Mailing List Archive: [mod_backhand-users] Difficulty getting some stats to show on /backhand/ stats page

Hi all,

I'm having some difficulty getting the #ready servers/# total servers to
display on my stats page. I read the FAQ and simply restarting apache
doesn't help. I am running a large number of modules, so perhaps that
is causing some of the problem.

I have two different clusters I'm playing with.

The first one: a three-machine Linux/Intel cluster I'm just setting
up. Two machines are Cobalt RAQ3's running Apache 1.3.6, the other is a
dual-PIII running Apache 1.3.12. I'm using backhand 1.08.

Strangely, after about 30 minutes or so, the two RAQ's started showing
me the stats for #servers...! However, the PIII has yet to show me
anything. I can see the proc's running, and in the error log backhand
reports the children as you'd expect on all three machines.

In addition, I have yet to see the ~md/req column show anything at all.

The second one: a two-machine Solaris cluster. Solaris 7, apache
1.3.12, backhand 1.08 on both machines. One is an SS20/dual processor,
the other is an LX. Neither of these machines report their #servers
either, but report some strange numbers in the ~ms/req column (negative
on the dual processor machine) which don't ever seem to change. Also,
on the Solaris machines, I don't get anything in the age or CPU idle
columns (always "0" and "0.000000", respectively).

Any insight as to what I could try to get accurate statistics?

Thanks!
Jason
jaherring@usa.net

Jason Herring wrote:
>
> Hi all,
>
> I'm having some difficulty getting the #ready servers/# total servers to
> display on my stats page. I read the FAQ and simply restarting apache
> doesn't help. I am running a large number of modules, so perhaps that
> is causing some of the problem.

That shouldn't cause any problems... make sure that you gracefully
restart not restart. kill -USR1 not kill -HUP -- or -- apachectl
graceful NOT apachectl restart.

> The first one: a three-machine Linux/Intel cluster I'm just setting
> up. Two machines are Cobalt RAQ3's running Apache 1.3.6, the other is a
> dual-PIII running Apache 1.3.12. I'm using backhand 1.08.
>
> Strangely, after about 30 minutes or so, the two RAQ's started showing
> me the stats for #servers...! However, the PIII has yet to show me
> anything. I can see the proc's running, and in the error log backhand
> reports the children as you'd expect on all three machines.

That isn't so strange... As soon a child dies, its replacement will be
fine. a graceful restart will immediately replace all existing children
with new ones.

> In addition, I have yet to see the ~md/req column show anything at all.

It should be 0/0, but I have noticed that it has garbage sometimes.
Still working on that one ;)

> The second one: a two-machine Solaris cluster. Solaris 7, apache
> 1.3.12, backhand 1.08 on both machines. One is an SS20/dual processor,
> the other is an LX. Neither of these machines report their #servers
> either, but report some strange numbers in the ~ms/req column (negative
> on the dual processor machine) which don't ever seem to change. Also,
> on the Solaris machines, I don't get anything in the age or CPU idle
> columns (always "0" and "0.000000", respectively).
> Any insight as to what I could try to get accurate statistics?

The ~ms/req is a known bug (annoying but not damaging). The age
"should" always be 0 (maybe 1), but usually 0. If it is 0, then you are
getting information (meaning the last received message for that node was
less than one second ago). the CPU idle should not be 0. Does this
only happen on the dual processor box?

On a side note, we have found that the byCPU candidacy function is less
than useless. It doesn't appear to do anything useful because the idle
isn't on scale with the load. That said, there is no excuse for it not
working... Except FreeBSD because it is different than BSDI for some
strange reason?!

The load, #a/#t server and a/t memory are the useful statistics.

--
Theo Schlossnagle
33131B65/2047/71 F7 95 64 49 76 5D BA 3D 90 B9 9F BE 27 24 E7