You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@beam.apache.org by "Maximilian Michels (JIRA)" <ji...@apache.org> on 2018/12/17 15:22:00 UTC
[jira] [Closed] (BEAM-2873) Detect number of shards for file sink
in Flink Streaming Runner
[ https://issues.apache.org/jira/browse/BEAM-2873?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Maximilian Michels closed BEAM-2873.
------------------------------------
Resolution: Fixed
Fix Version/s: 2.10.0
Fixed with a68f2099a978ce5a4a2af89c5e1524198c8c0fa0
> Detect number of shards for file sink in Flink Streaming Runner
> ---------------------------------------------------------------
>
> Key: BEAM-2873
> URL: https://issues.apache.org/jira/browse/BEAM-2873
> Project: Beam
> Issue Type: Improvement
> Components: runner-flink
> Reporter: Aljoscha Krettek
> Assignee: Dawid Wysakowicz
> Priority: Major
> Fix For: 2.10.0
>
> Time Spent: 6h
> Remaining Estimate: 0h
>
> [~reuvenlax] mentioned that this is done for the Dataflow Runner and the default behaviour on Flink can be somewhat surprising for users.
> ML entry: https://www.mail-archive.com/dev@beam.apache.org/msg02665.html:
> This is how the file sink has always worked in Beam. If no sharding is specified, then this means runner-determined sharding, and by default that is one file per bundle. If Flink has small bundles, then I suggest using the withNumShards method to explicitly pick the number of output shards.
> The Flink runner can detect that runner-determined sharding has been chosen, and override it with a specific number of shards. For example, the Dataflow streaming runner (which as you mentioned also has small bundles) detects this case and sets the number of out files shards based on the number of workers in the worker pool [Here|https://github.com/apache/beam/blob/9e6530adb00669b7cf0f01cb8b128be0a21fd721/runners/google-cloud-dataflow-java/src/main/java/org/apache/beam/runners/dataflow/DataflowRunner.java#L354] is the code that does this; it should be quite simple to do something similar for Flink, and then there will be no need for users to explicitly call withNumShards themselves.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)