named out of swap on NetBSD/amd64

Hi, I have a local caching resolver running bind 9.16.30 on NetBSD/amd64 9.3. I'm currently hitting it on localhost with approximately 200 qps, and it reliably gets killed after approximately 3 hours with "out of swap" messages in dmesg. The system in question is a Xen VPS with 6 GB RAM and 256

Re: named out of swap on NetBSD/amd64

Greg Choules wrote: > Since the queries are unique the responses should be NXDOMAIN Well, _some_ of them will be NXDOMAIN, many others will be NOERROR or NODATA etc., no? But yes, they all ended up contributing to the cache growing, and it seems that 90% of physical memory all in use by bind

Re: named out of swap on NetBSD/amd64

Greg Choules wrote: > Point taken. Unique does not necessarily mean non-existent and *something* > will end up in cache. So restricting your max-cache-size would seem to be > the thing for you. If it were my server, I would monitor just how much RAM > is getting used in total and adjust

Re: named out of swap on NetBSD/amd64

Michal Nowak wrote: > In your named log you may see a "max-cache-size" calculation like the one > below (I don't have "max-cache-size" set in the config explicitly, implicit > value of "90%" is used): > > 'max-cache-size 90%' - setting to 1729MB (out of 1922MB) Good call - I do see that:

Re: named out of swap on NetBSD/amd64

Jan Schaumann via bind-users wrote: > Greg Choules wrote: > > - Are you stuck on 9.16.30 for some reason? If not, grab the latest 9.18 > > package. It will be less memory hungry generally and contain fixes for > > recent issues. > > Yeah, will give that a try. Upg

Re: named out of swap on NetBSD/amd64

"John W. Blue via bind-users" wrote: > At the risk of stating the obvious .. have you tried 9.16.37 or 9.18.11? I haven't yet, but will give that a try. Thanks! -Jan -- Visit https://lists.isc.org/mailman/listinfo/bind-users to unsubscribe from this list ISC funds the development of this

Re: named out of swap on NetBSD/amd64

Greg Choules wrote: > There could be SO many things going on here. I have a few questions: All good questions, thanks! :-) > - Do you mean 200 QPS or 200,000 QPS? Actually only around 200. I'm effectively looping over a list of names and calling res_query(3). > 200 QPS is background noise

Re: [OFF-TOPIC] Question about ClouDNS (and others') ALIAS records

Karl Auer wrote: > I'm puzzled by the ClouDNS "ALIAS" record. I was wondering if anyone > knows how it is handled "under the hood"? Many DNS service providers have some sort of variation of this, since "aliases at the apex" is a feature many customers need: Akamai uses "Zone apex mapping":