[jira] [Commented] (HBASE-21573) More sensible client default timeout values

["Peter Somogyi (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HBASE-21573?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16732061#comment-16732061 ] 

Peter Somogyi commented on HBASE-21573:
---------------------------------------

Additionally to the ones that [~lhofhansl] mentioned earlier I also found hbase.rpc.shortoperation.timeout.

From all of these properties users should mainly focus on hbase.rpc.timeout, hbase.client.operation.timeout and hbase.client.scanner.timeout.period (for scan operations). Only users who want to fine tune read or write rpc timeout should set hbase.rpc.read.timeout and hbase.rpc.write.timeout values, when these properties are not set these will fall back to hbase.rpc.timeout. This is similar to hbase.client.operation.timeout and hbase.client.meta.operation.timeout values. So hbase.client.meta.operation.timeout will fall back to hbase.client.operation.timeout.

For quick RPC operations within cluster HBase relies on hbase.rpc.shortoperation.timeout property but this does not have client related effect.

There is a deprecated hbase.regionserver.lease.period config which is marked as being removed after 0.96 release. I'll remove this property and also pointers in the reference guide.

I'm not against the fail fast approach but I don't know what values are still reasonable as a default configuration.

> More sensible client default timeout values
> -------------------------------------------
>
>                 Key: HBASE-21573
>                 URL: https://issues.apache.org/jira/browse/HBASE-21573
>             Project: HBase
>          Issue Type: Wish
>          Components: Client
>    Affects Versions: 2.1.1
>            Reporter: Cosmin Lehene
>            Priority: Critical
>             Fix For: 2.1.3
>
>
> I guess the goal is to have operations allow enough time to recover from major failures.
> While this may make sense for large jobs, it's a PITA for OLTP scenarios and could probably benefit from a faster failure mode in default
>  
> hbase.rpc.timeout = 60000
> hbase.client.operation.timeout = 1200000
> hbase.client.meta.operation.timeout = 1200000
> The client meta ops timeout is not in defaults-xml and not documented in the book either.
> [https://hbase.apache.org/book.html#config_timeouts]
>  
> Would it make sense to have aggressive defaults instead?
>  



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)