Expand description
Object-store checkpoint layout with UUID v7 identifiers. Object-store checkpoint layout for distributed checkpoints.
Defines the directory structure, manifest format, and path conventions for checkpoints stored in object stores (S3, GCS, Azure Blob, local FS).
§Directory Layout
checkpoints/
├── _latest # Pointer to latest manifest
├── {checkpoint_id}/
│ ├── manifest.json # CheckpointManifestV2
│ ├── operators/
│ │ └── {operator_name}/
│ │ └── partition-{id}.snap # Full state snapshot
│ │ └── partition-{id}.delta # Incremental delta
│ └── offsets/
│ └── {source_name}.json # Source offset data
└── {checkpoint_id}/
└── ...§Checkpoint IDs
CheckpointId wraps a UUID v7, which is time-sortable. This means
lexicographic sorting of checkpoint directories equals chronological
ordering — no need to parse timestamps or sequence numbers.
Structs§
- Checkpoint
Id - A time-sortable checkpoint identifier based on UUID v7.
- Checkpoint
Manifest V2 - V2 checkpoint manifest for distributed object-store checkpoints.
- Checkpoint
Paths - Deterministic path generator for checkpoint artifacts in object stores.
- Operator
Snapshot Entry - Per-operator snapshot metadata in a V2 manifest.
- Partition
Snapshot Entry - Per-partition snapshot or delta entry.
- Source
Offset Entry - Per-source offset entry for exactly-once recovery.