coder / coder/vscode-coder

telemetry: schema-first event registry (typed catalog + generated docs)

Open
#999 1 comment 0 reactions 0 assignees View on GitHub
Dominant language
TypeScript
Stars
130
Forks
48
Avg merge
3d 3h
Merged PRs (30d)
15

Description

## Problem

Telemetry events, properties, and value unions are declared per-domain in `src/instrumentation/*` and documented by hand in `src/instrumentation/EVENTS.md`. Nothing structurally prevents:

- emitting an event name that isn't cataloged anywhere,
- adding/renaming a property without updating the docs,
- drifting from the conventions in `src/instrumentation/CONVENTIONS.md` (a recent audit found camelCase keys, kebab-case enum values, and caller-set `result` that review missed).

## Proposal

Adopt a schema-first registry as the single source of truth for the telemetry surface, the way OTel itself defines semantic conventions in a registry and generates code + docs from it (Weaver).

TS-native sketch:

- A central `TelemetryEventMap` interface declaring every event with its properties and measurements (value unions spelled out).
- `TelemetryService.trace`/`log`/`logError` constrained to `keyof TelemetryEventMap`; `Span` becomes generic (`Span`) so `setProperty`/`setMeasurement` keys and values are type-checked per event. An undeclared event or attribute key becomes a compile error.
- `EVENTS.md` generated from the registry (script walking the interface + JSDoc), with a CI `--check` mode so the doc can never go stale.

## Known costs / open questions

- Every instrumentation class and its tests are touched by the `Span` generic.
- Phase names compose at runtime (`parent.child`), so child phases need registry entries or looser typing.
- The registry partially duplicates per-domain unions the instrumentation classes already export; decide whether the registry imports those types or replaces them.

## Cheaper interim guards (could land independently)

- Type the event-name parameter of `TelemetryService.trace`/`log` as a union of registered names (names are already top-level string literals; near-zero churn).
- ESLint restriction banning raw `telemetry.trace(`/`.log(` outside `src/instrumentation` (the layering rule from CONVENTIONS.md).
- Golden event catalog aggregated from the test suite's sinks, checked like the existing OTLP goldens.

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.