Repository navigation
Conversation
improveChain/reparent/root-kind upgrades queued a deferred invalidation that drainPendingChainEvents applied as an erase at dump time. If the search then terminated via the canary, nothing could rebuild the erased chains (candidate and discovered state are zeroed, search pain-budget blocked ~11-15s), losing the correlation for the rest of the window - against the restartSearch() contract that resolved chains stay cached across a restart. Split deferred invalidations into EVICT (noise eviction, filter: erased at dump time as before) and REBUILD (path changed, source still live: applied by the BFS-thread poll, which rebuilds the chain from the intact frontier and keeps the old chain if the rebuild fails). Kind is encoded in-band in the existing bounded pending buffers; no new allocation or locks. Reproducer: ReferenceChainsBfsTest.ImprovedPathInvalidationKeepsLeakChainAcrossTerminalReset (fails before, passes after). PROF-16316.
The reference-chain walks are paced by the CPU pain budget; on slow cells the second pass can land late in the tests' dump windows, leaving only the first pass's uncorrelated chains (TL/LeakTag loose variants report out-of-pool or not-found in CI nightlies, PROF-16305). Test-side schedule changes within the existing bounds (no assertion weakened, no retries/skips, static entry caps kept): - cache-leak: growth completes by round 10, rounds 11-16 poll with a small top-up so the BFS drains its frontier backlog and intercepts a leak-tagged payload with slack (<= 36,601 entries, down from 69,001) - failure-only referenceChainDiagnostics() helper appended to the assertNotNull messages of the cache-leak and gc-root tests (debug- instrumented builds only; release returns empty string) - ThreadLocalLeakScenario: keeps full per-round chunk growth (the late-variant correlation needs the whole interception surface; verified on JDK 11) but keeps the filler at a quarter size so pass 0's pain-budget debt is repaid well inside the window - LeakTagCorrelationScenario: front-loaded growth with sparse top-ups, tail rounds quiet; keep-best ladder unchanged
All four reference-chain slow e2e tests leave quarantine: the pr-889 regime plus the W5 test-side pacing fixes and the PROF-16316 improveChain invalidation fix (PR #898) cleared the failures. The gc-root musl-aarch64 line is dropped after a 10/10 clean musl-aarch64 JDK11 container validation of the full stack.
Contributor
CI Test ResultsRun: #40192745607 | Commit:
Status Overview
Legend: ✅ passed | ❌ failed | ⚪ skipped | 🚫 cancelled Summary: Total: 32 | Passed: 32 | Failed: 0 Updated: 2026-10-25 12:44:29 UTC |
Contributor
This branch has not been deployed
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.
What does this PR do?:
Fixes the intermittently-red reference-chain slow e2e family (PROF-16305) with one product fix plus test-side hardening, and removes the PROF-16305 quarantine added in #888. The branch carries four commits:
2277cb539— PROF-16316: keep resolved leak chains across improve-path invalidations (EVICT/REBUILD invalidation kinds, in-band sign-bit encoding, generation-guarded BFS rebuild from the intact frontier; replace-only, no behavior change elsewhere).4b0879ed4— fix reference-chain revisits and leak-candidate admission.2908999d6— make the reference-chain slow e2e tests robust to walk-pacing latency (schedule reshape, front-loaded reference pressure, richer failure diagnostics).37fc26255— remove all four PROF-16305 quarantine lines fromddprof-test/quarantine.txt.Motivation:
The nightly sanitized run 39893765385 had the reference-chain slow family failing in ~20 of ~32 eligible HotSpot 11+ debug/asan jobs, which is what #888 quarantined. Root causes are now fixed rather than quarantined:
improveChain's deferred invalidation erased cached leak chains at dump time, and the terminal pass zeroed candidate/discovered state so nothing could rebuild them — defeating therestartSearch()keep-across-restart contract. Fixed by2277cb539.4b0879ed4revisit/admission regime plus the2908999d6test-side pacing fixes.Additional Notes:
Dependencies — this PR is stacked and should not merge before its dependencies:
2277cb539in this branch is that change, carried here so the whole stack validates together. If Reference-chain cache: rebuild improved-path chains instead of erasing them (PROF-16316) #898 merges first, this PR's effective diff shrinks to commits 2–4.4b0879ed4is its replay on top of Reference-chain cache: rebuild improved-path chains instead of erasing them (PROF-16316) #898, for the same reason. Fix reference-chain discovery and leak test prerequisites #889 is currently an open draft (Paul Fournillon).How to test the change?:
2277cb539and passes after (full:ddprof-lib:gtestDebug189/189).ddprof-test/quarantine.txtvalidate script passes; regular (non-slow) suite 61/61 unaffected.For Datadog employees:
credentials of any kind, I've requested a security review (run the
dd:platform-security-reviewskill, or file a request via the PSEC review form).
bewairealso runs automatically on every PR.