[
https://issues.apache.org/jira/browse/HBASE-15165?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15116948#comment-15116948
]
Heng Chen commented on HBASE-15165:
-----------------------------------
I met similar problem in our production cluster. I am not sure it is due to the
same reason.
As current logic, when all RPC handlers on RS are exhausted, client still sends
requests to this server. Then, the requests will timeout and client do retry.
It makes the situation worse.
IMO client should not do retry for old requests and not accept new requests
when it finds that the RS is in heavy load. To to it, we can throw some kind
exception to client when the situation of no handlers to use last long enough.
wdyt? [~eclark]
> AsyncProcess can spin wait indefinitly
> --------------------------------------
>
> Key: HBASE-15165
> URL: https://issues.apache.org/jira/browse/HBASE-15165
> Project: HBase
> Issue Type: Bug
> Affects Versions: 2.0.0, 1.2.0, 1.3.0
> Reporter: Elliott Clark
> Priority: Critical
>
> When the max outstanding requests per region or per server is reached, all
> threads trying to send more requests to that server will spin and will spin
> forever with no sleep, and no regard for timeouts.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
