Modernizing Prometheus: Native Storage for Composite Types

The Prometheus team is working on modernizing its storage system by moving from a “classic” model of storing composite metrics (like histograms) as multiple individual time series to a “native” model that stores them as single, atomic samples. This shift aims to improve efficiency, functionality, and transactionality, addressing limitations of the current system where partial data sends could lead to inaccurate alerts. While initially focused on histograms, the team is exploring extending this native storage to all composite types, potentially offering significant benefits despite challenges around legacy compatibility and data conversion.