Re: [Havp-users] is the error reporting when MAXSERVERS reached triggering?
Status: Beta
Brought to you by:
havp
From: Philippe L. <st...@fr...> - 2009-12-04 21:37:56
|
Hi Jason, My humble experience let me suppose that you correctly identified both of the problems : 1) Your clients->Squid resquest are higher than the "redirect_program" (HAVP) ones (which could cause some "redirector" connection to simply fail) 2) The CURL sessions you try to test to non-existant URLs means that Squid and/or HAVP seeks - and yes, wait (a long time by default) to handle the DNS/Webserver timeout To resolve part of these issues, you can try to "fine" filter some extensions or MIME types, or file sizes to be scanned by HAVP to decrease the redirector loads, and/or raise the redirector_program processes number. For the second point, you should try by reducing the DNS/Requests default values in both your Squid and HAVP. Note that this could also be due to a systemwide DNS configuration (unavailable DNS servers and so on...) Hope this will help a bit. Jason Haar a écrit : > Hi there > > We've had an issue where we think a HAVP proxy "hung" during a network > outage due to saturation, and I think MAXSERVERS was hit - but the logs > show nothing. This is havp-0.91 under CentOS4 > > So to test, I configured a server with "SERVERNUMBER 1" and "MAXSERVERS > 2" and kicked off 30 curl sessions to different URLs on a non-existent > web server (so that they will hang the proxies until a timeout occurs). > I saw havp climb from 2 processes to 10 - but there's no mention of "All > childs busy" like the source code says? (btw, "children are all busy" > would be better English :-) > > Should that error msg have occurred, and does my suspicion that network > saturation would lead to havp stopping working as all it's "slots" > filled sound correct? My point is that there's nothing in the logs to > show havp was in difficulty, but I'm suspecting there was a problem and > havp wasn't reporting it ("USESYSLOG true") > > This is client->squid->havp->Internet - and all we know is that when the > link gets saturated for 10 minutes, "the proxy" stops working for 30 > minutes (I'm making the times up, but hopefully you get the gist). I > thinking during the saturation, havp fills up with lots of outbound > SYN's, then even though the network clears, havp remains hung as those > SYN's take 3(?) minutes to timeout and free up some slots for new > sessions. Squid shows a few " TCP connection to 127.0.0.1/8080 failed" - > which also should be associated with some form of havp error - but isn't > > Thanks! > > |