ADR-0026: Connector Health Monitor

Decision: ConnectorHealthJob — a FeedJob subclass that runs in the shared FeedScheduler; schema drift tracked via JSON fingerprint comparison.

Why FeedJob subclass: Health checks are just another scheduled job. Reusing FeedScheduler means health monitoring gets retry logic, overlap protection, and on-success/on-failure callbacks for free.

Drift detection algorithm:

  1. Sample list_objects(limit=1) from each connector.
  2. Extract field names from the response (recursive key extraction).
  3. Compute a frozenset fingerprint and serialize to SchemaSnapshot JSON.
  4. On each run, compare current fingerprint against the stored baseline.
  5. If changed fields exceed drift_threshold (default 20%), emit a DriftReport.

Slack alerts: _post_slack_webhook() is called when drift is detected. The webhook URL is read from [health] INI section. This provides zero-dependency alerting without requiring a monitoring framework.

Baseline command: gnat health baseline runs one sampling pass and writes the initial SchemaSnapshot files. Production deployments should run baseline after initial connector setup and after planned API migrations.

Licensed under the Apache License, Version 2.0