Ivan . wrote:
Hi Had this running on a RedHat EL5 64bit OS running Squid v3.0.STABLE16 for about 3 days, with 8GB of memory. Slowly but surely "top" shows the availble memory dropping down to 500MB, which concerned me a great deal. I am not caching, using the cache_dir null directive, so not sure what is going on other a memory leak. Restarting the squid process didn't help, so I bounced the box and low and behold available memory is around 7GB.
Um ... Restarting Squid drops all the memory it has allocated, whether leaked or not. Same as killing the process.
This sounds very much like something I saw back in the 2.6.31 kernel last year. Any app that used a lot of memory or connections slowly (relative) leaked RAM into the kernel space somehow. Only a system restart or kernel upgrade to 2.6.32 fixed it here.
Amos -- Please be using Current Stable Squid 2.7.STABLE9 or 3.1.1