Avoid overflow in statitics.mean

[mrocklin]

I don't know why, but for some reason statistics.mean was overflowing in CI. See https://github.com/dask/distributed/actions/runs/3741526593/jobs/6351258185

I'm trying a naive implementation instead. It also happens to be faster and simpler.

In [1]: from statistics import mean

In [2]: x = list(range(1000))

In [3]: %timeit mean(x)
196 µs ± 777 ns per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

In [4]: %timeit sum(x) / len(x)
4.82 µs ± 15.2 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)

[github-actions]

Unit Test Results

See test report for an extended history of previous test failures. This is useful for diagnosing flaky tests.

       20 files  +         8         20 suites  +8   9h 0m 52s ⏱️ + 4h 41m 15s
  3 271 tests ±         0    3 184 ✔️ +         9       85 💤  -   10  2 ❌ +1 
33 337 runs  +13 057  31 907 ✔️ +12 397  1 426 💤 +657  4 ❌ +3 

For more details on these failures, see this check.

Results for commit 4e90750. ± Comparison against base commit 3ac8631.

[mrocklin]

This is trivial enough and coming up enough in current PRs' CI that I plan to merge tomorrow US-time if there are no comments.

[mrocklin]

Woot. Thanks

[fjetter]

This is still an issue but now in the new code...

  File "d:\a\distributed\distributed\distributed\dashboard\components\shared.py", line 569, in update
    self.label_source.data["memory"] = [
  File "d:\a\distributed\distributed\distributed\dashboard\components\shared.py", line 571, in <listcomp>
    f.__name__, dask.utils.format_bytes(f(self.source.data["memory"]))
  File "d:\a\distributed\distributed\distributed\dashboard\components\shared.py", line 562, in mean
    return sum(x) / len(x)

Looks like x is sometimes a numpy array and I assume we're overflowing int64??? (or we're using a different dtype somewhere, int64 overflow sounds crazy even if not impossible)

(the above screenshot is not reproducing, just a snapshot showing the data, I believe this is RSS memory, and the dtype)

[fjetter]

At the very least, this is giving exactly the same warning

In [1]: import numpy as np

In [2]: sum(np.array([2**63-1, 1], dtype=np.int64))