Re: [BProc] error mesages..cont

[Erik A. H. <er...@he...>]

On Wed, Jan 23, 2002 at 04:13:33PM -0500, henken wrote:
> Here is the latest error message. I am getting this with 3.1.5 + patch.
> Feb  4 17:55:30 master /usr/sbin/bpmaster: write(ghost): missing process
> for message type 14 req; to=3,7576 from=1,7576 result=0
> 
> Here is the relevant ps output
> [henken@master henken]$ ps -jx
>  PPID   PID  PGID   SID TTY      TPGID STAT   UID   TIME COMMAND
> 22401  7573 22385 22385 ?           -1 S    27659   0:00 bpsh 0
> /home/henken/cvs/jobs/bin/noop
>  7573  7576 22385 22385 ?           -1 RW   27659   0:00 [noop]
> 20876  7624  7624 20826 pts/2     7624 R    27659   0:00 ps -jx
> 
> 
> You can see that the messag is tryin to get to the process that bpsh
> started on the remote node.

Actually, it's an EXIT message from the remote process back to the
ghost on the front end.

message type 14 req; to=3,7576 from=1,7576 result=0
14   = EXIT
3    = route to ghost
7576 = pid of ghost
1    = route to real   (from real in this case)
7576 = pid of real process
0    = result - in this case indicating exit status 0 - normal exit.

That's perfectly normal since "noop" exits on the remote node and
needs to tell the ghost to do the same.  The question is why the
process on the front end doesn't seem to be a ghost.  Looking at your
ps output, it does look like it's probably ghosted since it's has no
memory space swapped in.  It's in the W state.

Does "noop" stay in that state pretty much forever?  What's in
/proc/<pid>/maps for noop at this point?

I really wish I could see a message trace for this.  A trace of just
the move and exit messages would be pretty useful here if possible.
Hopefully that would get the message bulk down far enough to be
managable.  You can filter by doing something like:

bpmaster -d -m - | egrep 'move|exit' > file

Basically, I want to know what the result on the move response for
"noop" was in this case.  If it's non-zero and the remote went on for
some reason, or if there's no error and the ghost screwed up somehow.

I don't think it should be a problem, but I could imagine this being
caused by the exit and move response getting out of order.

- Erik
-- 
Erik Arjan Hendriks          Printed On 100 Percent Recycled Electrons
er...@he...                   Contents may settle during shipment