Deploy Monitor: PR #8082 — Parakeet built-in embedding

[beastoin]

Deploy Monitor — PR #8082 (Parakeet built-in speaker embedding)

PR: #8082
Merge commit: cd7d932
Image: gcr.io/based-hardware/parakeet:3ed1eb7
Deploy time: 2026-06-21 09:29 UTC

T+0 Checkpoint (09:33 UTC)

• Pod: prod-omi-parakeet-7df54ff54f-gx9fx — 1/1 Ready, 0 restarts
• Health: {"status":"healthy","ready":true}
• Batch metrics: 9 requests processed, 5 batches, 0 rejected, 0 pending
• Smoke test: /v2/transcribe returns 200 with speaker labels — PASS
• Traffic context: off-peak (Saturday morning UTC)
• Status: PASS

Monitoring schedule

Checkpoint	Time (UTC)	Status
T+0	09:33	PASS
T+20m	~09:53	pending
T+30m	~10:03	pending
T+1h	~10:33	pending
T+2h	~11:33	pending
T+4h	~13:33	pending
T+8h	~17:33	pending
T+12h	~21:33	pending
T+16h	01:33	pending
T+20h	05:33	pending
T+24h	09:33	pending

[beastoin]

T+0 Log Scan (mon)

Built-in embedding: NOT ACTIVE — silently falling back to HTTP diarizer.

Every request logs: WARNING:transcribe:pyannote.audio not installed, built-in embedding unavailable

Root cause: torch_audiomentations missing from Dockerfile --no-deps install list. pyannote.audio is pip-installed but fails to import: pyannote.audio → pyannote.audio.core.task → torch_audiomentations → ModuleNotFoundError

Impact: Zero regression — all requests fall back to existing HTTP diarizer (same behavior as before #8082). The code change is correct but the feature is dormant.

Fix: PR #8085 — one-line Dockerfile addition. Needs rebuild + redeploy to activate.

• Pod health: stable, 0 restarts
• Errors: 0 tracebacks
• httpx volume: unchanged (57 lines/5min, all POST to diarizer /v2/embedding)
• Status: PASS (no regression, feature inactive)

by AI for @beastoin

[beastoin]

Closing — monitoring updates moved to PR #8082 comments per manager preference.