Struct SegmentedVolume 

pub struct SegmentedVolume { /* private fields */ }

WAL namespace: the zonal replica factories for segment data (one per zone) plus one regional factory hosting the manifest control register.

Implementations

impl SegmentedVolume

pub fn new( factories: Vec<GrpcReplicaFactory>, manifest_factory: GrpcReplicaFactory, prefix: impl Into<String>, client_config: ClientConfig, ) -> Result<Self, Error>

Bind 1, 3, or 5 GCS Rapid zonal buckets to one WAL object prefix. Any other factory count returns crate::Error. Durability and availability follow the strict-majority quorum: a single replica tolerates no zone loss, three tolerate one, five tolerate two.

Each element must target a different zone. Keep both the list order and the factory’s diagnostic zone index stable across restarts; recovery treats list position as the replica identity. The prefix is dedicated to this WAL. The zonal data prefix contains opaque, creation-ordered segment ids; the same prefix in the regional bucket contains the manifest control object.

The client validates replica count and bucket-name consistency but cannot verify physical placement. Operators must provision Rapid buckets in distinct zones of one region. Pass full v2 bucket resource names and use a gRPC-compatible endpoint such as https://storage.googleapis.com; Cloud Storage regional JSON/XML endpoints do not support gRPC.

Registers persist the ordered zonal bucket names in chorus.buckets; its length is the replica count. A later wrong, reordered, or duplicate set is rejected, and registers without the binding fail closed.

Independent processes may recover the same prefix concurrently. GCS append takeover and conditional segment creation elect one writer; the loser receives a terminal fencing or conditional-create error. Rotation policy is supplied later through crate::WalEngineConfig, not through this storage description.

pub fn new_with_metrics_recorder( factories: Vec<GrpcReplicaFactory>, manifest_factory: GrpcReplicaFactory, prefix: impl Into<String>, client_config: ClientConfig, metrics_recorder: Arc<dyn MetricsRecorder>, ) -> Result<Self, Error>

Bind a WAL namespace and emit metrics through metrics_recorder.

Chorus registers all handles during construction and updates them directly on event paths. The recorder owns aggregation and export; the WAL does not retain a readable metrics registry.

Most handles have no labels. When one backend recorder serves several volumes, wrap it per volume and inject a stable volume label into every registration; otherwise those volumes share indistinguishable series.

pub fn new_with_manifest_store( factories: Vec<GrpcReplicaFactory>, manifest_store: Arc<dyn ManifestStore>, prefix: impl Into<String>, client_config: ClientConfig, ) -> Result<Self, Error>

Bind a WAL namespace whose manifest register lives in a caller-supplied ManifestStore instead of the default regional GCS object.

The store must be scoped to exactly this WAL (one register per WAL prefix) and provide the linearizable compare-and-swap semantics the trait documents; Firestore, Spanner, or a SQL row with optimistic locking all qualify. Segment data still lives in the zonal factories. The required bucket binding covers the ordered zonal data factories; the caller remains responsible for scoping the custom store to this WAL.

pub fn new_with_manifest_store_and_metrics_recorder( factories: Vec<GrpcReplicaFactory>, manifest_store: Arc<dyn ManifestStore>, prefix: impl Into<String>, client_config: ClientConfig, metrics_recorder: Arc<dyn MetricsRecorder>, ) -> Result<Self, Error>

Self::new_with_manifest_store with a metrics recorder.

pub async fn recover(&self, checkpoint: WalSeqNo) -> Result<Recovery, Error>

Fence and seal the prior writer, then prepare database startup replay.

checkpoint is the first record not represented by the database’s durable state. The returned Recovery emits the fixed range [checkpoint, Recovery::end) and does not create an appendable segment yet. Apply every emitted record, consume the stream through None, then call Recovery::start to conditionally create the manifest-selected active segment and begin admission.

pub async fn recover_from_committed_floor(&self) -> Result<Recovery, Error>

Recover from the truncation floor committed in the manifest.

This is intended for diagnostic and maintenance tools that do not own a database checkpoint. Database integrations should use Self::recover with their durable replay boundary.

pub async fn repair_sealed_segments(&self) -> Result<RepairReport, Error>

Recover the committed chain and repair immutable sealed segment copies.

This volume-level primitive is intended for diagnostic and maintenance tools that need a detailed RepairReport. Applications should run the WAL engine, which performs the same repair automatically in the background. The operation claims the WAL writer epoch and creates the next active segment, just like normal recovery startup.

Trait Implementations

impl Clone for SegmentedVolume

fn clone(&self) -> SegmentedVolume

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.