On 05/24/2016 02:32 PM, Amos Jeffries wrote: > On 25/05/2016 5:46 a.m., Heiler Bemerguy wrote: >> Is this normal? > Depends on your normal. > AIUI, that "ERROR:" message means thehre are more than 1024 concurrent > transactions being collapsed by your proxy and the overflowed one is > failing. This is a CF notification queue overflow, not CF table overflow. The size of the former is hard-coded to 1K. In trunk, the size of the latter is configurable via collapsed_forwarding_shared_entries_limit and defaults to 16K. CF queue overflow is usually a sign of one or more of these problems: 1. UDS messages informing workers about certain important queue events are being lost. This is likely if you have not configured your OS UDS memory [correctly]. Unfortunately, Squid does not complain about lost UDS messages [loudly enough]. 2. Squid workers processing queued notifications get blocked on something. This is likely if Squid is overloaded, running out of RAM, and/or using blocking cache_dirs. 3. Squid workers processing queued notifications are slower than workers adding new notifications. This is unlikely, especially if all workers are configured the same. 4. Squid bugs. If you can reproduce the problem, enabling UDS logging may help (debug options ALL,1 54,7), but you will need to sift through a lot of messages (or find somebody willing to look at your logs). HTH, Alex. >> conf: >> >> client_idle_pconn_timeout 30 seconds >> client_persistent_connections on >> server_persistent_connections on >> collapsed_forwarding on _______________________________________________ squid-users mailing list squid-users@xxxxxxxxxxxxxxxxxxxxx http://lists.squid-cache.org/listinfo/squid-users
