[ https://issues.apache.org/jira/browse/MESOS-4613?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15136148#comment-15136148 ]
Klaus Ma commented on MESOS-4613: --------------------------------- It makes sense to improve fetcher's log; one option is to let fetcher only log error message to {{stderr}} and let slave log it into slave's log as fetcher's logic is simple and the error message should be enough for us. > Mesos when used with --log_dir generates hundreds of thousands of log files > per day > ----------------------------------------------------------------------------------- > > Key: MESOS-4613 > URL: https://issues.apache.org/jira/browse/MESOS-4613 > Project: Mesos > Issue Type: Bug > Affects Versions: 0.25.0 > Reporter: Lukas Loesche > > We're using mesos with --log_dir=/var/log/mesos > Lately in addition to the mesos-master and mesos-slave log there's also been > mesos-fetcher logs written into this directory. > It seems that every process generates a new log file with a unique file name > containing the date and pid. For mesos-master and mesos-slave this makes > sense. For mesos-fetcher not so much. > On a moderately busy agent it's currently generating 200k log files per day. > On our particular system this would cause logrotate to segfault. And standard > tools like 'rm mesos-fetcher*' won't work because there's too many files to > expand the command. > I also noted that a lot of the created files are zero bytes. So for now we're > running a cron every minute > {noformat} > find /var/log/mesos -size 0 -name 'mesos-fetcher*' -delete > {noformat} > as a workaround. > Anyways it would be nice if there was an option to make the mesos-fetcher > write into a single log file instead of creating thousands of individual > files. > Or if that's easier to implement an option to only write the master and slave > log but not the fetcher logs. -- This message was sent by Atlassian JIRA (v6.3.4#6332)