๐ Overview
Weโre transitioning our metric naming from always underscore-based (e.g., system_memory_usage
) to unnormalized (e.g., system.memory.usage
i.e metric data stored as it arrives) across our cloud observability pipeline. This change aligns with OpenTelemetry semantic conventions, enables us to build seamless correlations.
To facilitate a seamless migration, weโve initiated a dual write phase where both the normalized and unnormalized metrics are written. During this phase, unnormalized metrics remain in the backgroundโhidden from user viewsโso you will only see underscore-based metrics (system_memory_usage
).
๐ Why We Are Migrating
- Semantic Alignment: Dot-separated metric names follow the OpenTelemetry semantic conventions, providing clearer structure and hierarchy (e.g.,
system.memory.usage
). - Cross-Signal Compatibility: Consistent naming across signal reduces confusion and simplifies correlations.
๐ Migration Timeline
Dual Export Phase (In Progress)
- Start: Already underway
- Duration: Until next week
- Behavior: Both the normalized and unnormalized metrics are written, you won't be able to see any change during this period.
Final Migration (Next Week)
- We will retire the normalized metrics entirely.
- You will continue to see only unnormalized metric names moving forward.
๐ฃ Final Rollout
From next week onwards, we will roll out the final migration. You will begin to see unnormalized metric names exclusively; normalized metrics will no longer appear in your views.
โ What You Need to Do
Absolutely nothing. The migration is fully managed by our team โ your dashboards, alerts, and queries will continue to operate without any changes.