[jira] [Updated] (CASSANDRA-2755) ColumnFamilyRecordWriter fails to throw a write exception encountered after the user begins to close the writer

["Mck SembWever (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/CASSANDRA-2755?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Mck SembWever updated CASSANDRA-2755:
-------------------------------------

    Attachment: CASSANDRA-2755.patch

In RangeClient i cannot see why close() needs to be called before lastException is assigned. The following patch should work: I have tested it against various jobs but i have no reproducible testcase to confirm this bug against.

Also in the patch is a slight cleanup to ColumnFamilyRecordWriter's close() methods: keeping implementation out of deprecated methods.

> ColumnFamilyRecordWriter fails to throw a write exception encountered after the user begins to close the writer
> ---------------------------------------------------------------------------------------------------------------
>
>                 Key: CASSANDRA-2755
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-2755
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Hadoop
>    Affects Versions: 0.8.0
>            Reporter: Greg Katz
>            Assignee: Mck SembWever
>         Attachments: CASSANDRA-2755.patch
>
>
> There appears to be a race condition in {{ColumnFamilyRecordWriter}} that can result in the loss of an exception. Here is how it can happen (W stands for the {{RangeClient}}'s worker thread; U stands for the {{ColumnFamilyRecordWriter}} user's thread):
> # W: {{RangeClient}}'s {{run}} method catches an exception originating in the Thrift client/socket, but doesn't get a chance to set it on the {{lastException}} field before it the thread is preempted.
> # U: The user calls {{close}} which calls {{stopNicely}}. Because the {{lastException}} field is null, {{stopNicely}} does not throw anything. {{close}} then joins on the worker thread.
> # W: The {{RangeClient}}'s {{run}} method sets the {{lastException}} field and exits.
> # U: Although the thread in {{close}} is waiting for the worker thread to exit, it has already checked the {{lastException}} field so it doesn't detect the presence of the last exception. Instead, {{close}} returns without throwing anything.
> This race condition means that intermittently write failures will go undetected.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira