Trait DaemonDispatch 

pub trait DaemonDispatch:
    Clone
    + Send
    + Sync
    + 'static {
    // Required methods
    fn plan(&self, ops: &str) -> String;
    fn dispatch<'life0, 'async_trait>(
        &'life0 self,
        ops: String,
        presentation: Option<String>,
        presentation_per_op: Option<Vec<Option<String>>>,
        format: Option<String>,
        format_per_op: Option<Vec<Option<String>>>,
        from_wire: bool,
        identity: Option<RequestIdentity>,
    ) -> Pin<Box<dyn Future<Output = Result<String, String>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait;
    fn warm_all<'life0, 'async_trait>(
        &'life0 self,
    ) -> Pin<Box<dyn Future<Output = ()> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait;
    fn namespace(&self) -> &str;
    fn config_id(&self) -> &str;

    // Provided methods
    fn idle_retirement_blockers(&self) -> Vec<String> { ... }
    fn request_read_timeout(&self, _ops: &str) -> Duration { ... }
    fn dispatch_with_error_detail<'life0, 'async_trait>(
        &'life0 self,
        ops: String,
        presentation: Option<String>,
        presentation_per_op: Option<Vec<Option<String>>>,
        format: Option<String>,
        format_per_op: Option<Vec<Option<String>>>,
        from_wire: bool,
        identity: Option<RequestIdentity>,
    ) -> Pin<Box<dyn Future<Output = Result<String, DaemonDispatchError>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait { ... }
    fn pool_for_checkpoint(&self) -> Option<Arc<ConnectionPool>> { ... }
    fn secondary_pools_for_checkpoint(&self) -> Vec<Arc<ConnectionPool>> { ... }
    fn event_store_for_checkpoint(&self) -> Option<Arc<dyn EventStore>> { ... }
}

Transport-agnostic dispatch interface for the daemon server.

The MCP crate implements this by dispatching through the shared request body while honoring DaemonRequestFrame::from_wire (so subhandler visibility is gated by request origin, not by transport); any future transport can do the same.

Required Methods

fn plan(&self, ops: &str) -> String

Describe syntax and loaded catalog membership without dispatching.

fn dispatch<'life0, 'async_trait>( &'life0 self, ops: String, presentation: Option<String>, presentation_per_op: Option<Vec<Option<String>>>, format: Option<String>, format_per_op: Option<Vec<Option<String>>>, from_wire: bool, identity: Option<RequestIdentity>, ) -> Pin<Box<dyn Future<Output = Result<String, String>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Dispatch a verb-DSL request string and return the rendered result.

from_wire carries the origin discriminator from DaemonRequestFrame::from_wire: when true, the implementor enforces verb visibility (rejects Visibility::Subhandler verbs); when false, the request is from a trusted operator surface and subhandlers pass.

identity is the per-request identity context threaded from the frame (ADR-096): Some(..) when serving a request forwarded over the daemon socket (built from frame.namespace / frame.actor_id / frame.visible_namespaces by the connection handler), None for any other dispatch path. Implementors should mint the storage/gate token from identity when present and fall back to their own construction-baked identity when absent, so pure local (non-daemon) dispatch is unchanged.

fn warm_all<'life0, 'async_trait>( &'life0 self, ) -> Pin<Box<dyn Future<Output = ()> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Warm every pack’s in-memory state (ANN indexes, etc.).

fn namespace(&self) -> &str

The namespace this dispatcher was configured for.

fn config_id(&self) -> &str

Fingerprint of this dispatcher’s resolved runtime config (packs, db target, embedders). Used to reject forwarded requests from clients whose config differs, so a restricted client cannot dispatch through a broader daemon.

Provided Methods

fn idle_retirement_blockers(&self) -> Vec<String>

Named retained resources or unknown inventory that prevents idle exit. An implementor must explicitly account for its resources before retiring.

fn request_read_timeout(&self, _ops: &str) -> Duration

Read-deadline ceiling for one request. The default is the operator ceiling; an implementor that understands the request may grant a longer bounded allowance (a long poll’s declared wait plus a margin).

fn dispatch_with_error_detail<'life0, 'async_trait>( &'life0 self, ops: String, presentation: Option<String>, presentation_per_op: Option<Vec<Option<String>>>, format: Option<String>, format_per_op: Option<Vec<Option<String>>>, from_wire: bool, identity: Option<RequestIdentity>, ) -> Pin<Box<dyn Future<Output = Result<String, DaemonDispatchError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Preserve structured dispatch errors without breaking string-only implementors.

fn pool_for_checkpoint(&self) -> Option<Arc<ConnectionPool>>

Return the pool to use for background WAL checkpointing, if available.

Implementors backed by a file-based SQLite database should return Some(pool_arc). In-memory or test dispatchers that have no pool return None and the checkpoint task is not spawned.

The default implementation returns None.

fn secondary_pools_for_checkpoint(&self) -> Vec<Arc<ConnectionPool>>

File-backed backend pools beyond Self::pool_for_checkpoint’s pool (ADR-091 Amendment 3): one checkpoint task is spawned per entry here, in addition to the one spawned for the primary pool, so a multi-backend deployment gets PASSIVE/TRUNCATE checkpointing and sidecar enumeration on every file-backed backend it wired, not only the main one.

The default implementation returns an empty Vec — an implementor with only one backend (or none) needs no override.

fn event_store_for_checkpoint(&self) -> Option<Arc<dyn EventStore>>

Return the audit EventStore the checkpoint task should append ADR-094 lifecycle events (CheckpointOutcomeRecorded) to, if any.

Mirrors Self::pool_for_checkpoint’s default-None shape: an implementor with no configured event store (or no pool at all) simply gets a checkpoint task that never appends events — the checkpoint task itself remains fully functional either way.

The default implementation returns None.

Dyn Compatibility

This trait is not dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety".

Implementors