// the find
trickstercache/trickster
Open Source HTTP Reverse Proxy Cache and Time Series Dashboard Accelerator
Trickster is a Go-based reverse proxy that understands time series query semantics — it caches TSDB results by time range and only fetches the missing intervals on cache miss, rather than re-fetching the whole window. Aimed at teams running Grafana against Prometheus, ClickHouse, or InfluxDB who are getting hammered by redundant queries across dashboard refreshes. v1.1.x is the production release; main is v2.0 beta.
Delta proxy cache is genuinely smart: it knows the time range of each query, diffs it against what's cached, and only hits the TSDB for the gap. This makes dashboard auto-refresh nearly free. Step boundary normalization is the right complement — aligning all cached points to clean step boundaries means cache hits are actually reusable across different users hitting the same dashboard. The ALB with TSM (time series merge) can aggregate across multiple Prometheus instances and merge the results, which is a real use case for federated monitoring. The integration test suite is extensive, including chaos tests and race detection tests in a separate module with real backends.
The production/beta split is a real operational hazard: v1.1.x is stable on a separate branch, v2.0 has been 'beta' on main for long enough that the README still calls it out explicitly — you need to know which branch to run, and new users will pull main by default and get beta software. TSDB support is limited to Prometheus, ClickHouse, and InfluxDB; if you're running VictoriaMetrics, Thanos, Mimir, or anything else in the ecosystem, you're out of luck. No clustering or HA story for Trickster itself — it's a single-point-of-failure in front of your observability stack. Config is purely YAML with no schema enforcement, and the feature surface (ALB, rules engine, per-path config, tracing, auth) means that full.yaml is a wall of text you have to get right by hand.