mirror of
https://github.com/abhigyanpatwari/GitNexus.git
synced 2026-09-11 22:53:04 +00:00
2 commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
Gergő Magyar
|
95f87fc12a
|
perf(ingestion): Linux-kernel-scale analysis — worker-pool parse + finalize O(n²) + scope-resolution memory wall (#1983) (#2038)
* fix(ingestion): reduce parse-phase memory for huge repos (#1983)
Stop retaining full parse-cache chunks in RAM alongside the merged graph,
slim on-disk shards, defer worker ParsedFile emission for scope-resolver
languages, and add GITNEXUS_DEBUG_HEAP probes for OOM diagnosis.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix(ingestion): address #2038 tri-review findings (parse-phase memory)
Resolves the confirmed review findings on PR #2038:
- P1: thread exportedTypeMap through the sequential parse path
(processParsingSequential) so a no-worker run over a partially-warm
cache no longer silently drops the sequential-miss files' exported
types. Cache hits made exportedTypeMap.size > 0, suppressing the
end-of-loop buildExportedTypeMapFromGraph rebuild, but the sequential
path never populated the map. Regression test added (fails on the
pre-fix tree, passes after) plus a fully-sequential differential oracle.
- P2: saveParseCache builds its on-disk index from hashes actually
written/copied (writtenKeys), never a usedKeys hash whose shard write
or copy was skipped — no more phantom index entries.
- P2: add a unit test asserting SCOPE_RESOLUTION_LANGUAGES stays in sync
with SCOPE_RESOLVERS (asymmetric drift would lose a language's ParsedFile).
- Backfill cache coverage: loadParseCacheChunk missing/corrupt -> undefined,
pruneCache onDiskKeys branch, slim preserves nodes, saveParseCache
copy-evicted-shard round-trip.
- Cleanups: single-source heap-probe gating via isDebugHeapEnabled();
hoist the per-chunk mkdir in persistParseCacheChunk behind a
process-scoped Set; gate COBOL's unused worker-side ParsedFile
extraction (graph nodes still come from cobolPhase) while keeping
fileCount/progress unconditional.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(ingestion): remove dead worker-side ParsedFile extraction
After #2038 gated worker `ParsedFile` emission behind `!isScopeResolutionLanguage(language)`, and with all 16 SupportedLanguages registered in SCOPE_RESOLVERS, that gate was structurally always true — the worker already produced no ParsedFiles and scope-resolution re-extracts each file from source on the main thread (run.ts). Remove the now-dead machinery:
- Drop both worker `extractParsedFile` call-sites (tree-sitter processFileGroup + the standalone-provider branch) and the `result.parsedFiles.push`. The standalone branch keeps fileCount/onFileProcessed per file. `result.parsedFiles` stays declared but empty (field removal deferred).
- Remove the now-orphaned `scopeSourceKind` var + `ScopeCaptureSourceKind`/`extractParsedFile`/`isScopeResolutionLanguage` imports.
- Delete the consumerless `migrated-languages.ts` (isScopeResolutionLanguage + SCOPE_RESOLUTION_LANGUAGES) and its drift-guard test — parse-worker was their only importer. Also improves AGENTS.md "shared ingestion code must not name languages" compliance.
`extractParsedFile` and the scope-extractor-bridge stay (scope-resolution/run.ts + Vue resolver use them). Behavior-preserving: worker-sequential-parity passes before and after; tsc/eslint clean; no baseline/golden drift.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor(ingestion): worker-pool-only parsing; remove sequential parser (#1983)
Completes the #1983 huge-repo parse-OOM effort by making the worker pool
GitNexus's sole parse path.
Parallel serialization (the perf core): workers serialize their ParsedFiles to
a disk store in parallel and stream them back to scope-resolution, so the main
thread no longer re-parses every file (the tree-sitter native-memory leak that
caused the OOM). Adds chunk merge-pipelining + work-proportional chunk sizing so
the pool stays saturated.
Remove the sequential parser: `--workers 0`, `GITNEXUS_WORKER_POOL_SIZE=0`, and
`skipWorkers` now hard-error (no silent degrade — #1741); the small-repo
threshold no longer selects an in-process path; pool creation stays lazy /
cache-miss-gated so warm all-hit runs never spawn workers.
Worker-path parity fixes — removing sequential surfaced two pre-existing gaps
that tiny-fixture tests had masked by running below the worker threshold, both
fixed by carrying per-file metadata as DATA across the worker boundary (never
re-parsing on the main thread, preserving the OOM fix):
- C++: templateConstraints wired into worker node identity (SFINAE overload
disambiguation) + ADL / inline-namespace capture side-channel serialized
onto the ParsedFile.
- Kotlin: companion-scope side-channel serialized the same way (companion /
static dispatch).
Validation: tsc + build clean; full suite green (10,190 pass — the only
deterministic failures were the now-fixed C++/Kotlin worker-path gaps; the 2
remaining full-run failures are pre-existing load flakiness, green in
isolation); cpp-pipeline benchmark stays linear on a 1-worker pool.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(ingestion): wire C static-linkage side-channel + ADL O(1) collect + tri-review cleanups (#1983)
Follow-up to the worker-pool-only refactor, from a tri-review of the parse path.
- C static-linkage side-channel (P1): cProvider had no collect/applyCaptureSideChannel,
so on the now-sole worker path C `static` file-local marks were lost across the worker
boundary -> false cross-file CALLS edges + over-broad #include wildcard visibility on
every C analysis (the Linux kernel is C). Mirror the C++/Kotlin wiring: serialize
`staticNames` per file onto ParsedFile.captureSideChannel and restore it on the main
thread (no re-parse). + a worker-path regression test (the existing c-static-isolation
fixture passed vacuously — its collision resolves via #include before the global
free-call fallback ever consults static-linkage).
- captureSideChannel `kind` discriminant: add `kind:'cpp'`/`kind:'c'` tags + guards
(Kotlin already had one) now that C/C++/Kotlin share the single generic field.
- Perf: collectCppAdlSideChannel scanned the whole argInfoBySite/noAdlSites maps per file
(O(F^2) per sub-batch, ~100M parseSiteKey calls at kernel scale). Add per-filePath
lockstep indexes -> O(1) collect; serialized snapshot byte-identical.
- Cleanups: inline the one-line processParsingWithWorkers wrapper into processParsing;
drop the always-empty WorkerExtractedData.calls/assignments/constructorBindings fields;
remove the voided astCache param from processParsing; refresh stale "sequential
fallback" JSDoc.
Validation: tsc + build clean; cpp 297/297, c 8/8 (incl. the new worker-path
static-linkage guard), typescript + parsedfile-store green; cpp ADL benchmark stays linear.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* perf(scope-resolution): index C/C++ #include resolution in finalize (O(n²)→O(n))
Kernel-scale C/C++ analysis ground in finalizeScopeModel because three
per-#include operations each did a full O(F) scan with no index — the
finalize O(n²) that surfaced once the #1983 parse-phase OOM was fixed:
- expand{C,Cpp}WildcardNames: parsedFiles.find() per wildcard edge → O(R·F)
- resolveImportTarget: new Set(allFilePaths) rebuilt per #include
- resolveCImportTarget: suffix-match scanned all workspace paths
Each is replaced with a WeakMap-per-pass index keyed on the stable
parsedFiles/allFilePaths references that scope-resolution run.ts passes
once per pass:
- Map<ScopeId,ParsedFile> for wildcard expansion (c/static-linkage.ts +
cpp/file-local-linkage.ts)
- memoized augmented header set (c/scope-resolver.ts + cpp/scope-resolver.ts)
- basename-bucketed suffix index in resolveCImportTarget (c/import-target.ts),
shared by C and C++ since resolveCppImportTarget delegates to it
Collapses the C/C++ finalize from O(R·F) to O(R+F). Pure-perf, byte-identical
edge output: 962 targeted tests green (490 C + 472 C/C++ scope-resolution);
the basename index preserves the exact endsWith('/'+target) match and the
fewest-path-components-then-lexicographic tie-break.
The kernel's ~25-30k .h headers are classified C++, so both providers must
be fixed. Proven on the Linux kernel: the C finalize completed
(sr-post-finalize lang=c → sr-end lang=c), which the pre-fix run never
reached in 16+ min of grinding.
Build-independent follow-ups (separate from this finalize fix), documented
for later: emitFreeCallFallback same-name buckets (emit phase),
buildGraphNodeLookup + precount global setup, the ParsedFile store-load,
the dart/go/ruby expand-wildcards .find siblings, and the ~26GB
scope-resolution memory floor (full kernel completion needs >~40GB RAM).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(bench): regenerate C scope-capture baseline for the #1983 c-static-linkage-worker fixture
bench/scope-capture/measure.mjs fingerprints emitCScopeCaptures over the
lang-resolution/c-* fixture corpus. The #1983 PR added the
c-static-linkage-worker fixture (caller.c/lib.c/lib.h/local.c — the
worker-path static-linkage side-channel test) but did not regenerate the C
baseline, so `--check` has been red on this branch (main, lacking the
fixture, still matches 0de009b).
Pure fixture-corpus drift — no c/captures.ts or query change branch-vs-main,
existing fixtures' captures byte-identical (c-captures.test.ts 45/45),
scaling stays linear (~0.97). Regenerated: 0de009b -> 39f3a83. Bench now
PASS (14 languages). Unrelated to the finalize O(n²) fix.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* perf(scope-resolution): lower kernel-scale resident memory floor + setup cost
Reduce the scope-resolution resident-memory floor and setup throughput on
huge repos (Linux kernel), the wall that remains after #1983 (parse OOM) and
the finalize O(n^2) fix (
|
||
|
Gergő Magyar
|
c4b69402e1
|
feat(workers): self-healing worker pool + deferred-resolution observability (#1741) (#1947)
* fix(workers): fail fast instead of silently degrading on worker-pool startup failure (#1741) When an explicitly-sized worker pool (--workers <N>) fails to start because every worker crashes during top-of-script init, the parse phase used to log a swallowed `logger.warn` and silently fall back to the ~10x slower sequential parser. In #1741 (rc99) that turned a worker-startup regression into a 123-minute "stuck" parse with no explanation. This change: - Surfaces the real crash: the pool now spawns workers with `{ stderr: true }`, tees + captures each worker's stderr, and attaches the tail to its readiness-failure messages (propagated via WorkerPoolInitializationError.readinessFailures). "did not report ready" now carries the underlying native-binding/import error. - Gates the fallback: when --workers was explicit and fallback was not opted into, a total startup failure throws an actionable error instead of degrading. Auto-sized pools still fall back, but loudly (logger.error + progress warning). New --allow-sequential-fallback flag (+ i18n) opts back in. - Adds env-gated worker bootstrap-stage logging (GITNEXUS_WORKER_BOOTSTRAP / --verbose): imports+grammars loaded -> ready sent -> first task received, so a slow/crashing startup is diagnosable. Tests: all-workers-failed gating (fatal vs loud degrade), stderr surfacing, and the updated lazy-cache fallback contract (opt-in flag + fail-fast). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(ingestion): always-on slow-file watchdog for deferred call resolution (#1741) The original #1741 symptom is a run that appears stuck at "Resolving calls (all chunks)... (9000/18066 files)" — the progress bar freezes inside a single file's call resolution and nothing reaches the log. Rich per-file deferred diagnostics already exist, but only behind --verbose / GITNEXUS_PROFILE_DEFERRED, so a plain `analyze` run gives the user a frozen bar and silence. Add an always-on (not verbose-gated) per-file watchdog in processCallsFromExtracted: when a single file's call resolution exceeds alwaysOnSlowFileWarnMs() (default 15s, override GITNEXUS_SLOW_FILE_WARN_MS, 0 disables) it emits a throttled logger.warn naming the culprit file and the files-resolved-so-far — turning the silent stall into one actionable line. Throttled (>=30s between warnings) so a genuinely slow repo can't storm the log. The watchdog is observation-only; resolution behavior is unchanged. Note: deliberately did NOT add a heritage child x parent product cap — the name lookups are O(1) (type-registry Map.get) and the product is bounded, so the heritage build is not the bottleneck; a cap would risk dropping real edges for no measured gain. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * test(ingestion): worker-vs-sequential parity guard for binding/edge collapse (#1741) rc99 produced almost no bindings/edges (13 bindings vs rc91's 106,305) because a worker-path failure left extracted results unmerged while the run still reported success. Rather than an arbitrary "implausibly low" runtime threshold (which false-positives on legitimately low-binding repos/languages), pin the invariant directly: for the same repo, worker mode and sequential mode must produce the same graph. The test runs the ts-simple cross-file fixture through worker mode (workerPoolSize + lowered threshold) and sequential mode (skipWorkers), and asserts: usedWorkerPool is true/false respectively (guards the test itself against a silent fallback masking divergence), identical CALLS/IMPORTS/DEFINES/ HAS_METHOD edge sets and Class/Function/Method defs, and non-zero CALLS/IMPORTS (the rc99 collapse signature). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(workers): arm fail-fast for env-sized pools + fix watchdog /0 denominator (#1741) Addresses two review findings on the #1741 worker-startup PR: - Fail-fast gate missed the env channel. `explicitWorkers` keyed only off the `--workers` flag, so a pool sized via `GITNEXUS_WORKER_POOL_SIZE` (with no `--workers`) silently degraded to sequential on a total worker-startup crash — reproducing the original #1741 symptom for env-channel operators. The gate now arms on a non-zero size from either channel, via a single-source `envWorkerPoolSize()` helper exported from worker-pool.ts (also rewired through resolveAutoPoolSize). The fatal message now names the channel actually used instead of "--workers undefined". - Always-on slow-file watchdog printed "Resolved N/0 files". `resolvedTotal` was pre-counted only on the profile path, but the watchdog reads it on every run, so a plain `analyze` showed a bogus /0 denominator on exactly the unprofiled hang the watchdog exists to explain. Pre-count now runs whenever its result is read (profile path OR watchdog active). Tests: strengthened the watchdog test to assert "1/1" (not "/0"); added env-channel fail-fast/degrade cases and made the gating suite hermetic against an ambient GITNEXUS_WORKER_POOL_SIZE. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(workers): self-healing worker pool replaces the fail-fast flag (#1741) Replaces the interim --allow-sequential-fallback flag with automatic, bounded self-healing in the worker pool — industry-standard supervision (OTP restart-intensity, systemd StartLimit, circuit-breaker, AWS jittered backoff) translated to the Node worker_threads pool. worker-pool.ts — bounded startup self-heal (the missing layer): - A worker that crashes during top-of-script init is now RETRIED with capped, full-jitter backoff (BASE 250ms, CAP 2s) up to a small per-slot budget, so a transient blip heals itself with no operator action. The prior code dropped an unready initial slot on its first crash. - A DETERMINISTIC crash-loop (>=2 fresh workers crash with the same normalized signature before any reaches ready — the #1741 missing native-binding case) is detected and short-circuited, so the pool gives up in ~1s instead of burning every slot's budget. Correctness rests on the STRUCTURAL signal (zero workers ever ready + budget exhausted), so a missed signature only costs a few seconds, never a misfire; even a stderr-less crash groups via its normalized "exited with code N" message. - Backoff sleeps are cancellable (unref'd timer + abort on terminate), so terminate() can't be wedged for the backoff duration. - WorkerPoolInitializationError now carries a crashClass for an accurate, flag-free message. The runtime respawn/breaker path is unchanged. parse-impl.ts — collapse to automatic fail-fast: - handleWorkerStartupFailure always logs the real cause then THROWS with the captured crash + `--workers 0` as the explicit sequential escape. No more degrade branch; no dependence on how the pool was sized. This is reached only after the bounded self-heal is exhausted, so it can't resurrect the #1741 silent 123-minute sequential grind. Construction failure (broken install) also fails fast instead of degrading silently. Removed --allow-sequential-fallback end to end (CLI, run-analyze, pipeline, i18n). --workers 0 remains the explicit "parse sequentially" path; one flag removed, none added. Grounded in a research+critique pass; the critique's hazards (N-parallel race, empty-stderr timing, non-cancellable sleep, runtime-breaker regression) are addressed or scoped out by design. Tests: startup self-heal (transient recovers; deterministic fails fast without burning the budget); gating test rewritten to the fail-fast-always contract; obsolete degrade test removed. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(workers): ref + cancel startup backoff so transient retries aren't dropped (#1741 U1) abortableSleep unref'd its backoff timer, so a transient startup retry could be silently dropped if that timer was the last ref'd handle on the event loop — the process could exit mid-recovery. Keep the timer ref'd (a pending retry is necessary work) and register a cancel fn in a pool-scoped set; terminate() now clears pending backoffs so it can't be wedged for the backoff cap. A normally fired timer self-deregisters (clear-on-settle), so no timer lingers after a slot's retry loop exits. Exposes pendingStartupTimers in getStats. Tests: terminate-during-backoff cancels + spawns nothing after (R2); the recovery test now asserts no startup timer lingers after settle (R1). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(workers): route GITNEXUS_WORKER_POOL_SIZE=0 to sequential, not a phantom fail-fast (#1741 U2) env=0 (no --workers) built a size-0 pool that threw a fabricated "retry budget exhausted / native binding" crash. The shouldUseWorkers gate now routes env=0 to the sequential path before pool construction — but only when no explicit --workers <N> was given, so an explicit positive size wins over an ambient env=0. The route emits one log line so the undocumented (possibly accidental) env=0 case is observable instead of a silent degrade. envWorkerPoolSize is un-exported (module-internal sizing reader); a new workerPoolDisabledByEnv() predicate serves the gate. Empty/whitespace env is now treated as unset (auto formula), not 0 — an empty assignment is an accident, not a request for zero workers. Reattached the detached resolveAutoPoolSize JSDoc and corrected the stale docstring. Tests: env=0 → sequential (no spawn); explicit --workers wins over env=0; workerPoolDisabledByEnv unit (0=true, positive/empty/invalid=false); getStats shape updated for pendingStartupTimers. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(workers): make deterministic crash-loop detection conservative (#1741 U3) The old tally counted crash EVENTS in a shared signature->count map, so a simultaneous transient crash storm (e.g. spawn EAGAIN under fork pressure) or a single slot crashing identically twice falsely tripped "deterministic" and hard-aborted work that would have self-healed. Replace it: a crash counts toward deterministic only after its signature REPRODUCES across a respawn on the same slot, and the short-circuit fires once >=2 distinct slots reproduced (or 1 for a size-1 pool). Every slot now gets >=1 self-heal attempt before any short-circuit; the structural budget floor still bounds the worst case. crashSignature now also collapses Windows backslash paths and bare (no-0x) hex runs so the fast-path fires on those platforms; exported for unit testing. Tests: simultaneous storm self-heals (the discriminator vs an attempt-0 rule); distinct-per-attempt crashes classify transient-exhausted; single-slot reproduction classifies deterministic; crashSignature normalization unit. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(workers): class-aware startup failure hint + reattach detached JSDoc (#1741 U4) The "often a missing/broken native binding" hint was appended to every failure class, including a pool *construction* failure where no worker ever ran (a missing build / bad worker path). Make the hint class-aware: keep it for the readiness/init classes, use a construction-specific hint otherwise, and surface the construction error (e.g. "Worker script not found: …") verbatim. Reattach the waitForWorkerReady JSDoc that the stderr-capture block had detached from its function. (The abortableSleep docstring was already corrected in U1.) Tests: construction message surfaces the real error + drops the native-binding guess; deterministic/transient messages keep the hint (regression guard). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> |