feat(lib): single telemetry row per tool call with duration and outcome

[SavioBS629]

Telemetry currently writes one row when a tool starts and a second row if it fails, so no row carries how long the call took, and a tool that returns an error result without throwing is recorded as a success.

This makes it one row per call, written when the call settles.

• instrumentToolLatency wraps each registered handler. The existing trackMCP calls at entry and in catch blocks record into a request-scoped context instead of posting, and the wrapper writes the single row when the handler finishes. No tool files change.
• The row adds duration_ms and outcome (ok / error_result / threw) to the existing fields.
• success keeps its meaning: false when the handler reported an error or threw. Failure counts are unchanged; a failed call is now one row instead of two.
• trackMCP outside a wrapped call behaves as before.
• Request-scoped context, so concurrent calls never share state.
• withToolCall and instrumentToolLatency are exported for hosts that register their own tools.

Tests cover ok / error-result / throw paths, concurrency, idempotency, and that telemetry failures never affect a tool call.

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Central YAML (base), Organization UI (inherited), Workspace UI (inherited)

Review profile: ASSERTIVE

Plan: Enterprise

Run ID: d38b02b2-c69a-4fcc-b2e5-8a7357fbb0fb

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Central YAML (base), Organization UI (inherited), Workspace UI (inherited)

Review profile: ASSERTIVE

Plan: Enterprise

Run ID: 6d3222f1-1890-46ef-a7b1-3e2016f68ab1

📥 Commits

Reviewing files that changed from the base of the PR and between 203967b and 4135d38.

📒 Files selected for processing (7)

• src/index.ts
• src/lib/instrumentation.ts
• src/lib/tool-latency.ts
• src/server-factory.ts
• tests/lib/error-class.test.ts
• tests/lib/instrumentation-single-row.test.ts
• tests/lib/tool-latency.test.ts

Included review availability: Your plan provides up to 12 included reviews per hour; 11 remain after this review.

📜 Recent review details

🔇 Additional comments (7)

src/lib/instrumentation.ts (1)

1-1: LGTM!

Also applies to: 10-25, 36-39, 43-89, 91-128, 129-176, 178-184, 186-231

src/lib/tool-latency.ts (1)

1-28: LGTM!

src/server-factory.ts (1)

24-24: LGTM!

Also applies to: 80-86

src/index.ts (1)

56-57: LGTM!

tests/lib/error-class.test.ts (1)

1-107: LGTM!

tests/lib/instrumentation-single-row.test.ts (1)

1-135: LGTM!

tests/lib/tool-latency.test.ts (1)

1-192: LGTM!

---

📝 Walkthrough

Walkthrough

The change adds consolidated tool-call telemetry with error classification, latency, outcomes, and call-time client information. It wraps registered tool handlers, wires instrumentation into server setup, exports the new APIs, and adds coverage for success, failure, concurrency, and telemetry faults.

Changes

Tool call telemetry

Layer / File(s)	Summary
Instrumentation lifecycle src/lib/instrumentation.ts	Adds error classes, shared event construction and sending, async-local tracking, result-error detection, and withToolCall, which emits one settled-call event with duration and outcome.
Tool wrapper integration src/lib/tool-latency.ts, src/server-factory.ts, src/index.ts	Wraps eligible tool handlers once, reads client information at call time, instruments registered tools, and exports the new APIs.
Telemetry behavior validation tests/lib/error-class.test.ts, tests/lib/instrumentation-single-row.test.ts, tests/lib/tool-latency.test.ts	Tests error classification, event fields, concurrent calls, argument forwarding, idempotent wrapping, non-function handlers, and swallowed telemetry failures.

Priority: ➖ Normal

Estimated code review effort: 4 (Complex) | ~45 minutes

Change: Feature

Sequence Diagram(s)

sequenceDiagram
  participant ToolHandler
  participant withToolCall
  participant trackMCP
  participant InstrumentationAPI
  ToolHandler->>withToolCall: execute tool handler
  ToolHandler->>trackMCP: record call context or error
  withToolCall->>withToolCall: calculate duration and outcome
  withToolCall->>InstrumentationAPI: post consolidated telemetry event
  withToolCall-->>ToolHandler: return result or rethrow error

Loading

Merge Risk: ⚪ Minimal · up to 4135d

Each tool call now produces one telemetry row with duration, outcome and error class. Real tool calls reach this instrumentation. Telemetry failures do not affect tool results. The only visible change is that dashboards now count failed calls separately, which the PR describes. No blocking risk remains.

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main change: producing one telemetry row per tool call with duration and outcome.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create a new PR

---

_{A rabbit reads each line,
The patch grows clear beneath the moon,
Small changes hop in place,
Tests guard the garden path,
Reviews bloom before the dawn.}

_{Comment @coderabbitai help to get the list of available commands.}