Struct mz_controller::Controller

pub struct Controller<T = Timestamp> {
    pub storage: Box<dyn StorageController<Timestamp = T>>,
    pub compute: ComputeController<T>,
    /* private fields */
}

A client that maintains soft state and validates commands, in addition to forwarding them.

Fields

storage: Box<dyn StorageController<Timestamp = T>>

compute: ComputeController<T>

Implementations

impl<T> Controller<T>

where T: Timestamp + Lattice, ComputeGrpcClient: ComputeClient<T>,

pub fn create_cluster( &mut self, id: ClusterId, config: ClusterConfig ) -> Result<(), Error>

Creates a cluster with the specified identifier and configuration.

A cluster is a combination of a storage instance and a compute instance. A cluster has zero or more replicas; each replica colocates the storage and compute layers on the same physical resources.

pub fn drop_cluster(&mut self, id: ClusterId)

Drops the specified cluster.

Panics

Panics if the cluster still has replicas.

pub async fn create_replicas( &mut self, replicas: Vec<CreateReplicaConfig>, enable_worker_core_affinity: bool ) -> Result<(), Error>

Creates a replica of the specified cluster with the specified identifier and configuration.

This method is NOT idempotent; It can fail between processing of different replicas and leave the controller in an inconsistent state. It is almost always wrong to do anything but abort the process on Err.

pub async fn drop_replica( &mut self, cluster_id: ClusterId, replica_id: ReplicaId ) -> Result<(), Error>

Drops the specified replica of the specified cluster.

pub async fn remove_orphaned_replicas( &mut self, next_user_replica_id: u64, next_system_replica_id: u64 ) -> Result<(), Error>

Remove orphaned replicas.

pub fn events_stream(&self) -> BoxStream<'static, ClusterEvent>

impl<T: Timestamp> Controller<T>

pub fn active_compute(&mut self) -> ActiveComputeController<'_, T>

pub fn set_arrangement_exert_proportionality(&mut self, value: u32)

pub fn connection_context(&self) -> &ConnectionContext

Returns the connection context installed in the controller.

This is purely a helper, and can be obtained from self.storage.

pub fn storage_configuration(&self) -> &StorageConfiguration

Returns the storage configuration installed in the storage controller.

This is purely a helper, and can be obtained from self.storage.

pub fn dump(&self) -> Result<Value, Error>

Returns the state of the Controller formatted as JSON.

The returned value is not guaranteed to be stable and may change at any point in time.

impl<T> Controller<T>

where T: TimestampManipulation, ComputeGrpcClient: ComputeClient<T>,

pub fn update_orchestrator_scheduling_config( &mut self, config: ServiceSchedulingConfig )

pub fn initialization_complete(&mut self)

Marks the end of any initialization commands.

The implementor may wait for this method to be called before implementing prior commands, and so it is important for a user to invoke this method as soon as it is comfortable. This method can be invoked immediately, at the potential expense of performance.

pub async fn ready(&mut self)

Waits until the controller is ready to process a response.

This method may block for an arbitrarily long time.

When the method returns, the owner should call Controller::ready to process the ready message.

This method is cancellation safe.

pub fn install_compute_watch_set( &mut self, objects: BTreeSet<GlobalId>, t: T, token: Box<dyn Any> )

Install a watch set in the controller.

A watch set is a request to be informed by the controller when all of the frontiers of a particular set of objects have advanced at least to a particular timestamp.

When all the objects in objects have advanced to t, the object token is returned to the client on the next call to Self::process.

pub fn install_storage_watch_set( &mut self, objects: BTreeSet<GlobalId>, t: T, token: Box<dyn Any> )

Install a watch set in the controller.

A watch set is a request to be informed by the controller when all of the frontiers of a particular set of objects have advanced at least to a particular timestamp.

When all the objects in objects have advanced to t, the object token is returned to the client on the next call to Self::process.

pub async fn process( &mut self, storage_metadata: &StorageMetadata ) -> Result<Option<ControllerResponse<T>>, Error>

Processes the work queued by Controller::ready.

This method is guaranteed to return “quickly” unless doing so would compromise the correctness of the system.

This method is not guaranteed to be cancellation safe. It must be awaited to completion.

pub fn recent_timestamp( &self, source_ids: impl Iterator<Item = GlobalId> ) -> BoxFuture<'static, T>

Produces a timestamp that reflects all data available in source_ids at the time of the function call.

impl<T> Controller<T>

where T: Timestamp + Lattice + TotalOrder + TryInto<i64> + TryFrom<i64> + Codec64 + Unpin + From<EpochMillis> + TimestampManipulation + Display + Into<Timestamp>, <T as TryInto<i64>>::Error: Debug, <T as TryFrom<i64>>::Error: Debug, StorageCommand<T>: RustType<ProtoStorageCommand>, StorageResponse<T>: RustType<ProtoStorageResponse>,

pub async fn new( config: ControllerConfig, envd_epoch: NonZeroI64, persist_txn_tables: PersistTxnTablesImpl, storage_txn: &dyn StorageTxn<T> ) -> Self

Creates a new controller.

For correctness, this function expects to have access to the mutations to the storage_txn that occurred in prepare_initialization.

Panics

If this function is called before prepare_initialization.