Scheduled Jobs System

[Radiergummi]

Description

Background job scheduling for recurring tasks like metadata updates, cleanup, and maintenance.

Key Features

• Job Scheduling: Cron-like scheduling
• Job Types: Metadata refresh, cleanup, backups
• Monitoring: Job status and history
• Error Handling: Retry logic and alerting
• Admin UI: View and manage scheduled jobs

Implementation Areas

• Job scheduler (node-cron, bull, etc.)
• Job definitions and handlers
• Status persistence
• Admin management UI

Files

• Plan: plans/scheduled-jobs.md
• Related: plans/background-metadata-queue.md

Getting Started

1. Evaluate scheduling libraries
2. Design job configuration schema
3. Implement core job runner

[m13v]

we run a scheduling system for recurring agent tasks and went through the same evaluation. ended up using launchd plists on macOS instead of node-cron or bull because the OS-level scheduler survives process crashes and restarts without any extra infrastructure. if your users are on Linux, systemd timers give you the same reliability.

for the job configuration schema, the most important thing we learned was separating the schedule definition from the job handler. our jobs are just prompts with metadata (interval, last_run, project_name) stored in postgres, and the scheduler just picks up whatever is due and dispatches it. that made it trivial to add new job types without touching the scheduler code.

the monitoring piece is critical - we log every run to a database with status, duration, and output summary. without that you end up with silent failures where a job has been broken for days and nobody notices. a simple dashboard query showing last_success_at per job catches most issues early.

[m13v]

our launchd plist-based scheduling system for recurring agent tasks: https://github.com/m13v/social-autoposter/tree/main/launchd

and the tmux-based agent orchestration that manages multiple concurrent scheduled agents: https://github.com/m13v/tmux-background-agents/blob/main/SKILL.md