Operational Excellence with Insights
Resources
When a NATS cluster spikes its memory or pegs a CPU, the hardest question isn’t what happened — it’s why. Standard monitoring shows you the symptom. Closing the gap to the cause usually means a scramble of pprof captures, NATS CLI commands, and cross-referencing dashboards that were never designed to answer the question you actually have.
In this RethinkConn workshop, the Synadia team that builds and operates NATS every day — Byron Ruth (Engineering), Jared (Principal Support Engineer), and Eric Hansen (Senior Operations Engineer) — walks through Synadia Insights, a new observability tool born directly out of that pain. Insights scrapes all of the data the NATS monitoring endpoints expose (most of it untapped by traditional tooling), indexes it in DuckDB for fast analytical queries, and runs more than 100 checks distilled from years of real incident response. You point it at your system with a single read-only system account credential, and you get a complete, navigable view of your cluster in minutes.
What you’ll learn
Why traditional NATS monitoring hits a wall. We unpack the cardinality problem head-on — why per-stream, per-account, and per-connection metrics overwhelm time-series databases, why that forces you to drop the very data you need during an incident, and how an OLAP approach changes the economics of “monitor everything.”
How Insights actually works. A clear look at the architecture: the scraper that pulls from the system account endpoints, the indexer that consumes a stream and writes to DuckDB, the built-in HA model, and the read-only design that makes it safe to hand to operators, support, and platform teams alike.
The checks that catch problems before they page you. Jared shares the war stories behind the framework — runaway subject cardinality that quietly grows until a cluster falls over with head-of-line blocking, consumers hammering the JetStream API, and max_bytes storage reservations silently exhausting your ability to create new streams. You’ll also see the “hygiene” checks that surface silent failures like inactive JWT imports and orphaned exports that never log an error.
Triage in practice — a live fault simulator. Watch Insights detect injected faults in real time: high subject cardinality crossing a threshold, and a mirror going stale behind a network partition — a failure that’s normally invisible until someone downstream complains.
Where this is going. A walkthrough of the roadmap: federated collectors that bring observability to edge fleets without shuttling all the data back to the hub, a millisecond-granularity real-time view for catching auth-failure “disconnect storms,” geo-IP mapping of your global connections, and an agentic layer that lets AI assistants query the deterministic SQL schema to summarize incidents and generate reports — grounded in real data, not guesswork.
How it fits your existing stack. Insights isn’t another dashboard to babysit. See how it integrates through a Prometheus endpoint, outbound and inbound web hooks, an MCP server, and a Claude/Codex skill — all over a NATS API you control and authenticate yourself.
Who it’s for
Operations engineers, SREs, support teams, and platform owners running NATS in production — especially anyone managing multi-tenant systems, large superclusters, or distributed hub-to-edge deployments where today’s tools can’t see far enough.
Keep learning
All workshopsGo deeper with NATS training
Take your skills further with our live, instructor-led courses delivered by our experts to your team of developers or operators.