[
https://issues.apache.org/jira/browse/BEAM-5520?focusedWorklogId=150039&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-150039
]
ASF GitHub Bot logged work on BEAM-5520:
----------------------------------------
Author: ASF GitHub Bot
Created on: 01/Oct/18 13:53
Start Date: 01/Oct/18 13:53
Worklog Time Spent: 10m
Work Description: mxm commented on a change in pull request #6524:
[BEAM-5520] Flink pipeline option to run SDK harness per subtask.
URL: https://github.com/apache/beam/pull/6524#discussion_r221613616
##########
File path:
runners/flink/src/main/java/org/apache/beam/runners/flink/FlinkPipelineOptions.java
##########
@@ -184,4 +186,14 @@
Long getLatencyTrackingInterval();
void setLatencyTrackingInterval(Long interval);
+
+ @Description(
+ "SDK process parallelism for portable pipelines. Currently supported
options are "
+ + "'[auto]' (Let the runner decide) or "
+ + "'[pipeline]' (single SDK harness process per pipeline and task
manager JVM) or "
+ + "'[stage]' (separate SDK harness for every executable stage.)")
+ @Default.String(AUTO)
+ String getSdkWorkerParallelism();
Review comment:
I think this change is great but I wonder whether we should expose this as a
general PortablePipelineOptions?
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
Issue Time Tracking
-------------------
Worklog Id: (was: 150039)
Time Spent: 50m (was: 40m)
> Flink runner per operator SDK harness option
> --------------------------------------------
>
> Key: BEAM-5520
> URL: https://issues.apache.org/jira/browse/BEAM-5520
> Project: Beam
> Issue Type: Improvement
> Components: runner-flink
> Reporter: Thomas Weise
> Assignee: Thomas Weise
> Priority: Major
> Labels: portability, portability-flink
> Time Spent: 50m
> Remaining Estimate: 0h
>
> For streaming pipelines, the runner currently uses a single SDK harness
> process for all subtasks of a job that get deployed on the same task manager.
> In common deployments with 16 or more tasks slots, many executable stage
> operators all use the same SDK harness process. To scale, we need an option
> to run separate harness processes per subtask.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
