hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ted Tuttle <...@mentacapital.com>
Subject Invalid HFile block magic
Date Tue, 13 Jun 2017 18:02:43 GMT

We are running HBase 1.2.0 from cdh 5.7.0 distribution.

For about 10 of our 200 tables we are seeing client errors like this: https://pastebin.com/wDdyd1cZ

This stack trace led me to this JIRA: https://issues.apache.org/jira/browse/HBASE-11625

It seems this is a bug that has been fixed but we don't have the patch. Unfortunately, immediate
patching of our cluster is not possible.

The 2 comments in the JIRA mention major compaction of involved regions as a workaround. 
Is this a legitimate workaround?  While I've had some point success here, compacting a large
table w/ this problem was not effective.

What else can we do to clean up the cluster?  hbck identifies no issues w/ tables that have
this error.


PS We moved our datacenter some weeks back.  Post-move the cluster was not immediately stable.
 At least once we lost enough nodes that there weren't enough nodes to host all our regions
and the cluster when down completely. After settling in, we experienced 2 weeks of stability
before aforementioned problems began.

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message