feat: add retry with exponential backoff for LLM API calls #40

Merged

pook merged 1 commit from feature/llm-retry-mechanism into main

2026-04-09 06:13:49 -04:00

pook commented

2026-04-08 11:19:17 -04:00

Owner

Summary

Adds a reusable withRetry() utility with exponential backoff (1s, 2s, 4s, max 3 retries)
Wraps the OpenAI chat.completions.create call in llm.ts with the retry mechanism
Retries only on transient errors: HTTP 429, 500, 502, 503, 504, and network timeouts (ECONNREFUSED, ECONNRESET, ETIMEDOUT)
Does NOT retry client errors (400, 401, 403) — these fail immediately
Logs each retry attempt with attempt number and error reason

Files changed

packages/api/src/services/retry.ts — new retry utility module
packages/api/src/services/llm.ts — wraps OpenAI call with withRetry()
packages/api/tests/unit/retry.test.ts — 23 unit tests covering all retry scenarios

Test plan

23 unit tests pass (bun test packages/api/tests/unit/retry.test.ts)
Tests cover: retryable status codes, non-retryable client errors, network errors, retry exhaustion, exponential backoff timing, logging

🤖 Generated with Claude Code

Closes #37

## Summary - Adds a reusable `withRetry()` utility with exponential backoff (1s, 2s, 4s, max 3 retries) - Wraps the OpenAI `chat.completions.create` call in `llm.ts` with the retry mechanism - Retries only on transient errors: HTTP 429, 500, 502, 503, 504, and network timeouts (ECONNREFUSED, ECONNRESET, ETIMEDOUT) - Does NOT retry client errors (400, 401, 403) — these fail immediately - Logs each retry attempt with attempt number and error reason ## Files changed - `packages/api/src/services/retry.ts` — new retry utility module - `packages/api/src/services/llm.ts` — wraps OpenAI call with `withRetry()` - `packages/api/tests/unit/retry.test.ts` — 23 unit tests covering all retry scenarios ## Test plan - [x] 23 unit tests pass (`bun test packages/api/tests/unit/retry.test.ts`) - [x] Tests cover: retryable status codes, non-retryable client errors, network errors, retry exhaustion, exponential backoff timing, logging 🤖 Generated with [Claude Code](https://claude.com/claude-code) Closes #37

pook added 1 commit

2026-04-08 11:19:17 -04:00

feat: add retry with exponential backoff for LLM API calls c5d82d938f

Wraps OpenAI API calls in a retry mechanism that handles transient
failures (429, 500, 502, 503, 504, network timeouts) with exponential
backoff (1s, 2s, 4s, max 3 retries). Client errors (400/401/403) are
not retried. Each retry attempt is logged with attempt number and reason.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

pook referenced this pull request

2026-04-08 18:00:49 -04:00

Close stale issues that already have open PRs addressing them #93

pook commented

2026-04-08 18:03:13 -04:00

Author

Owner

Review notes (agent-bot):

Issue link: Added Closes #37 to the PR description. ✓
Retry logic looks well-scoped: correct transient error codes, no retry on 4xx client errors.
Consider adding jitter to the exponential backoff to avoid thundering herd on recovery (multiple callers retrying at the same interval).
23 unit tests is good coverage for the retry utility.

**Review notes (agent-bot):** - Issue link: Added `Closes #37` to the PR description. ✓ - Retry logic looks well-scoped: correct transient error codes, no retry on 4xx client errors. - Consider adding jitter to the exponential backoff to avoid thundering herd on recovery (multiple callers retrying at the same interval). - 23 unit tests is good coverage for the retry utility.

pook merged commit 86614a4d83 into main

2026-04-09 06:13:49 -04:00

pook referenced this pull request from a commit

2026-04-09 06:13:51 -04:00

Merge pull request 'feat: add retry with exponential backoff for LLM API calls' (#40) from feature/llm-retry-mechanism into main

pook referenced this pull request

2026-04-09 06:15:40 -04:00

PR pipeline unblocked: 18 PRs merged, 48 remaining need rebase #254