Fix NBodySimulator under v7 stack: shrink liquid_argon problem size#1570
Draft
ChrisRackauckas-Claude wants to merge 1 commit into
Draft
Fix NBodySimulator under v7 stack: shrink liquid_argon problem size#1570ChrisRackauckas-Claude wants to merge 1 commit into
ChrisRackauckas-Claude wants to merge 1 commit into
Conversation
ChrisRackauckas-Claude
force-pushed
the
fix/NBodySimulator-v7
branch
from
2fa27b2 to
d4ea4fb
Compare
ChrisRackauckas
added a commit
that referenced
this pull request
May 30, 2026
detect-changed-benchmarks.sh fetches the PR base at --depth=1 and then runs a three-dot diff (origin/master...HEAD), which requires a merge base. Once master advances past the PR's branch point, the depth=1 base graft shares no ancestor with HEAD, so git aborts with "fatal: origin/master...HEAD: no merge base" (exit 128) and the Detect Changed Benchmarks job fails. This broke re-runs of e.g. #1581 and #1570. Suppress the three-dot error and fall back to diffing the PR merge commit against its first parent (HEAD~1 = base tip), which yields exactly the PR's changes without needing a merge base. This mirrors the existing robust pattern already used on the push path in the same script. Co-Authored-By: Chris Rackauckas <accounts@chrisrackauckas.com>
The NBodySimulator/liquid_argon.jmd weave job ran for ~47h on amdci1-1 under the v7 stack before the runner workspace was reaped mid-run (observed in PR SciML#1559, run 25912449328, job 76160856440). Chunk 5 (`run_benchmark!(symplectic, 10 τs, t=40)`) alone accounted for the bulk of the runtime. The integrator-comparison benchmark just needs enough particles/steps to exercise force kernels and integrate at a range of timesteps; the absolute N and t are not load-bearing. Reduce the workload in both .jmd files so the suite finishes in a sensible window: - liquid_argon.jmd: N=350 → 128, t=40.0 → 10.0 - liquid_argon_long.jmd: N=350 → 128, t=200.0 → 50.0 Net work reduction is roughly (350/128)² × (40/10) ≈ 30× for the short suite and ≈ 30× for the long suite. The hardcoded `c_symplectic` / `c_adaptive` cost ratios remain valid since they are per-step ratios normalised to VelocityVerlet, essentially N-independent. Co-Authored-By: Chris Rackauckas <accounts@chrisrackauckas.com>
ChrisRackauckas-Claude
force-pushed
the
fix/NBodySimulator-v7
branch
from
d4ea4fb to
dafe640
Compare
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
2 participants
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.
Summary
NBodySimulator/liquid_argon was the only folder in PR #1559 (run 25912449328, job 76160856440) that ran for ~47h before the runner workspace was reaped mid-run on amdci1-1. The failing logs trace a
SystemError: opening file ... markdown/NBodySimulator/figures/liquid_argon_6_1.png: No such file or directoryafter Weave reports "Weaved all chunks / progress = 1" — i.e. the .jmd was still progressing when the workspace got cleaned out from under it.Timeline from the CI log:
2026-05-16T01:57:36run_benchmark!(symplectic, 10 τs, t=40)sweep) started immediately after@dfplot) started at2026-05-18T02:06:13— i.e. chunk 5 alone ran for ~47 hoursSo the .jmd is not "broken" in the v7-API sense; the problem-size simply exploded out of any reasonable CI budget under the v7 solver stack.
What this PR does
Reduce the workload in both .jmd files in this folder so the suite finishes in a sensible window. The integrator-comparison benchmark just needs enough particles/steps to exercise force kernels and integrate at a range of timesteps; the absolute N and t are not load-bearing.
liquid_argon.jmd:N = 350 → 128,t = 40.0 → 10.0liquid_argon_long.jmd:N = 350 → 128,t = 200.0 → 50.0Net work reduction is roughly
(350/128)² × (40/10) ≈ 30×for both suites (the long suite is dominated by step count, so reducing t by 4× nets a similar factor).The hardcoded
c_symplectic/c_adaptivecost ratios remain valid: they are per-step cost ratios normalised toVelocityVerlet, and per-step cost is essentially N-independent (all symplectic methods scale the same way in the pair-force kernel cost).Local verification
julia +1.11 --project=benchmarks/NBodySimulatorinstantiates cleanly. A microbenchmark atN=64, dt=1e-3, t=1.0(1000 steps) on this branch's stack:Extrapolating with
(N/64)² × (steps/1000), the full liquid_argon.jmd chunk 5 under the new parameters (N=128, t=10, 10 τs) lands at ~20–30 minutes, vs the observed ~47h previously. End-to-end I expect the whole folder under ~3h on amdci1-class hardware.Note for review
Please ignore until reviewed by @ChrisRackauckas.