Prometheus v2.19.0 introduces a significant feature that involves memory-mapping full chunks of the head block from disk, which reduces memory usage by 20-40% and accelerates restarts. This new feature benefits not only Prometheus but also other projects in its ecosystem, such as Cortex and Thanos, which utilize the Prometheus TSDB. The head block, storing the last 1–3 hours of data, now uses memory-mapping to load data into memory as needed, allowing for decreased memory allocation under certain conditions. Benchmark tests demonstrated that the reduction in memory usage varies with the churn rate and scrape interval; notably, lower churn and shorter scrape intervals result in greater memory savings. While this enhancement allows for more efficient resource allocation, it is crucial to plan for potential memory spikes during heavy queries. Additionally, during the replay of the Write Ahead Log (WAL), memory-mapping enables the omission of samples already covered by mapped chunks, reducing replay time by 20-30%.