fix(google): restart the realtime session on abnormal WebSocket close

[anzemur]

What

When the Gemini Live WebSocket dies abnormally, the session never recovers. In our production voice agent this branch fires for two distinct error groups: bare WebSocket closed with code 1006 (a few per day) and closes carrying Google's The operation was aborted. reason (TLS connection dropped server-side) - the latter at ~2,800 events / ~1,800 users over two months, ~30 live sessions dying per day. Every one of them passed our teardown filter, i.e. genuinely mid-conversation:

• onerror (network-level failures) calls markRestartNeeded(), so #mainTask's reconnect loop tears down, reconnects, and re-seeds the chat context.
• onclose with a non-normal code only logs, emits an unrecoverable error, and marks the generation done. Nothing sets sessionShouldClose, so the main task stays parked on sessionShouldClose.wait() with a dead socket — the user keeps talking into a session that will never answer. Outbound sends on the closed ws socket don't throw, so the retry path in #mainTask's catch never triggers either.

Fix

Mirror onerror in onclose: on a non-normal close while the session isn't already closing, call markRestartNeeded() so the main task reconnects and re-seeds history. The emitted error is marked recoverable: true in that case — a restart is coming, and apps that treat unrecoverable errors as fatal shouldn't tear down a session that's about to recover. When the session is already closing (sessionShouldClose set), behavior is unchanged (recoverable: false, no restart).

Same as the existing onerror path, these restarts don't consume the retry budget; a repeatedly failing endpoint behaves as it does today for network errors.

Tests

realtime_reconnect.test.ts drives the live-connect callbacks through a mocked @google/genai:

• abnormal close (1006) → a second connect happens and the error is emitted with recoverable: true (fails without the fix: no reconnect, recoverable: false),
• normal close (1000) → no reconnect, no error emitted.

Full @livekit/agents-plugin-google suite passes (59 tests).

Live verification

Applied this change to the installed 1.7.0 dist of a production voice agent and killed the real Gemini socket mid-session (ws.terminate() on the underlying ws@8.19.0 socket — the client observes exactly a 1006), with a probe question whose answer (47) existed only in the seeded chat history:

• Unfixed: WebSocket closed with code 1006 logged, no reconnect ever, and one step worse than the parked loop: the recoverable: false error makes AgentSession close outright (AgentSession is closing due to an unrecoverable error — ev.type is undefined so _onError's maxUnrecoverableErrors counting is skipped). The session dies mid-conversation.
• Fixed: reconnect fires 3 ms after the close (sessionShouldClose:true → new Connecting to Gemini Realtime API...), recoverable: true keeps the AgentSession alive, and the model answers 47. — proving both the reconnect and the chat-context re-seed.
• Full agent smoke suite passes on the fixed build (opening, barge-in, seeded history, typed text, tool injection, closing TTS).

So beyond un-parking the main task, recoverable: true also prevents the framework-level session teardown on a transient socket drop.

[changeset-bot]

🦋 Changeset detected

Latest commit: b616a87

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 39 packages

Name	Type
@livekit/agents-plugin-google	Patch
@livekit/agents	Patch
@livekit/agents-plugin-anam	Patch
@livekit/agents-plugin-anthropic	Patch
@livekit/agents-plugin-assemblyai	Patch
@livekit/agents-plugin-azure	Patch
@livekit/agents-plugin-baseten	Patch
@livekit/agents-plugin-bey	Patch
@livekit/agents-plugin-cartesia	Patch
@livekit/agents-plugin-cerebras	Patch
@livekit/agents-plugin-deepgram	Patch
@livekit/agents-plugin-did	Patch
@livekit/agents-plugin-elevenlabs	Patch
@livekit/agents-plugin-fishaudio	Patch
@livekit/agents-plugin-hedra	Patch
@livekit/agents-plugin-hume	Patch
@livekit/agents-plugin-inworld	Patch
@livekit/agents-plugin-krisp	Patch
@livekit/agents-plugin-lemonslice	Patch
@livekit/agents-plugin-liveavatar	Patch
@livekit/agents-plugin-livekit	Patch
@livekit/agents-plugin-minimax	Patch
@livekit/agents-plugin-mistral	Patch
@livekit/agents-plugin-mistralai	Patch
@livekit/agents-plugin-neuphonic	Patch
@livekit/agents-plugin-openai	Patch
@livekit/agents-plugin-perplexity	Patch
@livekit/agents-plugin-phonic	Patch
@livekit/agents-plugin-protoface	Patch
@livekit/agents-plugin-resemble	Patch
@livekit/agents-plugin-rime	Patch
@livekit/agents-plugin-runway	Patch
@livekit/agents-plugin-sarvam	Patch
@livekit/agents-plugin-silero	Patch
@livekit/agents-plugin-soniox	Patch
@livekit/agents-plugin-tavus	Patch
@livekit/agents-plugin-trugen	Patch
@livekit/agents-plugin-xai	Patch
@livekit/agents-plugins-test	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[devin-ai-integration]

🔍 Devin Review: 1 flag

Not posted on this PR by your GitHub settings — view it in Devin Review. (Configure)