You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@beam.apache.org by "Wout Scheepers (JIRA)" <ji...@apache.org> on 2018/11/07 14:32:00 UTC
[jira] [Commented] (BEAM-4796) SpannerIO waits for all input before
writing
[ https://issues.apache.org/jira/browse/BEAM-4796?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16678297#comment-16678297 ]
Wout Scheepers commented on BEAM-4796:
--------------------------------------
Hey [~nielm], thanks for implementing the streaming feature!
Unfortunately, I'm still having the following issue when running my streaming pipeline on dataflow:
My lag starts building up and it appears that the `Filter Unbatchable Mutations` step keeps on waiting for the schema PCollectionView to be passed. When I look into the `Schema View` step, the `Combine.GloballyAsSingletonView` ->` Combine.globally(Singleton)`step has no output, which I think is the cause for my lag.
Any idea on this? Does it have to do with windowing?
I've been looking into the tests to see how I can make a reproducible, but I have the feeling the SpannerWriteIT test covers my use case more or less.
Seems like the issue is similar is the following one:
[https://stackoverflow.com/questions/49273528/spannerio-dataflow-2-3-0-stuck-in-createdataflowview]
Thanks in advance
- Wout
> SpannerIO waits for all input before writing
> --------------------------------------------
>
> Key: BEAM-4796
> URL: https://issues.apache.org/jira/browse/BEAM-4796
> Project: Beam
> Issue Type: Bug
> Components: io-java-gcp
> Affects Versions: 2.5.0, 2.6.0, 2.7.0, 2.8.0
> Reporter: Niel Markwick
> Assignee: Niel Markwick
> Priority: Major
> Fix For: 2.9.0
>
> Time Spent: 50m
> Remaining Estimate: 0h
>
> SpannerIO.Write waits for all input in the window to arrive before getting the schema:
> [https://github.com/apache/beam/blame/release-2.5.0/sdks/java/io/google-cloud-platform/src/main/java/org/apache/beam/sdk/io/gcp/spanner/SpannerIO.java#L841]
>
> In streaming mode, this is not an issue, but in batch mode, this causes the pipeline to stall until all input is read, which could be a significant amount of time (and temp data).
>
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)