[jira] Commented: (HBASE-1636) disable and drop of table is flakey still

["Bassam Tabbara (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HBASE-1636?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12790376#action_12790376 ] 

Bassam Tabbara commented on HBASE-1636:
---------------------------------------

I just ran into an interesting case after disabling a table and dropping it. Stack asked me to add a comment here and send the logs. Here are the steps I did to the best of my knowledge:

1) Created a table "fragment" that has 250M rows in it spread over 3 region servers.
2) Started the rowcounter program and aborted it with CTRL-C (I did not do a hadoop kill)
3) In the hbase shell  I ran "disable fragment" which failed with a timeout
4) The fragment table was left enabled but all regions were disabled (128 regions over 3 RS)
5) I manually enable a the first region using hbase shell
6) Disabled the table again which succeeded
7) Dropped the table which succeeded.
8) Created a new table with the same name which succeeded.
9) Dropped the table.

There are still 8 regions being hosted by the 3 region servers, however the master does not know about any of these regions.

Here is a scan of the .META.:

hbase(main):001:0> scan '.META.'
ROW                          COLUMN+CELL
0 row(s) in 0.0640 seconds

And the master seems to assign regions to the RS servers:

2009-12-14 13:02:30,630 INFO org.apache.hadoop.hbase.master.RegionManager: Assigning region fragment,\xB1\xF5U6\xCD\xAC\xB5\x2B\x05\xB3\x9D3\xEF\xD8\x83\xD6x\x88\x3F\xAE,1260740563365 to ip-10-212-66-112.ec2.internal,60020,1260824549509

Logs attached.


> disable and drop of table is flakey still
> -----------------------------------------
>
>                 Key: HBASE-1636
>                 URL: https://issues.apache.org/jira/browse/HBASE-1636
>             Project: Hadoop HBase
>          Issue Type: Bug
>            Reporter: stack
>             Fix For: 0.21.0
>
>
> Just now, cheddar up on IRC had table of 2k regions.  A disable and drop gave him 2k rows in meta of historian info.  He couldn't progress.  Had to make below script for him:
> {code}
> meta = HTable.new(".META.")
> historian = "historian:"
> scanner = meta.getScanner([historian].to_java(java.lang.String))
> while (result = scanner.next())
>       meta.deleteAll(result.getRow())
> end
> exit 0
> {code}
> This flakey disable/enable/drop is frustrating users.  Need to fix.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.