feat: sink health monitoring — database schema & migrations (Issue #1)

[joejohnson123]

What

Implements Issue #1 — foundational database tables and Ecto schemas for the sink health monitoring feature.

This is the first PR in the sink health monitoring series. All subsequent work (evaluation worker, notification dispatcher, UI, etc.) builds on these schemas.

New Tables (all in config_schema)

sink_monitors

Per-sink monitor configuration. Supports two types:

• Threshold: fires when metric ≥ value for a sustained window
• Anomaly: fires when metric deviates statistically from rolling average

Fields: name, enabled, metric (pending/failed messages), monitor_type, threshold_value, threshold_window_seconds, anomaly_sensitivity (low/medium/high), notify_on_recovery, cooldown_seconds

Unique constraint on (sink_consumer_id, name).

sink_monitor_notification_channels

Alert destinations per monitor. Supports: webhook, Slack, Discord, PagerDuty, Incident.io.

Config is encrypted via Sequin.Encrypted.Map. Each channel type has validation for required config fields (e.g. webhook requires url, Slack requires webhook_url).

sink_monitor_alerts

Alert history — tracks triggered and resolved incidents with metric values and timestamps.

sink_monitor_metric_samples

Rolling metric snapshots for anomaly detection (auto-pruned to 2 hours).

Schemas

All 4 schemas follow existing Sequin patterns:

• use Sequin.ConfigSchema + typed_schema
• Ecto.Enum for enum fields
• Changeset validation with type-specific field requirements
• Query helpers (where_account_id, where_enabled, etc.)

Build Note

⚠️ Could not run mix compile --warnings-as-errors in the sandbox environment (thread/process limits prevent native NIF compilation for ezstd, crc32cer, argon2_elixir). Schemas follow all existing conventions exactly — please verify compilation locally.

Closes

Closes #1

---

Summary by cubic

Introduces the database schema and Ecto models for sink health monitoring (Issue #1). Establishes monitors, alerts, metric samples, and notification channels as the foundation for evaluation, notifications, and UI.

• New Features

  • New tables (config_schema): sink_monitors, sink_monitor_notification_channels, sink_monitor_alerts, sink_monitor_metric_samples.
  • Monitors: threshold or anomaly on pending/failed metrics; validations; unique name per sink; cooldown and recovery options.
  • Channels: webhook, Slack, Discord, PagerDuty, Incident.io with encrypted config and per-type validation.
  • Alerts: triggered/resolved status with timestamps; metric samples for anomaly detection.
  • Schemas use enums and include simple query helpers.

• Migration

  • Run migrations to create enum types, tables, and indexes under config_schema.

^{Written for commit a1bf6c7. Summary will update on new commits.}

[cubic-dev-ai]

No issues found across 5 files