Re: [HTCondor-devel] Need for dynamic slots in top-level collector?


Date: Mon, 13 Mar 2017 16:19:46 +0000
From: John M Knoeller <johnkn@xxxxxxxxxxx>
Subject: Re: [HTCondor-devel] Need for dynamic slots in top-level collector?
No, no current way to reset stats without a daemon restart.   I've been thinking about ways to add a reset. 

What do you think about a knob.  RESET_DC_STATS_ON_RECONFIG.    There are multiple stats collections, so having that knob reset ALL of the stats would take some work.  But just resetting the DC stats and collector stats would be easy.

-tj

-----Original Message-----
From: Brian Bockelman [mailto:bbockelm@xxxxxxxxxxx] 
Sent: Monday, March 13, 2017 11:12 AM
To: John M Knoeller <johnkn@xxxxxxxxxxx>
Cc: Greg Thain <gthain@xxxxxxxxxxx>; Todd Tannenbaum <tannenba@xxxxxxxxxxx>; htcondor-devel@xxxxxxxxxxx
Subject: Re: [HTCondor-devel] Need for dynamic slots in top-level collector?

Yeah - but you can safely ignore all those "maximum" statistics.  The machine was heavily swapping, meaning all sorts of hell was breaking loose.

Is there a way to clear the statistics counters without restarting the process?

Brian

> On Mar 13, 2017, at 11:10 AM, John M Knoeller <johnkn@xxxxxxxxxxx> wrote:
> 
> To be clear.  of the 20227 forked queries, the slowest was 19 seconds.  That's 19 seconds forking, there is no stat for how long the child was alive and handling the query.  The child knows how long it spent, and that info goes into the log, but the top level collector is only recording how long it spent forking the child.  
> 
> Also that 19 seconds might have been a child that was OOM killed and never finish forking. 
> 
> -tj
> 
> 
> 

[← Prev in Thread] Current Thread [Next in Thread→]