Long sleeping in HConnectionManager after thread is interrupted
---------------------------------------------------------------
Key: HBASE-3064
URL: https://issues.apache.org/jira/browse/HBASE-3064
Project: HBase
Issue Type: Bug
Components: client, ipc
Reporter: Bruno Dumon
Attachments: connectionmgr-interruptedexc-patch.txt
We run sometimes into the problem that when a thread running HBase client code
is interrupted, it hangs. The problem is it is sleeping in HConnectionManager,
in the methods locateRegionInMeta and getRegionServerWithRetries, where there
is code like this:
{code}
try{
Thread.sleep(getPauseTime(tries));
} catch (InterruptedException e) {
// continue
}
{code}
which is located in a for-loop, so it will keep retrying even when someone
requested the thread to stop its work.
The attached patch proposes as fix to re-assert the interrupted status of the
thread and to throw an IOException. Some other cases of
InterruptedException-handling in the same class do a similar thing, though
sometimes returning null or breaking. I found returning null causes NPE's in
other locations so I think it is better to throw an informative exception.
Side thought: I would not be against propagating the InterruptedException all
the way up to the client APIs (HTable/HBaseAdmin), so that users who want to
support interruptable threads do not have to check the interrupted flag. I'd
need to check some more but I have the impression that now sometimes methods
like HTable.get() simply return null when a thread is interrupted.
Some background on good ways of handling InterruptedExceptions can be found
here:
http://www.ibm.com/developerworks/java/library/j-jtp05236.html
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
