Skip to Content.
Sympa Menu

perfsonar-user - Re: [perfsonar-user] Getting rid of old tasks

Subject: perfSONAR User Q&A and Other Discussion

List archive

Re: [perfsonar-user] Getting rid of old tasks


Chronological Thread 
  • From: Mark Feit <>
  • To: "Niederberger, Ralph" <>, 'Andrew Lake' <>, "''" <>
  • Subject: Re: [perfsonar-user] Getting rid of old tasks
  • Date: Fri, 29 Mar 2019 15:10:08 +0000

Niederberger, Ralph writes:

 

Unfortunately all “latencybg” seem to be “(Canceled)” or are “(Pending)”.

 

For throughput tests, I do not see any “Finished” tasks. All outputs of the

 

pscheduler schedule  -PT96H

 

concerning “throughput” state: “(Missed)” or “(Non-Starter)”. Also, none of them are “(Canceled)”

 

Non-starters are often a result of there not being a time slot available to run tests like throughput that require exclusive use of the system.  There’s a schedule parameter called “slip” that allows the scheduler to let the start time be up to a specified amount of time later than the “ideal” start time.  If that’s missing, the scheduler will try to schedule runs of exclusive tests on top of each other and only the first one will get a slot.  Adding “slip”: “PT1H” (or whatever you find appropriate) to your every_4_hours schedule should help relieve some of the congestion.  We recommend putting as much slip as you’re willing to tolerate to maximize the number of runs that get scheduled.  There’s a command called “plot-schedule” that works like “schedule” and produces a PNG that shows what’s on the schedule under what scheduling categories.  That can be useful in visualizing what’s going on.

 

Missed, which means a run never started, is another story and can be the result of a couple of problems.   One is that the runner, which is the part of the system that makes the measurements happen, failed to get the run underway when expected.  The other is when a run has been started but doesn’t end when scheduled, in which case it goes overdue and eventually to missed.  These can happen if your system is exceptionally loaded down, short on resources or simply having problems.

 

  

I also thought it could be related to “rights/priviledges”.

Which processes need which priviledges.

My system is using ldap. Maybe I need to have root priviledges for some processes, which ldap does not know about?  

 

That shouldn’t be an issue.  Any parts of pScheduler that need special privileges configure the system to grant them.  Neither throughput nor latency/latencybg fall into that category.

 

--Mark

 




Archive powered by MHonArc 2.6.19.

Top of Page