On Wed, May 20, 2015 at 06:25:54AM -0600, Kevin Fenzi wrote: > On Wed, 20 May 2015 14:20:47 +0200 > Adrian Reber <adrian@xxxxxxxx> wrote: > > > Can I get two +1 for this change? > > > > Switch to another URL for mirrormanager haproxy check > > > > The haproxy check URL for the MirrorManager web frontend was a URL > > which resulted in a large DB query. Every proxy, every minute. > > This resulted in two much memory and CPU consumption. This switches > > the check to a small static file to reduce the load on mm-frontend01. > > Well, the problem then is that we don't test that the db is working at > all and just that loading a static file works. ;( > > So, haproxy could mark the service up when the db is down and the page > doesn't work. Yes, but that was how it was before (I know that is bad argument ;-). The old MM landing page was a simple login page without any DB access. The current constellation is good that it tests that the whole thing is actually working, but it does it from all proxies. One check if the DB is working would be enough. The new nagios check for the publiclist which is redirected to /mirrormanager/ is also testing if it correctly works. Most accesses to /mirrormanager are from the haproxies and they are basically taking down MM every few hours. Adrian
Attachment:
pgpX2r5ZwCCqN.pgp
Description: PGP signature
_______________________________________________ infrastructure mailing list infrastructure@xxxxxxxxxxxxxxxxxxxxxxx https://admin.fedoraproject.org/mailman/listinfo/infrastructure