Michael Ströder wrote:
HI!
I'm adapting my mtail program for log-based slapd metrics for release 2.5.
2.5 introduces qtime= and etime= in RESULT lines. Great!
I could easily grab histogram metrics for both but that doubles time-series data in Prometheus.
So I wonder what's the difference? Is it worth to always look at both?
qtime is the amount of time an op spent queued before executing. If this number ever gets larger than some milliseconds, that usually means slapd is too busy. Increasing worker threads might help then, but that assumes there are still CPU resources available. (I.e., it would only help if all existing slapd threads are waiting for external responses, e.g. back-ldap waiting for a remote reply. If all worker threads are actively using CPU then there's nothing else that can be done.)
At one point we had a patch that would simply return LDAP_BUSY to the client for this condition, but it seems to have disappeared.
Ciao, Michael.