impala-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Dan Hecht (Code Review)" <ger...@cloudera.org>
Subject [Impala-ASF-CR] IMPALA-5386: Fix ReopenCachedHdfsFileHandle failure case
Date Wed, 31 May 2017 19:01:58 GMT
Dan Hecht has posted comments on this change.

Change subject: IMPALA-5386: Fix ReopenCachedHdfsFileHandle failure case
......................................................................


Patch Set 2:

> > Did you look into the conditions which triggered the failure to
 > > begin with ? Is there any way to trigger similar error locally
 > with
 > > debug action or stress flag ? It would be good to add a test for
 > > this case.
 > 
 > I know the sequence of events:
 > 1. File is deleted using hdfs command line
 > 2. Run a query over the table that has the deleted file
 > 3. ScanRange::Open succeeds (!!)
 > 4. ScanRange::Read tries hdfsRead and fails, destroys the file
 > handle, and reopening the file handle fails. The ScanRange's file
 > handle reference is now invalid, but it is also non-null.
 > 5. Query is aborted, leading to ScanRange::Cancel
 > 6. ScanRange::Cancel calls ScanRange::Close, which sees that the
 > file handle reference is non-null and tries to release it. The
 > release fails, because the file handle reference is invalid.
 > 
 > The problem with reproducing this on normal Hdfs is that when a
 > file is deleted, the subsequent Open in #2 fails, so the query
 > never even has a file handle. If the Open succeeded, it is a file
 > handle to a local file, so POSIX guarantees that the file stays
 > around. I have tried modifying the code and modifying the test to
 > produce this sequence, but it is difficult to get this particular
 > combination. I'm looking at what it would take.

Maybe a stress option that artificially causes hdfsRead() to appear to have failed? Given
that this path doesn't get exercised, it does seem like we need some way to exercise it.

-- 
To view, visit http://gerrit.cloudera.org:8080/7020
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-MessageType: comment
Gerrit-Change-Id: Iee982fa5e964f6c8969b2eb7e5f3eca89e793b3a
Gerrit-PatchSet: 2
Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-Owner: Joe McDonnell <joemcdonnell@cloudera.com>
Gerrit-Reviewer: Dan Hecht <dhecht@cloudera.com>
Gerrit-Reviewer: Joe McDonnell <joemcdonnell@cloudera.com>
Gerrit-Reviewer: Michael Ho <kwho@cloudera.com>
Gerrit-HasComments: No

Mime
View raw message