Skip to content

test: standalone it_minmax bin + aws-gate smoke_minio (vbench follow-ups)#1486

Open
aaj3f wants to merge 3 commits into
perf/r2rml-pr2-decode-leversfrom
chore/vbench-followups
Open

test: standalone it_minmax bin + aws-gate smoke_minio (vbench follow-ups)#1486
aaj3f wants to merge 3 commits into
perf/r2rml-pr2-decode-leversfrom
chore/vbench-followups

Conversation

@aaj3f

@aaj3f aaj3f commented Jul 13, 2026

Copy link
Copy Markdown
Contributor

Test-infra follow-ups from the perf-series gate batteries (no engine code):

  1. it_minmax_fast_path_fired → standalone test bin. Its thread-local set_default capture requires sole ownership of the process (its own doc says so); the grp_* consolidation bundled it with 241 siblings and parallel interleaving flips its event capture (bit us in the PR-2 gates). A process-global subscriber is unsound here — the test asserts the absence of the shared fast-path decline event, which siblings also emit → nondeterministic false-fail under global capture. Standalone bin matches the existing it_cyclic_bgp_probe convention. Full api suite green; parallel grp_misc green.
  2. smoke_minio gated on required-features=["aws"] — bare cargo test -p fluree-db-iceberg is now green (was failing to compile the example under default features; pre-existing, proven by stash-to-base during PR-1 gates).
  3. q009/q010 perf flags closed environmental: idle-machine 5-rep recheck = 1.10×/1.03× vs blessed (the ~2.3× outliers during PR-2 gates were concurrent-build contention).

Separate pre-existing flake surfaced (NOT addressed here, out of scope): it_values_type_deleted_repro::values_bound_type_after_delete_incremental_index_timeline fails ~1/6 parallel grp_misc runs with Internal("failed to load binary index: dict branch: too small for header") — a binary-index reload race, passes single-threaded, structurally unrelated to these edits. Flagged for the binary-index owner; filed as #1489.

Stacks on #1482 (PR-2); orthogonal to PR-3/4/4b (no file overlap).

aaj3f added 2 commits July 13, 2026 16:18
…feature

it_minmax_fast_path_fired documents that its thread-local set_default
tracing capture requires being the only test in its binary, but the grp_*
consolidation bundled it into grp_misc with 241 siblings — parallel
interleaving flips its event capture (surfaced during the PR-2 gates).
A process-global subscriber is NOT a sound fix here: the test asserts the
ABSENCE of 'fast path declined; running fallback', which siblings also
emit from the shared aggregate fast path, so global capture would
false-fail nondeterministically. Restore the standalone [[test]] bin
(matches the existing it_cyclic_bgp_probe convention). Full api suite
green; standalone bin passes; grp_misc passes in parallel.

smoke_minio needs the aws-gated S3IcebergStorage: mark the example
required-features=["aws"] so bare cargo test -p fluree-db-iceberg is green.

Also closes the PR-2-era q009/q010 native perf flags as environmental:
idle-machine 5-rep recheck = 1.10x / 1.03x vs blessed (was ~2.3x under
concurrent build load).
…e note (review follow-up)

The #1486 review's one optional code item: the header documented BOTH
interference mechanisms but not their scope, leaving the rationale open to
a false disproof ("nextest-run bundled siblings don't interfere"). Both
mechanisms are process-scoped — callsite-interest cache poisoning and
sibling decline-event capture bite under thread-parallel cargo test (one
shared process), not under nextest's process-per-test model. The
standalone bin keeps the test sound under both runners, and cargo test is
what contributors reach for first.

Also from the review: the surfaced binary-index reload flake is now filed
as #1489 (was "issue-filing decision with AJ" in the PR body; body
updated with the number).

Gates: standalone bin 1/1; cargo fmt --all --check green.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant