You are viewing a plain text version of this content. The canonical link for it is here.

Posted to common-dev@hadoop.apache.org by "Daryn Sharp (JIRA)" <ji...@apache.org> on 2014/08/28 20:24:10 UTC

[jira] [Created] (HADOOP-11019) Queued IPC calls are not aborted if the connection drops

Daryn Sharp created HADOOP-11019:
------------------------------------

             Summary: Queued IPC calls are not aborted if the connection drops
                 Key: HADOOP-11019
                 URL: https://issues.apache.org/jira/browse/HADOOP-11019
             Project: Hadoop Common
          Issue Type: Bug
          Components: ipc
    Affects Versions: 2.0.0-alpha, 3.0.0
            Reporter: Daryn Sharp
            Assignee: Daryn Sharp


Once a call is read from the wire and queued, it will be executed even if the connection has already dropped.  If the client closes the connection due to timeout, perhaps because the server is overloaded, the client's retry will only exacerbate the problem.  One specific example is DNs with large block reports overwhelming an already unhealthy NN.

Ideally calls should be cancelled when the connection is dropped and/or connection state should be checked when the call is extracted from the callq, prior to decoding and invoking the call.



--
This message was sent by Atlassian JIRA
(v6.2#6252)