topic Re: DCL to detect runaway processes in Operating System - OpenVMS

DCL to detect runaway processes

Niall Godwin — Tue, 18 Sep 2007 04:10:17 GMT

I'm looking for a way to detect and report runaway processes (i.e. a process in constant COM state) using DCL.

Thank you.

Re: DCL to detect runaway processes

Duncan Morris — Tue, 18 Sep 2007 04:25:56 GMT

Hi Niall,

you can use the attached command procedure as a starter template.

It optionally kills/reports the CPU hogs. You should customise it for your own environment!

Regards,

Duncan

Re: DCL to detect runaway processes

Duncan Morris — Tue, 18 Sep 2007 04:32:07 GMT

Niall,

the last attachment was nor properly formatted.

Hopefully this one is!

Duncan

Re: DCL to detect runaway processes

Jeffrey Goodwin — Tue, 18 Sep 2007 13:13:48 GMT

Duncan,

I think you're missing a procedure:

MONITOR_CPU_CHECK_CAPTURE.COM

-Jeff

Re: DCL to detect runaway processes

Jon Pinkley — Tue, 18 Sep 2007 15:30:37 GMT

Niall,

Are you interested in all processes in COM state, or just ones that are running and competing with interactive users (i.e. ones that are getting CPU time)? The point being that a process being in COM state doesn't necessarily mean it is consuming a lot of resources, in fact, if it is being starved of CPU, that's the state it will be in most of the time.

You may also be more interested in interactive jobs than batch jobs.

Re: DCL to detect runaway processes

Paul Jerrom — Tue, 18 Sep 2007 21:48:23 GMT

Howdy,
I created a similar command file some time ago, but I found that defining a runaway was the most important thing. For example, you want to detect COM states - but on a multiprocessor you might also be looking for CUR states, or even COMO. Also, you might want your target process to be using over a certain threshold of CPU - but what if there are multiple runaways? Probably , your target process will not be clocking any IOs (otherwise it won't be COM/COMO/CUR all the time). So from memory I think my DCL was doing a MONITOR PROC/TOPCPU to an output file for a few seconds, reading the output file to get the top 2 or 3 processes, then looking at each process using f$getjpi to get the state and direct and buffered IOs, then storing those values, hibernating for a while then comparing the figures again. So my definition of a runaway was a process that, after x minutes, was a top 2 or 3 cpu user, was in com, como or cur state , and hadn't clocked any IOs in the intervening period.
Have fun!
PJ

Re: DCL to detect runaway processes

Duncan Morris — Wed, 19 Sep 2007 07:43:45 GMT

Hi,

here is the missing routine as an attachment.

Regards,

Duncan

Re: DCL to detect runaway processes

Steve Schultz_1 — Wed, 19 Sep 2007 09:28:34 GMT

Here is a procedure I have been using. It is based on a process using the cpu more than half of the interval check time. For example, the interval of each check is 30 minutes. If during that time a process has used more than 15 minutes of cpu, then it is flagged. This seems to work well, but note that some backup processes can be cpu intensive.

Steve

Re: DCL to detect runaway processes

Niall Godwin — Wed, 19 Sep 2007 10:45:46 GMT

Many thanks to all for your responses, and especially to Duncan and Steve for their solutions.
I have tested both procedures and they're exactly what I'm looking for.
Duncan's procedure is the more sophisticated, but Steve's is equally effective.
Nice work !
Niall