fix(voice): align outbound WhatsApp call lifecycle with real pickup

A pile of related fixes around the dashboard's WhatsApp call flow:

- ConversationHeader / Contacts/VoiceCallButton: drop the immediate
  setCallActive at initiate time. The call sits in incomingCalls
  (callDirection: outbound) until the backend signals real pickup, so
  the duration timer never starts pre-pickup. Phone button is disabled
  whenever there is an active or incoming call.

- FloatingCallWidget:
  * Loop a ringtone (bell.mp3) for inbound ringing only.
  * Hide the green Join button for outbound — the agent has nothing to
    "join", and clicking it routed through acceptIncomingCall →
    prepareInboundAnswer → cleanup() and tore down the live outbound
    session before the API 409 ("already accepted by another agent").
  * Auto-join watcher skips whatsapp outbound (Twilio's joinConference
    flow only).

- useCallSession:
  * joinCall short-circuits for outbound calls — defense-in-depth so
    no future surface can re-trigger the destroyed-session bug.
  * endCall + outbound rejectIncomingCall pass call.callId to
    endActiveCall, so terminate fires even if module state was wiped.

- useWhatsappCallSession:
  * New recorderArmed flag, reset by cleanup. ontrack only calls
    setupRecorder when armed.
  * Inbound's acceptIncomingCall arms the recorder before the API
    round-trip (agent click = pickup).
  * armOutboundRecorder exported for the cable handler when ACCEPTED
    arrives.
  * endActiveCall accepts a callIdOverride to fall back when the
    module's activeCallId was nulled by an earlier cleanup.

- actionCable:
  * Split the cable contract: outbound_connected only applies the SDP
    answer (tunnel-up signal); outbound_accepted (new) is the real
    pickup signal — flips active and arms the recorder.
This commit is contained in:
Tanmay Deep Sharma
2026-05-05 14:15:40 +07:00
parent 4a79abb367
commit e6a35193f3
6 changed files with 156 additions and 44 deletions
@@ -93,6 +93,8 @@ const startWhatsappCall = async inboxId => {
}
const callsStore = useCallsStore();
// Stay non-active until the connect cable event arrives — flipping to active
// here would start the duration timer before the contact picks up.
callsStore.addCall({
callSid: response.call_id,
callId: response.id,
@@ -101,7 +103,6 @@ const startWhatsappCall = async inboxId => {
callDirection: 'outbound',
provider: 'whatsapp',
});
callsStore.setCallActive(response.call_id);
useAlert(t('CONTACT_PANEL.CALL_INITIATED'));
navigateToConversation(conversationId);
@@ -1,5 +1,5 @@
<script setup>
import { computed, ref, watch } from 'vue';
import { computed, onBeforeUnmount, ref, watch } from 'vue';
import { useRouter } from 'vue-router';
import { useStore } from 'vuex';
import { useCallSession } from 'dashboard/composables/useCallSession';
@@ -7,6 +7,8 @@ import { setWhatsappCallMuted } from 'dashboard/composables/useWhatsappCallSessi
import WindowVisibilityHelper from 'dashboard/helper/AudioAlerts/WindowVisibilityHelper';
import Avatar from 'dashboard/components-next/avatar/Avatar.vue';
const RINGTONE_URL = '/audio/dashboard/bell.mp3';
const router = useRouter();
const store = useStore();
@@ -97,12 +99,15 @@ const handleJoinCall = async call => {
}
};
// Auto-join outbound calls when window is visible
// Auto-join outbound calls when window is visible. WhatsApp outbound has no
// separate join step (the offer was sent at initiate time and the answer is
// applied directly by the cable handler), so this only covers Twilio.
watch(
() => incomingCalls.value[0],
call => {
if (
call?.callDirection === 'outbound' &&
call?.provider !== 'whatsapp' &&
!hasActiveCall.value &&
WindowVisibilityHelper.isWindowVisible()
) {
@@ -111,6 +116,37 @@ watch(
},
{ immediate: true }
);
// Loop the ringtone while an inbound call is unanswered. Stop the moment any
// call is active (we joined), every inbound call cleared, or the widget tears
// down. Browser autoplay may reject the first play() if the tab has no prior
// user gesture; that's fine — the visual widget still surfaces the call.
const ringtone = new Audio(RINGTONE_URL);
ringtone.loop = true;
ringtone.volume = 1;
const stopRingtone = () => {
ringtone.pause();
ringtone.currentTime = 0;
};
const ringingInbound = computed(() =>
incomingCalls.value.some(call => call.callDirection !== 'outbound')
);
watch(
() => ringingInbound.value && !hasActiveCall.value,
shouldRing => {
if (shouldRing) {
ringtone.play().catch(() => {});
} else {
stopRingtone();
}
},
{ immediate: true }
);
onBeforeUnmount(stopRingtone);
</script>
<template>
@@ -223,7 +259,9 @@ watch(
<i class="text-lg text-white i-ph-phone-x-bold" />
</button>
<button
v-if="!hasActiveCall"
v-if="
!hasActiveCall && incomingCalls[0]?.callDirection !== 'outbound'
"
class="flex justify-center items-center w-10 h-10 bg-n-teal-9 hover:bg-n-teal-10 rounded-full transition-colors"
@click="handleJoinCall(incomingCalls[0])"
>
@@ -107,6 +107,13 @@ const isWhatsappVoiceInbox = computed(
() => getVoiceCallProvider(inbox.value) === VOICE_CALL_PROVIDERS.WHATSAPP
);
const isWhatsappCallButtonDisabled = computed(
() =>
whatsappCallSession.isInitiating.value ||
callsStore.hasActiveCall ||
callsStore.hasIncomingCall
);
const startWhatsappCall = async () => {
if (whatsappCallSession.isInitiating.value) return;
try {
@@ -125,6 +132,8 @@ const startWhatsappCall = async () => {
return;
}
// Stay non-active until Meta delivers the connect webhook (sdp_answer);
// flipping to active here would start the duration timer before pickup.
callsStore.addCall({
callSid: response.call_id,
callId: response.id,
@@ -133,7 +142,6 @@ const startWhatsappCall = async () => {
callDirection: 'outbound',
provider: 'whatsapp',
});
callsStore.setCallActive(response.call_id);
} catch (error) {
useAlert(error?.message || t('CONVERSATION.HEADER.WHATSAPP_CALL_FAILED'));
}
@@ -207,7 +215,7 @@ const startWhatsappCall = async () => {
color="slate"
icon="i-lucide-phone"
:is-loading="whatsappCallSession.isInitiating.value"
:disabled="whatsappCallSession.isInitiating.value"
:disabled="isWhatsappCallButtonDisabled"
class="rounded-md hover:bg-n-alpha-2"
@click="startWhatsappCall"
/>
@@ -106,8 +106,11 @@ export function useCallSession() {
const findCall = callSid => callsStore.calls.find(c => c.callSid === callSid);
const endCall = async ({ conversationId, inboxId, callSid }) => {
if (isWhatsappCall(findCall(callSid))) {
await whatsappSession.endActiveCall();
const call = findCall(callSid);
if (isWhatsappCall(call)) {
// Pass call.callId so a wiped module state (e.g. a prior accept attempt
// tore down the WebRTC session) doesn't stop us hitting /terminate.
await whatsappSession.endActiveCall(call.callId);
durationTimer.stop();
callsStore.clearActiveCall();
return;
@@ -122,9 +125,15 @@ export function useCallSession() {
const joinCall = async ({ conversationId, inboxId, callSid }) => {
if (isJoining.value) return null;
const call = findCall(callSid);
// Outbound calls were initiated by this agent — there is no inbound offer
// to accept and the WebRTC session is already mid-handshake. Routing
// through acceptIncomingCall would call prepareInboundAnswer → cleanup()
// and destroy the live outbound session, then 409 from the backend.
if (call?.callDirection === 'outbound') return null;
isJoining.value = true;
try {
const call = findCall(callSid);
if (isWhatsappCall(call)) {
await whatsappSession.acceptIncomingCall({
callId: call.callId,
@@ -174,7 +183,14 @@ export function useCallSession() {
const rejectIncomingCall = callSid => {
const call = findCall(callSid);
if (isWhatsappCall(call) && call?.callId) {
whatsappSession.rejectIncomingCall(call.callId);
// Outbound calls that are still ringing must be terminated, not rejected
// (reject is the inbound-side verb on Meta's API). Pass call.callId so
// a wiped module state still hits /terminate.
if (call.callDirection === 'outbound') {
whatsappSession.endActiveCall(call.callId);
} else {
whatsappSession.rejectIncomingCall(call.callId);
}
} else {
TwilioVoiceClient.endClientCall();
}
@@ -13,6 +13,12 @@ let recorderChunks = [];
let audioContext = null;
let activeCallId = null;
let intentionallyClosing = false;
// Inbound calls record from the moment the agent clicks accept (their click =
// pickup). Outbound calls must wait — Meta's `connect` webhook (which lands
// during ringing) negotiates remote tracks ~20s before the contact actually
// answers, and we don't want pre-pickup audio in the recording. This flag is
// flipped to true by armOutboundRecorder() when the ACCEPTED status arrives.
let recorderArmed = false;
const ensureRemoteAudioElement = () => {
if (remoteAudioEl) return remoteAudioEl;
@@ -38,6 +44,7 @@ const playRemoteStream = stream => {
// that races cleanup still has data to upload.
const RECORDING_TIMESLICE_MS = 1000;
const ICE_GATHER_TIMEOUT_MS = 10000;
const RECORDER_MIME_CANDIDATES = [
'audio/webm;codecs=opus',
'audio/webm',
@@ -64,32 +71,6 @@ const waitForIceGatheringComplete = peer =>
});
});
const cleanup = () => {
if (mediaRecorder && mediaRecorder.state !== 'inactive') {
try {
mediaRecorder.stop();
} catch (_) {
/* noop */
}
}
if (audioContext && audioContext.state !== 'closed') {
audioContext.close().catch(() => {});
}
if (localStream) localStream.getTracks().forEach(t => t.stop());
if (remoteStream) remoteStream.getTracks().forEach(t => t.stop());
if (pc) pc.close();
if (remoteAudioEl) remoteAudioEl.srcObject = null;
pc = null;
localStream = null;
remoteStream = null;
mediaRecorder = null;
recorderChunks = [];
audioContext = null;
activeCallId = null;
intentionallyClosing = false;
};
const setupRecorder = () => {
if (!localStream || !remoteStream || mediaRecorder) return;
// createMediaStreamSource on a stream with no audio tracks wires up to
@@ -118,6 +99,33 @@ const setupRecorder = () => {
mediaRecorder.start(RECORDING_TIMESLICE_MS);
};
const cleanup = () => {
if (mediaRecorder && mediaRecorder.state !== 'inactive') {
try {
mediaRecorder.stop();
} catch (_) {
/* noop */
}
}
if (audioContext && audioContext.state !== 'closed') {
audioContext.close().catch(() => {});
}
if (localStream) localStream.getTracks().forEach(t => t.stop());
if (remoteStream) remoteStream.getTracks().forEach(t => t.stop());
if (pc) pc.close();
if (remoteAudioEl) remoteAudioEl.srcObject = null;
pc = null;
localStream = null;
remoteStream = null;
mediaRecorder = null;
recorderChunks = [];
audioContext = null;
activeCallId = null;
intentionallyClosing = false;
recorderArmed = false;
};
const buildPeerConnection = iceServers => {
const config = iceServers && iceServers.length ? { iceServers } : {};
pc = new RTCPeerConnection(config);
@@ -134,7 +142,10 @@ const buildPeerConnection = iceServers => {
remoteStream.addTrack(track);
});
playRemoteStream(remoteStream);
setupRecorder();
// Only arm the recorder when the call is actually accepted. For outbound
// this is the ACCEPTED status webhook; for inbound this is the agent's
// own click (acceptIncomingCall flips recorderArmed before returning).
if (recorderArmed) setupRecorder();
};
return pc;
};
@@ -255,6 +266,11 @@ export function useWhatsappCallSession() {
const sdpAnswer = await prepareInboundAnswer(offer, ice);
activeCallId = callId;
// Inbound: agent's click is the pickup. Arm the recorder before the API
// round-trip so when ontrack fires (triggered by setRemoteDescription
// back in prepareInboundAnswer) the recorder is already authorized.
recorderArmed = true;
setupRecorder();
await WhatsappCallsAPI.accept(callId, sdpAnswer);
};
@@ -287,16 +303,20 @@ export function useWhatsappCallSession() {
}
};
const endActiveCall = async () => {
if (!activeCallId) {
// callIdOverride is the call.id from the dashboard's calls store. Module
// `activeCallId` may be null after a prior accept attempt's cleanup() — but
// the call still exists on Meta and must still be terminated. Falling back
// to the override means hangup is robust to a wiped local session.
const endActiveCall = async (callIdOverride = null) => {
const callId = activeCallId || callIdOverride;
if (!callId) {
cleanup();
return;
}
intentionallyClosing = true;
const callIdSnapshot = activeCallId;
try {
await stopRecorderAndUpload(callIdSnapshot);
await WhatsappCallsAPI.terminate(callIdSnapshot).catch(() => {});
await stopRecorderAndUpload(callId);
await WhatsappCallsAPI.terminate(callId).catch(() => {});
} finally {
cleanup();
}
@@ -322,6 +342,15 @@ export const applyOutboundAnswer = async (callId, sdpAnswer) => {
await pc.setRemoteDescription({ type: 'answer', sdp: sdpAnswer });
};
// Called by the cable handler when Meta delivers status=ACCEPTED for the
// outbound call (real pickup). Flips the recorder gate and starts the
// MediaRecorder. Idempotent — safe if ontrack hasn't fired yet (setupRecorder
// bails until the remote stream has audio tracks; ontrack will retry).
export const armOutboundRecorder = () => {
recorderArmed = true;
setupRecorder();
};
export const cleanupWhatsappSession = () => cleanup();
export const handleWhatsappRemoteEnd = async callId => {
+22 -2
View File
@@ -7,6 +7,7 @@ import { useImpersonation } from 'dashboard/composables/useImpersonation';
import { useCallsStore } from 'dashboard/stores/calls';
import {
applyOutboundAnswer,
armOutboundRecorder,
handleWhatsappRemoteEnd,
} from 'dashboard/composables/useWhatsappCallSession';
@@ -42,6 +43,7 @@ class ActionCableConnector extends BaseActionCableConnector {
'copilot.message.created': this.onCopilotMessageCreated,
'voice_call.incoming': this.onVoiceCallIncoming,
'voice_call.outbound_connected': this.onVoiceCallOutboundConnected,
'voice_call.outbound_accepted': this.onVoiceCallOutboundAccepted,
'voice_call.ended': this.onVoiceCallEnded,
};
}
@@ -230,10 +232,28 @@ class ActionCableConnector extends BaseActionCableConnector {
});
};
// `connect` is the WebRTC tunnel-ready signal (fires ~20s before pickup
// for outbound). Apply the SDP answer so the handshake completes during
// ringing, but stay non-active until `outbound_accepted` arrives.
// eslint-disable-next-line class-methods-use-this
onVoiceCallOutboundConnected = data => {
onVoiceCallOutboundConnected = async data => {
if (data?.provider !== 'whatsapp' || !data.sdp_answer) return;
applyOutboundAnswer(data.id, data.sdp_answer).catch(() => {});
try {
await applyOutboundAnswer(data.id, data.sdp_answer);
} catch (_) {
/* noop */
}
};
// Real pickup signal — Meta sends status=ACCEPTED on the call when the
// contact answers. Flip active (timer starts) and arm the recorder.
// eslint-disable-next-line class-methods-use-this
onVoiceCallOutboundAccepted = data => {
if (data?.provider !== 'whatsapp') return;
const store = useCallsStore();
if (!store.calls.some(c => c.callSid === data.call_id)) return;
store.setCallActive(data.call_id);
armOutboundRecorder();
};
// eslint-disable-next-line class-methods-use-this