Performance Shift(s): `44fae030`

[github-actions[bot]]

Benchmark comparison has identified performance shifts at

• commit 44fae030 (#4840).

Please review the report below and take corrective/congratulatory action as appropriate :slightly_smiling_face:

Performance shift report

       before           after         ratio
     [07b223ae]       [44fae030]
     <main~1>         <main>    
+       73.5±10μs         92.4±7μs     1.26  cube.MeshCoord.time_remove(10000)
+      1.78±0.01s       8.15±0.07s     4.57  load.LoadAndRealise.time_load((2, 2, 1000), False, 'FF')
+      1.93±0.01s       8.24±0.09s     4.28  load.LoadAndRealise.time_load((2, 2, 1000), False, 'PP')
+      1.79±0.01s       8.15±0.08s     4.56  load.LoadAndRealise.time_load((2, 2, 1000), True, 'FF')
+      1.93±0.01s       8.25±0.06s     4.28  load.LoadAndRealise.time_load((2, 2, 1000), True, 'PP')
+      1.73±0.07s       8.21±0.08s     4.75  load.STASHConstraint.time_stash_constraint((2, 2, 1000), 'FF')
+      1.88±0.08s       8.30±0.06s     4.41  load.STASHConstraint.time_stash_constraint((2, 2, 1000), 'PP')
+      1.77±0.01s       8.25±0.07s     4.65  load.StructuredFF.time_structured_load((2, 2, 1000), False)

Generated by GHA run 2593231305

[bjlittle]

Well, this is a humdinger...

[trexfeathers]

@wjbenfold had some ideas for improving loading time for files with many fields - there was some looping that was inappropriate. I seem to remember that the proposed changes got watered down a lot, although I can't remember why. Perhaps this issue changes the landscape somewhat, could be worth another visit?

[jamesp]

Oof.

[jamesp]

How does the benchmark infra work @trexfeathers? Does it use the new environment file (python 3.10) included in this PR to generate the numbers in the after column?

It's probably worth checking whether this is an environment regression or a result of the code change.

[trexfeathers]

How does the benchmark infra work @trexfeathers? Does it use the new environment file (python 3.10) included in this PR to generate the numbers in the after column?

It's probably worth checking whether this is an environment regression or a result of the code change.

Correct. In normal circumstances this allows us to distinguish environment vs code change, since we have lockfile-specific commits. Bit difficult in this case!

[trexfeathers]

https://github.com/SciTools/iris/pull/4852#issuecomment-1175951526