fix(opencode): broaden transport errno coverage and stop retrying permanent DNS failures

[Astro-Han]

Summary

Broadens the stream transport-failure classifier and makes its retryability per-errno. Before, classifyStreamFailure recognized only 4 errno codes (ECONNRESET/ECONNREFUSED/ETIMEDOUT/UND_ERR_SOCKET) and hardcoded every transport disconnect as retryable. Two gaps:

• Common transient codes (DNS, undici timeouts, broken pipe) fell through to UnknownError.
• A permanent failure like ENOTFOUND (an unresolved host — usually a misconfigured base URL) was retried up to the cap, stalling the turn instead of surfacing the cause.

Changes:

• stream-failure-classifier.ts: broaden TRANSPORT_CODES (EAI_AGAIN, EPIPE, ECONNABORTED, EHOSTUNREACH, ENETUNREACH, ENOTFOUND, UND_ERR_CONNECT_TIMEOUT, UND_ERR_HEADERS_TIMEOUT, UND_ERR_BODY_TIMEOUT) + a tiny message fallback (socket hang up / premature close) for disconnects with no errno code. An error carrying an HTTP statusCode is short-circuited up front — an HTTP response was received, so it is an API error classified by status downstream (the APICallError branch in fromError, which runs after this) and must not be reclassified as transport by a transport-coded cause or a transport phrase in its message. "terminated" is intentionally not in the fallback (on its own it is undici's generic wrapper; only transport when its cause carries a transport code).
• Per-errno retryability: TransportDisconnect.retryable is now a boolean. ENOTFOUND is non-retryable; EAI_AGAIN (transient DNS) and all others stay retryable.
• message-v2.ts fromError: propagate transport.retryable to APIError.isRetryable instead of hardcoding true.
• retry.ts classifyRetry: transport_disconnect now honors isRetryable (removed from RETRY_TRANSIENT_KINDS, given its own branch), so a non-retryable transport error takes the existing "not retryable → stop" path. retrySignal.retryable also feeds the recorded transport failure, so run-observability's retry_safety reports provider_terminal_failure for ENOTFOUND via the existing fix(opencode): stop recommending auto-retry for terminal provider failures #1118 branch.

Why

Split out of the error-classification work (#1105 / #1123) as its own slice because it changes retry strategy, not just classification. ENOTFOUND from a misconfigured base URL should fail fast with the real cause, not retry-storm into a stall; and the narrow errno set was dropping recoverable transient disconnects into opaque UnknownError.

Related Issue

Refs #1105 (classification spine), #1123 (classify-then-render umbrella).

Human Review Status

Pending

Review Focus

• The statusCode short-circuit: it must not drop a real raw transport error (raw transport errors carry code/cause.code and no HTTP status; only HTTP responses have statusCode).
• classifyRetry now honors isRetryable for transport_disconnect — confirm existing transport errnos (all retryable) still retry; only ENOTFOUND stops.
• Dropping "terminated" from the message fallback (kept consistent with the existing over-match guard).

Risk Notes

Behavior change (intentional, tested): ENOTFOUND no longer auto-retries. classifyRetry's transient-kind contract changed for transport_disconnect (now honors isRetryable); one existing test that asserted "transport_disconnect retries even when isRetryable is false" was updated to the new per-errno contract. No data migration. No UI.

How To Verify

bun test src/session/stream-failure-classifier.test.ts src/session/retry.test.ts test/session/message-v2.test.ts → 111 pass, 0 fail
bun test test/session src/session test/provider src/provider → 1328 pass, 4 skip, 1 todo, 0 fail
tsgo --noEmit (packages/opencode) → clean
codex review (xhigh, 3 rounds) → final fresh-eye pass: no P0/P1/P2/P3

Pinned: ENOTFOUND → non-retryable transport (incl. nested cause); EAI_AGAIN / UND_ERR_HEADERS_TIMEOUT → retryable; bare "socket hang up" → retryable via message fallback; HTTP error with a transport phrase or transport-coded cause stays classified by status; classifyRetry honors isRetryable for transport_disconnect both ways.

Screenshots or Recordings

N/A — backend transport-classification change, no UI.

Checklist

• Type label — this PR carries exactly one of bug, enhancement, task, documentation. Type labels are author-added; the labeler bot does NOT assign them. Add the label in the GitHub UI, then tick this.
• Routing labels — this PR carries at least one of app, ui, platform, harness, ci. The labeler bot assigns these on PR open based on changed paths. Confirm the bot's choice (or override if wrong), then tick this.
• Priority label — this PR carries exactly one of P0, P1, P2, P3. The priority-triage bot suggests one on PR open. Confirm or override, then tick this.
• Human Review Status above is set to Pending, Approved by @<reviewer>, or Not required: <reason> (default is Pending; "not required" is restricted to bot-authored low-risk PRs).
• I linked the related issue, or stated in Summary why there is no issue.

https://claude.ai/code/session_015bW9JQSkuB156gkNQdxCzi

[coderabbitai]

Warning

Review limit reached

@Astro-Han, we couldn't start this review because you've reached your PR review rate limit.

More reviews will be available in 26 minutes and 4 seconds. Learn how PR review limits work.

Your organization has used up its prepaid credits, and credit purchases are no longer available. Enable the review add-on in the billing tab to keep reviews running — you're only billed for reviews past your plan's rate limits ($0.25/file).

⌛ How to resolve this issue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based credits.

🚦 How do rate limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan refill rate.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, the refill rate gradually slows as usage increases. The highest same-day bursts are limited more strictly.

Please see our Fair Usage Limits Policy for further information.

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: d71ce1b7-e42b-4a2d-9f72-59b37e9bb50f

📥 Commits

Reviewing files that changed from the base of the PR and between f433411 and 6e39ced.

📒 Files selected for processing (6)

• packages/opencode/src/session/message-v2.ts
• packages/opencode/src/session/retry.test.ts
• packages/opencode/src/session/retry.ts
• packages/opencode/src/session/stream-failure-classifier.test.ts
• packages/opencode/src/session/stream-failure-classifier.ts
• packages/opencode/test/session/message-v2.test.ts

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch claude/i1105b-transport-errno

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[github-actions]

Suggested priority: P2 (includes non-doc, non-test paths outside the low-risk bucket).

P1/P0 are reserved for maintainer confirmation. Please relabel manually if this is a release blocker, security issue, data-loss risk, or updater/runtime failure.