mstore

MStore is an experimental metric store built in Erlang, designed to efficiently handle time-series data with a focus on sequential access patterns. It leverages filesystem characteristics and a consistent hashing ring to organize and retrieve metric data.

Key Features

• Simple API — Provides core functions (open, new, get, put) for metric storage and retrieval.
• Consistent Hashing Ring — Distributes metrics across files based on a configurable ring size to manage open file limits.
• Sequential Read/Write Optimization — Assumes metrics are written nearly sequentially and reads are highly sequential, enabling efficient file access.
• Immutable Metric Storage — Treats metrics as immutable 64-bit integers, simplifying data management.
• Filesystem Integration — Relies on filesystem features like checksums, compression (e.g., ZFS), and caching for data integrity and performance.
• Indexed File Layout — Uses separate index and data files per chunk for quick metric lookups and sequential data reads.

Philosophy

MStore is built on the principle that time-series metrics have predictable patterns (regular intervals, immutability) and that modern filesystems can handle storage efficiently without complex application-level overhead.