Feature: Possibility to suppress slaves from output

When using large number of CPU e.g 720+, each execution of utility prints slaves.

-should this printing be optional ?

On/off would be the first improvement. Could also consider allowing other integer values for collating by host (with pids, or simply the count per host).

mentioned in commit d0898b436bd3a7e74c35d539d542d36adcfa6dcc

closed

Any thoughts about this for a compromise. Report slave information, but in a more compact form? Code attached as patch.

0001-ENH-adjust-infoSwitch-to-report-collated-slaves-rela.patch

@andy @Mattijs

reopened

looks good, but any limitation on number of cpus of any node printed over line?

e.g. 24 cpus of node1

"node1" (1 ... 24)

No line length imposed. I'm not sure how important the pids are anyhow. We could readily just print out the number of processes per node (as yet another selectable option).

Eg,

      nProcs : 18
      Slaves :
      (
          (node1 6)
          (node2 8)
          (node1 3)
      )

Or

      nProcs : 18
      Slaves :
      (
          (6 node1)
          (8 node2)
          (3 node1)
      )

Looks good. I like the 3 choices (print pids, collate per node, suppress) For 2 my preference is

      nProcs : 18
      Slaves :
      (
          (node1 6)
          (node2 8)
          (node1 3)
      )

• pids is only important if something goes wrong
• but knowing the nodes that are have been used is nice. Quite a few customer problems come from e.g. oversubscribing or using nodes that have problems

That is what I was thinking too. When collating with count, maybe makes more sense to include the master node in the count list too. This is a better indication of the subscription. Of course, shouldn't be labeled as Slaves in that case, but Hosts or Procs

mentioned in commit 5873caccfab09b5f2953dcf2d1eb31c489b94a46

mentioned in merge request !150 (merged)

mentioned in commit e6ac4035eca7ac8ace4aa0318f43073151f1d71f

closed