[ 
https://issues.apache.org/jira/browse/HBASE-15593?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15237457#comment-15237457
 ] 

stack commented on HBASE-15593:
-------------------------------

Patch looks great.

On these:

267       private static final String MIN_CLIENT_REQUEST_TIMEOUT = 
"hbase.ipc.min.client.request.timeout";
268       private static final int DEFAULT_MIN_CLIENT_REQUEST_TIMEOUT = 20;

Timeout is in what unit? milliseconds?

Another issue is that the one thing we do not need in hbase is more ways to 
specify timeout (smile). We have timeouts and retries and many different levels 
making it very hard for operators to figure what to tweak to do stuff like fail 
fast. How does this new config. fit into the current scheme of things? How does 
it relate to current timeout and retry mechanisms? For example, could this 
supercede all other timeouts? Can we deprecate them? How does this relate to 
retries? At a minimum, we need some more doc on these new parameters (can 
copy/paste as release note for this feature).

Otherwise, patch looks great. It has potential to clean up a messy, 
long-running story around client timeout setting. Thanks [~yangzhe1991]








> Time limit of scanning should be offered by client
> --------------------------------------------------
>
>                 Key: HBASE-15593
>                 URL: https://issues.apache.org/jira/browse/HBASE-15593
>             Project: HBase
>          Issue Type: Bug
>    Affects Versions: 1.2.0, 1.1.4
>            Reporter: Phil Yang
>            Assignee: Phil Yang
>         Attachments: HBASE-15593-v1.patch, HBASE-15593-v2.patch, 
> HBASE-15593-v3.patch
>
>
> In RSRpcServices.scan, we will set a time limit equaling to 
> Math.min(scannerLeaseTimeoutPeriod, rpcTimeout) / 2, and will response 
> heartbeat message if we reach this limit. However, two timeout settings 
> (hbase.client.scanner.timeout.period and hbase.rpc.timeout) are read from 
> RS's configure, which may be different from client's. If client's setting is 
> much less than server's, there may still be timeout at client side.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to