pcp
[Top] [All Lists]

Re: pmlc handshake timeout needed?

To: Nathan Scott <nscott@xxxxxxxxxx>
Subject: Re: pmlc handshake timeout needed?
From: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon, 23 Nov 2009 17:29:40 +1100
Cc: pcp <pcp@xxxxxxxxxxx>
In-reply-to: <1271437514.516191258585382571.JavaMail.root@xxxxxxxxxxxxxxxxxx>
References: <1271437514.516191258585382571.JavaMail.root@xxxxxxxxxxxxxxxxxx>
Reply-to: kenj@xxxxxxxxxxxxxxxx
OK, that means /var/tmp/pmlogger is probably OK, and the pmlogger
process still existed (else it would not accept the connection from pmlc
and you'd never get to hang pmlc).

I thought it might be possible to get away with a new -w timeout option
to pmlc, but they connection to the pmlogger involves a handshake
protocol and one PDU from pmlogger to pmlc ... and this is buried in
libpcp with no way to get a configurable timeout into the __pmGetPDU
call.

So I think this means we need another environment variable, and I
suggest $PMLOGGER_REQUEST_TIMEOUT.

Any objections or better suggestions?

On Thu, 2009-11-19 at 10:03 +1100, Nathan Scott wrote:
> Quick followup - you asked yesterday where the PIDs used here were
> coming from (first arg to pmlc in the ps output earlier) ... they
> are delivered via this shell function:
> 
> get_running_loggers()
> {
>     # Gets the list of pid+host pairs for all running pmlogger processes
>     pminfo -f pmcd.pmlogger.pmcd_host \
>         | perl -ne 'if (/^ +inst \[(\d+).*value "(.*)"/) { print "$1 $2\n" }'
> }
> 
> cheers.
> 


<Prev in Thread] Current Thread [Next in Thread>