accumulo-notifications mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Keith Turner (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (ACCUMULO-1570) Offline tablets that repeatedly fail to minor compact
Date Mon, 15 Jul 2013 13:54:48 GMT

     [ https://issues.apache.org/jira/browse/ACCUMULO-1570?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Keith Turner updated ACCUMULO-1570:
-----------------------------------

    Description: 
If a tablet repeatedly fails to minor compact because of a problem with a user iterator, it
would be nice if it could be taken offline.  The tablet could be unloaded without a minor
compaction (its data would still be in the write ahead log, but tserver memory would be freed)
and marked as broken in the metadata table.  Being marked as broken in the metadata table
would keep the master from loading it and make scans and writes to it fail immediately.  This
approach isolates the impact to the problem tablet/table.  The system could periodically try
to load broken tablets.  Could also provide a way for an admin to force loading a broken tablets
(would be useful if they updated iterators or changed iterator config).

Currently when this happens the system just keeps trying to compact the tablet and memory
fills up on tablet servers.  When memory fills up this starts to impact other tablets.   Also,
it causes writes to the tablet to hang.  





  was:
If a tablet repeatedly fails to minor compaction because of a problem with a user iterator,
it would be nice if it could be taken offline.  The tablet could be unloaded without a minor
compaction (its data would still be in the write ahead log, but tserver memory would be freed)
and marked as broken in the metadata table.  Being marked as broken in the metadata table
would keep the master from loading it and make scans and writes to it fail immediately.  This
approach isolates the impact to the problem tablet/table.  The system could periodically try
to load broken tablets.  Could also provide a way for an admin to force loading a broken tablets
(would be useful if they updated iterators or changed iterator config).

Currently when this happens the system just keeps trying to compact the tablet and memory
fills up on tablet servers.  When memory fills up this starts to impact other tablets.   Also,
it causes writes to the tablet to hang.  





    
> Offline tablets that repeatedly fail to minor compact
> -----------------------------------------------------
>
>                 Key: ACCUMULO-1570
>                 URL: https://issues.apache.org/jira/browse/ACCUMULO-1570
>             Project: Accumulo
>          Issue Type: Improvement
>            Reporter: Keith Turner
>
> If a tablet repeatedly fails to minor compact because of a problem with a user iterator,
it would be nice if it could be taken offline.  The tablet could be unloaded without a minor
compaction (its data would still be in the write ahead log, but tserver memory would be freed)
and marked as broken in the metadata table.  Being marked as broken in the metadata table
would keep the master from loading it and make scans and writes to it fail immediately.  This
approach isolates the impact to the problem tablet/table.  The system could periodically try
to load broken tablets.  Could also provide a way for an admin to force loading a broken tablets
(would be useful if they updated iterators or changed iterator config).
> Currently when this happens the system just keeps trying to compact the tablet and memory
fills up on tablet servers.  When memory fills up this starts to impact other tablets.   Also,
it causes writes to the tablet to hang.  

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message