apache 1.1.1 gives up under heavy loads

[David J MacKenzie]

We've found that apache 1.1.1 running on BSDI BSD/OS 2.1 can't handle
heavy loads. We run a hundreds of low to medium traffic servers which
seem to be fine, but every day a couple of our dozen or so high
traffic servers get into a state where all the child processes except
one are zombies, and that last one is sleeping on network I/O.
The server doesn't accept connections anymore, and we have to kill and
restart it manually.

We didn't have this problem with 1.0.5, configured similarly except
without the new info, status, and anon_ftp modules, and without using
mmap for the scoreboard.

Has anyone else seen this failure mode?

[David J MacKenzie]

On 12 Aug 1996 00:52:41 -0400, d...@va.pubnix.com (David J MacKenzie) said:

> We didn't have this problem with 1.0.5, configured similarly except
> without the new info, status, and anon_ftp modules, and without using
> mmap for the scoreboard.

And we turned keepalives on in 1.1:

# KeepAlive: The number of Keep-Alive persistent requests to accept
# per connection. Set to 0 to deactivate Keep-Alive support
KeepAlive 25

# KeepAliveTimeout: Number of seconds to wait for the next request
# (idle timeout)
KeepAliveTimeout 5

[David Miller]

David J MacKenzie (d...@va.pubnix.com) wrote:
: We've found that apache 1.1.1 running on BSDI BSD/OS 2.1 can't handle

: heavy loads. We run a hundreds of low to medium traffic servers which
: seem to be fine, but every day a couple of our dozen or so high
: traffic servers get into a state where all the child processes except
: one are zombies, and that last one is sleeping on network I/O.
: The server doesn't accept connections anymore, and we have to kill and
: restart it manually.

I'm seeing something very similar under 1.05. It was my assumption that
the child processes would stack up waiting for IO on dead connections.

HOWEVER, close examination with netstat and fstat shows that no processes
have the dead connections open. IE, I might have 20 stale sockets but no
httpd process has them open. In addition there's no correlation between #
of zombie processes and stale network connections.

: We didn't have this problem with 1.0.5, configured similarly except

: without the new info, status, and anon_ftp modules, and without using
: mmap for the scoreboard.

: Has anyone else seen this failure mode?

This does not encourage me:( I was planning to upgrade to 1.1.1 to make
the problems go away:) I'll at least upgrade to 1.1.1 before debugging -
there's no sense debugging an old version.

--
David Miller Usual disclaimers apply

[Parthiv Shah]

David J MacKenzie wrote:
>
> We've found that apache 1.1.1 running on BSDI BSD/OS 2.1 can't handle
> heavy loads. We run a hundreds of low to medium traffic servers which
> seem to be fine, but every day a couple of our dozen or so high
> traffic servers get into a state where all the child processes except
> one are zombies, and that last one is sleeping on network I/O.
> The server doesn't accept connections anymore, and we have to kill and
> restart it manually.
>

> We didn't have this problem with 1.0.5, configured similarly except
> without the new info, status, and anon_ftp modules, and without using
> mmap for the scoreboard.
>
> Has anyone else seen this failure mode?

I started running 1.1.1 but the problem is that whenever the cgi file is
used to post it says server encountered an error, post is not defined
or something.. 1.0 post was default defined, where is it defined in
1.1.1 how can I make it default????

Parthiv
--
Shah, Parthiv http://www.genious.com/
par...@admin.idt.net (201) 928 - 4414
System Administrator at IDT pgp? -> finger par...@haven.ios.com