On 20/06/2014 7:20 p.m., Per Jessen wrote: > Amos Jeffries wrote: > >> On 17/06/2014 9:40 p.m., Per Jessen wrote: >>> squid 2.7: >>> >>> Due to a kernel issue with ICMP redirects, my squid has been unused >>> for 5-6 months, but after having found the fix, it's now active >>> again. >>> >>> I'm now seeing quite a few TCP_SWAPFAIL_MISS, which I believe to >>> mean "apparently cached, but cache not found" ? It's not IO error, >>> so what else might this be caused by? >> >> SWAPFAIL is I/O error or corrupted on-disk object. > > Hi Amos, > > There is no other option at all? I am 100% certain there are no I/O > problems on this system. Also, without exception, the SWAPFAIL only > affected openSUSE mirror sites which I cache. I could probably even > reproduce and trace it. > Digging into the code it appears that it can also happen if you have max_open_disk_fds configured and Squid exceeds that limit. Otherwise, no, the only way to reach it is to get an error in the disk I/O event handler when trying to load bytes from a cache file. Amos
