Repository navigation
Rank-resolved setup timeline (statistical sampler) and its summary - #88
Merged
Merged
Conversation
…cross rank counts Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…the MPI_INIT flake too
Open MPI 5 on the macOS Intel runner fails to start a launch now and then: either PMIx
cannot find a network interface ("ptl_tool: problems getting address", exit 213) or a rank's
MPI_Init cannot reach the others ("It looks like MPI_INIT failed ... PML add procs failed",
exit 14). #88's two Intel runs each lost one test to the second message, once in
test_pro_cli (two ranks) and once in test/multilevel (four ranks). The retry that caught
the first message lived in test/multilevel/conftest.py, so it covered neither.
The retry now lives in test/conftest.py, wraps every subprocess.run from any test, and
matches both messages; nothing has run when either appears, so relaunching is safe, and
any other failure is returned as it is. The multilevel conftest keeps only its rank-slot
gate, layered over the root retry so a relaunch holds the same slots.
test_mpi_launch_retry.py drives the retry with a fake launcher that fails a given number
of times with a given message (PROTEUS_TEST_MPI_RETRY_DELAY=0 makes it quick).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Tooling for the setup-and-assembly scoping (P0), kept out of #75/#77. Two scripts in `test/multilevel/`, not collected by pytest:
What it found (ceds-model, one socket / 96 cores / 192 threads, hex Q1, ~8.3–9.3k fine nodes per rank): setup 3.06 s on 1 rank to 5.76 s on 64; the same single-rank problem run as 64 independent processes at once, with no MPI, already takes 4.62 s. So about 58% of the growth is the node (and 53% with batched callbacks, 1.26 -> 2.81 s against 2.08 s independent). The rest is the ghost layer (1.24 local nodes per owned), the collective PETSc vector/matrix constructors (0.2–1.2 s) and mesh generation. Written up in `CXX_REFACTOR_setup_and_assembly_SCOPING.md` section 3.3a.
🤖 Generated with Claude Code