The notification service was holding a PARTIAL_WAKE_LOCK indefinitely,
preventing the CPU from ever sleeping. Combined with OkHttp's 30-second
ping interval, this caused continuous CPU wake-ups — worse than the
previous polling approach.
Changes:
- Remove permanent WakeLock that kept CPU running 24/7
- Disable OkHttp's 30s ping interval (useless without WakeLock in Doze)
- Add AlarmManager keepalive using setAndAllowWhileIdle() every ~8 min
- Acquire brief WakeLock (15s timeout) only during keepalive checks
- Track connection liveness via relay message timestamps
- Reconnect automatically when Doze kills the TCP connection
- Missed events are recovered via the existing 'since' timestamp
Expected improvement: ~3 brief wake-ups/hour vs continuous CPU usage.
Notifications may be delayed up to ~8 min in deep sleep.
Chad Curtis
07123403cd