On Mon, 2009-11-09 at 15:06 +0100, Tremolo 3molo wrote: > Hi, > > We're running pound with a lot of backends on a lot of different > ListenHTTP/HTTPS directives (different certificates). A grep of BackEnd and > wc -l on pound.cfg reveals 732(!). > We also use extremely high TimeOut values, it's really something our > application need. > > Pound version: pound-2.4.4 > > When we for example patch the servers, it take ages (15-20 minutes?) for > some sites to get back up. This only seem to happen in production > environment. (FreeBSD) > > I tried mimic it locally (Linux 2.6) by using the production config file, > but binding to a lot of 127.0.0.0 IP's, but the backends are detected within > a few seconds. > > 1) placed an iptables rule to block all backends of a specific ListenHTTP > (and ListenHTTPS) directive > 2) did a GET / and of course got a 503 thrown > 3) removed the iptable rules and it takes about 5-10 seconds for pound to > detect them as up and deliver the "/" > > The only difference is the OS and the lack of the thousands of requests per > second that the prod env has. > > Could it be that the detection of dead backends is serialized, and our > massive number in conjunction with the high timeout value has this unwanted > effect?
Yes it is possible - if every back-end is particularly slow in responding. We have never considered a scenario with hundreds of back-ends, so we'll probably have to look at it. We'll add it to the 2.6 wish-list. -- Robert Segall Apsis GmbH Postfach, Uetikon am See, CH-8707 Tel: +41-44-920 4904 -- To unsubscribe send an email with subject unsubscribe to [email protected]. Please contact [email protected] for questions.
