[meikelmosby]

Top layer of #127337. Reads the header #127163 adds.

Problem

• The CDP events consumer parsed every event body to find its team, then dropped the event when that team had no destination and no workflow, which is most events.
• Nothing cheaper than the body carried the team, so the parse could not be skipped.

Changes

• Nothing looks different to a user.
• The team now comes from the Kafka header, and the body is parsed inside toGlobals, which runs only once a team turns out to have a destination or a workflow.
• A message with no usable header keeps the old path and reads the team from the body, so nothing depends on the header being present.
• An event whose body names a different team than its header is parked instead of built. The header decides which team's destinations load and the body carries the event data, so disagreement would send one team's event to another team's destinations.
• cdp_events_team_id_source counts which path each message took, so the rollout is visible.

Warning

A malformed body now reaches the dead-letter topic only when its team has a destination or a workflow. For any other team nothing reads the body, and CDP would have produced nothing for that event anyway.

Numbers

	Before	After
Events whose body is parsed	every event	only events whose team has a destination or a workflow
JSON parses per surviving event	3 (body, properties, person properties)	3, unchanged
JSON parses per skipped event	1	0

Parsing the event body accounted for about 6.5% of main-thread samples in a merged wall profile of the consumer, with roughly 4.5 points of that inside JSON.parse. Most events belong to a team with neither a destination nor a workflow, so most of that work was spent on events the consumer then dropped.

cdp_events_team_id_source gives the header-versus-body split once this ships, and the ratio of parsed to consumed events is what the saving tracks.

Not yet measured in production: nothing is deployed, so there is no after-profile.

How did you test this code?

Test rationale: readTeamIdHeader gets a case table, because a header value that is not a usable team id must never become one: the consumer would load another team's destinations. Two consumer cases follow: a header naming a team with no function plus an unparseable body, where a parked record would prove the body was read, and a header and body naming different teams, which must park. Existing cases in that file carry no header, so they already cover the fallback.

Not run: jest, oxlint and oxfmt. pnpm install cannot reach the npm registry in this sandbox. tsc --noEmit reports no error in the changed files.

Release status

• No feature flag controls this change

Docs update

• None.

🤖 Agent context

Autonomy: Human-driven (agent-assisted)

Agent: Claude Code, Opus 5

Skills: /writing-code-comments, /writing-tests, /writing-pr-descriptions, /stacking-prs, /debugging-ci-failures

More context for AI agents

• Session: Claude Code. This PR held four changes before review and was split into layers 2 to 5 of the stack.
• Decisions: The mismatch guard was added after review found that a header and body naming different teams would cross tenants.
• Decisions: The body is parsed inside toGlobals rather than behind a separate flag, because parseBatch already runs toGlobals only for a team that has work.
• Follow-ups: cdp_events_team_id_source shows when every message carries the header; the body fallback can go once it does.
• Read closely: the header path in cdp-events.consumer.ts.

🤖 Generated with Claude Code

[github-actions]

🤖 CI report

✅ Trunk lane — non-backend lane

This PR is assigned to the non-backend lane. It does not run backend Python tests and may merge in parallel with PRs in other lanes.

⚠️ Duplication (Python) — 1 new duplicated block (worst 85 tokens)

New Python code duplication introduced by this branch. Fails at 70+ tokens in app code, or 150+ tokens when both copies live in test files. Advisory while the gate proves itself: extract a shared helper instead of copying.

First copy	Second copy	Lines	Tokens
posthog/api/integration.py:2736	posthog/api/user_integration.py:573	19	85

✅ Duplication (TypeScript) — clean

New TypeScript code duplication introduced by this branch. Fails at 70+ tokens in app code, or 150+ tokens when both copies live in test files. Advisory while the gate proves itself: extract a shared helper instead of copying.

🚨 Comment density — 17% of added code lines are comments (22 of 128)

This section warns when comments are more than 3% of the code lines a PR adds, and alerts above 6%. Before agent-assisted PRs, the typical share was about 2%. Only full-line comments count. Docstrings, generated files, snapshots, migrations, and workflow files are left out.

Comments that restate the code, record how the change came about, or narrate the next line add noise for the next reader. Keep the comments that explain a reason the code cannot show, and remove the rest. See .agents/skills/writing-code-comments/SKILL.md for the house rules.

Files with the most added comment lines:

File	Comment lines	Added lines
nodejs/src/cdp/consumers/message-parsing.ts	8	20
nodejs/src/cdp/consumers/cdp-events.consumer.ts	7	34
nodejs/src/cdp/consumers/cdp-events-consumer.test.ts	5	36
nodejs/src/cdp/consumers/message-parsing.test.ts	2	29

This check does not block merging. It updates on every push and clears when the share drops.

[posthog]

PostHog Review · Standard · reviewed 063baa8: 1 Must fix, 3 Should fix.

This is the only automatic review while this pull request is a draft. Comment @posthog review to review new changes sooner. Otherwise, a review of the new changes runs when you mark it ready for review.

See this review and its findings in PostHog Review. Comment @posthog review --deep for a Deep review. A Deep review takes longer and looks deeper than a Standard review.

[parameterai]

Risk: No findings

The only change since the last review removes a test-only app-metrics flush-interval override (with its comment) from two beforeEach blocks in cdp-events-consumer.test.ts. No production code in this PR changed since the last review, so no new security surface was introduced.

_{Sentinel reviewed fb8cf2c · Review settings}

[posthog]

PostHog Review

Found 1 should fix.

Findings on the changed lines

nodejs/src/common/services/app-metrics-aggregator.ts:97

P2 · Serialize timed drains with outstanding flushes

When Kafka acknowledgments stall beyond the interval, each timer tick starts another produce while earlier drains remain pending. Because flush() clears the buffer before awaiting delivery, batch flushes return without backpressure. Pending messages accumulate and exhaust the producer queue, dropping metrics; track the in-flight flush and make subsequent callers await it.

[meikelmosby]

/trunk merge

[trunk-io]

😎 Stack merged successfully - details.

[deployment-status-posthog]

Deploy status

Environment	Status	Deployed At	Workflow
dev	✅ Deployed	2026-11-03 10:13 UTC	Run
prod-us	✅ Deployed	2026-11-03 10:28 UTC	Run
prod-eu	✅ Deployed	2026-11-03 10:36 UTC	Run