Service health and observability

Separate liveness from readiness

A running process is not necessarily ready for work. The current services expose different checks intentionally. For example, go-report has liveness and readiness endpoints; readiness verifies PostgreSQL, memory budget and free export disk before Laravel admits another heavy report.

Minimum operational signals

Monitor:

• Web request latency and error rate.
• PostgreSQL CPU, active connections, slow queries, locks, WAL/checkpoint pressure and disk growth.
• Redis memory, evictions policy, queue depth and persistence health.
• Horizon queue depth, failed jobs, job runtime and worker memory.
• go-sync ready state, push queue depth, stale batches and retry rate.
• go-report active/queued jobs, rows/bytes processed, memory and free disk.
• go-dashboard outbox lag and projection freshness.
• go-bulk-import stale operations and worker heartbeats.
• go-xlsform resource rejections and processing duration.

Keep internal /metrics and worker health endpoints private. Feed them to a trusted monitoring layer rather than publishing them on the public internet.