Our hlog writing 'corrupts' hdfs
--------------------------------
Key: HBASE-509
URL: https://issues.apache.org/jira/browse/HBASE-509
Project: Hadoop HBase
Issue Type: Bug
Affects Versions: 0.1.0, 0.2.0, 0.16.0
Reporter: stack
Priority: Minor
A couple of times during an upload, hdfs complains it is corrupt. Complaint is
as following:
{code}
/hbase/XX.XX.XX-2.u.powerset.com/log_XX.XX.XX.92_1205384328364_60020/hlog.dat.025:
Replica placement policy is violated for blk_2712323855504360379. Block
should be additionally replicated on 2 more rack(s).
/hbase/XX.XX.XX-2.u.powerset.com/log_XX.XX.XX.92_1205384328364_60020/hlog.dat.025:
MISSING 1 blocks of total size 0 B.
{code}
Now the odd thing is that the next time I do a fsck, I see that log number its
complaining about for the above server has increased inline with a new file
just rolled as in:
{code}
......92_1205384328364_60020/hlog.dat.026: Replica placement policy is
violated for blk_4062204433046618058. Block should be additionally replicated
on 2 more rack(s).
/hbase/aa0-005-2.u.powerset.com/log_XX.XX.XX.92_1205384328364_60020/hlog.dat.026:
MISSING 1 blocks of total size 0 B.
{code}
Its no longer complaining about hlog.dat.025. If I do an fsck on the file
hlog.dat.020, it says its healthy, replicated 7M file.
Likely an hdfs issue. Or its the way we're doing our logging? Restart reports
cluster HEALTHY (didn't run fsck with remove 'bad' blocks or files).
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
