[jira] [Created] (HBASE-15857) Explain client timeouts better

["Misty Stanley-Jones (JIRA)" <ji...@apache.org>]

Misty Stanley-Jones created HBASE-15857:
-------------------------------------------

             Summary: Explain client timeouts better
                 Key: HBASE-15857
                 URL: https://issues.apache.org/jira/browse/HBASE-15857
             Project: HBase
          Issue Type: Bug
          Components: documentation
    Affects Versions: 2.0.0
            Reporter: Misty Stanley-Jones
            Assignee: Misty Stanley-Jones


we should better explain the various client timeout internals and the various parameters. In the HBase Ref Guide.
> Caused by: org.apache.hadoop.hbase.ipc.RpcClient$CallTimeoutException: Call id=902666287, waitTime=110833, rpcTimetout=5000

What's the difference between waitTime and rpcTimeout

{quote}
It waited for up to 110s before throwing out the exception because of the multiplex nature of the calls made by the underlying connection - for 110s it was doing other Call work that was successful, and during this phase the particular call ID of 902666287 had already failed at 5s (x3 with pauses), but this was just one call ID that hung, while other async had succeeded in the same period over the same connection. 

The waitTime is printed when the connection is finally thrown away (i.e. Connection.cleanupCalls(…)) and we check each used Call's state and total time.

From what I see, the 5s is being respected just fine (in terms of a single connection). Of what the Call id was doing against the awd61 server specifically that took longer than 5s (plus retries) to run, that'd depend on the Scanner and RS details.
If they don't want connections to get reused across calls like this (causing this behaviour), I guess they can take a performance hit and make unique connections per request (vs. shared connection objects)?
{quote}



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