hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jean-Marc Spaggiari (JIRA)" <j...@apache.org>
Subject [jira] [Created] (HBASE-10702) HBase fails to respect Deletes
Date Fri, 07 Mar 2014 22:17:45 GMT
Jean-Marc Spaggiari created HBASE-10702:

             Summary: HBase fails to respect Deletes
                 Key: HBASE-10702
                 URL: https://issues.apache.org/jira/browse/HBASE-10702
             Project: HBase
          Issue Type: Bug
    Affects Versions: 0.94.17, 0.94.15, 0.94.2
            Reporter: Jean-Marc Spaggiari
            Priority: Critical

One of our user contacted me about an issue with Deletes.

Some of the deletes they do are not totally processed. Therefore, after the Delete, if they
do a Get, from time to time, the Get return the row when it should have been deleted and should
have returned nothing. After multiple Deletes, the row is finally deleted. If we don't retry
after the 1st attempt, the row stays there. Even after a flush, a major_compact, etc.

I have been able to reproduce the issue in 0.94.2 (CDH4.2.0 EC2), 0.94.15(CDH4.6.0 EC2) and
0.94.17 (Apache version bare metal)

Here is a simple output from my test app.

1736509 Doing a delete for 0000099676 failed. Start to count
puts=311 deletes=64 retries=2

2281712 Doing a delete for 0000027606 failed. Start to count
puts=3679 deletes=247 retries=2

2388305 Doing a delete for 0000018306 failed. Start to count
puts=4744 deletes=290 retries=2

2532943 Doing a delete for 0000030446 failed. Start to count
puts=5678 deletes=337 retries=2

2551421 Doing a delete for 0000046304 failed. Start to count
puts=5845 deletes=345 retries=2

2561099 Doing a delete for 0000019619 failed. Start to count
puts=5869 deletes=347 retries=3

First field is the time in ms since the test started.  So first error occurs after about 30
minutes. Below are the number of puts and deletes done, and the numbers of required retries
to get the value deleted.

Key is random number between  0000000000 and 0000100000.

Very simple test. Just doing more puts than deletes.

Tests are running on for almost 1h now so it seems to be fine, but it's not on the
same cluster, so I will keep that running for hours/days first.

This message was sent by Atlassian JIRA

View raw message