Lawrence Mitchell

I'm not sure that `jump` can automagically do the right thing in this case without inspecting the provided expression, which is a recipe for pain. Perhaps it would be OK,...

> Curious how was it determined compression was happening with GPU data? Asking as this is [explicitly disabled in Distributed](https://github.com/dask/distributed/blob/3551d1574c9cd72d60197cc84dd75702ebcfec54/distributed/protocol/cuda.py#L28) and has been this way for a while. Would be...

> Curious how was it determined compression was happening with GPU data? Asking as this is [explicitly disabled in Distributed](https://github.com/dask/distributed/blob/3551d1574c9cd72d60197cc84dd75702ebcfec54/distributed/protocol/cuda.py#L28) and has been this way for a while. Would be...

> > Curious how was it determined compression was happening with GPU data? Asking as this is [explicitly disabled in Distributed](https://github.com/dask/distributed/blob/3551d1574c9cd72d60197cc84dd75702ebcfec54/distributed/protocol/cuda.py#L28) and has been this way for a while. Would...

We can monkey patch past this error in dask-cuda, by patching `protocol.loads` and `comm.utils.from_frames`: ```from dask_cuda import LocalCUDACluster from dask_cuda.initialize import initialize from distributed import Client import cupy import dask.array...

> FAILED tests/test_local_cuda_cluster.py::test_pre_import_not_found This one seems to be because scaling up a speccluster now swallows any errors when scaling up. (https://github.com/dask/distributed/issues/8309)

> FAILED tests/test_local_cuda_cluster.py::test_death_timeout_raises This is the same cause.

/ok to test

Apologies, this is not really a concrete proposal, but some slightly woolly thoughts > There is a temptation to create RAPIDS specific subclasses of these here, for example we could...

In 23.04, importing RMM will no longer hook up cupy, numba, and pytorch to use the RMM allocators (that was https://github.com/rapidsai/rmm/pull/1221). That was mostly done for import time reduction, but...