[
https://issues.apache.org/jira/browse/SOLR-6485?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Varun Thacker updated SOLR-6485:
--------------------------------
Attachment: SOLR-6485.patch
Thanks everyone for reviewing!
New patch which does the following -
- Refactored the DirectoryFileStream and LocalFsFileStream to reuse code. The
throttling logic is folded in DirectoryFileStream instead of having a separate
implementation
- The init pause is needed for the future pause timings to work correctly.
Check TestRateLimiter.testPause() which does the same thing. Test fails when
removing the init in testPause() method. We should fix this in another Jira I
guess.
- Test case and some small refactoring in TestReplicationHandler
About reserving index commit points, currently we extend it by 10s for every 5
reads of 1MB ( default ) packet sizes. If one throttles a lot this could trip
like Ramkumar pointed out.
So we could take two approaches -
1. Instead of the hardcoded 10 seconds look at the throttle speed and estimate
how much time should the reserve point be extended for.
2. Add functionality to the RateLimiter that actually returns the time it needs
to pause before actually pausing. We could split the current pause method into
3 methods -
preparePause()- would return the pauseNS value.
pause(int pauseNS) - Pause for the specified time duration.
pause() - calls preparePause() and then pause(pauseNs) . This makes sure that
we are not breaking the public api.
We could then take the time it needs to pause for and add it to the default 10
seconds.
> ReplicationHandler should have an option to throttle the speed of replication
> -----------------------------------------------------------------------------
>
> Key: SOLR-6485
> URL: https://issues.apache.org/jira/browse/SOLR-6485
> Project: Solr
> Issue Type: Improvement
> Reporter: Varun Thacker
> Assignee: Noble Paul
> Attachments: SOLR-6485.patch, SOLR-6485.patch, SOLR-6485.patch
>
>
> The ReplicationHandler should have an option to throttle the speed of
> replication.
> It is useful for people who want bring up nodes in their SolrCloud cluster or
> when have a backup-restore API and not eat up all their network bandwidth
> while replicating.
> I am writing a test case and will attach a patch shortly.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
