[
https://issues.apache.org/jira/browse/BEAM-13137?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17475469#comment-17475469
]
Evan Galpin commented on BEAM-13137:
------------------------------------
[~echauchot] I'm wondering if the flakiness could be to do with non-determinism
of whether data has been flushed from the translog prior to making a size
estimate. That's not something I'd want to introduce into the core
functionality of the IO because I wouldn't want to alter what's happening in a
user's ES cluster, but if it can stabilize tests I think that would be
perfectly acceptable to add in the unit tests util methods. I'll play with that
a bit now.
> make ElasticsearchIO$BoundedElasticsearchSource#getEstimatedSizeBytes
> deterministic
> -----------------------------------------------------------------------------------
>
> Key: BEAM-13137
> URL: https://issues.apache.org/jira/browse/BEAM-13137
> Project: Beam
> Issue Type: Improvement
> Components: io-java-elasticsearch
> Reporter: Etienne Chauchot
> Assignee: Evan Galpin
> Priority: P2
>
> Index size estimation is statistical in ES and varies. But it is the base for
> splitting so it needs to be more deterministic because that causes flakiness
> in the UTests in _testSplit_ and _testSizes_ and maybe entails sub-optimal
> splitting in production in some cases.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
