Resolution stress test (>10 seconds offline with warm cache) #10438
Assignees
Labels
great writeup
A wonderful example of a quality contribution 💜
performance
Potential performance improvement
Comments
konstin
added
performance
|
Thank you the perfect reproducer! From the profile this is the same underlying problem as #10430: It's 87% in uv/crates/uv-resolver/src/candidate_selector.rs Lines 181 to 188 in 0d57d29
Crudely patching that out and applying [https://github.com//pull/10439], the bottleneck is still the same code snippet from above:
|
|
That's good news since this is CPU work which we can always improve :) |
konstin
added a commit
that referenced
this issue
Jan 9, 2025
Motivated by #10438. Ideally we wouldn't be calling this code that much in the first place, but it's the more idiomatic way to write this code. Crude-ish benchmark based on #10438 with the `without_markers` patched out: ``` hyperfine --warmup 1 "./uv-1 pip compile scripts/requirements/airflow2-req.in -c scripts/requirements/airflow2-constraints.txt --universal --python-version 3.8" "./uv-2 pip compile scripts/requirements/airflow2-req.in -c scripts/requirements/airflow2-constraints.txt --universal --python-version 3.8" Benchmark 1: ./uv-1 pip compile scripts/requirements/airflow2-req.in -c scripts/requirements/airflow2-constraints.txt --universal --python-version 3.8 Time (mean ± σ): 1.229 s ± 0.018 s [User: 1.206 s, System: 0.111 s] Range (min … max): 1.211 s … 1.269 s 10 runs Benchmark 2: ./uv-2 pip compile scripts/requirements/airflow2-req.in -c scripts/requirements/airflow2-constraints.txt --universal --python-version 3.8 Time (mean ± σ): 873.8 ms ± 5.1 ms [User: 851.2 ms, System: 112.5 ms] Range (min … max): 865.2 ms … 880.4 ms 10 runs Summary ./uv-2 pip compile scripts/requirements/airflow2-req.in -c scripts/requirements/airflow2-constraints.txt --universal --python-version 3.8 ran 1.41 ± 0.02 times faster than ./uv-1 pip compile scripts/requirements/airflow2-req.in -c scripts/requirements/airflow2-constraints.txt --universal --python-version 3.8 ```
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
First, I don't use these requirements anymore, they are based on requirements I used to use ~6 months ago, so this is not a priority for me, but I do use these as a smoke test for uv. These are reduced from the real requirements, which were built to match the Airflow dockerfile, the Python 3.10 Airflow constraints file, and other production and development requirements.
Secondly, there is a big performance difference between my work environment and my home environment I haven't been able to explain, in my home environment I am only able to produce slow performance with fork strategy Requires-Python and python version 3.8, but in my work environment I am able to produce slow performance with both fork strategies and python version 3.10, and in my work environment I see a 20+% slow down between 0.5.15 and 0.5.16 but in my home environment I see almost no difference. I am trying to identify anything other than hardware that might be the difference.
See requirements.in and constraints.txt:
requirements.in
constraints.txt
Run Tests:
uv pip compile requirements.in -c constraints.txt --universal --no-progress --python-version 3.8 > /dev/nulluv pip compile requirements.in -c constraints.txt --universal --no-progress --python-version 3.8 --offline > /dev/nullOn my home machine the second command takes ~13 seconds, on my work machine this takes ~123 seconds.
The text was updated successfully, but these errors were encountered: