beam-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Kenneth Knowles (JIRA)" <j...@apache.org>
Subject [jira] [Resolved] (BEAM-1396) GABWVOBDoFn expects grouped values to be ordered by their timestamp but there is no such guarantee
Date Mon, 06 Feb 2017 17:30:41 GMT

     [ https://issues.apache.org/jira/browse/BEAM-1396?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Kenneth Knowles resolved BEAM-1396.
-----------------------------------
       Resolution: Fixed
    Fix Version/s: 0.6.0

> GABWVOBDoFn expects grouped values to be ordered by their timestamp but there is no such
guarantee
> --------------------------------------------------------------------------------------------------
>
>                 Key: BEAM-1396
>                 URL: https://issues.apache.org/jira/browse/BEAM-1396
>             Project: Beam
>          Issue Type: Bug
>          Components: sdk-java-core
>            Reporter: Amit Sela
>            Assignee: Kenneth Knowles
>             Fix For: 0.6.0
>
>
> GABWVOBDoFn relies on the grouped values to be ordered by their timestamp but nothing
in the SDK guarantees this: 
> https://github.com/apache/beam/blob/master/runners/core-java/src/main/java/org/apache/beam/runners/core/GroupAlsoByWindowsViaOutputBufferDoFn.java#L86
> If such a chunk of timestamped values will be processed out-of-order I assume we'd end
up with an {{IllegalStateException}} thrown here:
> https://github.com/apache/beam/blob/master/runners/core-java/src/main/java/org/apache/beam/runners/core/InMemoryTimerInternals.java#L191
> I suggest we go ahead and add sorting before processing the bundle in chunks - this might
prove expensive in extreme cases where a very large bundle with very few keys is processed,
but it seems that timestamp order is necessary.
> As for runners who provide order guarantee, since GABW is optional I don't see an issue
here, though [~dhalperi@google.com] suggested we add a "shouldSort" flag.
> Also, probably worth creating a test for this, though it would prove difficult since
we would have to preset the order which is the problem to begin with :-)



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Mime
View raw message