Iaroslav Igoshev

> future = c.submit(f1, (f2, 100)) @noloerino, yes, now I see the same as you but I am not sure that is correct enough arguments pass to a remote function....

> Updated benchmarks on the above `df.abs()` code (avg of 10 runs for each): > > * **Ray, master (fd225c52):** 0.014s > * **Dask, master:** 0.329s > * **Ray, latest...

I suggest getting rid of the Invocable stuff for now because I am still seeing the significant performance penalties on the workload I am currently profiling. You can find it...

@noloerino, please make CI green, after that I will run the benchmark in order not to encounter any trouble.

**master** whole time: 236.58521463605575 **PR4715** whole time: 281.4897154511418

Let me look into the changes.

Now the perf is on par with master on Ray engine, but there is a little slowdown on Dask engine.

Related discussion on handling metadata (index and columns) in https://github.com/modin-project/modin/issues/3673.

If we do not want to pass lengths from the QC to the core dataframe, I would suggest renaming `func_may_change_complementary_index_size` to `preserve_opposite_lengths` and passing it from `apply_full_axis` to `broadcast_apply_full_axis`, where...

> `>2x gain on my machine with 8 cores, compared with master, both for Ray and Dask` @anmyachev, are we still slower than pandas on that bench?