On Wed, Feb 27, 2013 at 2:59 PM, Mark Lybarger <mlybarger@xxxxxxxxx> wrote: > We're using apache mod_proxy to proxy requests to jboss servers on > different hosts. We've noticed that we're getting sporadic http 502 > response codes. In the apache error file, I see LOTS of these errors: > > [Tue Feb 26 13:55:49 2013] [error] [client 10.16.0.211] (70014)End of > file found: proxy: error reading status line from remote server > remotejboss-b > [Tue Feb 26 13:55:49 2013] [error] [client 10.16.0.211] proxy: Error > reading from remote server returned by > /STRService/services/SoapToRestSOAP > [Tue Feb 26 13:55:49 2013] [error] [client 10.16.0.211] no acceptable > variant: /usr/local/apache2/error/HTTP_BAD_GATEWAY.html.var > > A bit of digging around, and I find this bug report: > > https://issues.apache.org/bugzilla/show_bug.cgi?id=37770 Danger Will Robinson! You've got a problem with reading responses from your backend. The people posting to that bug also had problems reading responses from backends. Do you all have precisely the same problems, or did you all just get the generic "I couldn't talk to this backend, it returned nothing" error message? > > This bug talks about setting 3 environment variables to resolve the issue. > > SetEnv force-proxy-request-1.0 1 This makes the proxy connection to your backend only use HTTP 1.0. Does your backend application have known problems responding to HTTP/1.1 requests? If not, this is unlikely to fix anything. > SetEnv proxy-nokeepalive 1 Similarly, this causes the proxy connection to not use keepalives. Does your backend have a known problem with keepalive requests? > SetEnv proxy-initial-not-pooled 1 This one may - MAY - be useful if your backend aggressively closes connections. It prevents a race condition where the backend closes the persistent connection after the proxy checks it is alive, but before the proxy sends information to the backend. This is quite rare, and if your backend did exhibit this behaviour, setting this would have fixed it. > > My main problem is that I'm being asked to recreate this in our test > environment before promoting this change to the production apache > server. Recreating this has proven to be VERY cumbersome. I look at > what I can do on both sides of the connection (jboss and apache) to > try and recreate the issue. I see things like connection timeout and > pool size, etc. Any suggestions on how I might be able to get this to > occur in my test area would be GREAT!. > The problem is without knowing why it goes wrong, it is hard to tell you how to reproduce it! You need to analyse your situation and ask the "5 whys". The first "why" is that the backend did not respond to a valid request. What's the next one? Did the backend receive the request at all? If it did, why didn't it produce a response? If it did why didn't it produce it in time? tcpdump between the proxy and the backend may give you an idea as to why. Cheers Tom --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscribe@xxxxxxxxxxxxxxxx For additional commands, e-mail: users-help@xxxxxxxxxxxxxxxx
|