Aller au contenu

🕐 Market Data Scheduler

LibreFolio embeds an automatic background daemon to fetch current prices, historical pricing, and exchange rates without requiring external scheduler systems (like crontab or Celery).

🏗️ Overview

The scheduler runs as an independent asyncio.Task inside the FastAPI lifespan loop. It periodically evaluates whether configured data sync jobs are due.

---
config:
  layout: elk
---
graph LR
    Start["🚀 Server Lifespan Startup"] --> Loop["🕐 Scheduler Loop (Every 60s)"]
    Loop --> Leader{"👑 am_i_leader?"}

    Leader -->|No| Sleep["💤 Sleep 60s"]
    Leader -->|Yes| Enabled{"⚙️ scheduler_enabled?"}

    Enabled -->|No| Sleep
    Enabled -->|Yes| CheckCurrent{"⏱️ Current price due?"}

    CheckCurrent -->|Yes| RunCurrent["⚡ run_current_price_refresh()"]
    CheckCurrent -->|No| CheckHistory{"📅 History sync due?"}

    RunCurrent --> CheckHistory

    CheckHistory -->|Yes| RunHistory["⚡ run_history_sync()"]
    CheckHistory -->|No| Sleep

    RunHistory --> Sleep
    Sleep --> Loop

🔄 Async Loop Daemon (scheduler.py)

The main entry point is scheduler_loop() located in backend/app/services/scheduler/scheduler.py.

  • Lifespan Task: It is spawned during FastAPI startup in backend/app/main.py using asyncio.create_task().
  • Shutdown Event: The loop listens to shutdown_event: asyncio.Event. When the event is set, it gracefully breaks the loop and completes.
  • Tick: After an initial 5-second delay, the leader re-reads the scheduler settings on every tick and skips both jobs while scheduler_enabled is false; the loop then sleeps 60 seconds in 5-second steps, so a shutdown is noticed quickly.
  • Environment Bypass: Any non-empty value of LIBREFOLIO_NO_SCHEDULER keeps the loop from starting. ./dev.py server --no-scheduler and the gallery screenshot generation set it to 1.

👑 Leader Election (leader.py)

To prevent multiple processes from running duplicate sync jobs when uvicorn runs several workers, only one process — the leader — executes the jobs. am_i_leader() decides on every tick with psutil, without any lock file:

  • Among the running, non-zombie children of the uvicorn master, the process with the lowest PID is the leader. If it dies, the next tick promotes the new lowest PID.
  • Fast paths return True: no parent process (e.g. PID 1 in a container), a parent started with --reload (one logical worker), a single sibling, or any introspection error — safe because the jobs are idempotent.

💼 Job Definitions (jobs.py)

The scheduler runs two types of background synchronization jobs defined in backend/app/services/scheduler/jobs.py:

⏱️ 1. Current Price Refresh

  • Purpose: Keeps the current prices of all active assets that have a price provider up-to-date. This updates the local pricing cache, feeding the frontend's live prices on the Assets list and the asset detail page.
  • Frequency: Defined by scheduler_current_price_frequency_minutes (default: every 10 minutes).
  • Execution: Calls AssetSourceManager.get_current_prices_bulk() (concurrency 3) on every active asset that has a provider assignment.

📅 2. History Sync

  • Purpose: Synchronizes historical price series (closing prices) for assets and FX rates.
  • Frequency: Configured for specific days of the week (e.g., mon,tue,wed,thu,fri,sat) and times of the day (e.g., 06:00 and 23:00), as local wall-clock values in scheduler_timezone (see Configuration Integration).
  • Execution: Calls AssetSourceManager.bulk_refresh_prices() for the same assets over [today − horizon, today] (scheduler_history_sync_horizon_days, default: 14 days), then sync_pairs_bulk() for every FX conversion route except those with a MANUAL step.

Both jobs classify their run as ok (every item succeeded, or there was nothing to process), partial, or error (no item succeeded).

📜 Execution Logs (joblog.py)

All executions are appended to logs/scheduler_jobs.jsonl in the data directory, one JSON object per run:

  • Header fields: ts (ISO timestamp with offset), job (current_price or history_sync), duration_s, and status (ok, partial or error).
  • summary: ok/err counts for the current-price refresh; assets_ok, assets_err, fx_ok and fx_err for the history sync.
  • Per-item detail: items for the current-price refresh (asset, ok, error); assets and fx for the history sync (status, provider, errors, and prices_changed/events_changed or points_changed).
  • Rotation: after each append, the file is trimmed to its last 500 entries (MAX_ENTRIES).

The last run of each job is also kept in scheduler_state.json at the root of the data directory (state.py, atomic write-then-rename). The admin UI reads both through admin-only endpoints: GET /api/v1/settings/scheduler/state (last runs, scheduler_timezone, the server's UTC clock) and GET /api/v1/settings/scheduler/log?since=ISO (newest first).

⚙️ Configuration Integration

Scheduler options are fully integrated into LibreFolio's database-backed Global Settings system. When an administrator modifies settings in the UI:

  1. Settings are saved to the GlobalSetting table.
  2. The frontend triggers a settings reload.
  3. The backend scheduler loop loads the settings dynamically on the next loop iteration (every 60 seconds), adapting the polling frequency and execution times instantly without requiring a server restart.

History-sync times and days are stored as local wall-clock values in scheduler_timezone. due_history_sync() builds each slot in that timezone on the current local date and converts it to a UTC instant only when deciding whether it is due, so the slots follow daylight saving time. A slot is due once its instant has passed and the last history run is older than it. An unknown timezone falls back to UTC; an empty or invalid day list falls back to Monday–Saturday. The backend does not validate these values: the configuration dialog enforces the ranges (1–1440 minutes, 1–365 days, at least one time and one day).

  • 📡 Live Prices — How the Assets list and the asset detail page poll current prices
  • 💰 Asset Architecture — Pricing provider pipelines and sync steps
  • ⚙️ Settings System — Dynamic global configuration variables