2025-01-22 · Sora Nguyen
Observability budgets for modest Postgres fleets
Choosing signals that stay readable when you are not a 24/7 NOC.
Large platforms export everything. Smaller teams drown. Pick five charts you would still open during a dinner interruption: replication lag tied to a business query, disk growth derivative, error rate from application drivers, checkpoint duration, and one customer-visible latency.
Instrument with labels your on-call understands without opening three wikis. If a chart needs a legend longer than the graph, split it.
Rotate charts quarterly. Traffic patterns shift; a brilliant dashboard in January can lie by July. We archive retired graphs instead of deleting outright — useful for postmortems.
Tags: Observability, SRE