> did you tune the tcp timing on the machine? time_wait? close ? etc... No, tcp timing is CentOS default which should be fine for a server at this utilization
> the timeout can be really off the chart sometimes leaving you with hundreds > if not K's of connections open for a reason and will make things stuck in > memeory. That would not account for squid's slow accumulation of memory over the course of days. > since your squid dosnt crash and just ran out of memory and since you dont > have cache_mem specific directive in you settings I think it's using the > basic 256MB which will not benefit you that much for a mem only cache. It will crash if I let it. And it did crash before I found the leak and put in a watchdog. Now I just restart it before it starts eating into swap. > how much ram do you have on that machine? Different machines, 8 GB minimum 32 GB max. Thanks for the GDB wiki, I will use that along with the valgrind one from Amos. On Sep 26, 2012, at 3:41 PM, Eliezer Croitoru <[email protected]> wrote: > On 9/27/2012 12:13 AM, [email protected] wrote: >> Just wish I had some squid development experience so I could easily get into >> a debugging environment and track this down... but squid's a big package and >> I know there will be a big learning curve to start debugging builds. >> >> Thanks >> -Ty > you dont jump into debugging some problem until you confirm there are not > other culprits around the corner. > > it's not just the amount of file-descriptors but maybe some of this memory > squid is using actually is from things like open FD\connections etc. > > did you tune the tcp timing on the machine? time_wait? close ? etc... > > The systems I have seen was using Gentoo 64 bit and Ubuntu 10.04LTS 64 bit. > > you are assuming that things are not the problem and you can try to first > change this since you are at the "Unknown" area yet. > > you can try to take a look at the FD thing at first and it will look like "ok > everything is fine and i'm not running out of them" but there are many things > you need to take in account. > > Most basic linux settings of sockets are not well tuned. > the timeout can be really off the chart sometimes leaving you with hundreds > if not K's of connections open for a reason and will make things stuck in > memeory. > else then mgr:info did you take a look at mgr:mem? which is a more detailed > data on this specific subject. > take a look at: > http://wiki.squid-cache.org/SquidFaq/BugReporting#Using_gdb_debugger_on_a_live_proxy_.28with_minimal_downtime.29 > > which will help you get debug data. > > since your squid dosnt crash and just ran out of memory and since you dont > have cache_mem specific directive in you settings I think it's using the > basic 256MB which will not benefit you that much for a mem only cache. > > how much ram do you have on that machine? > > Eliezer
