Skip to content

Tests twp-table stalls and twp-native has terrible performance #4805

Closed

Description

@BsAtHome

@grandixximo The tests introduced in #4730 have terrible performance and one stalls the test process completely.

$ time ../scripts/runtests -n ../tests/twp-native/
Running test: ../tests/twp-native
Runtest: 1 tests run, 1 successful, 0 failed, 0 skipped, 0 shmem errors

real	1m12.154s
user	0m3.742s
sys	0m1.705s

$ time ../scripts/runtests -n ../tests/twp-table/
Running test: ../tests/twp-table
^C^C^C [needed to kill from other terminal]Aborted

real	5m33.543s
user	0m1.607s
sys	0m2.054s

Activity

  1. BsAtHome commented on Nov 9, 2026

    @BsAtHome
    ContributorAuthor

    Got twp-table to finish a run in a stellar 4 minutes 56 seconds... And all this while we are trying to speed up tests.
    /me clears tears

    $ time ../scripts/runtests -n ../tests/twp-table/
    Running test: ../tests/twp-table
    Runtest: 1 tests run, 1 successful, 0 failed, 0 skipped, 0 shmem errors
    
    real	4m56.223s
    user	0m1.764s
    sys	0m2.416s
    
  2. grandixximo commented on Nov 9, 2026

    @grandixximo
    Contributor

    Yeah sorry, forgot how long those took, I'll slim them down, trying to go for sub 20s

  3. BsAtHome commented on Nov 9, 2026

    @BsAtHome
    ContributorAuthor

    Any test that needs more than 10 seconds needs to have a very, very good reason to do so. Otherwise, the test needs to be split or seriously rethought to become faster. There is no point in doing parallel tests when we just get them as a long tail of slow tests.

  4. BsAtHome commented on Nov 9, 2026

    @BsAtHome
    ContributorAuthor

    BTW, your change let the CI's tests part go from ~11:10 minutes to 19:10 minutes. That is almost a factor of two.

  5. grandixximo commented on Nov 9, 2026

    @grandixximo
    Contributor

    Thanks, fair point. PR #4807 brings every test from #4730 under 10 seconds here: twp-native and twp-table are split in two, and the longest is now 9 s. twp-table never hung; it finished after about 5.6 minutes, just past where it was stopped. Speeding them up also turned up a race in the kinematics switch at a 1 ms task cycle, which I will look into separately.

  6. added a commit that references this issue on Nov 9, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions