Grafana Pyroscope
Continuous profiling platform that collects CPU, memory, goroutine, and other profiles from running applications over time. Grafana Pyroscope (merged with Phlare) stores profiling data in object storage and enables flame graph visualization and comparison to identify performance regressions and hotspots. Integrates with Grafana for profile correlation with metrics and traces. Supports Python, Go, Java, Ruby, .NET, and more via pull-based profiling agents. Part of Grafana's LGTM+ observability stack.
Score Breakdown
⚙ Agent Friendliness
🔒 Security
AGPL-3.0 — review license compliance for commercial SaaS use. Profiling data reveals application internals and code paths — handle with appropriate access controls. Grafana Labs company with SOC2. No auth by default — add reverse proxy auth.
⚡ Reliability
Best When
You have performance-sensitive services in production and want to continuously track where CPU and memory are spent over time to identify regressions and optimize hot paths.
Avoid When
You're doing one-time profiling or don't need continuous profiling — the overhead of running a continuous profiler isn't justified for occasional investigations.
Use Cases
- • Continuously profile production services to identify CPU hotspots, memory leaks, and goroutine issues without performance overhead
- • Compare CPU flame graphs between deployments to detect performance regressions before they impact users
- • Correlate profiling data with Grafana traces and metrics to find root cause of performance issues in distributed systems
- • Profile Python and Go agent processes in production to identify optimization opportunities
- • Debug memory leaks in long-running agent services by analyzing heap allocation profiles over time
Not For
- • One-time performance profiling — use cProfile (Python) or pprof (Go) for ad-hoc profiling sessions
- • Application-level metrics monitoring — use Prometheus/Victoria for metrics, Pyroscope is specifically for profiling
- • Security analysis — profiling reveals code paths but not security vulnerabilities; use SAST/DAST tools for security
Interface
Authentication
Self-hosted: no auth by default (use reverse proxy). Grafana Cloud Profiles: API key auth. Multi-tenancy via X-Scope-OrgID header. Basic auth supported via configuration.
Pricing
AGPL-3.0 — note: AGPL requires open sourcing modifications if you run as a service. Grafana Cloud is the managed alternative. Self-hosted is free for internal use.
Agent Metadata
Known Gotchas
- ⚠ AGPL-3.0 license — if running Pyroscope as part of a SaaS offering, you must open source your modifications per AGPL terms
- ⚠ Continuous profiling has overhead — pyspy (Python) uses ~1% CPU, eBPF-based profilers are lower but require kernel headers
- ⚠ Profile resolution and sampling rate are configurable — too-low sampling rates miss short functions; too-high rates increase overhead
- ⚠ Profile data is high cardinality — storing per-commit or per-deployment profiles grows storage rapidly; configure retention appropriately
- ⚠ SDK push mode (agent pushes to server) is simpler but requires network access from profiled service to Pyroscope
- ⚠ Python profiling with pyspy requires same-host access or Kubernetes DaemonSet — cannot profile remote processes
- ⚠ Flame graph comparison API requires knowing specific label sets — agents must query label values before constructing comparison queries
Alternatives
Full Evaluation Report
Detailed scoring breakdown, competitive positioning, security analysis, and improvement recommendations for Grafana Pyroscope.
Scores are editorial opinions as of 2026-03-06.