+1
Looking forward to the functionality for a long time. With this feature,
our internal users can backfill
data with the same code and minimal change of configuration.
On Fri, Sep 18, 2020 at 3:08 AM Kostas Kloudas wrote:
> +1
>
> My only suggestion (although by no means a blocker) would be to
Aljoscha,
> Instead the sink would have to check for each set of committables
seperately if they had already been committed. Do you think this is
feasible?
Yes, that is how it works in our internal implementation [1]. We don't use
checkpointId. We generate a manifest file (GlobalCommT) to bundle
Hi Yang,
I saw this FLIP, it is very good feature, I think overall for Kubernetes, it is
preferred over “StatefulSet + PV + FileSystemHAService” approach, when it will
be available we plan to use it. On other hand looks like FileSystemHAService is
easier to implement, I thought about contributin
1. With `FSStateBackend`, we used to decide where to store the checkpoint
by the `state.backend.fs.memory-threshold` configuration, and we need to
decide how to align with this behavior with the new implementation.
I see this configuration available on the FileSystemStorage class. I've
added that
Having the watermark lag metric was the important part from my side.
So +1 to go ahead.
On Fri, Sep 18, 2020 at 4:11 PM Becket Qin wrote:
> Hey Stephan,
>
> Thanks for the quick reply. I actually forgot to mention that I also added
> a "watermarkLag" metric to the Source metrics in addition to
Hello,
I would like to propose an enrichment of existing Flink SQL MATCH_RECOGNIZE
syntax to cover for the case of the absence of an event. Such an enrichment
would help our company solve a business case containing timed-out patterns
handling. An example of usage of such a clause from Flink tra
+1(belated) for the latest FLIP document.
Thanks for addressing my belated questions/comments and updating the doc
accordingly, Dawid!
Best Regards,
Yu
On Thu, 17 Sep 2020 at 21:31, Yu Li wrote:
> I've just added some comments in the discussion thread [1]. Thanks.
>
> Best Regards,
> Yu
>
> [
*bq. I agree with your assessment of the CheckpointStorage interface but I
want to push back at including those changes as a part of this FLIP.*
Makes sense, will start a separate discussion around this topic when
prepared.
*bq. One option could be to rename "CheckpointStorage" to
"CheckpointStora
Timo Walther created FLINK-19294:
Summary: Support key and value formats in Kafka connector
Key: FLINK-19294
URL: https://issues.apache.org/jira/browse/FLINK-19294
Project: Flink
Issue Type:
Hi Jark,
the fieldNames map is not intended for users. I would also be fine to
make it a default scope constructor and access it with some internal
utility class next to the Row class. The fieldNames map must only be
used by serializers and converters. A user has no benefit in using it.
For
Hey Stephan,
Thanks for the quick reply. I actually forgot to mention that I also added
a "watermarkLag" metric to the Source metrics in addition to the
"currentFetchEventTimeLag" and "currentEmitEventTimeLag".
So in the current proposal, both XXEventTimeLag will be based on event
timestamps, whil
> Ok, got your point now. I agree that it makes more sense to
> make StateBackend return a contract instead of a particular
> implementation. How about we name the new interface as
> `CheckpointableKeyedStateBackend`? We could make
> `BoundedStreamStateBackend` implement
> `CheckpointableKeyedState
*bq. The problem is that I could not use this "state backend" in a
StreamOperator.*
Ok, got your point now. I agree that it makes more sense to
make StateBackend return a contract instead of a particular implementation.
How about we name the new interface as `CheckpointableKeyedStateBackend`?
We co
Thomas Wozniakowski created FLINK-19293:
---
Summary: RocksDB last_checkpoint.state_size grows endlessly until
savepoint/restore
Key: FLINK-19293
URL: https://issues.apache.org/jira/browse/FLINK-19293
Rui Li created FLINK-19292:
--
Summary: HiveCatalog should support specifying Hadoop conf dir
with configuration
Key: FLINK-19292
URL: https://issues.apache.org/jira/browse/FLINK-19292
Project: Flink
> ===
> /class BoundedStreamInternalStateBackend implements
> KeyedStateBackend,
> SnapshotStrategy>,
> Closeable,
> CheckpointListener {/
> ===/
> /
The problem is that I could n
Thanks for the clarification Dawid. Some of my thoughts:
*bq. The results are times for end-to-end execution of a job. Therefore the
sorting part is included. The actual target of the replacement is RocksDB,
which does the serialization and key bytes comparison as well.*
I see. Checking the FLIP m
+1 (binding)
Best,
Aljoscha
On 12.09.20 17:20, Alexey Trenikhun wrote:
We use union state to generate sequences, each operator generates offset0 +
number-of-tasks - task-index + task-specific-counter * number-of-tasks (e.g.
for 2 instances of operator -one instance produce even number, another odd).
Last generated seque
+1, thanks for driving through the FLIP, Xuannan!
Cheers,
Jiangjie (Becket) Qin
On Thu, Sep 17, 2020 at 3:23 PM Timo Walther wrote:
> +1 (binding)
>
> Looking forward to review the pull requests for this valuable feature.
>
> Regards,
> Timo
>
>
> On 16.09.20 07:35, Aljoscha Krettek wrote:
> >
Hi Becket!
I am wondering if it makes sense to do the following small change:
- Have "currentFetchEventTimeLag" be defined on event timestamps
(optionally, if the source system exposes it) <== this is like in your
proposal
this helps understand how long the records were in the source bef
Congratulations! Thanks for being the release manager Zhu Zhu and everyone
involved in!
Best,
Zhijiang
--
From:Lijie Wang
Send Time:2020年9月18日(星期五) 17:48
To:dev@flink.apache.org
Subject:Re:[ANNOUNCE] Apache Flink 1.11.2 released
Congrats Godfrey! Well deserved!
On Fri, Sep 18, 2020 at 5:12 PM Yun Gao
wrote:
> Congratulations Godfrey!
>
> Best,
> Yun
>
>
>
> --Original Mail --
> Sender:Dawid Wysakowicz
> Send Date:Thu Sep 17 14:45:55 2020
> Recipients:Flink Dev , 贺小令
> Subject:Re: [ANNO
+1
My only suggestion (although by no means a blocker) would be to remove
from the FLIP the `env.setRuntimeMode()` method.
I say that because this is syntactic sugar over the `env.configure()`
with the `execution.runtime-mode` option set to BATCH or STREAMING.
These methods can be nice but they se
Hi folks,
Thanks for all the great feedback. I have just updated FLIP-33 wiki with
the following changes:
1. Renaming. "currentFetchLatency" to "currentFetchEventTimeLag",
"currentLatency" to "currentEmitEventTimeLag".
2. Added the public interface code change required for the new metrics.
3. Add
Congratulations! Thanks @ZhuZhu for driving this release !
On 09/17/2020 13:29,Zhu Zhu wrote:
The Apache Flink community is very happy to announce the release of Apache
Flink 1.11.2, which is the second bugfix release for the Apache Flink 1.11
series.
Apache Flink® is an open-source stream pr
Steven,
we were also wondering if it is a strict requirement that "later"
updates to Iceberg subsume earlier updates. In the current version, you
only check whether checkpoint X made it to Iceberg and then discard all
committable state from Flink state for checkpoints smaller X.
If we go wit
Thanks Zhuzhu for driving the release!!!
Best,
Guowei
On Fri, Sep 18, 2020 at 5:10 PM Yun Gao wrote:
> Great! Very thanks @ZhuZhu for driving this and thanks for all contributed
> to the release!
>
> Best,
> Yun
>
> --Original Mail --
> *Sender:*Jingsong Li
>
I don't think the segment-size will help here.
If I understand the code correctly, then we have a fixed number of
segments (# = memory/segment size), and if all segments are full we
spill _all_ current segments in memory to disk into a single file, and
re-use this file for future spilling unti
Congratulations Godfrey!
Best,
Yun
--Original Mail --
Sender:Dawid Wysakowicz
Send Date:Thu Sep 17 14:45:55 2020
Recipients:Flink Dev , 贺小令
Subject:Re: [ANNOUNCE] New Apache Flink Committer - Godfrey He
Congratulations!
On 16/09/2020 06:19, Jark Wu wrote:
> H
Great! Very thanks @ZhuZhu for driving this and thanks for all contributed to
the release!
Best,
Yun
--Original Mail --
Sender:Jingsong Li
Send Date:Thu Sep 17 13:31:41 2020
Recipients:user-zh
CC:dev , user , Apache Announce
List
Subject:Re: [ANNOUNCE] Apac
Personally I think the fieldNames Map is confusing and not handy.
I just have an idea but not sure what you think.
What about adding a new constructor with List field names, this enables all
name-based setter/getters.
Regarding to List -> Map cost for every record, we can suggest users to
reuse the
xiaozilong created FLINK-19291:
--
Summary: Meeting `SchemaParseException` when I use
`AvroSchemaConverter` converts flink logical type
Key: FLINK-19291
URL: https://issues.apache.org/jira/browse/FLINK-19291
Tzu-Li (Gordon) Tai created FLINK-19290:
---
Summary: Add documentation for Stateful Function's Flink
DataStream SDK
Key: FLINK-19290
URL: https://issues.apache.org/jira/browse/FLINK-19290
Project:
Yi Tang created FLINK-19289:
---
Summary: K8s resource manager terminated pod garbage collection
Key: FLINK-19289
URL: https://issues.apache.org/jira/browse/FLINK-19289
Project: Flink
Issue Type: Bug
Dawid Wysakowicz created FLINK-19288:
Summary: Make InternalTimeServiceManager an interface
Key: FLINK-19288
URL: https://issues.apache.org/jira/browse/FLINK-19288
Project: Flink
Issue Ty
36 matches
Mail list logo