Hi Chirs,
I write one test case base on HDFS-6231 patch. You are right, no future is
canceled. But I find the while loop in hedgedFetchBlockByteRange method
execute 30,000 times. Please see hedged-read-test-case.patch in HDFS-6494
jira. I think that may be one problem.
2014-06-10 4:46 GMT+08:00
Lei:
If you can attach the test code from your first email to HDFS-6494, that
would help us know the scenario you were referring to.
Cheers
On Mon, Jun 9, 2014 at 12:06 PM, Chris Nauroth
wrote:
> Hi Lei,
>
> I just reviewed this code path on trunk again, and I couldn't find a
> problem. It ap
Hi Lei,
I just reviewed this code path on trunk again, and I couldn't find a
problem. It appears to me that if one future fails, then the exception
handling logic will allow the other future to proceed without canceling.
Also, I haven't been able to reproduce the infinite loop that you reported
Hi Chris,
I review the patch, I think there is problem in the patch.
Example there are two futures, if the first return futrue is failure and
then the the second future will be cancled.
2014-06-07 3:44 GMT+08:00 Chris Nauroth :
> Hello Lei,
>
> There is a known bug in 2.4.0 that can cause he
Hello Lei,
There is a known bug in 2.4.0 that can cause hedged reads to hang. I fixed
it in HDFS-6231:
https://issues.apache.org/jira/browse/HDFS-6231
This patch will be included in the forthcoming 2.4.1 release. I'm curious
to see if applying this patch fixes the problem for you. Can you try
I use hadoop2.4.
When I use "hedged read", If there is only one live datanode, the reading
from the datanode throw TimeoutException and ChecksumException., the
Client will infinite wait.
Example below test case:
@Test
public void testException() throws IOException, InterruptedException,
Exec