Performance Shift(s): `2d578e91`

[github-actions]

Benchmark comparison has identified performance shifts at commit 2d578e9 (#4706). Please review the report below and take corrective/congratulatory action as appropriate 🙂

Performance shift report

       before           after         ratio
     [9308806f]       [2d578e91]
     <main~3>         <main~2>  
-     1.26±0.05ms         808±20μs     0.64  experimental.ugrid.Mesh.time_add_connectivities(6)
-     1.34±0.04ms         905±20μs     0.67  experimental.ugrid.Mesh.time_create(6)
       before           after         ratio
     [9308806f]       [2d578e91]
     <main~3>         <main~2>  
+     5.31±0.07ms       7.25±0.2ms     1.36  cube.AuxFactory.time_add

Generated by GHA run 2223357684

[trexfeathers]

The reductions appear to be genuine

• The results pre/post 2d578e9 are consistent.
• The change is replicated for the larger parameter (1000000 vs 6), although as a smaller proportion which suggests a reduction in overhead rather than any code that scales.
• I have replicated it offline multiple times.

It's hard to know which dependency change has prompted this. Dask Core has changed, but there are no lazy objects involved in ugrid.Mesh.time_add_connectivities - neither any Coords/Connectivitys in the Mesh object, nor the Connectivity that gets added.

[trexfeathers]

Thanks @pp-mo for reminding me that the Connectivity object makes some other uses of Dask in its operations, which explains why even a 'realised' Mesh sees a performance improvement from Dask Core improvements.