[ 
https://issues.apache.org/jira/browse/CASSANDRA-11949?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15396404#comment-15396404
 ] 

Michael Shuler commented on CASSANDRA-11949:
--------------------------------------------

Are we talking about debian packages or the -bin.tar.gz? Both? 

The debian packages create the directories here: 
https://github.com/apache/cassandra/blob/trunk/debian/dirs
The logs directory is set to /var/log/cassandra, so looking at this: 
https://github.com/apache/cassandra/compare/trunk...mm-binary:11949-2.2
..that is incorrect for a deb installation. Logs should not be written to a 
directory under $CASSANDRA_HOME (/usr/share/cassandra in deb's case), nor 
should a directory be made there by a startup file. This would mean keeping 2 
different cassandra-env.sh files, which isn't the end of the world, but 
unfortunate.

I think the correct thing to do in the case of debian packages is to write 
gc.log to the correct log location, which already exists at deb install time, 
not create a new log dir in the wrong location in -env.sh.

> GC log directory should be created in startup scripts
> -----------------------------------------------------
>
>                 Key: CASSANDRA-11949
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-11949
>             Project: Cassandra
>          Issue Type: Bug
>            Reporter: Joel Knighton
>            Assignee: Mahdi Mohammadi
>            Priority: Minor
>             Fix For: 2.2.8, 3.0.9, 3.9
>
>
> In [CASSANDRA-10140], we enabled GC logging by default, since the overhead 
> was low and asking people providing diagnostics to restart can often make it 
> more difficult to diagnose problems.
> The default GC log path is set to {{$CASSANDRA_HOME/logs/gc.log}} in 
> {{cassandra-env.sh}}, a directory that is not present in a fresh 
> clone/install. Even if logback creates this directory later in startup, it is 
> not present when the JVM initiates GC logging, so GC logging will silently 
> fail for this first Cassandra run
> I haven't tested this in Windows but suspect the same problem may occur. 
> Since lots of tooling around Cassandra won't create this directory, we should 
> instead consider attempting to create it in our startup scripts.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to