Re:Re: Re: Multiple dataflow jobs management(lots of jobs)

[刘岩]

Hi  ThadThank you very much for your advice. Kettle can do the job for sure , 
but the metadata i was talking about is the metadata of the job descriptions 
used for kettle itself. The only option left for kettle is multiple instances , 
but that also means that we need to develop a master application to gather all 
the instances metadata. Moreover , Kettle does not have a Web Based GUI for 
designing and testing the job , that39s why we want NIFI , but again , multiple 
instances of nifi also leads to a HA problem for master node, so we turn to 
ambari metrics for that issue.Talend has a cloud server doing the similar 
thing, but it39s running on public cloud which is not accepted by our 
client.Kettle is a great ETL tool, but Web Based designer is really the master 
point for future.Thank you very muchYan LiuYan Liu 

Hortonworks Service Division 

Richinfo, Shenzhen, China (PR)

14/03/2016邮件原文发件人：Thad Guidry  <thadgui...@gmail.com>收件人：users 
<us...@nifi.apache.org>抄　送: dev  <dev@nifi.apache.org>发送时间：2016-03-13 
23:04:39主题：Re: Re: Multiple dataflow jobs management(lots of jobs)
Yan,



Pentaho Kettle (PDI) can also certainly handle your needs. But using 10K jobs 
to accomplish this is not the proper way to setup Pentaho.  Also, using MySQL 
to store the metadata is where you made a wrong choice.  PostgreSQL with data 
silos on SSD drives would be a better choice, while properly doing Async config 
[1] and other necessary steps for high writes.  Don39t keep Pentaho39s Table 
output commit levels at their default of 10k rows when your processing millions 
of rows!) For Oracle 11g or PostgreSQL, where I need 30 sec time slice windows 
for the metadata logging and where I typically have less than 1k of data on 
average per row, I typically will choose 200k rows or more in Pentaho39s table 
output commit option.



I would suggest you contact Pentaho for some adhoc support or hire some 
consultants to help you learn more, or setup properly for your use case.  For 
free, you can also just do a web search on "Pentaho best practices".  There39s 
a lot to learn from industry experts who already have used these tools and know 
their quirks.



[1] 
http://www.postgresql.org/docs/9.5/interactive/runtime-config-resource.html#RUNTIME-CONFIG-RESOURCE-ASYNC-BEHAVIOR





Thad+ThadGuidry






On Sat, Mar 12, 2016 at 11:00 AM, 刘岩 <liu...@richinfo.cn> wrote:Hi Aldrinsome 
additional information.it39s a typical ETL offloading user case each extraction 
job should foucs on 1 table and 1 table only.  data will be written on HDFS , 
this is similar to Database Staging. The reason why we need to foucs on 1 table 
for each job is because there might be database error or disconnection occur 
during the extraction , if it39s running as  a script like extraction job with 
expression langurage, then it39s hard to do the re-running or excape thing on 
that table or tables.once the extraction is done, a triger like action will do 
the data cleansing.  this is similar to ODS layer of Datawarehousingif the data 
quality has passed the quality check , then it will be marked as cleaned. 
otherwise , it will return to previous step and redo the data extraction, or 
send alert/email to the  system administrator.if certain numbers of tables were 
all cleaned and checked , then it will  call some Transforming  processor to do 
the transforming ， then push the data into a datawarehouse (Hive in our 
case)Thank you very much Yan Liu 

Hortonworks Service Division 

Richinfo, Shenzhen, China (PR)

13/03/2016邮件原文发件人："刘岩" <liu...@richinfo.cn>收件人：users  
<us...@nifi.apache.org>抄　送: dev  <dev@nifi.apache.org>发送时间：2016-03-13 
00:12:27主题：Re:Re: Multiple dataflow jobs management(lots of jobs)
Hi AldrinCurrently  we need to extract 60K tables per day , and the time window 
is limited to 8 Hours.  Which means that we need to run jobs concurrently , and 
we need a general description of what39s going on with all those 60K job flows 
and take further actions.  We have tried Kettle and Talend ,  Talend is a 
IDE-Based so not what we are looking for,  and Kettle was crashed due to the 
Mysql cannot handle the Kettle39s metadata with 10K jobs.So we want to use Nifi 
,  this is really the product that we are looking for , but  the missing piece 
here is a DataFlow jobs Admin Page.  so we can have multiple Nifi instances 
running on different nodes, but monitoring the jobs in one page.  If it can 
intergrate with Ambari metrics API,  then we can develop an Ambari View for 
Nifi Jobs Monitoring just like HDFS View and Hive View. Thank you very much Yan 
Liu 

Hortonworks Service Division 

Richinfo, Shenzhen, China (PR)

06/03/2016邮件原文发件人：Aldrin Piri  <aldrinp...@gmail.com>收件人：users 
<us...@nifi.apache.org>抄　送: dev  <dev@nifi.apache.org>发送时间：2016-03-11 
02:27:11主题：Re: Mutiple dataflow jobs management(lots of jobs)Hi Yan,
We can get more into details an

Re: Re: Multiple dataflow jobs management(lots of jobs)

[Thad Guidry]

Yan,

Pentaho Kettle (PDI) can also certainly handle your needs. But using 10K
jobs to accomplish this is not the proper way to setup Pentaho.  Also,
using MySQL to store the metadata is where you made a wrong choice.
PostgreSQL with data silos on SSD drives would be a better choice, while
properly doing Async config [1] and other necessary steps for high writes.
Don't keep Pentaho's Table output commit levels at their default of 10k
rows when your processing millions of rows!) For Oracle 11g or PostgreSQL,
where I need 30 sec time slice windows for the metadata logging and where I
typically have less than 1k of data on average per row, I typically will
choose 200k rows or more in Pentaho's table output commit option.

I would suggest you contact Pentaho for some adhoc support or hire some
consultants to help you learn more, or setup properly for your use case.
For free, you can also just do a web search on "Pentaho best practices".
There's a lot to learn from industry experts who already have used these
tools and know their quirks.

[1]
http://www.postgresql.org/docs/9.5/interactive/runtime-config-resource.html#RUNTIME-CONFIG-RESOURCE-ASYNC-BEHAVIOR


Thad
+ThadGuidry 

On Sat, Mar 12, 2016 at 11:00 AM, 刘岩  wrote:

> Hi Aldrin
>
> some additional information.
>
> it's a typical ETL offloading user case
>
> each extraction job should foucs on 1 table and 1 table only.  data will
> be written on HDFS , this is similar to Database Staging.
>
> The reason why we need to foucs on 1 table for each job is because there
> might be database error or disconnection occur during the extraction , if
> it's running as  a script like extraction job with expression langurage,
> then it's hard to do the re-running or excape thing on that table or tables.
>
> once the extraction is done, a triger like action will do the data
> cleansing.  this is similar to ODS layer of Datawarehousing
>
> if the data quality has passed the quality check , then it will be marked
> as cleaned. otherwise , it will return to previous step and redo the data
> extraction, or send alert/email to the  system administrator.
>
> if certain numbers of tables were all cleaned and checked , then it will
> call some Transforming  processor to do the transforming ， then push the
> data into a datawarehouse (Hive in our case)
>
>
> Thank you very much
>
> Yan Liu
>
> Hortonworks Service Division
>
> Richinfo, Shenzhen, China (PR)
> 13/03/2016
>
> 邮件原文
> *发件人：*"刘岩" 
> *收件人：*users  
> *抄 送: *dev  
> *发送时间：*2016-03-13 00:12:27
> *主题：*Re:Re: Multiple dataflow jobs management(lots of jobs)
>
>
> Hi Aldrin
>
> Currently  we need to extract 60K tables per day , and the time window is
> limited to 8 Hours.  Which means that we need to run jobs concurrently ,
> and we need a general description of what's going on with all those 60K job
> flows and take further actions.
>
> We have tried Kettle and Talend ,  Talend is a IDE-Based so not what we
> are looking for,  and Kettle was crashed due to the Mysql cannot handle the
> Kettle's metadata with 10K jobs.
>
> So we want to use Nifi ,  this is really the product that we are looking
> for , but  the missing piece here is a DataFlow jobs Admin Page.  so we can
> have multiple Nifi instances running on different nodes, but monitoring the
> jobs in one page.  If it can intergrate with Ambari metrics API,  then we
> can develop an Ambari View for Nifi Jobs Monitoring just like HDFS View and
> Hive View.
>
>
> Thank you very much
>
> Yan Liu
>
> Hortonworks Service Division
>
> Richinfo, Shenzhen, China (PR)
> 06/03/2016
>
>
> 邮件原文
> *发件人：*Aldrin Piri  
> *收件人：*users 
> *抄 送: *dev  
> *发送时间：*2016-03-11 02:27:11
> *主题：*Re: Mutiple dataflow jobs management(lots of jobs)
>
> Hi Yan,
>
> We can get more into details and particulars if needed, but have you
> experimented with expression language?  I could see a Cron driven approach
> which covers your periodic efforts that feeds some number of ExecuteSQL
> processors (perhaps one for each database you are communicating with) each
> having a table.  This would certainly cut down on the need for the 30k
> processors on a one-to-one basis with a given processor.
>
> In terms of monitoring the dataflows, could you describe what else you are
> searching for beyond the graph view?  NiFi tries to provide context for the
> flow of data but is not trying to be a sole monitoring, we can give
> information on a processor basis, but do not delve into specifics.  There
> is a summary view for the overall flow where you can monitor stats about
> the components and connections in the system. We support interoperation
> with monitoring systems via push (ReportingTask) and pull (REST API [2])
> semantics.
>
> Any other details beyond your list of how this all interoperates might
> shed some