Grafana's Pyroscope 2.0 Revolutionizes Continuous Profiling with Storage and Performance Breakthroughs

Grafana Labs has released Pyroscope 2.0, a fundamental rearchitecture of its open source continuous profiling database designed to address the escalating costs and complexity of profiling at scale. The release, announced in April 2026, represents a significant evolution in observability technology, targeting the substantial storage overhead, query limitations, and operational challenges that emerged in the original implementation.

Continuous profiling has emerged as the fourth pillar of observability, complementing metrics, logs, and traces by providing granular visibility into exactly which functions and code lines consume system resources. As Christian Simon, a staff engineer at Grafana Labs, explains, this level of detail is essential for targeted optimization in increasingly complex systems: "Continuous profiling captures these moments as they happen, so you don't have to rely on luck with a debugger."

The architectural shift in Pyroscope 2.0 mirrors similar transformations in other Grafana projects. Just as Mimir recently redesigned its architecture to eliminate write-path replication and decouple reads from writes, Pyroscope 2.0 applies these same principles to the unique characteristics of profiling data—large payloads, significant symbolic information, and bursty query patterns.

The most significant architectural improvement is the elimination of write-path replication. In the original version, every profile was written three times, creating substantial storage overhead given that a single profile can be tens of megabytes. Pyroscope 2.0 writes each profile once to object storage, immediately reducing storage requirements by two-thirds. This architectural change alone represents a major cost reduction for organizations running profiling at scale.

The second major optimization comes from data co-location, where profiles from the same service are stored together. This approach enables deduplication of symbolic information such as function names, source locations, and stack traces. In Grafana's own production environment, this technique has reduced the symbol storage footprint by up to 95%, creating substantial additional savings.

The read path has undergone equally significant redesign. In the original Pyroscope, query processing occurred within stateful components that could not scale elastically, forcing organizations to reserve capacity for peak loads that remained idle most of the time. Pyroscope 2.0 implements a fully stateless read path, allowing any querier to process any query while enabling automatic scaling based on demand. This design accommodates the naturally bursty nature of profiling access patterns—minimal baseline traffic with heavy concurrent use during incidents. The architecture also accommodates the emerging class of LLM-powered agents that increasingly query profiling data as part of automated investigations, providing elastic scaling for these new workloads.

Operationally, the reduction in stateful components translates to fewer failure modes and dramatically faster deployments. According to Grafana, deployments that previously required 8-12 hours now complete in minutes. The segment writer has been made diskless, and the store-gateway component has been entirely removed, simplifying the operational footprint.

These architectural improvements unlock new capabilities that were not feasible in the original version. Pyroscope 2.0 now supports metrics derived from profiles, which aggregate profiling data into fleet-wide comparisons across services or deployments without requiring individual profile queries. The system also enables inspection of single profile instances rather than only aggregates, and introduces heatmap queries for visualizing profile distributions over time. These features emerged naturally from the cleaner data model and stateless read path rather than requiring separate engineering efforts.

The release aligns with broader industry trends as profiling gains recognition as a standard observability signal. OpenTelemetry incorporated continuous profiling as a core telemetry signal in August 2024, with Elastic donating its continuous profiling agent to the project. Pyroscope 2.0 includes native support for the OpenTelemetry Protocol (OTLP), allowing teams to ingest profiles through the standard OpenTelemetry pipeline. This integration supports the growing adoption of OpenTelemetry across observability stacks.

The timing of this release coincides with increased focus on FinOps—financial operations within cloud environments. An InfoQ article from February 2024 identified FinOps as a key force shaping observability, alongside OpenTelemetry growth and AI integration. All three trends converge in Pyroscope 2.0: the architectural changes reduce the cost of running profiling at scale, the OTLP support aligns with OpenTelemetry adoption, and the system explicitly accommodates AI agents querying profiling data in production.

In the competitive landscape, Pyroscope is not alone in the continuous profiling space. Polar Signals builds Parca, an open source system for collecting continuous profiling data that uses eBPF for low-overhead collection. Frederic Branczyk, co-founder of Polar Signals, has discussed how eBPF and a custom time-series database called FrostDB address similar storage and query challenges. Commercial alternatives include Datadog, New Relic, Dynatrace, and Sentry, all offering managed profiling solutions. CubeAPM provides continuous profiling as part of a full-stack observability platform for organizations seeking simpler setup with less operational overhead.

Pyroscope's position—open source code combined with a hosted option in Grafana Cloud Profiles—differentiates it from these vendors, particularly for teams already using other Grafana stack components. This hybrid approach offers flexibility for organizations at various stages of observability maturity.

Real-world implementations demonstrate Pyroscope's practical value. Engineers from Monzo described using Pyroscope for continuous profiling to detect performance regressions as they are deployed, alongside a Slack channel called "Graph Trending Downwards" where improved performance is noted. Uber has published details of using profile-guided optimization in Go, a workflow that Grafana Labs cites as using Pyroscope in the same process.

Grafana Cloud Profiles, the hosted version powered by Pyroscope, has been running the 2.0 architecture in production since April 2025. The company rolled it out to every region by September 2025 and has since processed 19.5PB of profiling data, demonstrating the system's ability to handle substantial workloads. For existing Grafana Cloud Profiles users, the migration has already occurred transparently. For teams running Pyroscope themselves, the key change in upgrading from v1 is that object storage is required for distributed deployments, as it now serves as the single source of truth for all profile data.

The evolution of Pyroscope reflects a broader maturation of observability practices, moving beyond simple metrics and logs toward deeper system understanding. By making continuous profiling practical at scale, Pyroscope 2.0 enables organizations to optimize their systems more effectively, reduce operational costs, and gain insights that were previously impractical to obtain. As systems grow increasingly complex, this level of visibility becomes not just beneficial, but essential for maintaining performance and efficiency in production environments.

For organizations evaluating continuous profiling solutions, Pyroscope 2.0 presents a compelling option, particularly for those already invested in the Grafana ecosystem or seeking to balance open source flexibility with operational simplicity. The architectural improvements address fundamental limitations of earlier profiling systems, making it feasible to maintain comprehensive profiling coverage even in large, dynamic environments.

Migration instructions and detailed release notes are available in the Pyroscope documentation, allowing organizations to assess the upgrade path and benefits for their specific use cases.