// the find
pixie-io/pixie
Instant Kubernetes-Native Application Observability
Pixie is a CNCF observability platform that uses eBPF to auto-instrument Kubernetes workloads — full-body HTTP requests, DB queries, DNS, CPU profiles — without any code changes. It stores and queries all telemetry in-cluster using a Pythonic query language (PxL). It's for platform engineers and SREs who want deep visibility into k8s services without an instrumentation tax.
eBPF-based auto-telemetry is the real differentiator: you get full request/response bodies for HTTP, gRPC, Postgres, MySQL, and more with zero app-side changes. In-cluster storage means your request data never leaves the cluster, which matters for compliance-sensitive environments. PxL scripts are genuinely powerful — you can write custom analyses that combine metrics, traces, and profiles in a single query. The Bazel + hermetic build setup is serious engineering; this isn't a weekend project.
The build system is Bazel-heavy and assumes Linux kernel 4.14+; getting a local dev environment working is a multi-hour commitment and Windows/Mac are effectively unsupported for development. Data retention is limited to a rolling window (hours, not days) by default — it's a live debugging tool, not a long-term metrics store, which surprises people expecting Prometheus-style historical querying. The cloud component adds significant operational complexity if you want multi-cluster support; the self-hosted path is not well-documented compared to the managed px.dev path. Protocol support gaps (Kafka, Redis pub/sub, gRPC streaming) mean you'll still need conventional instrumentation for parts of your stack.