When the instance domain of a logged metric changes a little, pmlogger
appears to dump the whole reported set to the .meta file, not just the
differences. This can make the .meta file grow quite rapidly.
For example, take my config file, please:
% cat config.pmlogger.procs
log mandatory on default { proc.psinfo.cmd }
% pmlogger -r -u -t 10 -c config.pmlogger.procs foo &
% grep Metric pmlogger.log
Metric proc.psinfo.cmd logged every 10 sec: 4128 bytes or 34.01 Mbytes/day
[... after a few minutes kill the logger ...]
% pmloglabel -L foo
Log Label (Log Format Version 2)
Performance metrics from host very.elastic.org
commencing Mon Jan 27 22:00:22.769 2014
ending Mon Jan 27 22:07:42.772 2014
Archive timezone: EST+5
% ls -al foo*
-rw-r--r--. 1 fche users 183660 Jan 27 22:07 foo.0
-rw-r--r--. 1 fche users 312 Jan 27 22:07 foo.index
-rw-r--r--. 1 fche users 76675 Jan 27 22:07 foo.meta
% strings foo.meta |sort | uniq -c
[... note a great deal of duplication ... about eight sets in this one ...]