On Fri, 27 Apr 2007, Henrik Nordstrom wrote: > tor 2007-04-26 klockan 16:56 -0700 skrev Justin Dossey: > > > Some of my users complain that large POSTs in the neighborhood of > > 30-100MB, which worked before we introduced Squid into the mix, now fail > > at some point along the process (it varies). > > Do you know if the failure is after about the same amount of time, or if > that too varies? > > > Notwithstanding HTTP's suitability for uploads, how can I make this work? > > First it needs to be identified how the POSTs is terminated.. The browser receives a 502 Bad Gateway (zero-sized reply). Squid's error header is "ERR_ZERO_SIZE_OBJECT 0". As I mentioned before, bypassing squid results in a 200 instead (some time later). In my tests, I got the 502 after the following times: 709 seconds (11 minutes, 49 seconds) (about 20 MB) 1032 seconds (17 minutes, 12 seconds) (about 29 MB) 610 seconds (10 minutes, 10 seconds) (about 14 MB) I can reproduce this without trouble, but I cannot think of a good way to provide TCP dumps of the communication between the webserver and the squid server because there is so much similar traffic (>100 Mbit) and it takes more than an instant to produce. > Have you changed any of the timeouts in squid.conf? Specifically, what > is the lifetime_timeout set to? I have not changed any timeouts in squid.conf, so the client_lifetime is default at 1 day. I can find no reference to any "lifetime_timeout" configuration directive in the squid configuration manual-- what is this? -- Justin Dossey Operations, podOmatic (415) 975-9981
