dora_message/

daemon_to_node.rs

use std::{
    collections::{BTreeMap, BTreeSet},
    net::SocketAddr,
    path::PathBuf,
};

use std::sync::Arc;

use crate::{
    DataflowId,
    config::NodeRunConfig,
    descriptor::OperatorDefinition,
    id::{DataId, NodeId, OperatorId},
    metadata::Metadata,
};

pub use crate::common::{DataMessage, SharedMemoryId, Timestamped};

// Passed via env variable
#[derive(Debug, serde::Serialize, serde::Deserialize)]
pub struct RuntimeConfig {
    pub node: NodeConfig,
    pub operators: Vec<OperatorDefinition>,
}

#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
pub struct NodeConfig {
    pub dataflow_id: DataflowId,
    pub node_id: NodeId,
    pub run_config: NodeRunConfig,
    pub daemon_communication: Option<DaemonCommunication>,
    pub dataflow_descriptor: serde_yaml::Value,
    pub dynamic: bool,
    pub write_events_to: Option<PathBuf>,
    /// Number of times this node has been restarted. 0 on first run.
    #[serde(default)]
    pub restart_count: u32,
    /// Per-output data-plane routing for the startup handshake, computed by the
    /// daemon from the **actual** placement of the dataflow's nodes at spawn
    /// time (the descriptor's `deploy` section is intent, not placement — label
    /// scheduling can resolve differently).
    ///
    /// `None` means the node was spawned by an older daemon that doesn't
    /// provide routing; the node then keeps every output on the reliable daemon
    /// path (correct, just without the direct-zenoh fast path).
    #[serde(default)]
    pub output_routing: Option<BTreeMap<DataId, OutputRouting>>,
}

/// Data-plane routing for one node output — see
/// [`NodeConfig::output_routing`].
#[derive(Debug, Default, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
pub struct OutputRouting {
    /// Some consumer of this output runs under another daemon. All sends must
    /// then go through this node's daemon so its inter-daemon forwarding can
    /// reach them (dora #2738) — the direct node-to-node zenoh mesh is
    /// same-machine only.
    #[serde(default)]
    pub daemon_only: bool,
    /// The static same-daemon consumers whose startup acks the producer must
    /// collect before switching this output from the reliable daemon path to
    /// the direct node-to-node zenoh path. Dynamic consumers are never
    /// required: they join at arbitrary times (or never), and nothing may wait
    /// on them.
    ///
    /// The producer only collects these during a bounded startup window; an
    /// acker that is slow to answer costs this output the fast path for the
    /// whole run rather than upgrading it mid-stream (dora-rs/dora#2891).
    #[serde(default)]
    pub required_ackers: BTreeSet<RequiredAcker>,
}

/// Identity of one consumer input that must ack a producer's startup markers
/// before the producer may switch the corresponding output to the direct zenoh
/// path — see [`OutputRouting::required_ackers`].
#[derive(Debug, Clone, PartialEq, Eq, PartialOrd, Ord, serde::Serialize, serde::Deserialize)]
pub struct RequiredAcker {
    pub node_id: NodeId,
    pub input_id: DataId,
}

#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
pub enum DaemonCommunication {
    Tcp { socket_addr: SocketAddr },
    Interactive,
}

#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
#[must_use]
#[allow(clippy::large_enum_variant)]
#[non_exhaustive]
pub enum DaemonReply {
    Result(Result<(), String>),
    NextEvents(Vec<Timestamped<NodeEvent>>),
    NodeConfig {
        result: Result<NodeConfig, String>,
    },
    /// Reply to [`DaemonRequest::ExtensionLoad`]. `None` means the key is not
    /// in the table — either never stored, or already dropped.
    ExtensionValue {
        value: Option<Vec<u8>>,
    },
    Empty,
    /// Opaque reply to [`crate::node_to_daemon::DaemonRequest::ExtensionRequest`],
    /// produced by the extension's daemon half. dora does not interpret it.
    ///
    /// Appended last so existing variants keep their postcard indices: the
    /// Python node API ships separately (PyPI) from the daemon, so a
    /// mixed-version pair must not misdecode older replies.
    ExtensionReply {
        #[serde(with = "crate::bulk_bytes::vec")]
        payload: Vec<u8>,
    },
}

impl DaemonReply {
    /// Bulk bytes this reply will contribute to its encoding, for
    /// [`crate::encode_presized`].
    ///
    /// `NextEvents` is a batch, so this sums the per-event hints rather than
    /// relying on the single flat envelope allowance `encode_presized` adds:
    /// the daemon drains up to `NODE_EVENT_CHANNEL_CAPACITY` events into one
    /// reply, and a batch of a few dozen would otherwise realloc several times
    /// on envelopes alone.
    pub fn encode_size_hint(&self) -> usize {
        match self {
            DaemonReply::NextEvents(events) => {
                events.iter().map(|e| e.inner.encode_size_hint()).sum()
            }
            // The extension value is opaque bytes handed straight back, so
            // its own length is the whole hint.
            DaemonReply::ExtensionValue { value } => value.as_ref().map_or(0, |bytes| bytes.len()),
            DaemonReply::Result(_) | DaemonReply::NodeConfig { .. } | DaemonReply::Empty => 0,
            DaemonReply::ExtensionReply { payload } => payload.len(),
        }
    }
}

#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
#[non_exhaustive]
#[allow(clippy::large_enum_variant)]
pub enum NodeEvent {
    Stop,
    Reload {
        operator_id: Option<OperatorId>,
    },
    Input {
        id: DataId,
        metadata: Arc<Metadata>,
        data: Option<Arc<DataMessage>>,
    },
    InputClosed {
        id: DataId,
    },
    /// Notifies a node that a previously closed input has recovered and will receive data again.
    InputRecovered {
        id: DataId,
    },
    /// Notifies a node that an upstream node has restarted.
    ///
    /// Sent to downstream nodes when a node with a restart policy successfully
    /// restarts after a failure.
    NodeRestarted {
        id: NodeId,
    },
    /// Notifies a node that all its inputs have been closed.
    ///
    /// This event is only sent to nodes that have at least one input.
    AllInputsClosed,
    /// A runtime parameter has been updated.
    ///
    /// Sent when `dora param set` changes a parameter for this node.
    ///
    /// `value_json` carries JSON-encoded bytes rather than `serde_json::Value`:
    /// this message is serialized with postcard on the daemon↔node TCP channel,
    /// and `serde_json::Value::deserialize` uses `deserialize_any`, which
    /// postcard (like any non-self-describing format) does not support.
    ParamUpdate {
        key: String,
        value_json: Vec<u8>,
    },
    /// A runtime parameter has been deleted.
    ///
    /// Sent when `dora param delete` removes a parameter for this node.
    ParamDeleted {
        key: String,
    },
    /// An upstream node has failed.
    ///
    /// Sent to downstream nodes when an upstream node exits with a
    /// non-zero exit code.
    NodeFailed {
        affected_input_ids: Vec<DataId>,
        error: String,
        source_node_id: NodeId,
    },
    /// An extension key this node stored or loaded has been dropped, by
    /// another node or by the daemon reclaiming it.
    ///
    /// Delivered out of band: the event stream consumes it rather than
    /// surfacing it to user code, so a language binding polls
    /// `drain_dropped_extension_keys()` instead.
    ExtensionDropped {
        namespace: String,
        key: String,
    },
}

impl NodeEvent {
    /// Bulk bytes this event will contribute to the encoding of the
    /// [`DaemonReply`] that wraps it.
    ///
    /// Includes a flat per-event allowance for the `Timestamped` wrapper,
    /// `Metadata` and ids, because these are batched: see
    /// [`DaemonReply::encode_size_hint`].
    pub fn encode_size_hint(&self) -> usize {
        /// Measured at ~72 bytes for a typical `Input` (timestamp 25 +
        /// `Metadata` 27 + tags and ids); rounded up so a batch of small events
        /// still lands in one allocation.
        const PER_EVENT_ENVELOPE: usize = 128;

        let payload = match self {
            NodeEvent::Input { data, .. } => data.as_ref().map_or(0, |d| d.len()),
            NodeEvent::Stop
            | NodeEvent::Reload { .. }
            | NodeEvent::InputClosed { .. }
            | NodeEvent::InputRecovered { .. }
            | NodeEvent::NodeRestarted { .. }
            | NodeEvent::AllInputsClosed
            | NodeEvent::ParamUpdate { .. }
            | NodeEvent::ParamDeleted { .. }
            | NodeEvent::NodeFailed { .. } => 0,
            NodeEvent::ExtensionDropped { namespace, key } => namespace.len() + key.len(),
        };
        payload.saturating_add(PER_EVENT_ENVELOPE)
    }
}