amazon S3 similar
>>>
>>> spark_config_s3 = {
>>> "spark.kubernetes.authenticate.driver.serviceAccountName":
>>> "service_account_name",
>>> "spark.hadoop.fs.s3a.impl": "org.apache.hadoop.fs.s3a.S3AFileSystem",
>>> "spark.hadoop.fs.s3a.access.key": "s3_access_key",
>>> "spark.hadoop.fs.s3a.secret.key": "secret_key",
>>> }
>>>
>>>
>>> To implement these configurations and enable Spark applications to
>>> interact with GCS and S3, I guess we can approach it this way
>>>
>>> 1) Spark Repository Integration: These configurations need to be added
>>> to the Spark repository as part of the supported configuration options for
>>> k8s deployments.
>>>
>>> 2) Configuration Settings: Users need to specify these configurations
>>> when submitting Spark applications to a Kubernetes cluster. They can
>>> include these configurations in the Spark application code or pass them as
>>> command-line arguments or environment variables during application
>>> submission.
>>>
>>> HTH
>>>
>>> Mich Talebzadeh,
>>>
>>> Technologist | Solutions Architect | Data Engineer  | Generative AI
>>> London
>>> United Kingdom
>>>
>>>
>>>view my Linkedin profile
>>> 
>>>
>>>
>>>  https://en.everybodywiki.com/Mich_Talebzadeh
>>>
>>>
>>>
>>> *Disclaimer:* The information provided is correct to the best of my
>>> knowledge but of course cannot be guaranteed . It is essential to note
>>> that, as with any advice, quote "one test result is worth one-thousand
>>> expert opinions (Werner
>>> Von Braun
>>> )".
>>>
>>>
>>> On Sun, 7 Apr 2024 at 13:31, Vakaris Baškirov <
>>> vakaris.bashki...@gmail.com> wrote:
>>>
 There is an IBM shuffle 

Re: External Spark shuffle service for k8s

[Vakaris Baškirov]

I see that both Uniffle and Celebron support S3/HDFS backends which is
great.
In the case someone is using S3/HDFS, I wonder what would be the advantages
of using Celebron or Uniffle vs IBM shuffle service plugin
 or Cloud Shuffle Storage Plugin
from AWS

?

These plugins do not require deploying a separate service. Are there any
advantages to using Uniffle/Celebron in the case of using S3 backend, which
would require deploying a separate service?

Thanks
Vakaris

On Mon, Apr 8, 2024 at 10:03 AM roryqi  wrote:

> Apache Uniffle (incubating) may be another solution.
> You can see
> https://github.com/apache/incubator-uniffle
>
> https://uniffle.apache.org/blog/2023/07/21/Uniffle%20-%20New%20chapter%20for%20the%20shuffle%20in%20the%20cloud%20native%20era
>
> Mich Talebzadeh  于2024年4月8日周一 07:15写道：
>
>> Splendid
>>
>> The configurations below can be used with k8s deployments of Spark. Spark
>> applications running on k8s can utilize these configurations to seamlessly
>> access data stored in Google Cloud Storage (GCS) and Amazon S3.
>>
>> For Google GCS we may have
>>
>> spark_config_gcs = {
>> "spark.kubernetes.authenticate.driver.serviceAccountName":
>> "service_account_name",
>> "spark.hadoop.fs.gs.impl":
>> "com.google.cloud.hadoop.fs.gcs.GoogleHadoopFileSystem",
>> "spark.hadoop.google.cloud.auth.service.account.enable": "true",
>> "spark.hadoop.google.cloud.auth.service.account.json.keyfile":
>> "/path/to/keyfile.json",
>> }
>>
>> For Amazon S3 similar
>>
>> spark_config_s3 = {
>> "spark.kubernetes.authenticate.driver.serviceAccountName":
>> "service_account_name",
>> "spark.hadoop.fs.s3a.impl": "org.apache.hadoop.fs.s3a.S3AFileSystem",
>> "spark.hadoop.fs.s3a.access.key": "s3_access_key",
>> "spark.hadoop.fs.s3a.secret.key": "secret_key",
>> }
>>
>>
>> To implement these configurations and enable Spark applications to
>> interact with GCS and S3, I guess we can approach it this way
>>
>> 1) Spark Repository Integration: These configurations need to be added to
>> the Spark repository as part of the supported configuration options for k8s
>> deployments.
>>
>> 2) Configuration Settings: Users need to specify these configurations
>> when submitting Spark applications to a Kubernetes cluster. They can
>> include these configurations in the Spark application code or pass them as
>> command-line arguments or environment variables during application
>> submission.
>>
>> HTH
>>
>> Mich Talebzadeh,
>>
>> Technologist | Solutions Architect | Data Engineer  | Generative AI
>> London
>> United Kingdom
>>
>>
>>view my Linkedin profile
>> 
>>
>>
>>  https://en.everybodywiki.com/Mich_Talebzadeh
>>
>>
>>
>> *Disclaimer:* The information provided is correct to the best of my
>> knowledge but of course cannot be guaranteed . It is essential to note
>> that, as with any advice, quote "one test result is worth one-thousand
>> expert opinions (Werner
>> Von Braun
>> )".
>>
>>
>> On Sun, 7 Apr 2024 at 13:31, Vakaris Baškirov <
>> vakaris.bashki...@gmail.com> wrote:
>>
>>> There is an IBM shuffle service plugin that supports S3
>>> https://github.com/IBM/spark-s3-shuffle
>>>
>>> Though I would think a feature like this could be a part of the main
>>> Spark repo. Trino already has out-of-box support for s3 exchange (shuffle)
>>> and it's very useful.
>>>
>>> Vakaris
>>>
>>> On Sun, Apr 7, 2024 at 12:27 PM Mich Talebzadeh <
>>> mich.talebza...@gmail.com> wrote:
>>>

 Thanks for your suggestion that I take it as a workaround. Whilst this
 workaround can potentially address storage allocation issues, I was more
 interested in exploring solutions that offer a more seamless integration
 with large distributed file systems like HDFS, GCS, or S3. This would
 ensure better performance and scalability for handling larger datasets
 efficiently.


 Mich Talebzadeh,
 Technologist | Solutions Architect | Data Engineer  | Generative AI
 London
 United Kingdom


view my Linkedin profile
 


  https://en.everybodywiki.com/Mich_Talebzadeh



 *Disclaimer:* The information provided is correct to the best of my
 knowledge but of course cannot be guaranteed . It is essential to note
 that, as with any advice, quote "one test result is worth one-thousand
 expert opinions (Werner
 Von Braun
 )".


 On Sat, 6 Apr 2024 at 21:28, Bjørn Jørgensen 
 wrote:

> You can make a PVC on K8S call it 300GB
>
> make a folder in yours dockerfile
> 

How to get db related metrics when use spark jdbc to read db table?

[casel.chen]

Hello, I have a spark application with jdbc source and do some calculation. 
To monitor application healthy, I need db related metrics per database like 
number of connections, sql execution time and sql fired time distribution etc.
Does anybody know how to get them? Thanks!

Re: External Spark shuffle service for k8s

[roryqi]

Apache Uniffle (incubating) may be another solution.
You can see
https://github.com/apache/incubator-uniffle
https://uniffle.apache.org/blog/2023/07/21/Uniffle%20-%20New%20chapter%20for%20the%20shuffle%20in%20the%20cloud%20native%20era

Mich Talebzadeh  于2024年4月8日周一 07:15写道：

> Splendid
>
> The configurations below can be used with k8s deployments of Spark. Spark
> applications running on k8s can utilize these configurations to seamlessly
> access data stored in Google Cloud Storage (GCS) and Amazon S3.
>
> For Google GCS we may have
>
> spark_config_gcs = {
> "spark.kubernetes.authenticate.driver.serviceAccountName":
> "service_account_name",
> "spark.hadoop.fs.gs.impl":
> "com.google.cloud.hadoop.fs.gcs.GoogleHadoopFileSystem",
> "spark.hadoop.google.cloud.auth.service.account.enable": "true",
> "spark.hadoop.google.cloud.auth.service.account.json.keyfile":
> "/path/to/keyfile.json",
> }
>
> For Amazon S3 similar
>
> spark_config_s3 = {
> "spark.kubernetes.authenticate.driver.serviceAccountName":
> "service_account_name",
> "spark.hadoop.fs.s3a.impl": "org.apache.hadoop.fs.s3a.S3AFileSystem",
> "spark.hadoop.fs.s3a.access.key": "s3_access_key",
> "spark.hadoop.fs.s3a.secret.key": "secret_key",
> }
>
>
> To implement these configurations and enable Spark applications to
> interact with GCS and S3, I guess we can approach it this way
>
> 1) Spark Repository Integration: These configurations need to be added to
> the Spark repository as part of the supported configuration options for k8s
> deployments.
>
> 2) Configuration Settings: Users need to specify these configurations when
> submitting Spark applications to a Kubernetes cluster. They can include
> these configurations in the Spark application code or pass them as
> command-line arguments or environment variables during application
> submission.
>
> HTH
>
> Mich Talebzadeh,
>
> Technologist | Solutions Architect | Data Engineer  | Generative AI
> London
> United Kingdom
>
>
>view my Linkedin profile
> 
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> *Disclaimer:* The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner  Von
> Braun )".
>
>
> On Sun, 7 Apr 2024 at 13:31, Vakaris Baškirov 
> wrote:
>
>> There is an IBM shuffle service plugin that supports S3
>> https://github.com/IBM/spark-s3-shuffle
>>
>> Though I would think a feature like this could be a part of the main
>> Spark repo. Trino already has out-of-box support for s3 exchange (shuffle)
>> and it's very useful.
>>
>> Vakaris
>>
>> On Sun, Apr 7, 2024 at 12:27 PM Mich Talebzadeh <
>> mich.talebza...@gmail.com> wrote:
>>
>>>
>>> Thanks for your suggestion that I take it as a workaround. Whilst this
>>> workaround can potentially address storage allocation issues, I was more
>>> interested in exploring solutions that offer a more seamless integration
>>> with large distributed file systems like HDFS, GCS, or S3. This would
>>> ensure better performance and scalability for handling larger datasets
>>> efficiently.
>>>
>>>
>>> Mich Talebzadeh,
>>> Technologist | Solutions Architect | Data Engineer  | Generative AI
>>> London
>>> United Kingdom
>>>
>>>
>>>view my Linkedin profile
>>> 
>>>
>>>
>>>  https://en.everybodywiki.com/Mich_Talebzadeh
>>>
>>>
>>>
>>> *Disclaimer:* The information provided is correct to the best of my
>>> knowledge but of course cannot be guaranteed . It is essential to note
>>> that, as with any advice, quote "one test result is worth one-thousand
>>> expert opinions (Werner
>>> Von Braun
>>> )".
>>>
>>>
>>> On Sat, 6 Apr 2024 at 21:28, Bjørn Jørgensen 
>>> wrote:
>>>
 You can make a PVC on K8S call it 300GB

 make a folder in yours dockerfile
 WORKDIR /opt/spark/work-dir
 RUN chmod g+w /opt/spark/work-dir

 start spark with adding this

 .config("spark.kubernetes.driver.volumes.persistentVolumeClaim.300gb.options.claimName",
 "300gb") \

 .config("spark.kubernetes.driver.volumes.persistentVolumeClaim.300gb.mount.path",
 "/opt/spark/work-dir") \

 .config("spark.kubernetes.driver.volumes.persistentVolumeClaim.300gb.mount.readOnly",
 "False") \

 .config("spark.kubernetes.executor.volumes.persistentVolumeClaim.300gb.options.claimName",
 "300gb") \

 .config("spark.kubernetes.executor.volumes.persistentVolumeClaim.300gb.mount.path",
 "/opt/spark/work-dir") \

 .config("spark.kubernetes.executor.volumes.persistentVolumeClaim.300gb.mount.readOnly",
 

Re: External Spark shuffle service for k8s

[Enrico Minack]

There is Apache incubator project Uniffle:
https://github.com/apache/incubator-uniffle

It stores shuffle data on remote servers in memory, on local disk and HDFS.

Cheers,
Enrico


Am 06.04.24 um 15:41 schrieb Mich Talebzadeh:

I have seen some older references for shuffle service for k8s,
although it is not clear they are talking about a generic shuffle
service for k8s.

Anyhow with the advent of genai and the need to allow for a larger
volume of data, I was wondering if there has been any more work on
this matter. Specifically larger and scalable file systems like HDFS,
GCS , S3 etc, offer significantly larger storage capacity than local
disks on individual worker nodes in a k8s cluster, thus allowing
handling much larger datasets more efficiently. Also the degree of
parallelism and fault tolerance  with these files systems come into
it. I will be interested in hearing more about any progress on this.

Thanks
.

Mich Talebzadeh,

Technologist | Solutions Architect | Data Engineer  | Generative AI

London
United Kingdom


view my Linkedin profile


  https://en.everybodywiki.com/Mich_Talebzadeh



Disclaimer: The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner Von Braun)".

-
To unsubscribe e-mail: user-unsubscr...@spark.apache.org




-
To unsubscribe e-mail: user-unsubscr...@spark.apache.org

Spark UDAF in examples fail with not serializable error

[Owen Bell]

The type-safe example given at
https://spark.apache.org/docs/latest/sql-ref-functions-udf-aggregate.html
fails with a not serializable exception

Is this a known issue?

Re: Idiomatic way to rate-limit streaming sources to avoid OutOfMemoryError?

[Mich Talebzadeh]

OK,

This is a common issue in Spark Structured Streaming (SSS), where the
source generates data faster than Spark can process it. SSS doesn't have a
built-in mechanism for directly rate-limiting the incoming data stream
itself. However, consider the following:


   - Limit the rate at which data is produced. This can involve configuring
   the data source itself to emit data at a controlled rate or implementing
   rate limiting mechanisms in the application or system that produces the
   data.
   - SSS supports backpressure, which allows it to dynamically adjust the
   ingestion rate based on the processing capacity of the system. This can
   help prevent overwhelming the system with data. To enable backpressure, set
   the appropriate configuration properties such as
spark.conf.set("spark.streaming.backpressure.enabled",
   "true") and spark.streaming.backpressure.initialRate.
   - Consider adjusting the micro-batch interval to control the rate at
   which data is processed. Increasing the micro-batch interval and reduce the
   frequency of processing, allowing more time for each batch to be processed
   and reducing the likelihood of out-of-memory
   errors.. spark.conf.set("spark.sql.streaming.trigger.interval", "
   seconds"
   -  Dynamic Resource Allocation (DRA), Not implemented yet. DRA will
   automatically adjust allocated resources based on workload. This ensures
   Spark has enough resources to process incoming data within the trigger
   interval, preventing backlogs and potential OOM issues.


>From Spark UI, look at the streaming tab. There are various statistics
there. In general your Processing Time has to be less than your batch
interval. The scheduling Delay and Total Delay are additional indicator of
health.

HTH

Mich Talebzadeh,
Technologist | Solutions Architect | Data Engineer  | Generative AI
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Sun, 7 Apr 2024 at 15:11, Baran, Mert  wrote:

> Hi Spark community,
>
> I have a Spark Structured Streaming application that reads data from a
> socket source (implemented very similarly to the
> TextSocketMicroBatchStream). The issue is that the source can generate
> data faster than Spark can process it, eventually leading to an
> OutOfMemoryError when Spark runs out of memory trying to queue up all
> the pending data.
>
> I'm looking for advice on the most idiomatic/recommended way in Spark to
> rate-limit data ingestion to avoid overwhelming the system.
>
> Approaches I've considered:
>
> 1. Using a BlockingQueue with a fixed size to throttle the data.
> However, this requires careful tuning of the queue size. If too small,
> it limits throughput; if too large, you risk batches taking too long.
>
> 2. Fetching a limited number of records in the PartitionReader's next(),
> adding the records into a queue and checking if the queue is empty.
> However, I'm not sure if there is a built-in way to dynamically scale
> the number of records fetched (i.e., dynamically calculating the offset)
> based on the system load and capabilities.
>
> So in summary, what is the recommended way to dynamically rate-limit a
> streaming source to match Spark's processing capacity and avoid
> out-of-memory issues? Are there any best practices or configuration
> options I should look at?
> Any guidance would be much appreciated! Let me know if you need any
> other details.
>
> Thanks,
> Mert
>
>
> -
> To unsubscribe e-mail: user-unsubscr...@spark.apache.org
>
>

Re: External Spark shuffle service for k8s

[Mich Talebzadeh]

Thanks Cheng for the heads up. I will have a look.

Cheers

Mich Talebzadeh,
Technologist | Solutions Architect | Data Engineer  | Generative AI
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Sun, 7 Apr 2024 at 15:08, Cheng Pan  wrote:

> Instead of External Shuffle Shufle, Apache Celeborn might be a good option
> as a Remote Shuffle Service for Spark on K8s.
>
> There are some useful resources you might be interested in.
>
> [1] https://celeborn.apache.org/
> [2] https://www.youtube.com/watch?v=s5xOtG6Venw
> [3] https://github.com/aws-samples/emr-remote-shuffle-service
> [4] https://github.com/apache/celeborn/issues/2140
>
> Thanks,
> Cheng Pan
>
>
> > On Apr 6, 2024, at 21:41, Mich Talebzadeh 
> wrote:
> >
> > I have seen some older references for shuffle service for k8s,
> > although it is not clear they are talking about a generic shuffle
> > service for k8s.
> >
> > Anyhow with the advent of genai and the need to allow for a larger
> > volume of data, I was wondering if there has been any more work on
> > this matter. Specifically larger and scalable file systems like HDFS,
> > GCS , S3 etc, offer significantly larger storage capacity than local
> > disks on individual worker nodes in a k8s cluster, thus allowing
> > handling much larger datasets more efficiently. Also the degree of
> > parallelism and fault tolerance  with these files systems come into
> > it. I will be interested in hearing more about any progress on this.
> >
> > Thanks
> > .
> >
> > Mich Talebzadeh,
> >
> > Technologist | Solutions Architect | Data Engineer  | Generative AI
> >
> > London
> > United Kingdom
> >
> >
> >   view my Linkedin profile
> >
> >
> > https://en.everybodywiki.com/Mich_Talebzadeh
> >
> >
> >
> > Disclaimer: The information provided is correct to the best of my
> > knowledge but of course cannot be guaranteed . It is essential to note
> > that, as with any advice, quote "one test result is worth one-thousand
> > expert opinions (Werner Von Braun)".
> >
> > -
> > To unsubscribe e-mail: dev-unsubscr...@spark.apache.org
> >
>
>

Re: External Spark shuffle service for k8s

[Vakaris Baškirov]

There is an IBM shuffle service plugin that supports S3
https://github.com/IBM/spark-s3-shuffle

Though I would think a feature like this could be a part of the main Spark
repo. Trino already has out-of-box support for s3 exchange (shuffle) and
it's very useful.

Vakaris

On Sun, Apr 7, 2024 at 12:27 PM Mich Talebzadeh 
wrote:

>
> Thanks for your suggestion that I take it as a workaround. Whilst this
> workaround can potentially address storage allocation issues, I was more
> interested in exploring solutions that offer a more seamless integration
> with large distributed file systems like HDFS, GCS, or S3. This would
> ensure better performance and scalability for handling larger datasets
> efficiently.
>
>
> Mich Talebzadeh,
> Technologist | Solutions Architect | Data Engineer  | Generative AI
> London
> United Kingdom
>
>
>view my Linkedin profile
> 
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> *Disclaimer:* The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner  Von
> Braun )".
>
>
> On Sat, 6 Apr 2024 at 21:28, Bjørn Jørgensen 
> wrote:
>
>> You can make a PVC on K8S call it 300GB
>>
>> make a folder in yours dockerfile
>> WORKDIR /opt/spark/work-dir
>> RUN chmod g+w /opt/spark/work-dir
>>
>> start spark with adding this
>>
>> .config("spark.kubernetes.driver.volumes.persistentVolumeClaim.300gb.options.claimName",
>> "300gb") \
>>
>> .config("spark.kubernetes.driver.volumes.persistentVolumeClaim.300gb.mount.path",
>> "/opt/spark/work-dir") \
>>
>> .config("spark.kubernetes.driver.volumes.persistentVolumeClaim.300gb.mount.readOnly",
>> "False") \
>>
>> .config("spark.kubernetes.executor.volumes.persistentVolumeClaim.300gb.options.claimName",
>> "300gb") \
>>
>> .config("spark.kubernetes.executor.volumes.persistentVolumeClaim.300gb.mount.path",
>> "/opt/spark/work-dir") \
>>
>> .config("spark.kubernetes.executor.volumes.persistentVolumeClaim.300gb.mount.readOnly",
>> "False") \
>>   .config("spark.local.dir", "/opt/spark/work-dir")
>>
>>
>>
>>
>> lør. 6. apr. 2024 kl. 15:45 skrev Mich Talebzadeh <
>> mich.talebza...@gmail.com>:
>>
>>> I have seen some older references for shuffle service for k8s,
>>> although it is not clear they are talking about a generic shuffle
>>> service for k8s.
>>>
>>> Anyhow with the advent of genai and the need to allow for a larger
>>> volume of data, I was wondering if there has been any more work on
>>> this matter. Specifically larger and scalable file systems like HDFS,
>>> GCS , S3 etc, offer significantly larger storage capacity than local
>>> disks on individual worker nodes in a k8s cluster, thus allowing
>>> handling much larger datasets more efficiently. Also the degree of
>>> parallelism and fault tolerance  with these files systems come into
>>> it. I will be interested in hearing more about any progress on this.
>>>
>>> Thanks
>>> .
>>>
>>> Mich Talebzadeh,
>>>
>>> Technologist | Solutions Architect | Data Engineer  | Generative AI
>>>
>>> London
>>> United Kingdom
>>>
>>>
>>>view my Linkedin profile
>>>
>>>
>>>  https://en.everybodywiki.com/Mich_Talebzadeh
>>>
>>>
>>>
>>> Disclaimer: The information provided is correct to the best of my
>>> knowledge but of course cannot be guaranteed . It is essential to note
>>> that, as with any advice, quote "one test result is worth one-thousand
>>> expert opinions (Werner Von Braun)".
>>>
>>> -
>>> To unsubscribe e-mail: user-unsubscr...@spark.apache.org
>>>
>>>
>>
>> --
>> Bjørn Jørgensen
>> Vestre Aspehaug 4, 6010 Ålesund
>> Norge
>>
>> +47 480 94 297
>>
>

Idiomatic way to rate-limit streaming sources to avoid OutOfMemoryError?

[Baran, Mert]

Hi Spark community,

I have a Spark Structured Streaming application that reads data from a 
socket source (implemented very similarly to the 
TextSocketMicroBatchStream). The issue is that the source can generate 
data faster than Spark can process it, eventually leading to an 
OutOfMemoryError when Spark runs out of memory trying to queue up all 
the pending data.


I'm looking for advice on the most idiomatic/recommended way in Spark to 
rate-limit data ingestion to avoid overwhelming the system.


Approaches I've considered:

1. Using a BlockingQueue with a fixed size to throttle the data. 
However, this requires careful tuning of the queue size. If too small, 
it limits throughput; if too large, you risk batches taking too long.


2. Fetching a limited number of records in the PartitionReader's next(), 
adding the records into a queue and checking if the queue is empty. 
However, I'm not sure if there is a built-in way to dynamically scale 
the number of records fetched (i.e., dynamically calculating the offset) 
based on the system load and capabilities.


So in summary, what is the recommended way to dynamically rate-limit a 
streaming source to match Spark's processing capacity and avoid 
out-of-memory issues? Are there any best practices or configuration 
options I should look at?
Any guidance would be much appreciated! Let me know if you need any 
other details.


Thanks,
Mert


-
To unsubscribe e-mail: user-unsubscr...@spark.apache.org

Re: External Spark shuffle service for k8s

[Cheng Pan]

Instead of External Shuffle Shufle, Apache Celeborn might be a good option as a 
Remote Shuffle Service for Spark on K8s.

There are some useful resources you might be interested in.

[1] https://celeborn.apache.org/
[2] https://www.youtube.com/watch?v=s5xOtG6Venw
[3] https://github.com/aws-samples/emr-remote-shuffle-service
[4] https://github.com/apache/celeborn/issues/2140

Thanks,
Cheng Pan


> On Apr 6, 2024, at 21:41, Mich Talebzadeh  wrote:
> 
> I have seen some older references for shuffle service for k8s,
> although it is not clear they are talking about a generic shuffle
> service for k8s.
> 
> Anyhow with the advent of genai and the need to allow for a larger
> volume of data, I was wondering if there has been any more work on
> this matter. Specifically larger and scalable file systems like HDFS,
> GCS , S3 etc, offer significantly larger storage capacity than local
> disks on individual worker nodes in a k8s cluster, thus allowing
> handling much larger datasets more efficiently. Also the degree of
> parallelism and fault tolerance  with these files systems come into
> it. I will be interested in hearing more about any progress on this.
> 
> Thanks
> .
> 
> Mich Talebzadeh,
> 
> Technologist | Solutions Architect | Data Engineer  | Generative AI
> 
> London
> United Kingdom
> 
> 
>   view my Linkedin profile
> 
> 
> https://en.everybodywiki.com/Mich_Talebzadeh
> 
> 
> 
> Disclaimer: The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner Von Braun)".
> 
> -
> To unsubscribe e-mail: dev-unsubscr...@spark.apache.org
> 


-
To unsubscribe e-mail: user-unsubscr...@spark.apache.org

Re: External Spark shuffle service for k8s

[Mich Talebzadeh]

Splendid

The configurations below can be used with k8s deployments of Spark. Spark
applications running on k8s can utilize these configurations to seamlessly
access data stored in Google Cloud Storage (GCS) and Amazon S3.

For Google GCS we may have

spark_config_gcs = {
"spark.kubernetes.authenticate.driver.serviceAccountName":
"service_account_name",
"spark.hadoop.fs.gs.impl":
"com.google.cloud.hadoop.fs.gcs.GoogleHadoopFileSystem",
"spark.hadoop.google.cloud.auth.service.account.enable": "true",
"spark.hadoop.google.cloud.auth.service.account.json.keyfile":
"/path/to/keyfile.json",
}

For Amazon S3 similar

spark_config_s3 = {
"spark.kubernetes.authenticate.driver.serviceAccountName":
"service_account_name",
"spark.hadoop.fs.s3a.impl": "org.apache.hadoop.fs.s3a.S3AFileSystem",
"spark.hadoop.fs.s3a.access.key": "s3_access_key",
"spark.hadoop.fs.s3a.secret.key": "secret_key",
}


To implement these configurations and enable Spark applications to interact
with GCS and S3, I guess we can approach it this way

1) Spark Repository Integration: These configurations need to be added to
the Spark repository as part of the supported configuration options for k8s
deployments.

2) Configuration Settings: Users need to specify these configurations when
submitting Spark applications to a Kubernetes cluster. They can include
these configurations in the Spark application code or pass them as
command-line arguments or environment variables during application
submission.

HTH

Mich Talebzadeh,

Technologist | Solutions Architect | Data Engineer  | Generative AI
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Sun, 7 Apr 2024 at 13:31, Vakaris Baškirov 
wrote:

> There is an IBM shuffle service plugin that supports S3
> https://github.com/IBM/spark-s3-shuffle
>
> Though I would think a feature like this could be a part of the main Spark
> repo. Trino already has out-of-box support for s3 exchange (shuffle) and
> it's very useful.
>
> Vakaris
>
> On Sun, Apr 7, 2024 at 12:27 PM Mich Talebzadeh 
> wrote:
>
>>
>> Thanks for your suggestion that I take it as a workaround. Whilst this
>> workaround can potentially address storage allocation issues, I was more
>> interested in exploring solutions that offer a more seamless integration
>> with large distributed file systems like HDFS, GCS, or S3. This would
>> ensure better performance and scalability for handling larger datasets
>> efficiently.
>>
>>
>> Mich Talebzadeh,
>> Technologist | Solutions Architect | Data Engineer  | Generative AI
>> London
>> United Kingdom
>>
>>
>>view my Linkedin profile
>> 
>>
>>
>>  https://en.everybodywiki.com/Mich_Talebzadeh
>>
>>
>>
>> *Disclaimer:* The information provided is correct to the best of my
>> knowledge but of course cannot be guaranteed . It is essential to note
>> that, as with any advice, quote "one test result is worth one-thousand
>> expert opinions (Werner
>> Von Braun
>> )".
>>
>>
>> On Sat, 6 Apr 2024 at 21:28, Bjørn Jørgensen 
>> wrote:
>>
>>> You can make a PVC on K8S call it 300GB
>>>
>>> make a folder in yours dockerfile
>>> WORKDIR /opt/spark/work-dir
>>> RUN chmod g+w /opt/spark/work-dir
>>>
>>> start spark with adding this
>>>
>>> .config("spark.kubernetes.driver.volumes.persistentVolumeClaim.300gb.options.claimName",
>>> "300gb") \
>>>
>>> .config("spark.kubernetes.driver.volumes.persistentVolumeClaim.300gb.mount.path",
>>> "/opt/spark/work-dir") \
>>>
>>> .config("spark.kubernetes.driver.volumes.persistentVolumeClaim.300gb.mount.readOnly",
>>> "False") \
>>>
>>> .config("spark.kubernetes.executor.volumes.persistentVolumeClaim.300gb.options.claimName",
>>> "300gb") \
>>>
>>> .config("spark.kubernetes.executor.volumes.persistentVolumeClaim.300gb.mount.path",
>>> "/opt/spark/work-dir") \
>>>
>>> .config("spark.kubernetes.executor.volumes.persistentVolumeClaim.300gb.mount.readOnly",
>>> "False") \
>>>   .config("spark.local.dir", "/opt/spark/work-dir")
>>>
>>>
>>>
>>>
>>> lør. 6. apr. 2024 kl. 15:45 skrev Mich Talebzadeh <
>>> mich.talebza...@gmail.com>:
>>>
 I have seen some older references for shuffle service for k8s,
 although it is not clear they are talking about a generic shuffle
 service for k8s.

 Anyhow with the advent of genai and the need to allow for a larger
 volume of data, I was wondering if there has been any more work on
 this 

Example UDAF fails with "not serializable" exception

[Owen Bell]

https://spark.apache.org/docs/3.3.2/sql-ref-functions-udf-aggregate.html

I'm trying to run this example on Databricks, and it fails with the stacktrace 
below.  It's literally a copy-paste from the example, what am I missing?



Job aborted due to stage failure: Task not serializable: 
java.io.NotSerializableException: org.apache.spark.sql.TypedColumn 
Serialization stack: - object not serializable (class: 
org.apache.spark.sql.TypedColumn, value: 
myaverage(knownnotnull(assertnotnull(input[0, 
$line867a2458f6df413bb3920f7fc2dbb8c381.$read$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$Average,
 true])).sum AS sum, knownnotnull(assertnotnull(input[0, 
$line867a2458f6df413bb3920f7fc2dbb8c381.$read$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$Average,
 true])).count AS count, newInstance(class 
$line867a2458f6df413bb3920f7fc2dbb8c381.$read$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$$iw$Average),
 boundreference()) AS average_salary)

Re: External Spark shuffle service for k8s

[Mich Talebzadeh]

Thanks for your suggestion that I take it as a workaround. Whilst this
workaround can potentially address storage allocation issues, I was more
interested in exploring solutions that offer a more seamless integration
with large distributed file systems like HDFS, GCS, or S3. This would
ensure better performance and scalability for handling larger datasets
efficiently.


Mich Talebzadeh,
Technologist | Solutions Architect | Data Engineer  | Generative AI
London
United Kingdom


   view my Linkedin profile



 https://en.everybodywiki.com/Mich_Talebzadeh



*Disclaimer:* The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner  Von
Braun )".


On Sat, 6 Apr 2024 at 21:28, Bjørn Jørgensen 
wrote:

> You can make a PVC on K8S call it 300GB
>
> make a folder in yours dockerfile
> WORKDIR /opt/spark/work-dir
> RUN chmod g+w /opt/spark/work-dir
>
> start spark with adding this
>
> .config("spark.kubernetes.driver.volumes.persistentVolumeClaim.300gb.options.claimName",
> "300gb") \
>
> .config("spark.kubernetes.driver.volumes.persistentVolumeClaim.300gb.mount.path",
> "/opt/spark/work-dir") \
>
> .config("spark.kubernetes.driver.volumes.persistentVolumeClaim.300gb.mount.readOnly",
> "False") \
>
> .config("spark.kubernetes.executor.volumes.persistentVolumeClaim.300gb.options.claimName",
> "300gb") \
>
> .config("spark.kubernetes.executor.volumes.persistentVolumeClaim.300gb.mount.path",
> "/opt/spark/work-dir") \
>
> .config("spark.kubernetes.executor.volumes.persistentVolumeClaim.300gb.mount.readOnly",
> "False") \
>   .config("spark.local.dir", "/opt/spark/work-dir")
>
>
>
>
> lør. 6. apr. 2024 kl. 15:45 skrev Mich Talebzadeh <
> mich.talebza...@gmail.com>:
>
>> I have seen some older references for shuffle service for k8s,
>> although it is not clear they are talking about a generic shuffle
>> service for k8s.
>>
>> Anyhow with the advent of genai and the need to allow for a larger
>> volume of data, I was wondering if there has been any more work on
>> this matter. Specifically larger and scalable file systems like HDFS,
>> GCS , S3 etc, offer significantly larger storage capacity than local
>> disks on individual worker nodes in a k8s cluster, thus allowing
>> handling much larger datasets more efficiently. Also the degree of
>> parallelism and fault tolerance  with these files systems come into
>> it. I will be interested in hearing more about any progress on this.
>>
>> Thanks
>> .
>>
>> Mich Talebzadeh,
>>
>> Technologist | Solutions Architect | Data Engineer  | Generative AI
>>
>> London
>> United Kingdom
>>
>>
>>view my Linkedin profile
>>
>>
>>  https://en.everybodywiki.com/Mich_Talebzadeh
>>
>>
>>
>> Disclaimer: The information provided is correct to the best of my
>> knowledge but of course cannot be guaranteed . It is essential to note
>> that, as with any advice, quote "one test result is worth one-thousand
>> expert opinions (Werner Von Braun)".
>>
>> -
>> To unsubscribe e-mail: user-unsubscr...@spark.apache.org
>>
>>
>
> --
> Bjørn Jørgensen
> Vestre Aspehaug 4, 6010 Ålesund
> Norge
>
> +47 480 94 297
>

Re: External Spark shuffle service for k8s

[Bjørn Jørgensen]

You can make a PVC on K8S call it 300GB

make a folder in yours dockerfile
WORKDIR /opt/spark/work-dir
RUN chmod g+w /opt/spark/work-dir

start spark with adding this

.config("spark.kubernetes.driver.volumes.persistentVolumeClaim.300gb.options.claimName",
"300gb") \

.config("spark.kubernetes.driver.volumes.persistentVolumeClaim.300gb.mount.path",
"/opt/spark/work-dir") \

.config("spark.kubernetes.driver.volumes.persistentVolumeClaim.300gb.mount.readOnly",
"False") \

.config("spark.kubernetes.executor.volumes.persistentVolumeClaim.300gb.options.claimName",
"300gb") \

.config("spark.kubernetes.executor.volumes.persistentVolumeClaim.300gb.mount.path",
"/opt/spark/work-dir") \

.config("spark.kubernetes.executor.volumes.persistentVolumeClaim.300gb.mount.readOnly",
"False") \
  .config("spark.local.dir", "/opt/spark/work-dir")




lør. 6. apr. 2024 kl. 15:45 skrev Mich Talebzadeh :

> I have seen some older references for shuffle service for k8s,
> although it is not clear they are talking about a generic shuffle
> service for k8s.
>
> Anyhow with the advent of genai and the need to allow for a larger
> volume of data, I was wondering if there has been any more work on
> this matter. Specifically larger and scalable file systems like HDFS,
> GCS , S3 etc, offer significantly larger storage capacity than local
> disks on individual worker nodes in a k8s cluster, thus allowing
> handling much larger datasets more efficiently. Also the degree of
> parallelism and fault tolerance  with these files systems come into
> it. I will be interested in hearing more about any progress on this.
>
> Thanks
> .
>
> Mich Talebzadeh,
>
> Technologist | Solutions Architect | Data Engineer  | Generative AI
>
> London
> United Kingdom
>
>
>view my Linkedin profile
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> Disclaimer: The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner Von Braun)".
>
> -
> To unsubscribe e-mail: user-unsubscr...@spark.apache.org
>
>

-- 
Bjørn Jørgensen
Vestre Aspehaug 4, 6010 Ålesund
Norge

+47 480 94 297

Re: Re: [Spark SQL] How can I use .sql() in conjunction with watermarks?

[刘唯]

This indeed looks like a bug. I will take some time to look into it.

Mich Talebzadeh  于2024年4月3日周三 01:55写道：

>
> hm. you are getting below
>
> AnalysisException: Append output mode not supported when there are
> streaming aggregations on streaming DataFrames/DataSets without watermark;
>
> The problem seems to be that you are using the append output mode when
> writing the streaming query results to Kafka. This mode is designed for
> scenarios where you want to append new data to an existing dataset at the
> sink (in this case, the "sink" topic in Kafka). However, your query
> involves a streaming aggregation: group by provinceId, window('createTime',
> '1 hour', '30 minutes'). The problem is that Spark Structured Streaming
> requires a watermark to ensure exactly-once processing when using
> aggregations with append mode. Your code already defines a watermark on the
> "createTime" column with a delay of 10 seconds (withWatermark("createTime",
> "10 seconds")). However, the error message indicates it is missing on the
> start column. Try adding watermark to "start" Column: Modify your code as
> below  to include a watermark on the "start" column generated by the
> window function:
>
> from pyspark.sql.functions import col, from_json, explode, window, sum,
> watermark
>
> streaming_df = session.readStream \
>   .format("kafka") \
>   .option("kafka.bootstrap.servers", "localhost:9092") \
>   .option("subscribe", "payment_msg") \
>   .option("startingOffsets", "earliest") \
>   .load() \
>   .select(from_json(col("value").cast("string"),
> schema).alias("parsed_value")) \
>   .select("parsed_value.*") \
>   .withWatermark("createTime", "10 seconds")  # Existing watermark on
> createTime
>
> *# Modified section with watermark on 'start' column*
> streaming_df = streaming_df.groupBy(
>   col("provinceId"),
>   window(col("createTime"), "1 hour", "30 minutes")
> ).agg(
>   sum(col("payAmount")).alias("totalPayAmount")
> ).withWatermark(expr("start"), "10 seconds")  # Watermark on
> window-generated 'start'
>
> # Rest of the code remains the same
> streaming_df.createOrReplaceTempView("streaming_df")
>
> spark.sql("""
> SELECT
>   window.start, window.end, provinceId, totalPayAmount
> FROM streaming_df
> ORDER BY window.start
> """) \
> .writeStream \
> .format("kafka") \
> .option("checkpointLocation", "checkpoint") \
> .option("kafka.bootstrap.servers", "localhost:9092") \
> .option("topic", "sink") \
> .start()
>
> Try and see how it goes
>
> HTH
>
> Mich Talebzadeh,
>
> Technologist | Solutions Architect | Data Engineer  | Generative AI
>
> London
> United Kingdom
>
>
>view my Linkedin profile
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> Disclaimer: The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner Von Braun)".
>
> Mich Talebzadeh,
> Technologist | Solutions Architect | Data Engineer  | Generative AI
> London
> United Kingdom
>
>
>view my Linkedin profile
> 
>
>
>  https://en.everybodywiki.com/Mich_Talebzadeh
>
>
>
> *Disclaimer:* The information provided is correct to the best of my
> knowledge but of course cannot be guaranteed . It is essential to note
> that, as with any advice, quote "one test result is worth one-thousand
> expert opinions (Werner  Von
> Braun )".
>
>
> On Tue, 2 Apr 2024 at 22:43, Chloe He 
> wrote:
>
>> Hi Mich,
>>
>> Thank you so much for your response. I really appreciate your help!
>>
>> You mentioned "defining the watermark using the withWatermark function on
>> the streaming_df before creating the temporary view” - I believe this is
>> what I’m doing and it’s not working for me. Here is the exact code snippet
>> that I’m running:
>>
>> ```
>> >>> streaming_df = session.readStream\
>> .format("kafka")\
>> .option("kafka.bootstrap.servers", "localhost:9092")\
>> .option("subscribe", "payment_msg")\
>> .option("startingOffsets","earliest")\
>> .load()\
>> .select(from_json(col("value").cast("string"),
>> schema).alias("parsed_value"))\
>> .select("parsed_value.*")\
>> .withWatermark("createTime", "10 seconds")
>>
>> >>> streaming_df.createOrReplaceTempView("streaming_df”)
>>
>> >>> spark.sql("""
>> SELECT
>> window.start, window.end, provinceId, sum(payAmount) as totalPayAmount
>> FROM streaming_df
>> GROUP BY provinceId, window('createTime', '1 hour', '30 minutes')
>> ORDER BY window.start
>> """)\
>>   .withWatermark("start", "10 seconds")\
>>   .writeStream\
>>   .format("kafka") \
>>   .option("checkpointLocation", "checkpoint") \
>>   .option("kafka.bootstrap.servers", "localhost:9092") \
>>   .option("topic", "sink") \
>>   .start()
>>
>> AnalysisException: Append output mode not 

Re: [External] Re: Issue of spark with antlr version

[Bjørn Jørgensen]

[[VOTE] Release Plan for Apache Spark 4.0.0 (June 2024)](
https://lists.apache.org/thread/r0zn6rd8y25yn2dg59ktw3ttrwxzqrfb)

Apache Spark 4.0.0 Release Plan
===

1. After creating `branch-3.5`, set "4.0.0-SNAPSHOT" in master branch.

2. Creating `branch-4.0` on April 1st, 2024.

3. Apache Spark 4.0.0 RC1 on May 1st, 2024.

4. Apache Spark 4.0.0 Release in June, 2024.

tir. 2. apr. 2024 kl. 12:06 skrev Chawla, Parul :

> ++ Ashima
>
> --
> *From:* Chawla, Parul 
> *Sent:* Tuesday, April 2, 2024 9:56 AM
> *To:* Bjørn Jørgensen ; user@spark.apache.org <
> user@spark.apache.org>
> *Cc:* Sahni, Ashima ;
> user@spark.apache.org ; Misra Parashar, Jyoti <
> jyoti.misra.paras...@accenture.com>
> *Subject:* Re: [External] Re: Issue of spark with antlr version
>
> Hi Team,
> Any update on below query :when spark 4.x will be released to maven as on
> maven site i could see spark core 3.5.1 .
>
> Regards,
> Parul
>
> --
> *From:* Chawla, Parul 
> *Sent:* Monday, April 1, 2024 4:20 PM
> *To:* Bjørn Jørgensen 
> *Cc:* Sahni, Ashima ;
> user@spark.apache.org ; Misra Parashar, Jyoti <
> jyoti.misra.paras...@accenture.com>; Mekala, Rajesh <
> r.mek...@accenture.com>; Grandhi, Venkatesh <
> venkatesh.a.gran...@accenture.com>; George, Rejish <
> rejish.geo...@accenture.com>; Tayal, Aayushi 
> *Subject:* Re: [External] Re: Issue of spark with antlr version
>
> Hi Team,
>
> Can you let us know the when   this spark 4.x will be released to maven.
>
> regards,
> Parul
>
> Get Outlook for iOS 
> --
> *From:* Bjørn Jørgensen 
> *Sent:* Wednesday, February 28, 2024 5:06:54 PM
> *To:* Chawla, Parul 
> *Cc:* Sahni, Ashima ;
> user@spark.apache.org ; Misra Parashar, Jyoti <
> jyoti.misra.paras...@accenture.com>; Mekala, Rajesh <
> r.mek...@accenture.com>; Grandhi, Venkatesh <
> venkatesh.a.gran...@accenture.com>; George, Rejish <
> rejish.geo...@accenture.com>; Tayal, Aayushi 
> *Subject:* Re: [External] Re: Issue of spark with antlr version
>
> [image: image.png]
>
> ons. 28. feb. 2024 kl. 11:28 skrev Chawla, Parul <
> parul.cha...@accenture.com>:
>
>
> Hi ,
> Can we get spark version on whuich this is resolved.
> --
> *From:* Bjørn Jørgensen 
> *Sent:* Tuesday, February 27, 2024 7:05:36 PM
> *To:* Sahni, Ashima 
> *Cc:* Chawla, Parul ; user@spark.apache.org <
> user@spark.apache.org>; Misra Parashar, Jyoti <
> jyoti.misra.paras...@accenture.com>; Mekala, Rajesh <
> r.mek...@accenture.com>; Grandhi, Venkatesh <
> venkatesh.a.gran...@accenture.com>; George, Rejish <
> rejish.geo...@accenture.com>; Tayal, Aayushi 
> *Subject:* [External] Re: Issue of spark with antlr version
>
> *CAUTION:* External email. Be cautious with links and attachments.
> [SPARK-44366][BUILD] Upgrade antlr4 to 4.13.1
> 
>
>
> tir. 27. feb. 2024 kl. 13:25 skrev Sahni, Ashima
> :
>
> Hi Team,
>
>
>
> Can you please let us know the update on below.
>
>
>
> Thanks,
>
> Ashima
>
>
>
> *From:* Chawla, Parul 
> *Sent:* Sunday, February 25, 2024 11:57 PM
> *To:* user@spark.apache.org
> *Cc:* Sahni, Ashima ; Misra Parashar, Jyoti <
> jyoti.misra.paras...@accenture.com>
> *Subject:* Issue of spark with antlr version
>
>
>
> Hi Spark Team,
>
>
>
>
>
> Our application is currently using spring framrwork 5.3.31 .To upgrade it
> to 6.x , as per application dependency we must upgrade Spark and
> Hibernate jars as well .
>
> With Hibernate compatible upgrade, the dependent Antlr4 jar version has
> been upgraded to 4.10.1 but there’s no Spark version available with the
> upgraded Antlr4 jar.
>
> Can u please update when we can have updated version with upgraded antl4
> version..
>
>
>
>
>
> Regards,
>
> Parul
>
> --
>
> This message is for the designated recipient only and may contain
> privileged, proprietary, or otherwise confidential information. If you have
> received it in error, please notify the sender immediately and delete the
> original. Any other use of the e-mail by you is prohibited. Where allowed
> by local law, electronic communications with Accenture and its affiliates,
> including e-mail and instant messaging (including content), may be scanned
> by our systems for the purposes of information security, AI-powered support
> capabilities, and assessment of internal compliance with Accenture policy.
> Your privacy is important to us. Accenture uses your personal data only in
> compliance with data protection laws. For further information on how
> Accenture processes your personal data, please see our privacy statement at
> https://www.accenture.com/us-en/privacy-policy.
>
> __
>
> www.accenture.com
>
>
>
> --
> 

Unsubscribe

[rau-jannik]

Unsubscribe

-
To unsubscribe e-mail: user-unsubscr...@spark.apache.org

External Spark shuffle service for k8s

[Mich Talebzadeh]

I have seen some older references for shuffle service for k8s,
although it is not clear they are talking about a generic shuffle
service for k8s.

Anyhow with the advent of genai and the need to allow for a larger
volume of data, I was wondering if there has been any more work on
this matter. Specifically larger and scalable file systems like HDFS,
GCS , S3 etc, offer significantly larger storage capacity than local
disks on individual worker nodes in a k8s cluster, thus allowing
handling much larger datasets more efficiently. Also the degree of
parallelism and fault tolerance  with these files systems come into
it. I will be interested in hearing more about any progress on this.

Thanks
.

Mich Talebzadeh,

Technologist | Solutions Architect | Data Engineer  | Generative AI

London
United Kingdom


   view my Linkedin profile


 https://en.everybodywiki.com/Mich_Talebzadeh



Disclaimer: The information provided is correct to the best of my
knowledge but of course cannot be guaranteed . It is essential to note
that, as with any advice, quote "one test result is worth one-thousand
expert opinions (Werner Von Braun)".

-
To unsubscribe e-mail: user-unsubscr...@spark.apache.org

Clarification on what "[id=#]" refers to in Physical Plan Exchange hashpartitioning

[Tahj Anderson]

Hello,

While looking through spark physical plans generated by the spark history 
server log to find any bottle necks in my code, I stumbled across an ID that 
shows up in a partitioning stage.
My goal is to use the history server log to provide meaningful analysis on my 
spark system performance. With this goal in mind, I am trying to connect spark 
physical plans to StageIDs which house useful information that I can tie back 
to my code. Below is a snippet from one of the physical plans.
+- *(2) Sort [Column#46 ASC NULLS FIRST], true, 0
+- Exchange hashpartitioning(ColumnId#329, 200), ENSURE_REQUIREMENTS, 
[id=#278]


What exactly does [id=#278] refer to?
I have seen some examples that say this ID is a reference to a specific 
partition, a stage id, or a plan_id but I have not been able to confirm which 
one it is.

Thank you

Clarification on what "[id=#]" refers to in Physical Plan Exchange hashpartitioning

[Tahj Anderson]

Hello,

While looking through spark physical plans generated by the spark history 
server log to find any bottle necks in my code, I stumbled across an ID that 
shows up in a partitioning stage.
My goal is to use the history server log to provide meaningful analysis on my 
spark system performance. With this goal in mind, I am trying to connect spark 
physical plans to StageIDs which house useful information that I can tie back 
to my code. Below is a snippet from one of the physical plans.
+- *(2) Sort [Column#46 ASC NULLS FIRST], true, 0
+- Exchange hashpartitioning(ColumnId#329, 200), ENSURE_REQUIREMENTS, 
[id=#278]


What exactly does [id=#278] refer to?
I have seen some examples that say this ID is a reference to a specific 
partition, a stage id, or a plan_id but I have not been able to confirm which 
one it is.

Thank you,
Tahj

Re: [EXTERNAL] Re: [Spark]: Spark / Iceberg / hadoop-aws compatibility matrix

[Oxlade, Dan]

I don't really understand how Iceberg and the hadoop libraries can coexist in a 
deployment.

The latest spark (3.5.1) base image contains the hadoop-client*-3.3.4.jar. The 
AWS v2 SDK is only supported in hadoop*-3.4.0.jar and onward.
Iceberg AWS integration states AWS v2 SDK is 
required

Does anyone have a working combination of pyspark, iceberg and hadoop? Or, is 
there an alternative way to use pyspark to 
spark.read.parquet("s3a:///.parquet") such that I don't need the 
hadoop dependencies?

Kind regards,
Dan

From: Oxlade, Dan 
Sent: 03 April 2024 15:49
To: Oxlade, Dan ; Aaron Grubb 
; user@spark.apache.org 
Subject: Re: [EXTERNAL] Re: [Spark]: Spark / Iceberg / hadoop-aws compatibility 
matrix

Swapping out the iceberg-aws-bundle for the very latest aws provided sdk 
('software.amazon.awssdk:bundle:2.25.23') produces an incompatibility from a 
slightly different code path:

java.lang.NoSuchMethodError: 'void 
org.apache.hadoop.util.SemaphoredDelegatingExecutor.(java.util.concurrent.ExecutorService,
 int, boolean, org.apache.hadoop.fs.statistics.DurationTrackerFactory)'
at org.apache.hadoop.fs.s3a.S3AFileSystem.executeOpen(S3AFileSystem.java 
[s3afilesystem.java]:1767)
at org.apache.hadoop.fs.s3a.S3AFileSystem.open(S3AFileSystem.java 
[s3afilesystem.java]:1717)
at org.apache.hadoop.fs.FileSystem.open(FileSystem.java 
[filesystem.java]:976)
at 
org.apache.parquet.hadoop.util.HadoopInputFile.newStream(HadoopInputFile.java 
[hadoopinputfile.java]:69)
at org.apache.parquet.hadoop.ParquetFileReader.(ParquetFileReader.java 
[parquetfilereader.java]:774)
at org.apache.parquet.hadoop.ParquetFileReader.open(ParquetFileReader.java 
[parquetfilereader.java]:658)
at 
org.apache.spark.sql.execution.datasources.parquet.ParquetFooterReader.readFooter(ParquetFooterReader.java
 
[parquetfooterreader.java]:53)
at 
org.apache.spark.sql.execution.datasources.parquet.ParquetFooterReader.readFooter(ParquetFooterReader.java
 
[parquetfooterreader.java]:44)
at 
org.apache.spark.sql.execution.datasources.parquet.ParquetFileFormat$.$anonfun$readParquetFootersInParallel$1(ParquetFileFormat.scala:429)




From: Oxlade, Dan 
Sent: 03 April 2024 14:33
To: Aaron Grubb ; user@spark.apache.org 
Subject: Re: [EXTERNAL] Re: [Spark]: Spark / Iceberg / hadoop-aws compatibility 
matrix


[sorry; replying all this time]

With hadoop-*-3.3.6 in place of the 3.4.0 below I get 
java.lang.NoClassDefFoundError: com/amazonaws/AmazonClientException

I think that the below iceberg-aws-bundle version supplies the v2 sdk.

Dan


From: Aaron Grubb 
Sent: 03 April 2024 13:52
To: user@spark.apache.org 
Subject: [EXTERNAL] Re: [Spark]: Spark / Iceberg / hadoop-aws compatibility 
matrix

Downgrade to hadoop-*:3.3.x, Hadoop 3.4.x is based on the AWS SDK v2 and should 
probably be considered as breaking for tools that build on < 3.4.0 while using 
AWS.

From: Oxlade, Dan 
Sent: Wednesday, April 3, 2024 2:41:11 PM
To: user@spark.apache.org 
Subject: [Spark]: Spark / Iceberg / hadoop-aws compatibility matrix


Hi all,



I’ve struggled with this for quite some time.

My requirement is to read a parquet file from s3 to a Dataframe then append to 
an existing iceberg table.



In order to read the parquet I need the hadoop-aws dependency for s3a:// . In 
order to write to iceberg I need the iceberg dependency. Both of these 
dependencies have a transitive dependency on the aws SDK. I can’t find versions 
for Spark 3.4 that work 

Re: [EXTERNAL] Re: [Spark]: Spark / Iceberg / hadoop-aws compatibility matrix

[Oxlade, Dan]

Swapping out the iceberg-aws-bundle for the very latest aws provided sdk 
('software.amazon.awssdk:bundle:2.25.23') produces an incompatibility from a 
slightly different code path:

java.lang.NoSuchMethodError: 'void 
org.apache.hadoop.util.SemaphoredDelegatingExecutor.(java.util.concurrent.ExecutorService,
 int, boolean, org.apache.hadoop.fs.statistics.DurationTrackerFactory)'
at org.apache.hadoop.fs.s3a.S3AFileSystem.executeOpen(S3AFileSystem.java:1767)
at org.apache.hadoop.fs.s3a.S3AFileSystem.open(S3AFileSystem.java:1717)
at org.apache.hadoop.fs.FileSystem.open(FileSystem.java:976)
at 
org.apache.parquet.hadoop.util.HadoopInputFile.newStream(HadoopInputFile.java:69)
at 
org.apache.parquet.hadoop.ParquetFileReader.(ParquetFileReader.java:774)
at org.apache.parquet.hadoop.ParquetFileReader.open(ParquetFileReader.java:658)
at 
org.apache.spark.sql.execution.datasources.parquet.ParquetFooterReader.readFooter(ParquetFooterReader.java:53)
at 
org.apache.spark.sql.execution.datasources.parquet.ParquetFooterReader.readFooter(ParquetFooterReader.java:44)
at 
org.apache.spark.sql.execution.datasources.parquet.ParquetFileFormat$.$anonfun$readParquetFootersInParallel$1(ParquetFileFormat.scala:429)




From: Oxlade, Dan 
Sent: 03 April 2024 14:33
To: Aaron Grubb ; user@spark.apache.org 
Subject: Re: [EXTERNAL] Re: [Spark]: Spark / Iceberg / hadoop-aws compatibility 
matrix


[sorry; replying all this time]

With hadoop-*-3.3.6 in place of the 3.4.0 below I get 
java.lang.NoClassDefFoundError: com/amazonaws/AmazonClientException

I think that the below iceberg-aws-bundle version supplies the v2 sdk.

Dan


From: Aaron Grubb 
Sent: 03 April 2024 13:52
To: user@spark.apache.org 
Subject: [EXTERNAL] Re: [Spark]: Spark / Iceberg / hadoop-aws compatibility 
matrix

Downgrade to hadoop-*:3.3.x, Hadoop 3.4.x is based on the AWS SDK v2 and should 
probably be considered as breaking for tools that build on < 3.4.0 while using 
AWS.

From: Oxlade, Dan 
Sent: Wednesday, April 3, 2024 2:41:11 PM
To: user@spark.apache.org 
Subject: [Spark]: Spark / Iceberg / hadoop-aws compatibility matrix


Hi all,



I’ve struggled with this for quite some time.

My requirement is to read a parquet file from s3 to a Dataframe then append to 
an existing iceberg table.



In order to read the parquet I need the hadoop-aws dependency for s3a:// . In 
order to write to iceberg I need the iceberg dependency. Both of these 
dependencies have a transitive dependency on the aws SDK. I can’t find versions 
for Spark 3.4 that work together.





Current Versions:

Spark 3.4.1

iceberg-spark-runtime-3.4-2.12:1.4.1

iceberg-aws-bundle:1.4.1

hadoop-aws:3.4.0

hadoop-common:3.4.0



I’ve tried a number of combinations of the above and their respective versions 
but all fall over with their assumptions on the aws sdk version with class not 
found exceptions or method not found etc.



Is there a compatibility matrix somewhere that someone could point me to?



Thanks

Dan

T. Rowe Price International Ltd (registered number 3957748) is registered in 
England and Wales with its registered office at Warwick Court, 5 Paternoster 
Square, London EC4M 7DX. T. Rowe Price International Ltd is authorised and 
regulated by the Financial Conduct Authority. The company has a branch in Dubai 
International Financial Centre (regulated by the DFSA as a Representative 
Office).

T. Rowe Price (including T. Rowe Price International Ltd and its affiliates) 
and its associates do not provide legal or tax advice. Any tax-related 
discussion contained in this e-mail, including any attachments, is not intended 
or written to be used, and cannot be used, for the purpose of (i) avoiding any 
tax penalties or (ii) promoting, marketing, or recommending to any other party 
any transaction or matter addressed herein. Please consult your independent 
legal counsel and/or professional tax advisor regarding any legal or tax issues 
raised in this e-mail.

The contents of this e-mail and any attachments are intended solely for the use 
of the named addressee(s) and may contain confidential and/or privileged 
information. Any unauthorized use, copying, disclosure, or distribution of the 
contents of this e-mail is strictly prohibited by the sender and may be 
unlawful. If you are not the intended recipient, please notify the sender 
immediately and delete this e-mail.

T. Rowe Price International Ltd (registered number 3957748) is registered in 
England and Wales with its registered office at Warwick Court, 5 Paternoster 
Square, London EC4M 7DX. T. Rowe Price International Ltd is authorised and 
regulated by the Financial Conduct Authority. The company has a branch in Dubai 
International Financial Centre (regulated by the DFSA as a Representative 
Office).

T. Rowe Price (including T. Rowe Price International Ltd and its affiliates) 
and its associates do not provide legal or tax advice. 

Participate in the ASF 25th Anniversary Campaign

[Brian Proffitt]

Hi everyone,

As part of The ASF’s 25th anniversary campaign[1], we will be celebrating
projects and communities in multiple ways.

We invite all projects and contributors to participate in the following
ways:

* Individuals - submit your first contribution:
https://news.apache.org/foundation/entry/the-asf-launches-firstasfcontribution-campaign
* Projects - share your public good story:
https://docs.google.com/forms/d/1vuN-tUnBwpTgOE5xj3Z5AG1hsOoDNLBmGIqQHwQT6k8/viewform?edit_requested=true
* Projects - submit a project spotlight for the blog:
https://cwiki.apache.org/confluence/pages/viewpage.action?pageId=278466116
* Projects - contact the Voice of Apache podcast (formerly Feathercast) to
be featured: https://feathercast.apache.org/help/
*  Projects - use the 25th anniversary template and the #ASF25Years hashtag
on social media:
https://docs.google.com/presentation/d/1oDbMol3F_XQuCmttPYxBIOIjRuRBksUjDApjd8Ve3L8/edit#slide=id.g26b0919956e_0_13

If you have questions, email the Marketing & Publicity team at
mark...@apache.org.

Peace,
BKP

[1] https://apache.org/asf25years/

[NOTE: You are receiving this message because you are a contributor to an
Apache Software Foundation project. The ASF will very occasionally send out
messages relating to the Foundation to contributors and members, such as
this one.]

Brian Proffitt
VP, Marketing & Publicity
VP, Conferences

Re: [EXTERNAL] Re: [Spark]: Spark / Iceberg / hadoop-aws compatibility matrix

[Oxlade, Dan]

[sorry; replying all this time]

With hadoop-*-3.3.6 in place of the 3.4.0 below I get 
java.lang.NoClassDefFoundError: com/amazonaws/AmazonClientException

I think that the below iceberg-aws-bundle version supplies the v2 sdk.

Dan


From: Aaron Grubb 
Sent: 03 April 2024 13:52
To: user@spark.apache.org 
Subject: [EXTERNAL] Re: [Spark]: Spark / Iceberg / hadoop-aws compatibility 
matrix

Downgrade to hadoop-*:3.3.x, Hadoop 3.4.x is based on the AWS SDK v2 and should 
probably be considered as breaking for tools that build on < 3.4.0 while using 
AWS.

From: Oxlade, Dan 
Sent: Wednesday, April 3, 2024 2:41:11 PM
To: user@spark.apache.org 
Subject: [Spark]: Spark / Iceberg / hadoop-aws compatibility matrix


Hi all,



I’ve struggled with this for quite some time.

My requirement is to read a parquet file from s3 to a Dataframe then append to 
an existing iceberg table.



In order to read the parquet I need the hadoop-aws dependency for s3a:// . In 
order to write to iceberg I need the iceberg dependency. Both of these 
dependencies have a transitive dependency on the aws SDK. I can’t find versions 
for Spark 3.4 that work together.





Current Versions:

Spark 3.4.1

iceberg-spark-runtime-3.4-2.12:1.4.1

iceberg-aws-bundle:1.4.1

hadoop-aws:3.4.0

hadoop-common:3.4.0



I’ve tried a number of combinations of the above and their respective versions 
but all fall over with their assumptions on the aws sdk version with class not 
found exceptions or method not found etc.



Is there a compatibility matrix somewhere that someone could point me to?



Thanks

Dan

T. Rowe Price International Ltd (registered number 3957748) is registered in 
England and Wales with its registered office at Warwick Court, 5 Paternoster 
Square, London EC4M 7DX. T. Rowe Price International Ltd is authorised and 
regulated by the Financial Conduct Authority. The company has a branch in Dubai 
International Financial Centre (regulated by the DFSA as a Representative 
Office).

T. Rowe Price (including T. Rowe Price International Ltd and its affiliates) 
and its associates do not provide legal or tax advice. Any tax-related 
discussion contained in this e-mail, including any attachments, is not intended 
or written to be used, and cannot be used, for the purpose of (i) avoiding any 
tax penalties or (ii) promoting, marketing, or recommending to any other party 
any transaction or matter addressed herein. Please consult your independent 
legal counsel and/or professional tax advisor regarding any legal or tax issues 
raised in this e-mail.

The contents of this e-mail and any attachments are intended solely for the use 
of the named addressee(s) and may contain confidential and/or privileged 
information. Any unauthorized use, copying, disclosure, or distribution of the 
contents of this e-mail is strictly prohibited by the sender and may be 
unlawful. If you are not the intended recipient, please notify the sender 
immediately and delete this e-mail.
T. Rowe Price International Ltd (registered number 3957748) is registered in 
England and Wales with its registered office at Warwick Court, 5 Paternoster 
Square, London EC4M 7DX. T. Rowe Price International Ltd is authorised and 
regulated by the Financial Conduct Authority. The company has a branch in Dubai 
International Financial Centre (regulated by the DFSA as a Representative 
Office).

T. Rowe Price (including T. Rowe Price International Ltd and its affiliates) 
and its associates do not provide legal or tax advice. Any tax-related 
discussion contained in this e-mail, including any attachments, is not intended 
or written to be used, and cannot be used, for the purpose of (i) avoiding any 
tax penalties or (ii) promoting, marketing, or recommending to any other party 
any transaction or matter addressed herein. Please consult your independent 
legal counsel and/or professional tax advisor regarding any legal or tax issues 
raised in this e-mail.

The contents of this e-mail and any attachments are intended solely for the use 
of the named addressee(s) and may contain confidential and/or privileged 
information. Any unauthorized use, copying, disclosure, or distribution of the 
contents of this e-mail is strictly prohibited by the sender and may be 
unlawful. If you are not the intended recipient, please notify the sender 
immediately and delete this e-mail.

Re: [Spark]: Spark / Iceberg / hadoop-aws compatibility matrix

[Aaron Grubb]

Downgrade to hadoop-*:3.3.x, Hadoop 3.4.x is based on the AWS SDK v2 and should 
probably be considered as breaking for tools that build on < 3.4.0 while using 
AWS.

From: Oxlade, Dan 
Sent: Wednesday, April 3, 2024 2:41:11 PM
To: user@spark.apache.org 
Subject: [Spark]: Spark / Iceberg / hadoop-aws compatibility matrix


Hi all,



I’ve struggled with this for quite some time.

My requirement is to read a parquet file from s3 to a Dataframe then append to 
an existing iceberg table.



In order to read the parquet I need the hadoop-aws dependency for s3a:// . In 
order to write to iceberg I need the iceberg dependency. Both of these 
dependencies have a transitive dependency on the aws SDK. I can’t find versions 
for Spark 3.4 that work together.





Current Versions:

Spark 3.4.1

iceberg-spark-runtime-3.4-2.12:1.4.1

iceberg-aws-bundle:1.4.1

hadoop-aws:3.4.0

hadoop-common:3.4.0



I’ve tried a number of combinations of the above and their respective versions 
but all fall over with their assumptions on the aws sdk version with class not 
found exceptions or method not found etc.



Is there a compatibility matrix somewhere that someone could point me to?



Thanks

Dan

T. Rowe Price International Ltd (registered number 3957748) is registered in 
England and Wales with its registered office at Warwick Court, 5 Paternoster 
Square, London EC4M 7DX. T. Rowe Price International Ltd is authorised and 
regulated by the Financial Conduct Authority. The company has a branch in Dubai 
International Financial Centre (regulated by the DFSA as a Representative 
Office).

T. Rowe Price (including T. Rowe Price International Ltd and its affiliates) 
and its associates do not provide legal or tax advice. Any tax-related 
discussion contained in this e-mail, including any attachments, is not intended 
or written to be used, and cannot be used, for the purpose of (i) avoiding any 
tax penalties or (ii) promoting, marketing, or recommending to any other party 
any transaction or matter addressed herein. Please consult your independent 
legal counsel and/or professional tax advisor regarding any legal or tax issues 
raised in this e-mail.

The contents of this e-mail and any attachments are intended solely for the use 
of the named addressee(s) and may contain confidential and/or privileged 
information. Any unauthorized use, copying, disclosure, or distribution of the 
contents of this e-mail is strictly prohibited by the sender and may be 
unlawful. If you are not the intended recipient, please notify the sender 
immediately and delete this e-mail.

[Spark]: Spark / Iceberg / hadoop-aws compatibility matrix

[Oxlade, Dan]

Hi all,

I've struggled with this for quite some time.
My requirement is to read a parquet file from s3 to a Dataframe then append to 
an existing iceberg table.

In order to read the parquet I need the hadoop-aws dependency for s3a:// . In 
order to write to iceberg I need the iceberg dependency. Both of these 
dependencies have a transitive dependency on the aws SDK. I can't find versions 
for Spark 3.4 that work together.


Current Versions:
Spark 3.4.1
iceberg-spark-runtime-3.4-2.12:1.4.1
iceberg-aws-bundle:1.4.1
hadoop-aws:3.4.0
hadoop-common:3.4.0

I've tried a number of combinations of the above and their respective versions 
but all fall over with their assumptions on the aws sdk version with class not 
found exceptions or method not found etc.

Is there a compatibility matrix somewhere that someone could point me to?

Thanks
Dan
T. Rowe Price International Ltd (registered number 3957748) is registered in 
England and Wales with its registered office at Warwick Court, 5 Paternoster 
Square, London EC4M 7DX. T. Rowe Price International Ltd is authorised and 
regulated by the Financial Conduct Authority. The company has a branch in Dubai 
International Financial Centre (regulated by the DFSA as a Representative 
Office).

T. Rowe Price (including T. Rowe Price International Ltd and its affiliates) 
and its associates do not provide legal or tax advice. Any tax-related 
discussion contained in this e-mail, including any attachments, is not intended 
or written to be used, and cannot be used, for the purpose of (i) avoiding any 
tax penalties or (ii) promoting, marketing, or recommending to any other party 
any transaction or matter addressed herein. Please consult your independent 
legal counsel and/or professional tax advisor regarding any legal or tax issues 
raised in this e-mail.

The contents of this e-mail and any attachments are intended solely for the use 
of the named addressee(s) and may contain confidential and/or privileged 
information. Any unauthorized use, copying, disclosure, or distribution of the 
contents of this e-mail is strictly prohibited by the sender and may be 
unlawful. If you are not the intended recipient, please notify the sender 
immediately and delete this e-mail.