[prev in list] [next in list] [prev in thread] [next in thread] 

List:       apache-httpd-users
Subject:    [users@httpd] Re: intermittent timeouts
From:       Alex Harvey <alexharv074 () gmail ! com>
Date:       2011-10-30 12:44:34
Message-ID: CAF0Ep4UxmUOO3yxwqC+PGYAaecxQ237u+_mnusCnOBfX6YFnxg () mail ! gmail ! com
[Download RAW message or body]

Hi List,

We figured this out just in case anyone reads this in the archives.

It turned out we were doing a few things wrong in our setup - one was using
prefork on a busy webserver whereas worker is recommended, and the other
was running the Computer Associates Siteminder plugin despite vendor
recommendations to use the worker handler.  In more detail, it appears that
all incoming connections to our site - including unencrypted requests for
index.html - resulted in connections to a policy server.  So it appears
that three issues combined to cripple our site: (i) prefork simply isn't
the best handler for a busy webserver; (ii) Siteminder & prefork don't
agree with each other (because there is no connection pooling in the
prefork handler, the overheads relating to setting up & tearing down TCP
connections is expensive), and (iii) I suspect that our website code is
badly written so that calls to the policy server are being made
unnecessarily.

NOTE: it turned out that the EAGAIN messages observed in strace were
probably a red herring.  I have seen other hits on google suggesting these
are a problem but I have subsequently concluded that some of these errors
are actually normal - even on site that's not under high load.

Best regards,
Alex

[Attachment #3 (text/html)]

Hi List,<br><br>We figured this out just in case anyone reads this in the \
archives.<br><br>It turned out we were doing a few things wrong in our setup - one \
was using prefork on a busy webserver whereas worker is recommended, and the other \
was running the Computer Associates Siteminder plugin despite vendor recommendations \
to use the worker handler.  In more detail, it appears that all incoming connections \
to our site - including unencrypted requests for index.html - resulted in connections \
to a policy server.  So it appears that three issues combined to cripple our site: \
(i) prefork simply isn&#39;t the best handler  for a busy webserver; (ii) Siteminder \
&amp; prefork don&#39;t agree with  each other (because there is no connection \
pooling in the prefork  handler, the overheads relating to setting up &amp; tearing \
down TCP  connections is expensive), and (iii) I suspect that our website code is 
badly written so that calls to the policy server are being made 
unnecessarily.<br>
<br>NOTE: it turned out that the EAGAIN messages observed in strace were
 probably a red herring.  I have seen other hits on google suggesting these are a \
problem but I have subsequently concluded that some of these errors are actually \
normal - even on site that&#39;s not under high load.<br> <br>Best \
regards,<br>Alex<br>



[prev in list] [next in list] [prev in thread] [next in thread]