Guido Imperiale

Thanks @rikardn, this helps a lot. An important nitpick though: did you accidentally omit ```python def func(): client = get_client() ``` ? In other words, do `func` (which scatters) and...

> Sorry, yes it is also scattered, I did't think it was important. It isn't. What I meant was if res is actually just the raw output of client.get, e.g....

**Reproduced.** I'm willing to bet that your pseudocode is missing a detail: before you're calling `create_workflow2`, you're dereferencing `dsk1`. My reproducer falls apart almost immediately: ```python from time import sleep...

Ok, found the difference. the deep-copy is NOT tripped on the transfer of the task output from a to b; it's the random seed that's sent from the client to...

This issue applies to all embedded variables that are sent from client to scheduler, e.g. ```python x = client.submit(lambda x: x + 1, np.random.random(1024), key="x", workers=[a]) ```

I'm aiming to merge this PR with these tests still failing, as they are more involved: - test_dataframe_set_index_sync (#8561) - test_default_task_duration_splits (https://github.com/dask/distributed/pull/8563) - test_blocklist_shuffle_split (https://github.com/dask/distributed/pull/8563)

CC @fjetter @milesgranger

The raw XY plot from the system_monitor looks much better. The PR to fix prometheus is straightforward.  I think there's an additional problem Those points at zero during the...

- Follow-up: https://github.com/milesgranger/gilknocker/issues/23

> Will this restart the worker immediately or when the thing that is running releases the Gil again/finishes? > > What I want to know: If my program holds the...