fix: do not retry or eject endpoint on client-side timeout

[killme2008]

Problem

A gRPC DEADLINE_EXCEEDED surfaces as TimeoutError (see promise-adapter), and it was classified as retriable in both modes and as an endpoint-health failure. Both are wrong for a client-side timeout:

• Retry multiplies the latency budget. timeoutMs is the caller's hard latency budget for the write, but each attempt resets the deadline (unaryCall does new Date(Date.now() + deadlineMs) per attempt). Retrying a timeout therefore silently stretched the budget to maxAttempts × timeoutMs + backoff — a caller asking for 60s could wait minutes — and usually just timed out again.
• It ejected healthy endpoints. isEndpointFailure(TimeoutError) returned true, so a tight caller deadline reported the endpoint as failed and could eject it. A timeout reflects the caller's clock, not the endpoint's health.

This also aligns the TS client with the Go ingester, which already excludes DEADLINE_EXCEEDED from both retry and endpoint-failure classification.

Fix

• Make TimeoutError non-retriable in every mode via an early return (same treatment as AbortedError).
• Exclude TimeoutError from isEndpointFailure.
• Drop the dead DEADLINE_EXCEEDED (4) entries from CONSERVATIVE_RETRIABLE_GRPC_CODES and ENDPOINT_FAILURE_GRPC_CODES: promise-adapter always maps code 4 to TimeoutError, so a TransportError with code 4 never reaches those sets.

The server-side business DeadlineExceeded (status 1008) was already non-retriable and is unchanged.

Tests

• Updated test/unit/errors.test.ts: TimeoutError is non-retriable in both modes; raw transport code 4 is non-retriable; TimeoutError is not an endpoint failure.
• Full unit suite (207 tests), typecheck, and lint all pass.

[Copilot]

Pull request overview

This PR corrects retry and endpoint-health classification for client-side gRPC timeouts by treating DEADLINE_EXCEEDED surfaced as TimeoutError as neither retriable nor an endpoint failure, preventing inflated latency budgets and accidental ejection of healthy endpoints.

Changes:

• Make TimeoutError non-retriable in both retry modes and remove DEADLINE_EXCEEDED (4) from conservative retriable code sets.
• Exclude TimeoutError (and transport code 4) from endpoint-failure detection so client deadlines don’t affect endpoint health.
• Update unit tests to cover the new retry/health behavior and bump SDK version to 0.2.1.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/errors.ts	Updates retry and endpoint-failure classification to exclude TimeoutError / gRPC code 4 and adjusts related docs/comments.
test/unit/errors.test.ts	Adds/updates tests ensuring TimeoutError and transport code 4 are non-retriable and not endpoint failures.
test/unit/client-failover.test.ts	Adds a failover test asserting no retries and no endpoint-health reporting on TimeoutError.
src/version.ts	Bumps internal SDK version to 0.2.1.
package.json	Bumps package version to 0.2.1.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.