[
https://issues.apache.org/jira/browse/HBASE-10701?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13939875#comment-13939875
]
Enis Soztutar commented on HBASE-10701:
---------------------------------------
Thanks Devaraj for the review.
bq. It so extreme that I don't really know. I suppose that whatever you do it's
going to be difficult at the end . I'm +1 whatever the final choice here.
I think it is safer to send even the DoNotRetryIOException to replicas. If it
becomes a problem to wait for all results from replicas, we can fix it later.
bq. However, I notice that the test most of the time dies with OOM, cannot
create native thread, because the number of threads grow unbounded
One cause for the # threads to jump was that, meta's own location is not
cached, resulting in a zk request for every region location cache miss. In the
test we are doing 12K req/s from a single client, and with CM, we do a LOT of
zk requests causing multi second slowdowns because of zk contention.
HBASE-10785 attacks this issue.
[~nkeywal] I'll commit v3 if you are ok with it.
> Cache invalidation improvements from client side
> ------------------------------------------------
>
> Key: HBASE-10701
> URL: https://issues.apache.org/jira/browse/HBASE-10701
> Project: HBase
> Issue Type: Sub-task
> Reporter: Enis Soztutar
> Assignee: Enis Soztutar
> Fix For: hbase-10070
>
> Attachments: hbase-10701_v1.patch, hbase-10701_v2.patch,
> hbase-10701_v3.patch
>
>
> Running the integration test in HBASE-10572, and HBASE-10355, it seems that
> we need some changes for cache invalidation of meta entries from the client
> side in backup RPCs.
> Mainly the RPC's made for replicas should not invalidate the cache for all
> the replicas (for example on RegionMovedException, connection error etc).
--
This message was sent by Atlassian JIRA
(v6.2#6252)
