Sink Health Monitoring: Context Module & CRUD Operations

[joejohnson123]

Overview

Create the Sequin.Monitors context module providing all CRUD operations and business logic for sink monitors.

Depends on: #1 (Database Schema & Migrations)

Context: Sequin.Monitors

Monitor CRUD

• list_monitors_for_consumer(consumer_id) — returns all monitors for a sink, preloading notification channels
• list_monitors_for_account(account_id) — all monitors across all sinks for an account
• list_enabled_monitors() — all enabled monitors across all accounts (used by evaluation worker)
• get_monitor(id) — fetch by ID with preloads
• get_monitor!(id) — raising version
• create_monitor(sink_consumer_id, attrs) — creates monitor + nested notification channels
• update_monitor(monitor, attrs) — updates monitor + upserts/removes channels
• delete_monitor(monitor) — cascading delete
• toggle_monitor(monitor, enabled) — enable/disable shortcut

Notification Channel CRUD

• add_notification_channel(monitor, attrs)
• update_notification_channel(channel, attrs)
• remove_notification_channel(channel)

Alert Operations

• create_alert(monitor, attrs) — records a new alert
• resolve_alert(alert) — sets status: :resolved and resolved_at
• list_alerts_for_monitor(monitor_id, opts) — paginated, ordered by triggered_at desc
• get_active_alert(monitor_id) — gets current triggered (unresolved) alert if any
• prune_old_alerts(older_than) — cleanup alerts older than N days

Metric Samples (for anomaly detection)

• record_metric_sample(monitor, value) — inserts a sample
• get_recent_samples(monitor_id, window_seconds) — returns samples within window
• prune_old_samples(older_than) — removes samples older than 2 hours

Helper Functions

• current_metric_value(monitor) — fetches the live metric value:
  • :pending_messages → SlotMessageStore.count_messages(consumer)
  • :failed_messages → Consumers.count_messages_for_consumer(consumer, delivery_count_gte: 1)
• cooldown_active?(monitor) — checks if last alert was within cooldown_seconds

Files to Create

• lib/sequin/monitors/monitors.ex

Implementation Notes

• Preload notification_channels by default on monitor queries
• Use Repo.transact for create/update operations that touch channels
• Follow existing context patterns (see Sequin.Consumers, Sequin.Health)
• current_metric_value/1 needs to preload the sink_consumer association
• Consider caching monitor list for the evaluation worker (refreshed every cycle)

Acceptance Criteria

• All CRUD operations work correctly
• Nested channel create/update/delete within monitor operations
• current_metric_value/1 returns correct counts for both metrics
• cooldown_active?/1 correctly checks timing
• Proper error handling with Sequin.Error types
• Tests for all public functions