[
https://issues.apache.org/jira/browse/BEAM-7859?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16914638#comment-16914638
]
Kyle Weaver commented on BEAM-7859:
-----------------------------------
The error happens because work is happening on different docker containers,
which do not have access to each other's file systems. We need all the work to
happen in a single container for this to be fixed. This requires:
# BEAM-7600 is merged (allows the reuse of the SDK harness)
# The option --sdk_worker_parallelism=1 needs to be set, directly (which I
prefer, because tracking down defaults is too confusing – BEAM-7657) and/or by
modifying the Spark runner to default to 1, as Flink does [1]. (Right now, we
are instead using n_cores-1)
[1]
[https://github.com/apache/beam/blob/c2f0d282337f3ae0196a7717712396a5a41fdde1/runners/flink/src/main/java/org/apache/beam/runners/flink/FlinkJobInvoker.java#L73-L75]
> Portable Wordcount on Spark runner does not work in DOCKER execution mode.
> --------------------------------------------------------------------------
>
> Key: BEAM-7859
> URL: https://issues.apache.org/jira/browse/BEAM-7859
> Project: Beam
> Issue Type: Bug
> Components: runner-spark, sdk-py-harness
> Reporter: Valentyn Tymofieiev
> Assignee: Kyle Weaver
> Priority: Major
> Labels: portability-spark
>
> The error was observed during Beam 2.14.0 release validation, see:
> https://issues.apache.org/jira/browse/BEAM-7224?focusedCommentId=16896831&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-16896831
> Looks like master currently fails with a different error, both in Loopback
> and Docker modes.
> [~ibzib] [~altay] [~robertwb] [~angoenka]
--
This message was sent by Atlassian Jira
(v8.3.2#803003)
