Hi guys, I have received a bunch of comments and suggestions in my personal inbox from people that didn't want to participate in the thread directly. Some of them are absolutely valid points IMHO. So here they are: On Mon, 2008-11-10 at 20:09 +0100, Fabio M. Di Nitto wrote: > == Output == > > to_file: > > echo $(date "+%b %d %T") subsystem_or_daemon: entry_to_log > Nov 10 19:46:40 subsystem_or_daemon: entry_to_log Quoting: -- I've not read any of the threads but did you consider a date format that sorts easily i.e. YYYY MM DD-based? And for investigating races across files/nodes using sub-second timestamps (e.g. 19:46:40.123 ) and adding the node name. So you can just extract relevant sections from several files, cat them all together, sort them and then review the sequence of happenings. If you're generalising, make the log format string a customisable option similar to apache. -- I agree that a more precise time and sortable format is a major winner. He also has a good point to add the node name to the log. My reply to the "customisable format" request was based on the discussion we had at the Summit (that we want to avoid parsing lots of different logging etc.) and his reply was: Quoting: -- 1. Most people will use the default or one of the example lines you supply. 2. If it is a problem, you've got the corresponding config line so it should be straightforward to have a utility to convert it back to your preferred canonical output format. (And yes, "Nov 14" format can also be parsed and converted back into a sortable format without much difficulty, so it's no big deal.) 3. If lots of people change the log format, then you probably chose a poor default. 4. There's no reason why log-to-file should exactly match syslog format. If you want syslog format, use syslog. You want the best format to assist you in debugging problems etc. -- Fabio -- Linux-cluster mailing list Linux-cluster@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/linux-cluster