On Wed, 2017-02-01 at 23:55 +0300, Vitaly Lavrov wrote: > Periodically squid begins to linearly increase the use of the CPU. > Sometimes this process reaches 100%. At random moment of time the CPU > usage is reduced to 5-15%, > and in the presence of client requests can again start linearly > increasing use of CPU. > > In the protocols are no error messages. > > CPU consumption does not correlate with the number of requests and > traffic. > > The increase CPU consumption from 0 to 60% occurs in about 4-5 hours, > and to 100% for 6-8 hours. > > A typical graph of CPU usage can be viewed on http://devel.aanet.ru/t > mp/squid-cpu-x.png > > With the "perf record -p` pgrep -f squid-1` - sleep 30" I have > received the following information: > > At 100% CPU load most of the time took 3 calls > > 49.15% squid squid [.] MemObject :: dump > 25.11% squid squid [.] Mem_hdr :: freeDataUpto > 20.03% squid squid [.] Mem_hdr :: copy > > When loading CPU 30-60% most of the time took 3 calls > > 37.26% squid squid [.] Mem_node :: dataRange > 22.61% squid squid [.] Mem_hdr :: NodeCompare > 17.31% squid squid [.] Mem_hdr :: freeDataUpto > > What is it ? Is it possible to somehow fix it? > > System: slackware64 14.2 > > sslbump not used. http only. > > Part of config: > > memory_pools off > memory_pools_limit 512 MB > cache_mem 768 MB > maximum_object_size_in_memory 64 KB > cache_dir ufs /cache/sq_c1 16312 16 256 > cache_dir ufs /cache/sq_c2 16312 16 256 > cache_dir ufs /cache/sq_c3 16312 16 256 Hi Vitaly, It seems you faced known issue related to linear search through in- memory nodes. See bug report 4477 [1]. [1] http://bugs.squid-cache.org/show_bug.cgi?id=4477 Garri _______________________________________________ squid-users mailing list squid-users@xxxxxxxxxxxxxxxxxxxxx http://lists.squid-cache.org/listinfo/squid-users
