Pixie is an open-source observability platform for Kubernetes that uses eBPF to automatically capture telemetry data from clusters without requiring any manual instrumentation or code changes. It functions as an eBPF telemetry collector, a continuous application profiler, a network traffic analyzer, and a scriptable telemetry query engine, all within a single Kubernetes-native tool.
The platform distinguishes itself through several integrated capabilities. It continuously samples stack traces from compiled-language code to identify CPU performance bottlenecks, visualizing the results as interactive flamegraphs. Pixie also performs protocol-agnostic traffic decoding, automatically parsing full-body messages for application protocols including HTTP, DNS, and database protocols. A Python-like scripting language called PxL allows users to query and transform telemetry data through immutable dataframe operations, while a semantic type system guides data retrieval and visualization with appropriate units. Additionally, Pixie supports dynamic binary instrumentation for running Go binaries, edge-side machine learning for anomaly detection, and distributed deployment of custom bpftrace programs across all cluster nodes.
Beyond these core differentiators, Pixie provides comprehensive monitoring and debugging capabilities. It automatically collects resource metrics for every pod, traces inter-service requests with full-body payloads, and offers cluster-wide resource inspection across namespaces, pods, and nodes. Users can write custom observability scripts, execute pre-built telemetry queries, and export results in JSON or CSV format. The platform also includes service topology visualization, database query profiling, DNS request pattern inspection, and TCP packet drop detection. Pixie encrypts telemetry data in transit between the cluster and API clients, and supports deployment options ranging from managed cloud services to fully self-hosted instances.