fix: finalize native WS log on parse failure and clean close

[thiscantbeserious]

Summary

Every native-WS request that exchanges frames with an upstream must produce exactly one log row. Two independent code paths in tryNativeWSUpstream broke that invariant, leaving log entries stuck in "processing" until the 5-minute TTL sweeper silently evicted them.

Path A - parse failure: parseUpstreamWSEvent does a full sonic.Unmarshal into BifrostResponsesStreamResponse and returns nil on any error. A frame that is valid JSON but contains nested fields the struct cannot accept (observed with large ~57 KB response.completed frames from chatgpt.com/backend-api/codex) causes streamResp to stay nil. isTerminal never becomes true, StreamEndIndicator is never set, and PostHookRunner is never called with a terminal response, so the logging plugin never writes the DB row.

Path B - clean upstream close: when ReadMessage errors after at least one frame has been forwarded but no terminal post-hook has fired yet, the read loop wrote a 502 to the client and returned without calling PostHookRunner. The in-memory pending log entry was later silently evicted by the TTL sweeper with no DB trace.

Changes

• Add extractStreamEventType(data []byte) - performs a minimal single-field unmarshal to read only the type key as a fallback when the full struct parse fails. Returns an empty string on malformed JSON or missing type.
• Add synthesizeTerminalStreamResponse(provider, model, eventType) - builds a minimal BifrostResponsesStreamResponse with ExtraFields populated so downstream plugins (logging, tracing) can identify the request type, provider, and model.
• In tryNativeWSUpstream: add terminalPostHookFired guard. When the upstream WS closes (cleanly or with an error) after forwarding content but before a terminal post-hook fired, synthesize a response.completed event, set StreamEndIndicator=true, and call PostHookRunner once so the logging plugin writes the DB row. Clean closes (codes 1000, 1001, 1005) do not write a 502 to the client; abnormal closes still do.
• When the full struct parse fails on an in-flight frame, fall back to extractStreamEventType. If the raw type is a recognized terminal type, synthesize a minimal struct and fire the terminal post-hook so the log is finalized.
• Unit tests for extractStreamEventType (valid terminal, valid non-terminal, malformed JSON, missing type field, unknown nested fields), synthesizeTerminalStreamResponse field population, and isTerminalStreamType coverage.

Type

• Bug fix
• New feature
• Refactoring
• Documentation
• Other

Affected areas

• Transports (HTTP) - transports/bifrost-http/handlers/wsresponses.go

How to test

Run the unit tests directly:

go test ./transports/bifrost-http/handlers/... -run "TestExtractStreamEventType|TestSynthesizeTerminalStreamResponse|TestIsTerminalStreamType" -v
go test ./transports/bifrost-http/handlers/... -cover
go vet ./transports/bifrost-http/...

Manual reproduction for Path B (clean close without terminal event):

cat > /tmp/ws_mock_clean_close.py <<'PY'
import asyncio, json
import websockets

async def handler(ws):
    await ws.send(json.dumps({"type":"response.created","response":{"id":"resp_mock_1","status":"in_progress"},"sequence_number":0}))
    await asyncio.sleep(0.5)
    await ws.close(code=1000, reason="clean close")

async def main():
    async with websockets.serve(handler, "0.0.0.0", 9998):
        await asyncio.Future()

asyncio.run(main())
PY
python3 /tmp/ws_mock_clean_close.py &

sqlite3 bifrost-data/logs.db "SELECT COUNT(*) FROM logs"

printf '{"type":"response.create","model":"openai/gpt-4o","input":[{"type":"message","role":"user","content":[{"type":"input_text","text":"hi"}]}]}\n' \
  | websocat -n1 ws://localhost:8080/openai/v1/responses -H='Authorization: Bearer sk-fake-key'

sleep 2
sqlite3 bifrost-data/logs.db "SELECT COUNT(*) FROM logs"

Before this fix: count unchanged (no row written). After: count increases by one with status success.

Breaking changes

No.

Related

Closes #2997

This bug was discovered while working on PR #2775 in thiscantbeserious/bifrost. That branch contains a potential fix as a side-effect of the feature work. The fix has been extracted here as a focused standalone change.

[coderabbitai]

Important

Review skipped

Draft detected.

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: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 3556c8fc-da9d-42e1-9e07-a644dd929687

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

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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.}

[thiscantbeserious]

Superseded by #3018 (merged 2026-04-24), which bundles the fix for this issue and several other native WS reliability bugs. Closing this draft as redundant.