accumulo-notifications mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Josh Elser (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (ACCUMULO-3856) ProxyServer.updateAndFlush leaks BatchWriter
Date Thu, 28 May 2015 18:39:19 GMT

     [ https://issues.apache.org/jira/browse/ACCUMULO-3856?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Josh Elser updated ACCUMULO-3856:
---------------------------------
    Description: 
updateAndFlush creates a new BatchWriter, adds the updates, then tries to flush and close
the BatchWriter.

If the flush or addMutations fails (such as the case in SimpleProxyBase ITs where we make
sure that constraints fire), this will leak that BatchWriter threads.

The BatchWriter should be closed in a finally block.

  was:
Inspecting ProxyServer, saw the following:

{code}
  @Override
  public void update(String writer, Map<ByteBuffer,List<ColumnUpdate>> cells)
throws TException {
    try {
      BatchWriterPlusException bwpe = getWriter(writer);
      addCellsToWriter(cells, bwpe);
    } catch (UnknownWriter e) {
      // just drop it, this is a oneway thrift call and throwing a TException seems to make
all subsequent thrift calls fail
    }
  }
{code}

This means that calls to update which fail to get a BatchWriter (for whatever reason) will
silently fail and appear to the user to have written data successfully. This is horrible.

I'm not sure what the comment means either. I'm not sure why throwing a TException from the
server would invalidate the connection, but perhaps I'm just ignorant of some finer details.


> ProxyServer.updateAndFlush leaks BatchWriter 
> ---------------------------------------------
>
>                 Key: ACCUMULO-3856
>                 URL: https://issues.apache.org/jira/browse/ACCUMULO-3856
>             Project: Accumulo
>          Issue Type: Bug
>          Components: proxy
>            Reporter: Josh Elser
>            Priority: Critical
>             Fix For: 1.6.3, 1.8.0, 1.7.1
>
>          Time Spent: 1h 40m
>  Remaining Estimate: 0h
>
> updateAndFlush creates a new BatchWriter, adds the updates, then tries to flush and close
the BatchWriter.
> If the flush or addMutations fails (such as the case in SimpleProxyBase ITs where we
make sure that constraints fire), this will leak that BatchWriter threads.
> The BatchWriter should be closed in a finally block.



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

Mime
View raw message