Skip to content

Daemon performance seems slow #16745

Open
@shaperilio

Description

@shaperilio

I've written a simple little timing script:

import os
from timing import TicToc # our own timing context
with TicToc('stop daemon'):
    os.system('dmypy stop')
for _ in range(10):
    with TicToc('Run on scratch'):
        os.system('dmypy run scratch.py > boo.txt')
    with TicToc('Modify scratch'):
        with open('scratch.py', 'a') as f:
            f.write('\n\nx: int = 5\n')

scratch.py is a file in a repo with ~100k lines of code. It's a work in progress as far as typing goes; boo.txt ends with Found 2587 errors in 143 files (checked 1 source file)

The intention of the script is to simulate me working on code, e.g. scratch.py, and having type checking run frequently (e.g. every time I save the file).

With default mypy (1.8.0) settings, I get this:

Daemon stopped
[stop daemon] 0.161 seconds
[Run on scratch] 33.603 seconds
[Modify scratch] 0.002 seconds
[Run on scratch] 7.247 seconds
[Modify scratch] 0.002 seconds
[Run on scratch] 3.549 seconds
[Modify scratch] 0.001 seconds
[Run on scratch] 7.029 seconds
[Modify scratch] 0.002 seconds
[Run on scratch] 3.721 seconds
[Modify scratch] 0.001 seconds
[Run on scratch] 6.860 seconds
[Modify scratch] 0.002 seconds
[Run on scratch] 3.705 seconds
[Modify scratch] 0.001 seconds
[Run on scratch] 8.777 seconds
[Modify scratch] 0.002 seconds
[Run on scratch] 3.623 seconds
[Modify scratch] 0.001 seconds
[Run on scratch] 6.872 seconds
[Modify scratch] 0.002 seconds

In other words, the time it takes to check this one file (once the daemon has started) oscillates between ~4 seconds and ~7 seconds, just by adding a single line to scratch.py each time.

Without modifying scratch.py between runs, I get this:

Daemon stopped
[stop daemon] 0.161 seconds
[Run on scratch] 33.603 seconds
[Run on scratch] 4.002 seconds
[Run on scratch] 0.369 seconds
[Run on scratch] 0.364 seconds
[Run on scratch] 0.364 seconds
[Run on scratch] 0.371 seconds
[Run on scratch] 0.364 seconds
[Run on scratch] 0.366 seconds
[Run on scratch] 0.362 seconds
[Run on scratch] 0.362 seconds
  1. Is this oscillation in run time (seen on the first test) expected?
  2. Is it expected to see a second run be much slower than subsequent runs on the same file if nothing is modified?
  3. Is there anything I can do to speed this up, other than setting follow-imports to skip?

Metadata

Metadata

Assignees

No one assigned

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions