Florian Jetter

https://github.com/coiled/benchmarks/actions/runs/10679723905 The results of the A/B run are interesting. I will copy the test cases with some signal here There are a couple of notable improvements already. This shows test_swap_axes_in_memory...

The Client.map perf regression was trivial and should be fixed by now. I only migrated the complex version with kwargs but not the case without. Therefore, we went from "basically...

I'll have to look into this a little more, I think I repeated the benchmarks and got ## Wall time  ## Peak Memory (W)  The only thing I'm...

I suspect that the memory increase, particularly for the P2P cases, is caused by the cache building up but I haven't checked. yet. In absolute values, the increase is much...

So I briefly looked into `test_tiles_to_rows` (that one with the really bad memory result. I ran CPU and memory profiles but cannot find anything that's wrong with it. The caches...

I changed CI configuration to run against main again. https://github.com/dask/dask/pull/11248 should be fine to merge and afterwards we'll retrigger CI

needs https://github.com/dask/dask/pull/11429

Also this https://github.com/dask/dask/pull/11431

The mindeps builds are sad. Everything else seems unrelated

> Update: test_merge failures seem systemic (and thus related). I've been trying to reproduce but without any luck so far