Closed fjetter closed 8 months ago
Yikes and only for P2P, I'll take a look
I am confused, the dask and distributed commits are the same an the only thing that changed with dask-expr is https://github.com/dask/dask-expr/pull/971 which doesn't touch the code-paths of this benchmark. Even more confusing, the first regression above was with exactly the same dask-expr run as the expected runtime before
performance is back to normal on latest run, closing here
Haven't seen this pop up in another issue but browsing the data, this is a rather severe regression
cc @phofl @hendrikmakait