hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sergey Shelukhin (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-3787) Increment is non-idempotent but client retries RPC
Date Tue, 09 Apr 2013 00:34:15 GMT

    [ https://issues.apache.org/jira/browse/HBASE-3787?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13626024#comment-13626024
] 

Sergey Shelukhin commented on HBASE-3787:
-----------------------------------------

Other comments after discussion:
1) Note that this scheme has different behavior in case of nonces collected by timeout chore
- the hash one will make nonce available again, whereas this one will make nonce unavailable
(which is probably better, especially when there's a possibility of very delayed "retry",
such as during replication).
2) Storage of gaps could be optimized by storing number ranges [from, to) instead of individual
ops.
3) With regard to what result to return - it depends a lot on whether we think "atomic return
value" is important. I.e. should increment strictly return the incremented value, like AtomicLong,
or is it allowed to return some value in future? Ditto for append. I'd argue that for the
former case it will be an overkill to store return values of all append/increment operations
in memory for minutes, so the server should respond something to the effect of "the operation
succeeded, but I don't know what the result is". In the latter case the current value can
be returned, so only "get" part of increment/append is done. Perhaps the operation should
have a flag that determined which semantic to use.
                
> Increment is non-idempotent but client retries RPC
> --------------------------------------------------
>
>                 Key: HBASE-3787
>                 URL: https://issues.apache.org/jira/browse/HBASE-3787
>             Project: HBase
>          Issue Type: Bug
>          Components: Client
>    Affects Versions: 0.94.4, 0.95.2
>            Reporter: dhruba borthakur
>            Priority: Critical
>             Fix For: 0.95.1
>
>
> The HTable.increment() operation is non-idempotent. The client retries the increment
RPC a few times (as specified by configuration) before throwing an error to the application.
This makes it possible that the same increment call be applied twice at the server.
> For increment operations, is it better to use HConnectionManager.getRegionServerWithoutRetries()?
Another  option would be to enhance the IPC module to make the RPC server correctly identify
if the RPC is a retry attempt and handle accordingly.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message