Skip to content

GitLab

  • Menu
Projects Groups Snippets
  • Help
    • Help
    • Support
    • Community forum
    • Submit feedback
    • Contribute to GitLab
  • Sign in
  • T Trilinos
  • Project information
    • Project information
    • Activity
    • Labels
    • Members
  • Repository
    • Repository
    • Files
    • Commits
    • Branches
    • Tags
    • Contributors
    • Graph
    • Compare
  • Issues 936
    • Issues 936
    • List
    • Boards
    • Service Desk
    • Milestones
  • Merge requests 22
    • Merge requests 22
  • CI/CD
    • CI/CD
    • Pipelines
    • Jobs
    • Schedules
  • Deployments
    • Deployments
    • Environments
    • Releases
  • Monitor
    • Monitor
    • Incidents
  • Analytics
    • Analytics
    • Value stream
    • CI/CD
    • Repository
  • Wiki
    • Wiki
  • Snippets
    • Snippets
  • Activity
  • Graph
  • Create a new issue
  • Jobs
  • Commits
  • Issue Boards
Collapse sidebar
  • James Willenbring
  • Trilinos
  • Issues
  • #893

Closed
Open
Created Nov 30, 2016 by James Willenbring@jmwilleOwner

Address expensive PT pre-push BASIC Trilinos tests

Created by: bartlettroscoe

CC: @trilinos/framework, @trilinos/rol, @trilinos/teko, @trilinos/zoltan

Description

The initial PT pre-push CI test suite (see #482 (closed)) currently has some pretty expensive tests. Looking a the first post-push CI iteration this morning, we see the 2259 tests and their times at:

  • http://testing.sandia.gov/cdash/viewTest.php?onlypassed&buildid=2634468

We see that there are 9 tests that take over 1 minute to run and they drop down to under 30s once you get past the 18 most expensive tests. These tests are:

Test Name Status Time
KokkosContainers_UnitTest_MPI_1 Passed 2m 45s 30ms
ROL_example_PDE-OPT_topo-opt_elasticity_example_02_MPI_4 Passed 2m 35s 300ms
ROL_example_PDE-OPT_0ld_adv-diff-react_example_02_MPI_4 Passed 1m 59s 280ms
Teko_testdriver_tpetra_MPI_1 Passed 1m 36s 880ms
Zoltan_hg_simple_zoltan_parallel Passed 1m 28s 370ms
ROL_example_PDE-OPT_0ld_poisson_example_01_MPI_4 Passed 1m 14s 460ms
Zoltan_ch_simple_zoltan_parallel Passed 1m 7s 150ms
ROL_example_PDE-OPT_topo-opt_elasticity_example_01_MPI_4 Passed 1m 7s 40ms
ROL_example_burgers-control_example_06_MPI_1 Passed 1m 4s 170ms
MueLu_ParameterListInterpreterTpetra_MPI_1 Passed 58s 70ms
ROL_test_sol_solSROMGenerator_MPI_1 Passed 55s 390ms
Teko_testdriver_tpetra_MPI_4 Passed 54s 950ms
KokkosAlgorithms_UnitTest_MPI_1 Passed 48s 850ms
Rythmos_IntegratorBuilder_ConvergenceTest_MPI_1 Passed 44s 730ms
ROL_example_PDE-OPT_ginzburg-landau_example_02_MPI_4 Passed 35s 810ms
KokkosCore_UnitTest_Serial_MPI_1 Passed 35s 220ms
ROL_example_parabolic-control_example_03_MPI_1 Passed 31s 70ms
Rythmos_BackwardEuler_ConvergenceTest_MPI_1 Passed 29s 700ms
... ... ...

After that, all of the remaining tests 2241 of the 2259 tests take less than 30s to run in this build.

What I would like to propose is that we make all of these tests that take longer than 1 minute and either do some work to make them run in under 1 minute or we change their category from the default BASIC to CONTINUOUS using the argument CATEGORY CONTINUOUS.

And I would like to make it a policy that we try to keep the test times for pre-push BASIC tests to under 1 minute. If we can do that, then we can make the pre-push test suite run faster. If you have just a single test that takes 2m45s to run, then no matter how many processes your throw at it, the test suite will never run faster than 2m45s. That is the major motivation here for focusing on individual test times.

As part of this, we can consider how long each package's test suite takes to run by looking at the package-by-package summary this morning at:

  • http://testing.sandia.gov/cdash/index.php?project=Trilinos&parentid=2634468

which when sorted by Test Time looks like:

Package Test time
ROL 6m 34s
Kokkos 2m 45s
Zoltan 2m 1s
Teko 1m 37s
Panzer 1m 28s
MueLu 1m 25s
Zoltan2 55s
Tpetra 54s
Rythmos 51s
... ...

If we could cut down on these tests that take over 1 minute, I think it would also help to shorten the test time for each package as well.

Does anyone from any of these packages have any objections to relegating these tests from BASIC to CONTINUOUS using the argument CATEGORY CONTINUOUS? If so, is it possible to make these run a little faster for pre-push testing?

Note that changing from BASIC to CONTINUOUS does not impact what tests get enabled by default when you configure Trilinos manually since the default test category enabled is NIGHTLY. See:

  • https://tribits.org/doc/TribitsDevelopersGuide.html#test-test-category
Assignee
Assign to
Time tracking