[
https://issues.apache.org/jira/browse/ACCUMULO-2575?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13993638#comment-13993638
]
ASF subversion and git services commented on ACCUMULO-2575:
-----------------------------------------------------------
Commit 53fc90fcc783a37df666915cec45f3da3a290c7f in accumulo's branch
refs/heads/ACCUMULO-378 from [~elserj]
[ https://git-wip-us.apache.org/repos/asf?p=accumulo.git;h=53fc90f ]
ACCUMULO-2575 Fix bug in GC where files were being prematurely deleted and add
tests to catch
After realizing the deadlock potential when not having tservers write the
initial
replication entry to the metadata table, the garbage collection code was not
updated
for the shift in data layout. On a clean system, when first beginning
replication,
the replication table will not yet exist, but replication entries will exist in
the
metadata table.
The GC code still assuming that the lack of a replication table implied that
there was no data pending replication and short-circuited early. We need to
check
the metadata table always, and, when present, also the replication table.
> Update GC to not remove files before replication has happened
> -------------------------------------------------------------
>
> Key: ACCUMULO-2575
> URL: https://issues.apache.org/jira/browse/ACCUMULO-2575
> Project: Accumulo
> Issue Type: Sub-task
> Reporter: Josh Elser
> Assignee: Josh Elser
> Fix For: 1.7.0
>
>
> Replication adds to the criteria that define when a file is no longer
> referenced and can be safely deleted.
> Until we have replicated the file to each slave, the GC cannot delete the
> file.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
