samza-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jagadish Venkatraman <jagadish1...@gmail.com>
Subject Re: Review Request 51346: SAMZA-974 - Support finite datasources in Samza that have a notion of End-Of-Stream
Date Tue, 06 Sep 2016 15:56:30 GMT


> On Aug. 29, 2016, 11:54 p.m., Xinyu Liu wrote:
> > samza-core/src/main/java/org/apache/samza/task/AsyncRunLoop.java, line 375
> > <https://reviews.apache.org/r/51346/diff/5/?file=1488333#file1488333line375>
> >
> >     No need to resume. This is running on the main thread.

Not calling resume() will stall the runLoop, causing an infinite wait. I verified this using
an unit test.


> On Aug. 29, 2016, 11:54 p.m., Xinyu Liu wrote:
> > samza-core/src/main/java/org/apache/samza/task/AsyncRunLoop.java, line 370
> > <https://reviews.apache.org/r/51346/diff/5/?file=1488333#file1488333line370>
> >
> >     Do we need to consider the auto-commit case here?

Talked with Xinyu and Yi offline. We agreed that auto-commit is nothing more than a periodic
commit. It does not provide any guarantees that we'll commit at the end-of-stream. Users requiring
end-of-stream commit should invoke `commit` in the end-of-stream-listener callback.


- Jagadish


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/51346/#review147245
-----------------------------------------------------------


On Aug. 30, 2016, 12:32 a.m., Jagadish Venkatraman wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/51346/
> -----------------------------------------------------------
> 
> (Updated Aug. 30, 2016, 12:32 a.m.)
> 
> 
> Review request for samza, Boris Shkolnik, Chris Pettitt, Yi Pan (Data Infrastructure),
Navina Ramesh, and Xinyu Liu.
> 
> 
> Repository: samza
> 
> 
> Description
> -------
> 
> Samza currently works with unbounded data sources (kafka streams). However, for bounded
data sources like HDFS files, snapshot files which are not infinite, we need a notion of 'end-of-stream'.

> 
> This is a step towards realizing a 'finite' Samza job that terminates once data processing
is complete.(as opposed to an infinite stream job that keeps running)  
> 
> RB changes:
> - New interface for EndOfStreamListener
> - New 'end-of-stream' state in the state-machine of AsyncStreamTask (Invariant: When
end-of-stream is reached there are no buffered messages, no-callbacks are in-flight and no-window
or commit call shall be in progress)
> - Changes to allow clean shut-downs of the tasks/container/job for end-of-stream.
> 
> Design Doc and Implementation Notes: https://issues.apache.org/jira/secure/attachment/12825119/ProposalforEndofStreaminSamza.pdf
> 
> 
> Diffs
> -----
> 
>   build.gradle 1d4eb74b1294318db8454631ddd0901596121ab2 
>   samza-api/src/main/java/org/apache/samza/system/IncomingMessageEnvelope.java cc860cf7eb4d514736913c1dceaa80534b61d71a

>   samza-api/src/main/java/org/apache/samza/system/SystemStreamPartitionIterator.java
a8f858aa7e4f4ce436f450cf439fe1a102983c64 
>   samza-api/src/main/java/org/apache/samza/task/EndOfStreamListenerTask.java PRE-CREATION

>   samza-core/src/main/java/org/apache/samza/task/AsyncRunLoop.java a510bb0c5914c772438930d27f100b4d360c1296

>   samza-core/src/main/java/org/apache/samza/task/AsyncStreamTaskAdapter.java 1fc645673b7547b642830df5639c0b4fcd11c0d5

>   samza-core/src/main/scala/org/apache/samza/container/TaskInstance.scala 89f6857014489aba2db4129bc2e26dfec5b10652

>   samza-core/src/test/java/org/apache/samza/task/TestAsyncRunLoop.java ca913dea79fecbcecdfd1010dc794318055c5764

> 
> Diff: https://reviews.apache.org/r/51346/diff/
> 
> 
> Testing
> -------
> 
> Unit tests to test scenarios for inorder processing, out-of-order processing and commit
semantics.
> 
> 
> Thanks,
> 
> Jagadish Venkatraman
> 
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message