Very Computer

by **Charles F. Randa** » Thu, 30 Jul 1998 04:00:00

[Note: Cc'd to author and posted]

>From time to time the server doesn't seem to
>accept connections from the outside world, although there's no reason for
>it... network OK, system load OK (of course going down a lot then).

What's your MaxClients setting in httpd.conf? How many httpd processes
are running when you see this?

Have you checked error_log?

Charles

by **Charles F. Randa** » Sat, 01 Aug 1998 04:00:00

>From emailed replies I noticed that there are more people experiencing this
>problem, but so far nobody seems to know a real solution. I think it is an
>Apache 1.3-problem since the Apache is not reporting errors in the error_log
>about anything wrong with the other settings, say for the number of child
>processes. Accept for the fact that Apache doesn't accept as many connections
>as it should according to the MaxClients setting nothing appears to be wrong.

Are the other people experiencing this problem running Linux too? If so,
track kernel versions, patch levels etc. and try to find a common thread.

Quote:>I've been unable to find a group dedicated to the Apache webserver, but since
>it's used so often I guess I must have overlooked it. Can anyone tell me
>which group it is... If there's anyone outthere that has the same kind of
>experiences, or hints in what direction I should look to solve or identify
>the cause, please post a reply...

Were you running any version of Apache in this same environment prior
to Apache 1.3.x? If so, did you notice any similar problem? If you
really think this is 1.3.x, you could try downgrading to 1.2.6 and
see if your problem goes away.

If this problem doesn't occur in 1.2.6, then report it as a bug in 1.3.x by
using the Apache bug database:

http://bugs.apache.org/

However, I would guess that this is something more subtle related to your
configuration.

Charles Randall

by **Marc Slemk** » Sat, 01 Aug 1998 04:00:00

>From emailed replies I noticed that there are more people experiencing this
>problem, but so far nobody seems to know a real solution. I think it is an
>Apache 1.3-problem since the Apache is not reporting errors in the error_log
>about anything wrong with the other settings, say for the number of child
>processes. Accept for the fact that Apache doesn't accept as many connections
>as it should according to the MaxClients setting nothing appears to be wrong.
>It is a problem for us that is returning on more than a daily basis, but at
>random times... even at the times that our traffic is at the lowest level of
>the 24hr period.

Most of the time such problems have nothing to do with Apache but are
something else broken. Just because Apache isn't giving an error
doesn't mean a problem is an Apache problem!

You need to do some looking at the state of the system when you see
the problem, the exact behaviour you see (eg. if you can connect
but it doesn't respond, if it refuses connections, if connections
time out, etc.), what sort of state the processes are in according
to ps, a strace on a child to see what it is doing, etc.

by **Rasmus Lerdo** » Wed, 05 Aug 1998 04:00:00

Quote:> I've been running other webservers also (also 1.3-releases) on the same
> system configuration. These problems didn't occur then... The problem first
> started occurring when I modified the webserver configuration to run two
> webservers on the same system. When we noticed the problem for the first time
> we were running Apache 1.3b5 with this new webserver configuration. We'd been
> running Apache 1.3b5 for some time before without problems. Upgrading to
> Apache 1.3.0 using the same webserver configuration didn't solve the problem,
> completely changing the webserver config with Apache 1.3.0 didn't solve the
> problem. Upgrading from 1.3.0 to 1.3.1 didn't solve the problem either.

I wonder if somehow there are two accept queues fighting over the same
accept queue lock file. I would have to dive into the code to see if
this idea even makes sense, but as a quick check you could try explicitly
defining your lock files in your two httpd.conf files. Put

LockFile /tmp/accept1.lck

in one and

LockFile /tmp/accept2.lck

in the other.

-Rasmus

by **Marc Slemk** » Wed, 05 Aug 1998 04:00:00

No, the PID is appended to the name.

I'm betting more on this being some Linux bug somewhere... or some
libc thing.

Quote:>> I've been running other webservers also (also 1.3-releases) on the same
>> system configuration. These problems didn't occur then... The problem first
>> started occurring when I modified the webserver configuration to run two
>> webservers on the same system. When we noticed the problem for the first time
>> we were running Apache 1.3b5 with this new webserver configuration. We'd been
>> running Apache 1.3b5 for some time before without problems. Upgrading to
>> Apache 1.3.0 using the same webserver configuration didn't solve the problem,
>> completely changing the webserver config with Apache 1.3.0 didn't solve the
>> problem. Upgrading from 1.3.0 to 1.3.1 didn't solve the problem either.
>I wonder if somehow there are two accept queues fighting over the same
>accept queue lock file. I would have to dive into the code to see if
>this idea even makes sense, but as a quick check you could try explicitly
>defining your lock files in your two httpd.conf files. Put
>LockFile /tmp/accept1.lck
>in one and
>LockFile /tmp/accept2.lck
>in the other.
>-Rasmus

by **Marc Slemk** » Wed, 05 Aug 1998 04:00:00

>At times the server freezes only a small part of the number set in MaxClients
>is running. (say 40 out of 256). Doing a netstat shows lots of connections in
>FIN_WAIT1. Doing an strace of the main pid show absolutely nothing at times
>the webserver is frozen.

Does top show the parent occuping any significant amount of CPU time?

If a strace shows absolutely nothing, and the parent isn't using a lot of
CPU time then that is probably a Linux bug; strace should always
show _something_ (even if it is only one line) unless the code is
in a loop in userspace code. Of course, it could just be strace being broken.

by **Scot E. Wilcoxo** » Sat, 08 Aug 1998 04:00:00

I've also noticed Apache stalling for a few minutes.
During the stall, I can ping, telnet, or FTP between
the same machines. I also have not determined any
cause nor effect.

Very Computer

Unresponsive Apache webserver

Unresponsive Apache webserver

Unresponsive Apache webserver

Unresponsive Apache webserver

Unresponsive Apache webserver

Unresponsive Apache webserver

Unresponsive Apache webserver

Unresponsive Apache webserver