[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: accesslog database: overflow or data rotation?



--On Wednesday, May 15, 2019 12:33 PM +0200 Dieter Kluenter <dieter@dkluenter.de> wrote:

Would a small logpurge interval strongly influence the system's
performance?  If it would, a both ways on-the-fly change based on
automatic watch of the actual size of the accesslog database
might be interesting if the high rate writes are a seldom event.
If it wouldn't, setting a small enough interval from the beggining would
do the trick.

It all depends on the syncrepl intervals. The logpurge interval should
consider the timegap between 2 synchronization operations. It is hard to
say to what extend performance will be influenced, you should find out
yourself and set the system optimum, it might vary between a few ours
and a few days.

It depends on how many changes fit within a time period. I always suggest a frequent log purge setting, as this significantly reduces the impact on the system. For example, say that the log purge deletes all entries > 1 day old. If you have 10,000 changes per day that are about evenly distributed, then we end up with purges like this (depending on purge interval):

Purge once a day: 10,000 deletes per purge

Purge twice a day: 5,000 deletes per purge

Purge four times a day: 2,500 deletes per purge

etc.

I usually have the log purge interval set to 4 hours (6 times a day). As I've discussed before, it's better to purge frequently, as server processing comes to a halt while the purge is ongoing. With a single daily purge interval in a production environment that has thousands of changes, I've seen slapd pause for several minutes (I.e., stop processing requests). Changing the configuration to have a purge interval of every 4 hours made it so slapd only paused for less than a second and was not noticable to applications.

--Quanah

--

Quanah Gibson-Mount
Product Architect
Symas Corporation
Packaged, certified, and supported LDAP solutions powered by OpenLDAP:
<http://www.symas.com>