Pyroscope 2.0: Continuous Profiling, Now Scalable & Cost-Effective

The Evolution of Observability Profiling

Grafana's Pyroscope 2.0 represents a substantial leap forward in making continuous profiling a viable and cost-effective pillar of observability. The ground-up rearchitecture, moving away from Cortex and embracing object storage as the single source of truth, directly tackles the storage cost and performance bottlenecks that plagued earlier iterations. The shift to a stateless read path is particularly impactful, enabling elastic scaling to handle the notoriously bursty nature of profiling data, especially with the emergent trend of AI agents querying this data for automated diagnostics. This move not only democratizes access to granular performance insights but also aligns perfectly with the broader industry push towards FinOps and standardized telemetry through OpenTelemetry. The emphasis on deduplication of symbolic information is a clever optimization that significantly reduces the storage footprint, a crucial factor for large-scale deployments.

However, while the benefits are clear, the operational shift to requiring object storage for distributed deployments is a notable change that users will need to adapt to. The migration from v1 to v2 will require careful planning, especially for those managing their own Pyroscope instances. Furthermore, while the article highlights the scalability improvements and cost reductions, the practical experience of migrating and managing the new architecture at extreme scales will be the ultimate test. The success of this redesign will hinge on its ability to maintain stability and performance as adoption grows and the volume of profiling data continues to explode, driven by increasingly complex systems and the integration of AI-driven analysis. The open-source nature, coupled with the Grafana Cloud offering, positions it well, but the competitive landscape with solutions like Parca and managed offerings from Datadog and others will ensure continued innovation and user choice.

Key Points

• Pyroscope 2.0 is a ground-up rearchitecture of the open source continuous profiling database, addressing storage costs, query performance, and operational complexity.
• The new architecture establishes object storage as the single source of truth and reduces profile writes from 3x in v1 to 1x, significantly lowering storage costs.
• Data co-location and deduplication of symbolic information have reduced the symbol storage footprint by up to 95%.
• The read path has been redesigned to be entirely stateless, enabling elastic scaling to gracefully handle bursty query loads, including those from AI agents.
• Operational complexity is reduced, with deployments shrinking from hours to minutes and fewer failure modes.
• The new architecture enables new capabilities such as metrics derived from profiling data, inspection of single profile instances, and heatmap queries.
• Pyroscope 2.0 supports the OpenTelemetry Protocol (OTLP), aligning with OpenTelemetry's continuous profiling signal integration.
• The release aligns with industry trends like FinOps and AI integration, demonstrated by reduced profiling costs and support for AI-driven analysis.
• Existing Grafana Cloud Profiles users have automatically migrated to the 2.0 architecture.

---

📖 Source: Grafana's Pyroscope 2.0 Makes Continuous Profiling Practical at Scale