Over half year ago I’ve open sourced klag, a lightweight Kafka consumer lag exporter. I’ve been working in data streaming systems and data infrastructure for over a decade and it’s an ultimate tool, I wish I had sooner.
Klag (pronounced “kay-lag”) docker image has been downloaded over 2,000 times, repo has passed 70 ⭐️, contributions started flowing in. So what has been added recently?
🤖 Talk to your lag
klag now ships an MCP server. Point your AI agent at it and just ask: "which consumer groups are falling behind right now?"
No PromQL. No squinting at dashboards. Debugging a backlog at 2am is now a conversation instead of a query-writing exercise. This is the feature I'm most excited about — lag monitoring goes from read this to ask this.
⚡ Native builds — no JVM, instant startup
There's now a native image build. Same exporter, compiled ahead-of-time, near-instant startup, smaller footprint.
If klag is a tiny always-on sidecar whose whole job is to scrape offsets and emit numbers — and it should be — this is the build you want. No JRE dragging along.
🎯 Smarter group filtering
You can now exclude groups, not just include them. Want myapp-* but not the noisy canary groups? Done. Comma-separated globs, includes and excludes, exactly where you'd expect them.
📦 One-command install
The Helm chart is on ArtifactHub and the Grafana dashboard is public and ready to drop in. No git clone, no hardcoded datasource UIDs. helm install and you've got lag metrics flowing into Grafana in minutes.
🙌 ...and it's not just me anymore
People are showing up with PRs that fix real edges in real clusters — full KAFKA_* passthrough, older-broker compatibility, more accurate time-based lag, and more. That's exactly the feedback loop I wanted.
Top comments (0)