fix(db): lazy-reconnect module singleton to survive mid-cycle teardown (#1570)#1758
Closed
toilalesondev wants to merge 2 commits into
Closed
fix(db): lazy-reconnect module singleton to survive mid-cycle teardown (#1570)#1758toilalesondev wants to merge 2 commits into
toilalesondev wants to merge 2 commits into
Conversation
Two coupled bugs blocked code-callers/code-callees for TypeScript:
1. extractSymbolName (code.ts) didn't descend into variable_declarator, so
'export const Foo = () => {}' (dominant React/RN form) got symbol_name=null
and its edges were dropped at the import-file.ts:1100 qualified-name gate.
Fix: name lexical_declaration/variable_declaration via the declarator's
name field (exact node.type match — never substring, which would corrupt
Go's var_declaration etc.).
2. import-file.ts edge writes omitted source_id, so every code_edges_symbol
row landed source_id=NULL and was invisible to source-scoped reads
(code-callers/callees with --source). Fix: thread sourceId into the edge
input (default brain → null, matches unscoped reads).
Also bump EDGE_EXTRACTOR_VERSION_TS to force a re-walk past stale watermarks.
Verified on a real RN/TSX repo: code-callers useTheme → 64 resolved callers
(HeroChip, Chip, Card…), code-callees DiaryScreen → 66. Was 0/0 before.
33 chunker unit tests pass incl. 2 new const-arrow naming pins.
garrytan#1570) The dream cycle runs phases sharing the db.ts module-singleton connection. A mid-cycle callee (a module-style engine disconnect between phases) nulled the singleton, so the later sync + synthesize phases threw 'connect() has not been called' and the whole cycle reported 'partial' — which in turn left sources.config.last_full_cycle_at unwritten and doctor's cycle_freshness permanently FAILing. Fix: getConnection() now lazily rebuilds the pool from a remembered URL (lastConnectedUrl) instead of throwing when the singleton was torn down earlier in the same process. postgres() opens no socket until first query, so the rebuild is safe from the synchronous getConnection() path. Extracted buildPool() so connect() and the lazy path share identical pool options. Verified: gbrain dream --source {helmsman,perfeat-mobile,default} now run sync + synthesize to completion (exit 0, 0 failed phases); doctor cycle_freshness flips to OK across all 3 sources. 52 connection-resilience tests + 64 chunker tests pass. Refs garrytan#1570 (singleton-null disconnect investigation).
Author
|
Closing as superseded. Upstream already fixes this exact singleton-null mid-cycle bug (#1570) via two landed commits:
Upstream's caller-side fix is canonical; my lazy-reconnect in getConnection() is redundant. Thanks! |
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.
Problem
The dream cycle runs its phases sharing the
db.tsmodule-singleton postgres connection. A mid-cycle callee (a module-style engine disconnect between phases) nulls the singleton, so the later sync and synthesize phases throwNo database connection: connect() has not been calledand the whole cycle reportspartial.Knock-on effect:
partial(well, the failed phases) meant the freshness signal was unreliable, anddoctor'scycle_freshnesscheck stayed FAIL ("Source X has never completed a full cycle") because the broken phases dominated the run.This is the singleton-null bug the disconnect-audit instrumentation in #1570 was added to chase. I captured the culprit by logging full stacks in
db.disconnect()during a realgbrain dream --source <id>run (interpreted, unminified): a module-stylePostgresEngine.disconnect()fires mid-cycle, nullingsqlbefore sync/synthesize reachgetConnection().Fix
getConnection()now lazily rebuilds the pool from a remembered URL (lastConnectedUrl) instead of throwing when the singleton was torn down earlier in the same process.postgres()opens no socket until the first query, so rebuilding from the synchronousgetConnection()path is safe. ExtractedbuildPool()soconnect()and the lazy path share identical options.This is defensive: it self-heals regardless of which phase drops the connection, without re-plumbing every caller's engine.
Verification
gbrain dream --source {helmsman,perfeat-mobile,default}— sync + synthesize now run to completion, exit 0, 0 failed phases (were 2: sync + synthesize).doctorcycle_freshnessflips to OK across all 3 sources.Refs #1570.