thoughtspot / mcp-server / 25191489670

89%

push

github

SCAL-309983 Harden metrics request isolation (#129)

* SCAL-309983 Harden metrics request isolation

## Summary

Harden the metrics request wrapper so metrics setup, scheduling, and flush failures cannot
block MCP business logic or become visible to MCP clients.

## What Changed

- fall back to a noop sink if metrics config or sink construction throws
- keep `recorder.flush()` focused on metrics delivery, not Worker scheduling
- schedule request metrics flush through `ctx.waitUntil(...)` without awaiting delivery
- log waitUntil scheduling failures separately from metrics execution failures
- add tests for setup fallback, slow flushes, handler failures, waitUntil failures, and
  delivery failures

## Semantics

Direct `recorder.flush()` calls still await delivery. `withRequestMetrics(...)` only schedules
the request-scoped flush so slow, failing, or unavailable sinks do not delay responses.

Metrics failures remain best-effort side effects and are never propagated to MCP clients.

## Validation

- `npm run lint`
- focused runtime metrics tests
- `npm test`

* SCAL-309983 Return noop recorder on init fallback

## Summary

Use a stateless noop metrics recorder for initialization fallback so the fail-open path does
not construct a second request recorder.

## What Changed

- add a shared `NOOP_METRICS_RECORDER` implementation
- return the noop recorder when metrics config or sink setup throws
- update tests and log wording to reflect the recorder-level fallback

479 of 564 branches covered (84.93%)

Branch coverage included in aggregate %.

17 of 20 new or added lines in 2 files covered. (85.0%)

1187 of 1327 relevant lines covered (89.45%)

337.33 hits per line