tephra-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Poorna Chandra (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (TEPHRA-244) Invalid tx pruning does not handle deletion of tables well
Date Mon, 11 Sep 2017 05:04:00 GMT

    [ https://issues.apache.org/jira/browse/TEPHRA-244?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16160694#comment-16160694

Poorna Chandra commented on TEPHRA-244:

https://github.com/apache/incubator-tephra/pull/55 - This PR just removes regions of deleted
tables when computing prune upper bound. This ensures that transient tables do not block pruning.
However, if a transient table is created/deleted again with the same name and never gets a
chance to be compacted, then it has a potential to block pruning. Filed TEPHRA-254 to handle
such cases.

> Invalid tx pruning does not handle deletion of tables well
> ----------------------------------------------------------
>                 Key: TEPHRA-244
>                 URL: https://issues.apache.org/jira/browse/TEPHRA-244
>             Project: Tephra
>          Issue Type: Bug
>          Components: core, manager
>    Affects Versions: 0.12.0-incubating
>            Reporter: Andreas Neumann
>            Assignee: Poorna Chandra
>             Fix For: 0.13.0-incubating
> Suppose an application regularly creates and deletes tables (for example, temporary tables).
In such a scenario, there will always be such a temporary table when pruning runs, and its
regions will be recorded for that time. However, the region will be deleted before it ever
compacts or flushes, and it will never record prune information. Because all prune times have
such regions, there will never be a set of transactional regions that all have prune info,
and pruning will never happen. 
> The fix is to exclude deleted tables from the list of regions at each time. This is all
regions of deleted tables (tables that do not exist any more), not deleted regions: a region
may disappear due to a split and its data will be in new regions. 

This message was sent by Atlassian JIRA

View raw message