You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@zookeeper.apache.org by "Rakesh R (JIRA)" <ji...@apache.org> on 2015/01/27 09:08:35 UTC
[jira] [Commented] (ZOOKEEPER-1983) Append to zookeeper.out (not
overwrite) to support logrotation
[ https://issues.apache.org/jira/browse/ZOOKEEPER-1983?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14293154#comment-14293154 ]
Rakesh R commented on ZOOKEEPER-1983:
-------------------------------------
Thanks [~shyamal], patch looks fine. Just one thought, with append support the out file will keep grow. One way is to provide an option of {{MaxFileSize}} threshold and then rollover based on that, right ? Also, welcome better ideas.
> Append to zookeeper.out (not overwrite) to support logrotation
> --------------------------------------------------------------
>
> Key: ZOOKEEPER-1983
> URL: https://issues.apache.org/jira/browse/ZOOKEEPER-1983
> Project: ZooKeeper
> Issue Type: Bug
> Components: server
> Affects Versions: 3.3.5, 3.3.6, 3.4.6
> Environment: CentOS 5.x (and probably any Linux distribution for that matter)
> Reporter: Shyamal Prasad
> Assignee: Shyamal Prasad
> Fix For: 3.5.1
>
> Attachments: ZK1983.patch, ZOOKEEPER-1983.patch
>
>
> Currently zkServer.sh will redirect output to zookeeper.out using a simple shell redirect.
> When logrotate (and similar tools) are used to rotate the zookeeper.out file with the 'copytruncate' semantics (copy the file, truncate it to zero bytes) the next write results in a sparse file with the write at the offset of the last file. Effectively the log file is now full a null bytes and it is hard to read/use the file (and the rotated copies).
> Even worse, the result is zookeeper.out file only gets "larger" (though sparse) and after a while on a chatty system it takes significant CPU resources to compress the file (which is all nulls!)
> The simple fix is to append to the file (>>) instead of a simple redirection (>)
> This issue was found in a 3.3.5 production system, however code in trunk has the same issue.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)