Re: [Queue-developers] Queue 1.40.1 beta released

[Mike C. <da...@ix...>]

On Fri, May 11, 2001 at 07:26:15PM -0400, Werner G. Krebs wrote:
> Yes, on re-reading your older post I noticed that it was in configure.in and have
> now fixed that in the CVS version (along with every last reference to NFS except
> the historical ones in the ChangeLog. So, I think the beast is now dead. I hope.)

Ok.  Things are better now, but I can still get queue hung on a regular
basis.

Now that I have both machines using separate spool directories, I can stay
up.  But queue stops working.

For reference, I am using a test case of compiling the zlib-1.1.3 package.

On my particular set of machines (older P5-233 and a dual P5-90, don't
laugh), I do the following:

make CC=mygcc -jX

Where X varies from run to run, or just a -j.

The reason for this test is as follows:  make -j by itself creates too many
processes and you end up thrashing.  However, make -j going through queue
has the benefit of serializing the jobs, but also spreading them out over
machines.  I use zlib because it's less than 20 files, so doesn't take long
to run through a test cycle.

Now, I use the default profile.  This is what I do:

If I do a make CC=mygcc -j4, it takes just under 24 wall seconds to complete 
(btw, this beats my personal best using my own simple queuing system that I
posted here a few days ago, so queue definitely has possible value).

I do a make clean, then make -j4 again.  And this time it never finishes.
I saw a defunct process and the main queued was stuck in a read (according
to strace).

Or course I can't recreate it using -j4, so instead I do a -j and let them
all run at once.

I see the following now:
:/usr/src/queue/queue-development# ps -ef | grep queued
root      5331     1  0 23:47 ?        00:00:02 queued
nexus     8604  5331  0 23:53 ?        00:00:00 [queued <defunct>]
nexus     8607  5331  0 23:53 ?        00:00:00 [queued <defunct>]
root      8815   753  0 23:53 pts/2    00:00:00 grep queued

And things hang for about, oh, 120 seconds or so (that number sounds
familiar  ;-).

So it seems that queued is losing SIGCHLD's.

I'll run with -D tomorrow.  Midnight.  Bed time.

mrc
-- 
       Mike Castle       Life is like a clock:  You can work constantly
  da...@ix...  and be right all the time, or not work at all
www.netcom.com/~dalgoda/ and be right at least twice a day.  -- mrc
    We are all of us living in the shadow of Manhattan.  -- Watchmen