[jira] [Commented] (BEAM-14171) CoGroupByKey loses values with large groups on Dataflow v1

["Daniel Oliveira (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/BEAM-14171?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17512130#comment-17512130 ] 

Daniel Oliveira commented on BEAM-14171:
----------------------------------------

How does this issue look in terms of timeline to getting a fix in and regarding blocking the release? I see this issue recently got reported and a PR just went up, but not sure if that PR addresses the root cause or is tangentially related.

It looks like this affected the previous two versions, which by precedent is a reason for us _not_ to block since this wouldn't be a new regression. But it also sounds like a pretty major problem if it affects all CoGroupByKeys, which I think might be worth making an exception _if_ we think a fix can be implemented in a timely manner. Like within the next three workdays as a rough target. Alternatively, if there are simple workarounds/mitigations for this, then I think we can just list it as a known issue and describe how users can mitigate it.

> CoGroupByKey loses values with large groups on Dataflow v1
> ----------------------------------------------------------
>
>                 Key: BEAM-14171
>                 URL: https://issues.apache.org/jira/browse/BEAM-14171
>             Project: Beam
>          Issue Type: Bug
>          Components: runner-dataflow, sdk-java-core
>    Affects Versions: 2.36.0, 2.37.0
>            Reporter: Niel Markwick
>            Assignee: Robert Bradshaw
>            Priority: P1
>             Fix For: 2.38.0
>
>          Time Spent: 20m
>  Remaining Estimate: 0h
>
> CoGroupByKey can lose elements - replacing them with null values when a group is large (>10,000 elements).
>  
> This only occurs in dataflow v1, not dataflow-v2 runner
> Possibly related to BEAM-13541.
>  
> https://lists.apache.org/thread/5y56kbgm3q0m1byzf7186rrkomrcfldm
>  
>  
>  



--
This message was sent by Atlassian Jira
(v8.20.1#820001)