Schedulers and enablement
Toolora starts its in-process schedulers from the FastAPI lifespan in
backend/server.py and stops them during shutdown. Registration establishes
which work this API process can schedule. It does not prove a job has completed,
a tenant is eligible or an external action succeeded.
Global and component enablement
Section titled “Global and component enablement”TOOLORA_DISABLE_SCHEDULERS skips these scheduler starts. Runtime heartbeat writes
have a separate TOOLORA_DISABLE_RUNTIME_HEARTBEATS control. A declared app state
of running is not evidence that every optional component started. Each module
also applies its own configuration and existing-instance guard.
| Scheduler family | Declared work/cadence | Additional gate |
|---|---|---|
| Fortnox | Outbox/pull, sweeps, backfills, walks, projections and reconciliation below | Configured integration; FORTNOX_SCHEDULER_DISABLED can opt out |
| Purchase invoices | Legacy mail 60 s; AP/payroll outboxes 30 s; payment feed 60 s; optional cost review 60 s | Legacy intake defaults enabled unless explicitly disabled; connector workers need connector URL/token; cost review needs the pilot tenant |
| Unified mail intake | One tick every 300 s, covering bounded sync/original-copy/interpretation work | Explicit MAIL_INTAKE_SCHEDULER_ENABLED; binding policy and feature flags still apply |
| Imports | Schedule scan every 30 s | IMPORTS_SCHEDULER_DISABLED; individual enabled/due import schedules determine actual work |
| Customer notifications | Durable notification outbox every 15 s | CUSTOMER_NOTIFICATIONS_SCHEDULER_DISABLED; recipient/sender/operation policy still applies |
| Planning AI | Recommendation reconciliation every 600 s | PLANNING_AI_SCHEDULER_DISABLED; the tick evaluates current feature/configuration eligibility |
The purchase and unified-mail paths coexist under guarded rollout settings. A communications-only mailbox does not acquire document intake simply because the mail scheduler runs. Check the binding and feature configuration before claiming that a specific document should have been interpreted.
Fortnox job inventory
Section titled “Fortnox job inventory”These jobs are registered by the configured Fortnox scheduler. Times are UTC, not Swedish wall time; daylight-saving changes alter their local presentation. The scheduled functions still select eligible companies and resource work.
| Job ID | Declared cadence |
|---|---|
fortnox_outbox |
Every 30 seconds |
fortnox_pull |
Every 5 seconds |
fortnox_lastmodified_sweep |
Every 300 seconds |
fortnox_invoice_payments |
Every 300 seconds |
fortnox_open_ar_backstop |
Daily 00:30 UTC |
fortnox_backfill_tick |
Every 60 seconds |
fortnox_partition_maintenance |
Daily 00:05 UTC |
fortnox_partition_audit |
Daily 03:30 UTC |
fortnox_walk_invoices |
Daily 01:30 UTC |
fortnox_walk_customers |
Daily 02:00 UTC |
fortnox_walk_suppliers |
Daily 02:30 UTC |
fortnox_walk_vouchers |
Daily 02:45 UTC |
fortnox_walk_supplier_invoices |
Sunday 01:00 UTC |
fortnox_walk_accounts |
Sunday 03:00 UTC |
fortnox_walk_cost_centers |
Sunday 04:00 UTC |
fortnox_parties_sync |
Daily 05:00 UTC |
fortnox_supplier_invoice_gold |
Daily 05:30 UTC |
fortnox_reconciliation |
Daily 03:00 UTC |
A registered backfill tick can do no work when no active backfill exists. Daily list walks, incremental sweeps, local projections and bookkeeping pushes retain different source/receipt meanings. Registration is not permission to test a real Fortnox write. Accounting owns the active adapter/connector split and test-company boundary.
Concurrency and progress
Section titled “Concurrency and progress”APScheduler’s reviewed defaults coalesce missed runs and use one instance per job within the scheduler. Misfire windows and thread pools differ by family. Those process-local controls do not coordinate multiple API processes. Durable work uses its own database claim, lease, lock or version predicate where implemented; those mechanisms are owned by the worker/RPC, not by APScheduler.
Selected required workers record successful completed ticks in WorkerProgress.
A lock-skipped tick or repeated exception must not advance success merely because
the scheduler is alive. Readiness compares expected components and current-instance
heartbeats; it is not a receipt for every pending invoice or source generation.
Stalled-work diagnosis explains the distinction.
For a synthetic pending document, first check global/component/binding enablement, then whether the appropriate tick completed and its durable job progressed. Do not start another scheduler instance to overcome a source or daily-budget refusal. Scheduler/progress tests use isolated process state and substituted workers. The PR records executions and makes no claim to have observed every deployed flag, cron run or provider result.