[
https://issues.apache.org/jira/browse/BEAM-1294?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15871624#comment-15871624
]
ASF GitHub Bot commented on BEAM-1294:
--------------------------------------
GitHub user aviemzur opened a pull request:
https://github.com/apache/beam/pull/2033
[BEAM-1294] Long running UnboundedSource Readers via Broadcasts
Be sure to do all of the following to help us incorporate your contribution
quickly and easily:
- [ ] Make sure the PR title is formatted like:
`[BEAM-<Jira issue #>] Description of pull request`
- [ ] Make sure tests pass via `mvn clean verify`. (Even better, enable
Travis-CI on your fork and ensure the whole test matrix passes).
- [ ] Replace `<Jira issue #>` in the title with the actual Jira issue
number, if there is one.
- [ ] If this contribution is large, please file an Apache
[Individual Contributor License
Agreement](https://www.apache.org/licenses/icla.txt).
---
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/aviemzur/beam unbounded-reader-cache
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/beam/pull/2033.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #2033
----
commit 3813deeb439b4898876cde11ab7fb087f1d63500
Author: Aviem Zur <[email protected]>
Date: 2017-02-17T10:35:49Z
[BEAM-1294] Long running UnboundedSource Readers via Broadcasts
----
> Long running UnboundedSource Readers via Broadcasts
> ---------------------------------------------------
>
> Key: BEAM-1294
> URL: https://issues.apache.org/jira/browse/BEAM-1294
> Project: Beam
> Issue Type: Improvement
> Components: runner-spark
> Reporter: Amit Sela
> Assignee: Aviem Zur
>
> When reading from an UnboundedSource, current implementation will cause each
> split to create a new Reader every micro-batch.
> As long as the overhead of creating a reader is relatively low, it's
> reasonable (though I'd still be happy to get rid of), but in cases where the
> creation overhead is large it becomes unreasonable forcing large batches.
> One way to solve this could be to create a pool of lazy-init readers to serve
> each executor, maybe via Broadcast variables.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
