Project

General

Profile

Feature #2184

Different idle timeouts for startup and tail

Added by Parag Mhashilkar about 8 years ago. Updated about 7 years ago.

Status:
Closed
Priority:
Normal
Assignee:
Douglas Strain
Category:
-
Target version:
Start date:
11/21/2011
Due date:
% Done:

0%

Estimated time:
Stakeholders:
Duration:

Description

Hi guys.

I was toying with the idea of changing how the fast the glidein shuts down if idle.
The reason why we have 20 mins now is so the glideins have time to be matched the first round around (packet loss, negotiator busy, etc.)

However, once the glidein is matched, it does not go back to the negotiator, unless there are no more jobs for that user on the schedd.
So if the glidein is not matched again after the schedd runs out of job, it is not a big deal (although we should still want that it happens).
So I would propose we reduce the idle_timeout for the re-match to 400s. (1 min for negotiation cycle + 5 min for one packet loss)

We should be able to do this easily now that we are using the DAEMON_SHUTDOWN expression.

What do you think?
It would certainly reduce drastically the wasted time on the tails... but maybe I forgot some negative aspects? Also, does 400s sound reasonable? Should we make it shortr? Longer?

Thanks,
Igor

History

#1 Updated by Douglas Strain over 7 years ago

  • Assignee set to Douglas Strain

#2 Updated by Douglas Strain over 7 years ago

This has been done and committed:

branch_v2plus_2184 13a2c7a

I just realized I need to update documentation. So, I will do that then assign for review.

#3 Updated by Douglas Strain over 7 years ago

  • Status changed from New to Feedback
  • Assignee changed from Douglas Strain to Parag Mhashilkar

I have added documentation with the second commit:
branch_v2plus_2184 7af878d

Please review.

#4 Updated by Parag Mhashilkar over 7 years ago

  • Assignee changed from Parag Mhashilkar to Douglas Strain
  • Target version set to v2_6_2

looks ok to merge

#5 Updated by Douglas Strain over 7 years ago

  • Status changed from Feedback to Resolved

Merged into v2plus and master.

#6 Updated by Douglas Strain about 7 years ago

Added bcf737f commit with additional term: TotalTimeClaimedBusy =!= UNDEFINED.
Also cherry-pick to master.

#7 Updated by Parag Mhashilkar about 7 years ago

  • Status changed from Resolved to Closed


Also available in: Atom PDF