Stage 4: honest health, explicit error policy, event-bus integrity

Engine mode is now explicit: USE_MOCK_SIP=true is the only way to get
the mock engine; an unconfigured trunk fails startup with guidance
instead of silently degrading. Root-caused why the engine always ran
mock: nested pydantic-settings never read .env (no env_file on the
sub-settings classes) — all 8 now declare it.

/health stops lying: reports engine mode (sippy|mock), a live DB
SELECT 1, trunk registration state with reason, and TTS/STT
availability from their last real request; "healthy" now requires
ready + db + sippy + registered trunk.

Error policy: leaf services (tts/transcription/llm_client) raise and
track availability; call-loop callers catch, publish EventType.ERROR
naming the failed service, and apply an explicit fallback. Persistence
writes get one bounded 3x exponential retry, then an ERROR log — no
more silent data loss.

Event bus: a full subscriber queue drops its oldest event (counted)
instead of silently evicting the subscription; subscribe(replay_last=N)
delivers the advertised history replay, used by /ws/events (25).

Receptionist correctness: a matched TAKE_MESSAGE rule beats the LLM;
voicemail polls for early hangup and stops/transcribes/hangs up in
finally; RecordingSession finally keeps its leg_ids so taps detach.

Dead code removed: models/contact.py + Contact table, dtmf_buffer,
transcribe_stream stub, SMS stub in notification.py.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
2026-07-10 07:01:45 -04:00
parent 67a00defc3
commit 4048ce1db6
21 changed files with 492 additions and 357 deletions

View File

@@ -27,6 +27,8 @@ class TranscriptionService:
def __init__(self, settings: SpeachesSettings):
self.settings = settings
self._client: Optional[httpx.AsyncClient] = None
# Last-known reachability, surfaced by /health (None = no requests yet)
self.available: Optional[bool] = None
async def _get_client(self) -> httpx.AsyncClient:
"""Get or create the HTTP client."""
@@ -60,6 +62,9 @@ class TranscriptionService:
# Convert raw PCM to WAV format for the API
wav_data = self._pcm_to_wav(audio_data)
# Raises on failure — callers decide the per-call fallback and
# publish a service-error event; swallowing here made a down
# Speaches look like "the AI is deciding badly".
try:
response = await client.post(
"/v1/audio/transcriptions",
@@ -72,44 +77,13 @@ class TranscriptionService:
},
)
response.raise_for_status()
text = response.text.strip()
logger.debug(f"Transcription: '{text}'")
return text
except httpx.HTTPStatusError as e:
logger.error(f"Speaches API error: {e.response.status_code} {e.response.text}")
return ""
except httpx.ConnectError:
logger.error(f"Cannot connect to Speaches at {self.settings.url}")
return ""
except Exception as e:
logger.error(f"Transcription failed: {e}")
return ""
async def transcribe_stream(
self,
audio_data: bytes,
language: str = "en",
):
"""
Stream transcription — for real-time results.
Uses Speaches streaming endpoint if available,
falls back to chunked transcription.
Yields:
str: Partial transcription chunks
"""
# For now, do chunked transcription
# TODO: Implement WebSocket streaming when Speaches supports it
chunk_size = 16000 * 2 * 3 # 3 seconds of 16kHz 16-bit mono
for i in range(0, len(audio_data), chunk_size):
chunk = audio_data[i:i + chunk_size]
if len(chunk) > 0:
text = await self.transcribe(chunk, language)
if text:
yield text
except Exception:
self.available = False
raise
self.available = True
text = response.text.strip()
logger.debug(f"Transcription: '{text}'")
return text
async def close(self) -> None:
"""Close the HTTP client."""