syrupy >5.5 running slow/stalling for large snapshots

[ubdbra001]

Hi,

I'm working on an inherited codebase that has a large snapshot file (~33MB). Versions of syrupy prior to 5.5 had no issue dealing with this (test suite completed on GHA in ~ 5-10 mins), now the test suite is running into the hours. I've verified this is specifically due to syrupy by skipping the snapshot tests, and that it only occurs in version 5.5 by comparing it to older versions.

Looking at the release notes didn't give me an idea of why this may have happened. Would you be able to shed any light on what's changed in 5.5 that would cause this?

Thanks!

[noahnu]

Are you using xdist in your project?

What other plugins do you have enabled? What's the nature of the large snapshot file? Is it an amber file? Single test case?

[ubdbra001]

Yes, using xdist.

Other plug-ins: pytest-cov, pytest-github-actions-annotate-failures, pytest-mpl, pytest-profiling, pytest-xdist, filetype

It is an amber file, and it covers multiple tests cases (all test cases in a single file). There are a lot of repeated 0 values (see below) in there that I suspect 1. are not needed, and 2. causing the large file size. So, medium term I'm looking into how the size can be cut down.

But in the meantime I want to work out if I should just pin 5.4 as a dependency, or if there is something I can do with 5.5 to go back to previous performance.

---

Example:

# name: test_add_branches_to_labelled_image[catenane].10
  0
# ---
# name: test_add_branches_to_labelled_image[catenane].100
  0
# ---
# name: test_add_branches_to_labelled_image[catenane].1000
  0
# ---
# name: test_add_branches_to_labelled_image[catenane].10000
  0
# ---

Note: name: test_add_branches_to_labelled_image[catenane] and name: test_add_branches_to_labelled_image[catenane].1 contain arrays

[noahnu]

To clarify, the individual snapshots for the test cases are small there's just a lot of tests? Or are there some very large multi-MB individual test cases?

I suspect there's a perf bottleneck around the way we now write to the xdist worker output and merge reports in #1132. Will have to profile that. I can take a look a bit later.

Breaking up the file may work as a short term solution. It should be doable with a custom snapshot extension if you want to leave the test file itself unsplit.

[noahnu]

#1156

[ubdbra001]

There are 4 test cases in the file: snapshots for two are small (2 and 10), the other two are very large (100k each, all as in the example above aside from the first 2-3 snapshots for that test).

Breaking up the file may work as a short term solution. It should be doable with a custom snapshot extension if you want to leave the test file itself unsplit.

Okay, I'll have a look into doing that. Thanks!