Bugzilla – Bug 228
performance degredation/dropping queries after ~30min-1h
Last modified: 2009-01-14 08:40:25 CET
Created attachment 93 [details]
see ticket body
Hi, as I was talking to you on email yesterday, unbound 1.1.2 prerelease that you sent me seems to have some issues. I've changed the config to be 4 threads, set *-caches-slabs into 4 and set each thread to answer 2048 queries outgoing-range 2048 too. However after ~30min-1h of running (~10m queries), unbound reaches ~350mb resident and cpu usage doubles from 15-20% to 30-60%. After this period (around 10 minutes), cpu usage dropped again to 15-20% but the number of queries it processed seemed to decrease. I've not got any proper stats from the server, but have attached a graph of what happens to the mail system. I switched to the new server around 11:00 (there is a slight spike from there to 11:45). Then from 11:45-12, something seems to have been happening with the dns server which causes more messages than normal to be rejected. At 12 I reverted to the old dns server and there is a spike in messages as external systems retry.
When I try switching back to the new server, it seems good for the first minute or two but after that mail slows again. I'll try to get a cache dump and do some more research into this.
Sorry I don't have any more useful information but I'm pretty sure there is a problem here (compared to 1.0.2 anyway).
Hi Mark, how do you measure the number of messages it processes?
The requestlist is different from the number of cache responses.
What do you mean with message getting rejected?
Mail should not be slowing. Can you explain more? Perhaps we should be using email for this.
Fixed in 1.2.0