fix(logger): never let structured serialization throw into the caller

[waleedlatif1]

Context

Production structured logging went fully silent for hours across all five sim-production-us-east-1-app ECS tasks — INFO/WARN/ERROR all dropped to zero per 30 min, and a raw substring search for \"module\": returned nothing. Confirmed as genuine loss via AWS/Logs IncomingLogEvents (~25k → ~2-6k per 30 min), not a query artifact. Non-JSON output (Next.js internals, Better Auth, raw stack traces) kept shipping, and the socket service on the same cluster and same awslogs driver was unaffected — so it was localized to the app containers' own output, not the driver or the account.

This PR fixes one concrete, reproducible defect on that path. It is not claimed to be the whole root cause.

The bug

In the production JSON branch, Logger.log merged caller-supplied arguments into the entry with Object.assign and then called JSON.stringify, with no error handling on either step. If that metadata contained a cyclic reference, a BigInt, or a throwing getter, the call raised a TypeError out of logger.info(...) into the caller — the line was lost and whatever was being logged about was aborted.

Verified against the current staging logger under NODE_ENV=production:

RESULT: circular -> THREW: TypeError | JSON.stringify cannot serialize cyclic structures.
RESULT: bigint   -> THREW: TypeError | JSON.stringify cannot serialize BigInt.

Cyclic metadata is easy to hit unintentionally — errors carrying request/response/cause chains, ORM rows with back-references, socket handles.

The development path was never affected: the colorized branch routes objects through formatObject, which already has a try/catch. So this class of bug is invisible locally and only appears in production, where its signature is exactly structured logs vanish while raw stack traces keep shipping.

The fix

Entry construction and serialization now go through serializeEntry, which degrades instead of throwing:

1. normal JSON.stringify
2. retry with a replacer that tolerates cycles ([Circular]) and BigInt (stringified)
3. minimal entry preserving timestamp/level/module/message, flagged serializationError: true

A logger must never be able to break its caller.

Tests

Three regression tests in packages/logger/src/index.test.ts covering cyclic, BigInt, and throwing-getter metadata. All three were confirmed to fail against the unfixed source (git stash on index.ts → 3 failed | 25 passed) and pass with it (28 passed).

bun run lint and tsc --noEmit are clean.

Follow-ups (not in this repo)

Alarms and ECS task definitions live in the private sibling CDK repo, not here. Two gaps found there and reported separately:

• No alarm exists on AWS/Logs IncomingLogEvents for the app log group. Every metric filter over that group sees zero matches during a blackout and sits in OK — a silent alarm indistinguishable from healthy. A log-absence alarm with treatMissingData: BREACHING is specified in the report.
• Both app and socket LogDrivers.awsLogs({ ... }) calls omit mode, so both default to blocking.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
docs	Ready	Preview	Aug 6, 2026 8:32pm

[cursor]

PR Summary

Medium Risk
Touches the core production logging path used for observability; behavior changes only on previously failing payloads, but every log line now goes through new serialization logic.

Overview
Production JSON logging no longer throws back into callers when metadata or extra args are cyclic, contain BigInt, or use hostile getters/proxies/toJSON. The structured branch now routes through serializeEntry, which tries plain JSON.stringify, retries with a path-aware replacer ([Circular], stringified bigints), then emits a minimal line flagged serializationError: true.

withMetadata copies input through materializeMetadata so spreads and key reads degrade ([Unreadable], metadataError) instead of raising during child logger setup.

Regression tests cover cycles, BigInt, throwing getters, repeated non-cyclic refs, hostile toJSON, and hostile proxies.

^{Reviewed by Cursor Bugbot for commit 1a4b662. Configure here.}

[greptile-apps]

Greptile Summary

The PR makes production structured logging degrade safely when caller-supplied metadata cannot be serialized.

• Adds tolerant handling for cyclic references and BigInt values.
• Adds guarded metadata materialization and a minimal last-resort log entry.
• Adds regression coverage for getters, proxies, toJSON, repeated references, cycles, and BigInt.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
packages/logger/src/index.ts	Adds layered, non-throwing metadata materialization and structured-entry serialization fallbacks; the previously reported escape paths are addressed.
packages/logger/src/index.test.ts	Adds focused regression tests covering hostile metadata and serialization edge cases.

_{Reviews (5): Last reviewed commit: "fix(logger): keep repeated references ou..." | Re-trigger Greptile}

[waleedlatif1]

@greptile

[waleedlatif1]

@cursor review

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 2463ff6. Configure here.}

[waleedlatif1]

@greptile

[waleedlatif1]

@cursor review

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 22e6b9b. Configure here.}

[waleedlatif1]

@greptile

[waleedlatif1]

@cursor review

[waleedlatif1]

@greptile

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 1a4b662. Configure here.}