Repository metrics
- Stars
- (2,443 stars)
- PR merge metrics
- (PR metrics pending)
Description
Context
LinkedIn has been adopting SPIRE extensively to power our internal PKI infrastructure and a variety of production use cases. Throughout this process, we identified several integration gaps—both within the LinkedIn ecosystem and when interoperating with other open-source projects. While our current workarounds are not ideal for long-term maintenance, they give us concrete insight into where SPIRE can be strengthened.
Upstreaming these enhancements provides a sustainable, community-supported solution rather than isolated fixes. By contributing them to SPIRE, we can improve interoperability, reduce fragmentation, and help other organizations avoid duplicating the same work. Our goal is to collaborate with the community to deliver maintainable improvements that align with SPIRE’s long-term direction.
OpenTelemetry Metrics for Latency
At LinkedIn, the observability platform is evolving toward a push-based metrics model, with OpenTelemetry (OTel) as the standard for metrics collection and export. Metrics designed for pull-based monitoring systems do not integrate cleanly with LinkedIn’s OTel-based dashboards and alerting pipelines, reducing their effectiveness for latency observability.
Change to OSS SPIRE
To align with this direction, LinkedIn proposes adding configurable support for emitting latency metrics using OpenTelemetry metrics alongside the existing metrics. This additive approach enables seamless integration with the OTel observability stack while preserving compatibility for current metric consumers, allowing both approaches to coexist during the transition.
Histogram-Type Metrics Support for Latency
At LinkedIn, Summary-type metrics are considered deprecated and are not supported. As a result, LinkedIn’s internal dashboards and alerts do not function correctly for Summary-based latency metrics.
Change to OSS SPIRE
Replacing Summary-type metrics with Histogram-type metrics would introduce backward-compatibility concerns. To avoid this, LinkedIn will contribute a small code change to add Histogram-type latency metrics alongside the existing Summary-type metrics.