How it runs

The scheduler is started in the FastAPI lifespan hook. Every job is registered via scheduler.add_job(fn, trigger, id=..., replace_existing=True), which means the process can be restarted safely without orphaned duplicates.

pythonapp/core/scheduler.py (abbreviated)

scheduler = AsyncIOScheduler()
scheduler.add_job(
    booking_reminders,
    IntervalTrigger(hours=1),
    id="booking_reminders",
    replace_existing=True,
)
# ... 39 more jobs
scheduler.start()

Timezone

Business-hour jobs use WITA (Central Indonesia Time, UTC+8) so that nightly and daily jobs fire at sensible Bali-local hours regardless of where the container actually runs. Pure interval jobs are timezone-agnostic.

All 39 jobs

Job ID	Schedule	Purpose
`booking_reminders`	`interval · 1h`	Send upcoming-booking reminders to users.
`re_engagement`	`cron · Mon 10:00 WITA`	Weekly re-engagement nudge to lapsed users.
`venue_daily_summary`	`cron · 22:00 WITA daily`	Evening ops summary for every venue owner.
`monthly_invoice`	`cron · 1st of month 09:00 WITA`	Generate monthly subscription invoices.
`retry_failed_sends`	`interval · 60s`	Retry outbound messages in the send queue.
`b2b_session_cleanup`	`cron · 03:00 WITA daily`	Purge expired B2B onboarding sessions.
`deliver_reminders`	`interval · 5m`	Deliver due reminders via the channel sender.
`retry_pending_fees`	`interval · 30m`	Retry Stripe fee collections that failed.
`reap_stale_envelopes`	`interval · 15m`	Mark PENDING envelopes past TTL as EXPIRED.
`sweep_orphaned_responses`	`interval · 30m`	Close out venue responses with no matching booking.
`auto_confirm_stale_bookings`	`interval · 1h`	Auto-confirm bookings venues forgot to accept.
`purge_deleted_users`	`cron · 04:00 WITA daily`	Hard-delete users past the retention window.
`mark_completed_bookings`	`interval · 1h`	Transition past-slot bookings to COMPLETED.
`reconcile_subscription_state`	`cron · 05:00 WITA daily`	Reconcile local subscription rows with Stripe state.
`reconcile_fee_status`	`interval · 30m`	Reconcile fee collection statuses against the ledger.
`recover_stuck_bookings`	`interval · 15m`	Unstick bookings caught between A2A states.
`cancel_orphaned_bookings`	`interval · 30m`	Cancel bookings whose envelope has been abandoned.
`retry_dispatch_failed`	`interval · 30m`	Re-dispatch envelopes stuck in DISPATCH_FAILED.
`expire_stale_referrals`	`cron · 06:00 WITA daily`	Expire unredeemed referral codes past their TTL.
`mark_abandoned_b2b_sessions`	`cron · 03:30 WITA daily`	Mark half-finished B2B onboarding flows as abandoned.
`reconcile_venue_compound_state`	`cron · 04:00 WITA daily`	Reconcile venue compound state (is_active, is_claimed, etc).
`check_price_watchers`	`interval · 6h`	Refresh price watchers on Shopee/Tokopedia/Lazada products.
`daily_nutrition_summary`	`cron · 21:00 WITA daily`	Send users their calorie diary summary.
`light_dreaming`	`cron · 02:00 WITA daily`	Lightweight nightly reflection over recent interactions.
`deep_dreaming`	`cron · Mon 04:00 WITA`	Weekly deep reflection producing new agent insights.
`dlq_cleanup`	`cron · 02:15 WITA daily`	Trim the dead-letter queue of processed failures.
`dlq_weekly_report`	`cron · Mon 06:00 WITA`	Weekly DLQ health report to the alerts channel.
`faq_compilation`	`cron · Sun 03:00 WITA`	Cluster support conversations into draft FAQ entries.
`memory_eviction`	`cron · 03:30 WITA daily`	Evict stale user memory entries past the keep window.
`reset_api_key_queries_today`	`cron · 08:00 WITA (00:00 UTC)`	Reset the per-API-key daily query counter.
`charge_unbilled_usage`	`cron · 05:00 WITA daily`	Batch metered ApiKeyUsageEvent rows to Stripe.
`drain_webhook_retries`	`interval · 2m`	Retry failed outbound webhook deliveries.
`prune_error_logs`	`cron · 05:30 WITA daily`	Delete ErrorLog rows past the retention window.
`platform_health_monitor`	`interval · 15m`	Run platform health checks and emit alerts on regression.
`rating_prompt`	`interval · 1h`	Prompt users to rate recent bookings.
`spend_digest`	`cron · 09:00 WITA daily`	Send daily user-spend digest.
`rotate_delta_log`	`cron · Sun 04:30 WITA`	Rotate the fleet delta log.
`faq_extraction`	`cron · Sun 23:59 WITA`	Extract FAQ candidates from the week's conversations.
`rebuild_entity_graph`	`interval · 1h`	Rebuild derived edges in the knowledge graph.

Inspecting running jobs

The /admin/scheduler endpoint (protected by INTERNAL_API_TOKEN) returns the current job registry with next run times. Use it to verify a deploy didn't silently drop a job.

bash

curl -H "X-Internal-Token: $INTERNAL_API_TOKEN" \
  https://api.agntdot.com/admin/scheduler

Nightly heavy work

A few jobs under cron/ are too big to live in scheduler.py directly. They are still registered as scheduler jobs but their implementation lives in dedicated files:

cron/dreaming.py— light + deep dreaming reflections
cron/graph_builder.py— knowledge graph rebuild
cron/faq_compile.py— weekly FAQ clustering
cron/wiki_reconcile.py— fleet wiki reconciliation

Scheduler.

How it runs

Timezone

All 39 jobs

Inspecting running jobs

Nightly heavy work

Related