Guido Imperiale

> Additionally, when the worker process is so borked that it won't even notice that the batched comms have been shut down (e.g. a GIL-holding C package went into an...

Reproduced. Fairly sure it's not safe_sizeof though. ```python import distributed client = distributed.Client(n_workers=1) def f(): d = {} d[0] = d return d fut = client.submit(f, key="x") distributed.wait(fut) # So...

Reproduced with dask=2023.9.3 msgpack=1.0.5 (before #8214). This is not a recent regression.

Even more minimal reproducer: ```python >>> from distributed.protocol import serialize >>> d = {} >>> d[0] = d >>> serialize(d) RecursionError: maximum recursion depth exceeded >>> from collections import UserDict...

Rewriting it as dicts won't work either: ``` ctx = { 'gang1': { 'gangsters': [ {'name': 'Al' }, {'name': 'John'}, {'name': 'Jack'}] }, 'gang2': { 'gangsters': [ {'name': 'Tim'}, {'name':...

Thanks, this works fine: ``` def rstrip(self): "Remove last character" return lambda s: copy.deepcopy(self.renderer).render(s, self.renderer.context)[:-1] ```

> Dask tries to pickle a partial-like class from dask-awkward in order to tokenize it You can prevent pickle serialization by either adding a `__dask_tokenize__` method to your class or...

> d = {"what": b} > normalized = sorted(d.items(), key=str) We can fix this easily.

I'm seeing two different analyses above, and I suspect that only one is correct: 1. @lgray states that the issue is that dask is calling str() on the graph values,...

I typically run `pre-commit run --all-files` several times per minutes and I personally never experience pain with linters other than mypy (which ruff doesn't replace). So for me this is...