OpenStack Swift -- Object Storage Service

Swift provides distributed object storage with an HTTP-accessible API. Unlike block storage (Cinder), which provides attached devices, Swift stores objects (files) in containers (buckets) accessible via REST API calls. Swift is S3-compatible when the s3api middleware is enabled, making it a drop-in replacement for AWS S3 in many applications.

Architecture

Swift uses a ring architecture to determine where objects are stored. Even in a single-node deployment, understanding rings is important because they control data placement and replication.

• Rings: Swift maintains three rings -- account, container, and object. Each ring maps a partition to a set of devices (disks). The ring builder calculates this mapping based on the number of replicas, partition power, and device weights.
• Partition power: Determines how many partitions exist (2^power). Higher partition power = more granular distribution but more memory usage. Single-node typically uses partition power 10 (1024 partitions).
• Replicas: The number of copies of each object. Production uses 3 replicas across failure zones. Single-node uses 1 replica (no redundancy benefit with one node).
• Eventual consistency: Swift is designed around eventual consistency. After a write, reads from different nodes may temporarily return different results. Replicator and auditor processes converge state over time.

Service components: