traverser: set iter count when request_feasible returns < 0

milroy commented 3 months ago

When diagnosing a recent scheduling problem, the max-match-iters count was observed to remain at its default value:

"succeeded": {
   "njobs": 323,
   "njobs-reset": 323,
   "max-match-jobid": 499941450105867264,
   "max-match-iters": -1,
   "stats": {
    "min": 0.001378509,
    "max": 6.2174810100000002,
    "avg": 1.9132730813622281,
    "variance": 2.3300616684658966
   }
  },

This early return can prevent the count from getting updated: https://github.com/flux-framework/flux-sched/blob/7f0105fcbcb3173be4d105d7ffabb6fa633e58f6/resource/traversers/dfu.cpp#L155

This PR fixes the behavior with a goto to jump to the end of schedule and set perf.tmp_iter_count.

milroy commented 3 months ago

Thanks for the quick review @trws! Setting MWP.

codecov[bot] commented 3 weeks ago

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 75.3%. Comparing base (7f0105f) to head (f3df3c0). Report is 119 commits behind head on master.

Additional details and impacted files

```diff @@ Coverage Diff @@ ## master #1263 +/- ## ====================================== Coverage 75.3% 75.3% ====================================== Files 107 107 Lines 15128 15129 +1 ====================================== + Hits 11405 11406 +1 Misses 3723 3723 ``` | [Files with missing lines](https://app.codecov.io/gh/flux-framework/flux-sched/pull/1263?dropdown=coverage&src=pr&el=tree&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=flux-framework) | Coverage Δ | | |---|---|---| | [resource/traversers/dfu.cpp](https://app.codecov.io/gh/flux-framework/flux-sched/pull/1263?src=pr&el=tree&filepath=resource%2Ftraversers%2Fdfu.cpp&utm_medium=referral&utm_source=github&utm_content=comment&utm_campaign=pr+comments&utm_term=flux-framework#diff-cmVzb3VyY2UvdHJhdmVyc2Vycy9kZnUuY3Bw) | `86.1% <100.0%> (+<0.1%)` | :arrow_up: |

flux-framework / flux-sched

traverser: set iter count when request_feasible returns < 0 #1263

Codecov Report