Work-around for upstream_fair bug?

[russellneufeld]

Hi all,

  We have nginx set up to reverse proxy to a couple of apache instances.  Every so often apache takes longer than 60 seconds to respond, and then nginx pulls that server out of rotation and does not add it back in.  We are using the upstream_fair plugin, as it does a nicer job of round robining, however we are clearly running into this bug:

"Upstream servers are not reintroduced to the pool after becoming available"
http://nginx.localdomain.pl/ticket/3

  This bug has been open for 16 months.  Doesn't seem like it's going to get fixed any time soon.  Are there any known workarounds?  I tried bumping up max_fails from the default of one, however that didn't seem to help - any single slow response from apache pulls that server out of rotation.  Here is the relevant section of our nginx.conf: 

upstream www.foo.com {
    server public-web1:8000 weight=2  max_fails=2;
    server public-web2:8000 weight=3  max_fails=2;
    fair;
}

  Any suggestions?  Thanks in advance,

    Russ

Posted at Nginx Forum: http://forum.nginx.org/read.php?2,65269,65269#msg-65269