[
https://issues.apache.org/jira/browse/BEAM-12776?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sam Whittle reopened BEAM-12776:
--------------------------------
There is a large buffer by default for GCS writes and closing all windows in
parallel can increase memory usage and trigger OOMs. I plan on changing this
to limit parallelism to a certain amount, and allow that parallelism to be
controlled by an option.
> Improve parallelism of closing files in FileIO
> ----------------------------------------------
>
> Key: BEAM-12776
> URL: https://issues.apache.org/jira/browse/BEAM-12776
> Project: Beam
> Issue Type: Bug
> Components: io-java-files
> Reporter: Sam Whittle
> Assignee: Sam Whittle
> Priority: P2
> Time Spent: 2h
> Remaining Estimate: 0h
>
> Currently close happens in processElement which is per-window.
> If there are many windows firing this can throttle throughput waiting for IO
> instead of closing in parallel in finishBundle.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
