[mountainowl]

Requested by Ajay Singh · project thread

What changed

• Before: bubo service calls poll() in one process every service_poll_interval_seconds, which defaults to 900. Each poll sent session_start, so an idle install sent about 96 liveness events a day, and the event name suggested a human session that wasn't there.
• After: session_start is sent once per process per UTC day. A long-running idle service still checks in daily, so it keeps showing as alive. run-once, the Action and CLI polls are separate processes, so each of them still sends one.
• How:
  • record_session_start remembers the UTC day it last sent under a lock.
  • It only marks the day once analytics consent is on, so turning consent on later in the same process still sends that day's event.

Verification

• tests/test_session_start.py drives the real poll() with an idle config and counts the queued PostHog events.
• On main, 3 polls sent 3 events and 4 polls across 2 days sent 4. With this change they send 1 and 2. The third test checks that a poll with consent off doesn't use up the day; it passes on both.
• The full suite on Python 3.14 gives 1930 passed. The only failures are the 2 container-only test_ui_export failures, which also fail on main.
• ruff check and cz check pass.

---

Generated by Claude Code