hyperhive/hive-c0re/src/job_queue/mod.rs

//! Generic job-DAG queue + desired-state reconciliation — the host-side
//! wrapper over the domain-agnostic [`hive_jobq`] scheduler. Jobs are nodes in
//! per-request DAGs (see [`templates`]); the special cases (graceful-stop
//! watcher, deferred-start follow-up, meta-update cascade) collapse into DAG
//! *shapes* over the shared node primitives ([`model::NodeKind`]).
//!
//! [`hive_jobq`] owns the graph, the two-class resource pool, and the roll-up
//! settle loop; this module maps hive-c0re's concepts onto it:
//! - [`model::NodeKind`] **is** the crate payload `N` directly — each variant
//!   carries the agent it targets ([`NodeKind::agent`]); the two resource
//!   classes are [`resource::Resource`] (`BuildSlot` node-held, `Agent` lease
//!   subtree-held), declared per node at its construction site;
//! - a **DAG is a single container node** ([`NodeKind::Dag`], `parent = None`)
//!   carrying the group's metadata, with the work nodes hung under it as
//!   its subtree (the **parent axis** groups; `deps` order). So the container's
//!   `NodeId` is the DAG id, its rolled-up state is the DAG state, and membership
//!   is a graph walk — there are no host grouping side-tables. The lease is owned
//!   by a subtree root and borrowed by its descendants (continuity);
//! - per-DAG terminal work is an ordinary **tail node**
//!   ([`NodeKind::ResolveApproval`] / [`NodeKind::EmitRebuilt`]) that the builder
//!   appends in [`templates`], edged onto the DAG's other group roots by the
//!   outcome it reports. Templates emit one tail per outcome and the graph runs
//!   exactly one, so nothing branches at runtime.
//!
//! The queue is runtime-only (no persistence): an empty graph on boot; desired
//! state is re-derived by the reconcile sweep. A single scheduler task
//! ([`scheduler::run_worker`]) drives it; concurrency comes from the build-slot
//! capacity, not multiple workers. Design: `docs/coordinator.md::Job queue`.

pub mod exec;
pub mod model;
pub mod resource;
pub mod scheduler;
pub mod submit;
pub mod templates;
#[cfg(test)]
mod tests;

use std::sync::{Arc, Mutex};

use chrono::{DateTime, Utc};
use hive_host_sock::jobs::NodeView;
use hive_jobq::resources::ResourceTable;
use hive_jobq::scheduler::{Outcome, Scheduler};
use hive_jobq::{Dep, Graph, NodeId};
use tokio::sync::Notify;

pub use hive_jobq::TerminalState;
pub use model::{DagSpec, DagView, NodeKind, PermPayload, Source, State};
use resource::Resource;

/// A job under construction: `hive_jobq`'s builder over this queue's payload
/// ([`NodeKind`]) and resource ([`Resource`]) types. Templates declare into a
/// borrowed one; only `hive_jobq` can make or insert it.
pub type Job = hive_jobq::JobBuilder<NodeKind, Resource>;

/// A handle to one node a template declared — where its edges, grouping and
/// resources are declared. `Copy`; naming a node as a dependency does not
/// consume the ability to name it again.
pub type Handle<'a> = hive_jobq::NodeRef<'a, NodeKind, Resource>;

/// How many terminal DAGs (`Done` / `Failed` / `Cancelled`) the snapshot
/// retains, newest first. A flat cap over the whole sorted list: the
/// dashboard renders one recent-builds list, so one number bounds it.
const MAX_HISTORY_DAGS: usize = 50;

/// Cap on stored node error strings.
const MAX_ERROR_LEN: usize = 2_000;

/// One live transient pill, derived from a running node.
///
/// A named struct rather than a tuple because three of its four fields are
/// easy to confuse at a call site: two are strings and two answer questions
/// nobody should have to guess at ("is this the agent or the label?", "does
/// this bool mean deliberate or running?").
#[derive(Debug, Clone)]
pub struct RunningTransient {
    /// The agent whose lease the node declared.
    pub agent: String,
    /// The node's own wire tag, rendered as the pill.
    pub label: String,
    /// Whether this operation is expected to take the container down — the
    /// crash watcher's input. See [`NodeKind::takes_container_down`].
    pub takes_container_down: bool,
    /// When the node started running, so the dashboard can tick elapsed
    /// seconds. Taken from the node itself, which is the true start of the
    /// operation rather than the moment a watcher noticed it.
    pub since: DateTime<Utc>,
}

/// An owned read-view of a DAG container's carried metadata ([`NodeKind::Dag`]).
/// Derived on read from the container node — the data has a single home (the
/// node payload); this is not a stored side-table.
struct DagMeta {
    source: Source,
    reason: String,
    created_at: DateTime<Utc>,
}

/// The crate scheduler, specialised to this host's node + resource types.
///
/// A **DAG is a single container node** ([`NodeKind::Dag`], `parent = None`)
/// whose subtree is the DAG's work — so the container's `NodeId` is the DAG id,
/// its rolled-up state is the DAG state, and there are no grouping side-tables:
/// membership + meta are graph queries ([`container`] / [`dag_meta`] + the
/// `hive_jobq::Graph` accessors). One shared crate [`Graph`] holds every DAG.
///
/// There is deliberately **no wrapper struct and no per-node side map**. The
/// last map held the `build_logs` row id; that link now lives on the log row
/// itself (`build_logs.node_id`). With nothing else to guard, the mutex holds
/// the scheduler *directly* — which is what lets `hive_jobq` drive the run loop
/// (it takes `&Arc<Mutex<Scheduler<..>>>`, a type a host-side wrapper could not
/// satisfy).
type Sched = Scheduler<NodeKind, Resource>;

/// The queue. Lives on `Coordinator` (one per hive-c0re process); a single
/// scheduler task ([`scheduler::run_worker`]) drives it.
pub struct JobQueue {
    /// The scheduler, held directly rather than behind a host-side wrapper —
    /// `hive_jobq`'s run-loop seam takes `&Arc<Mutex<Scheduler<..>>>`, so this
    /// *is* the type the crate drives.
    sched: Arc<Mutex<Sched>>,
    /// Wakes the scheduler when something new arrives or state changed.
    pub(crate) notify: Notify,
}

impl std::fmt::Debug for JobQueue {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("JobQueue").finish_non_exhaustive()
    }
}

impl Default for JobQueue {
    fn default() -> Self {
        Self::new(1)
    }
}

/// A node runner's `Result` as the scheduler's [`Outcome`].
///
/// The failure reason + `finished_at` are stamped onto the graph `Node` by the
/// scheduler (the reason rides `Outcome::Failed`); there is no host-side copy,
/// so nothing needs clearing on success.
fn outcome_of(result: Result<(), String>) -> Outcome {
    match result {
        Ok(()) => Outcome::Done,
        Err(e) => Outcome::Failed(truncate_error(&e)),
    }
}

/// Insert a declared `job` into the shared graph and record its per-node
/// `node_rt`, returning the inserted ids.
///
/// A node that declared no parent hangs under `group_parent` — the DAG
/// container for a template, the emitting node for a runtime-appended
/// subgraph. Templates declare the parent axis + sibling ordering directly, so
/// there is no dep-on-root to drop and no lease to hoist: each node declares
/// its own resources, and the crate's borrow model keeps a resource continuous
/// across a subtree (a root owns it, descendants borrow it). Independent group
/// roots carry no cross-links, so a multi-agent DAG's per-agent subgraphs run
/// concurrently, each on its own lease.
///
/// # Errors
/// Propagates a crate graph-insert error (malformed dep/parent / dep-scope).
fn insert_group(
    inner: &mut Sched,
    declare: impl FnOnce(&Job),
    group_parent: Option<NodeId>,
) -> anyhow::Result<()> {
    inner
        .insert_job(group_parent, |b| {
            declare(b);
            // c0re names no handles: a DAG is addressed by its container node,
            // which `submit` inserts itself, and nothing downstream looks an
            // individual step up by id.
            Vec::new()
        })
        .map_err(|e| anyhow::anyhow!("job_queue: graph insert failed: {e}"))?;
    Ok(())
}

impl JobQueue {
    #[must_use]
    pub fn new(build_slots: usize) -> Self {
        let mut table = ResourceTable::new();
        table.set_capacity(
            Resource::BuildSlot,
            u32::try_from(build_slots.max(1)).unwrap_or(u32::MAX),
        );
        Self {
            sched: Arc::new(Mutex::new(Scheduler::new(Graph::new(), table))),
            notify: Notify::new(),
        }
    }

    fn lock(&self) -> std::sync::MutexGuard<'_, Sched> {
        self.sched.lock().expect("job_queue mutex poisoned")
    }

    /// Submit a DAG: insert a [`NodeKind::Dag`] **container node** carrying the
    /// group's metadata, then insert the template's nodes as its subtree (their
    /// roots re-parented to the container). Returns the container's id as the
    /// DAG id — its rolled-up state is the DAG state.
    ///
    /// Takes the spec's recipe by generic, not as a boxed closure: a spec
    /// travels from the template that built it directly into this call, so
    /// there is nothing to allocate for.
    ///
    /// # Errors
    /// Propagates a graph-insert error (dependencies that aren't
    /// dependency-topological).
    pub fn submit<F: FnOnce(&Job)>(&self, spec: DagSpec<F>) -> anyhow::Result<u64> {
        let mut inner = self.lock();
        let container = inner
            .append(
                NodeKind::Dag {
                    source: spec.source,
                    reason: spec.reason,
                    created_at: Utc::now(),
                },
                Vec::new(),
                None,
            )
            .map_err(|e| anyhow::anyhow!("job_queue: container insert failed: {e}"))?;
        insert_group(&mut inner, spec.declare, Some(container))?;
        // Settle the container's own (no-op) logic immediately so it parks in
        // `Finishing` and its children become runnable — it never needs claiming
        // or executing, and stays out of `claim_ready`. It rolls up terminal when
        // its whole subtree settles (that's the DAG-done signal).
        inner.complete(container, Outcome::Done);
        drop(inner);
        self.notify.notify_one();
        Ok(container.get())
    }

    /// The scheduler itself, for `hive_jobq`'s run-loop seam
    /// (`Scheduler::claim_next`), which takes exactly this type.
    ///
    /// Handing out the `Arc` rather than wrapping each crate call keeps the
    /// host from growing a parallel API: the run loop uses `hive_jobq`'s
    /// functions directly, and this module stays the thin glue it is being
    /// reduced to.
    pub(crate) fn sched(&self) -> &Arc<Mutex<Sched>> {
        &self.sched
    }

    /// The DAG container id owning `node`, for log lines and the dashboard.
    /// Derived from the graph rather than carried alongside the node — the
    /// parent axis already knows it.
    #[must_use]
    pub fn dag_of(&self, node: NodeId) -> Option<u64> {
        self.lock().graph().root_of(node).map(NodeId::get)
    }

    /// Cancel a DAG that hasn't started yet: every work node is still `Pending`,
    /// so each is cancelled. `false` once any work node is running or terminal —
    /// an in-flight nix build isn't interruptible.
    ///
    /// **Nodes that explicitly observe cancellation are spared** — a node whose
    /// edge names [`hive_jobq::TerminalState::Cancelled`] is asking to run when
    /// the work it follows was dropped, which is exactly what an approval tail
    /// needs: cancel the work, and the tail still fires to resolve the approval
    /// row rather than leaving it dangling forever.
    ///
    /// Nothing is special-cased by node kind. `AFTER_ANY` deliberately does *not*
    /// accept `Cancelled`, so an ordinary weak-edged step (rebuild's `Reconcile`,
    /// say) is cancelled along with everything else — there is nothing to converge
    /// when no node ever ran. Only a node that named `Cancelled` survives, and it
    /// survives because it asked to.
    ///
    /// `id` names **any node**, not specifically a DAG. Cancelling a group root
    /// drops that whole group (the cascade is the scheduler's), which is what
    /// the dashboard's whole-DAG cancel does; cancelling an interior node drops
    /// just that branch. Nothing here knows about DAGs.
    pub fn cancel(&self, id: u64) -> bool {
        let mut inner = self.lock();
        let Some(node) = inner.graph().resolve_id(id) else {
            return false;
        };
        if !inner.cancel_node(node) {
            return false;
        }
        drop(inner);
        self.notify.notify_one();
        true
    }

    /// The first failed node's error in `dag_id`, if any has failed yet.
    ///
    /// Unlike the roll-up summary this is readable *mid-flight*, which is the
    /// point: a compensation node runs `AfterAny` its subject, so when it asks,
    /// the DAG is still `Finishing` (the compensation node itself is running)
    /// while the node it is compensating for has already settled `Failed`. That
    /// lets the compensation annotate its bookkeeping with the reason the deploy
    /// failed, instead of having the error handed down from the node that hit
    /// it. `None` when nothing has failed — the ordinary success path.
    #[must_use]
    pub fn first_error(&self, dag_id: u64) -> Option<String> {
        let inner = self.lock();
        let container = container(&inner, dag_id)?;
        inner.graph().first_error(container).map(ToOwned::to_owned)
    }

    /// `(agent, label, takes_container_down)` for the live transient-pill set,
    /// recomputed from the nodes **actually running** — not from an intent a
    /// template declared at submit time. (A rebuild used to report `rebuilding`
    /// for its whole life: prebuild, stop, swap, tail and reconcile alike.)
    ///
    /// A node lights a pill when it is `Running` **and declares the agent's
    /// resource itself**. Declaring is the test, not targeting — `Prebuild` /
    /// `MetaSync` name an agent but are lease-exempt on purpose, since the
    /// container keeps serving through them. Nor is it the lease *owner*:
    /// `resource_state()` answers "who holds the slot", a different question.
    ///
    /// `label` is the node's own wire tag ([`NodeKind::as_str`]), the vocabulary
    /// [`NodeView::kind`] already ships, so a pill and a DAG node name an
    /// operation identically. `takes_container_down` is the crash watcher's
    /// input, carried rather than inferred from the label — a `Start` pill and a
    /// `Stop` pill are both pills; only one means a vanished container is
    /// expected.
    ///
    /// Read off the node's **declared** resource edges, not off its kind. Those
    /// are the same thing now that every construction site states what it holds,
    /// and the distinction is the whole point: `Start` / `Stop` / `PostSwap` run
    /// inside a lease-holding ancestor, and while the declaration was derived
    /// from the kind they re-declared nothing and lit no pill. Asking the node
    /// what it holds cannot go stale that way. An agent's lease is cap-1, so at
    /// most one entry per agent.
    #[must_use]
    pub fn running_transients(&self) -> Vec<RunningTransient> {
        let inner = self.lock();
        inner
            .graph()
            .nodes()
            .filter(|n| matches!(n.state, State::Running))
            .filter_map(|n| {
                let agent = n.deps.iter().find_map(|dep| match dep {
                    hive_jobq::Dep::Resource {
                        name: Resource::Agent(a),
                        ..
                    } => Some(a.clone()),
                    _ => None,
                })?;
                Some(RunningTransient {
                    agent,
                    label: n.payload.as_str().to_owned(),
                    takes_container_down: n.payload.takes_container_down(),
                    // `started_at` is set when a node enters `Running`, and this
                    // only sees `Running` nodes — the fallback is unreachable in
                    // practice, and "just now" is the honest answer if it isn't.
                    since: n.started_at.unwrap_or_else(Utc::now),
                })
            })
            .collect()
    }

    /// Snapshot every live + retained DAG for `/api/state` + `RebuildQueueChanged`.
    #[must_use]
    pub fn snapshot(&self) -> Vec<DagView> {
        let inner = self.lock();
        let mut ids = visible_dags(&inner);
        ids.sort_unstable_by_key(|c| c.get());
        ids.into_iter()
            .filter_map(|c| dag_view(&inner, c))
            .collect()
    }

    /// Number of live (non-terminal) DAGs — tests + diagnostics.
    #[cfg(test)]
    #[must_use]
    pub fn live_count(&self) -> usize {
        let inner = self.lock();
        containers(&inner)
            .into_iter()
            .filter(|&c| inner.graph().is_settled(c) == Some(false))
            .count()
    }
}

/// The container node of `dag_id` — the `NodeKind::Dag` root whose id equals
/// `dag_id`. `NodeId` is un-fabricable from a raw `u64`, so this is a search.
fn container(sched: &Sched, dag_id: u64) -> Option<NodeId> {
    sched.graph().nodes().find_map(|n| {
        (n.parent.is_none() && n.id.get() == dag_id && matches!(n.payload, NodeKind::Dag { .. }))
            .then_some(n.id)
    })
}

/// The container's carried domain metadata as an owned read-view. The data
/// lives solely in the [`NodeKind::Dag`] payload — this is a derived read,
/// not a stored side-table.
fn dag_meta(sched: &Sched, container: NodeId) -> Option<DagMeta> {
    let NodeKind::Dag {
        source,
        reason,
        created_at,
    } = &sched.graph().node(container)?.payload
    else {
        return None;
    };
    Some(DagMeta {
        source: *source,
        reason: reason.clone(),
        created_at: *created_at,
    })
}

/// Project a DAG into its wire [`DagView`]: a near-raw view of the
/// container's work nodes, with `Done` nodes excluded. Lifecycle
/// (`state` / `started_at` / `finished_at` / `error`) is read straight
/// off each `hive_jobq::Node`; the client derives the DAG label, roll-up
/// state, and DAG timestamps from the node set. Non-derivable per-node
/// payload (`approval_id`, meta `inputs`) rides the owning node. Returns
/// `None` when every work node is `Done` or `Skipped` — a fully-settled
/// DAG drops out of the snapshot entirely (a `Failed` one lingers until
/// aged out).
fn dag_view(sched: &Sched, container: NodeId) -> Option<DagView> {
    let meta = dag_meta(sched, container)?;
    let mut nodes = Vec::new();
    // Whether anything in this DAG still has an outcome worth showing.
    // Kept separate from `nodes` being non-empty: skipped nodes ride the
    // wire so the dashboard can mark the branches that weren't taken, but
    // they must not by themselves hold a finished DAG in the snapshot.
    let mut any_unsettled = false;
    // DAG-level timestamps are taken over *all* subtree nodes (including the
    // `Done` ones excluded from the wire) — the client can't derive them
    // from a `Done`-filtered node set, so the host computes them here.
    let mut started: Vec<DateTime<Utc>> = Vec::new();
    let mut finished: Vec<DateTime<Utc>> = Vec::new();
    for node in sched.graph().descendants(container) {
        let id = node.id;
        if let Some(s) = node.started_at {
            started.push(s);
        }
        if let Some(f) = node.finished_at {
            finished.push(f);
        }
        // `Done` nodes drop off the wire — a finished step isn't
        // interesting. `Skipped` ones stay: which branch a run *didn't*
        // take is the readable half of an outcome-branched DAG.
        if matches!(node.state, State::Done) {
            continue;
        }
        any_unsettled |= !matches!(node.state, State::Skipped);
        let deps: Vec<u64> = node
            .deps
            .iter()
            .filter_map(|d| match d {
                Dep::Node { id, .. } => Some(id.get()),
                Dep::Resource { .. } => None,
            })
            .collect();
        // Non-derivable per-node payload rides the node that owns it. Every
        // deploy phase carries the approval id, but only the subtree root
        // projects it onto the wire — hanging the approval link off all of
        // them would render the same card once per phase.
        let approval_id = match &node.payload {
            NodeKind::DeployWindow { approval_id, .. } => Some(*approval_id),
            _ => None,
        };
        let inputs = match &node.payload {
            NodeKind::MetaLock { inputs, .. } => inputs.clone(),
            _ => Vec::new(),
        };
        // Looked up from the log row itself (`build_logs.node_id`), not a
        // host-side map. One indexed query per node in the snapshot; the
        // node set is bounded by `MAX_HISTORY_DAGS` and the store is a
        // local sqlite file, so this is cheaper than the lock contention
        // a second shared map would reintroduce.
        let build_log_id = crate::build_logs::global().and_then(|h| h.id_for_node(id.get()));
        // `node.parent` is the structural jobq parent. Top-level nodes
        // have `parent == Some(container)` (direct children of the Dag
        // container); those become `parent: None` on the wire since the
        // container itself is not part of the work-node payload. Sub-nodes
        // carry the id of their containing parent work-node.
        let parent = node
            .parent
            .filter(|&p| p != container)
            .map(hive_jobq::NodeId::get);
        nodes.push(NodeView {
            id: id.get(),
            agent: node.payload.agent().to_owned(),
            kind: node.payload.as_str().to_owned(),
            deps,
            state: node.state,
            started_at: node.started_at,
            finished_at: node.finished_at,
            error: node.error.clone(),
            approval_id,
            inputs,
            build_log_id,
            parent,
        });
    }
    if !any_unsettled {
        return None;
    }
    let is_terminal = sched.graph().is_settled(container) == Some(true);
    Some(DagView {
        id: container.get(),
        source: meta.source,
        reason: meta.reason.clone(),
        created_at: meta.created_at,
        started_at: started.into_iter().min(),
        finished_at: is_terminal.then(|| finished.into_iter().max()).flatten(),
        nodes,
    })
}

/// When a DAG's work node finishes on `finished_at` — the max over its
/// subtree (read off the graph `Node`, as unix seconds), for the history
/// cap ordering.
fn dag_finished_at(sched: &Sched, container: NodeId) -> i64 {
    sched
        .graph()
        .descendants(container)
        .filter_map(|n| n.finished_at)
        .map(|t| t.timestamp())
        .max()
        .unwrap_or(0)
}

/// Every DAG container node id in the graph.
fn containers(sched: &Sched) -> Vec<NodeId> {
    sched
        .graph()
        .nodes()
        .filter(|n| n.parent.is_none() && matches!(n.payload, NodeKind::Dag { .. }))
        .map(|n| n.id)
        .collect()
}

/// The **visible** DAG set for the snapshot: every live (non-terminal) DAG,
/// plus the newest [`MAX_HISTORY_DAGS`] terminal ones. Crate nodes for
/// evicted DAGs linger in the graph (bounded-prune is a Stage-C follow-up);
/// this filter is what bounds what the dashboard sees.
fn visible_dags(sched: &Sched) -> Vec<NodeId> {
    let mut live: Vec<NodeId> = Vec::new();
    let mut terminal: Vec<(NodeId, i64)> = Vec::new();
    for c in containers(sched) {
        if sched.graph().is_settled(c) == Some(true) {
            terminal.push((c, dag_finished_at(sched, c)));
        } else {
            live.push(c);
        }
    }
    // Newest first, so truncating to the cap keeps the most recent.
    terminal.sort_by(|a, b| b.1.cmp(&a.1).then(b.0.get().cmp(&a.0.get())));
    terminal.truncate(MAX_HISTORY_DAGS);
    let mut kept = live;
    kept.extend(terminal.into_iter().map(|(c, _)| c));
    kept
}

/// Truncate a node error to [`MAX_ERROR_LEN`] on a char boundary, appending `…`.
fn truncate_error(e: &str) -> String {
    if e.len() <= MAX_ERROR_LEN {
        return e.to_owned();
    }
    let cut = (0..=MAX_ERROR_LEN)
        .rev()
        .find(|i| e.is_char_boundary(*i))
        .unwrap_or(0);
    let mut msg = e[..cut].to_owned();
    msg.push('…');
    msg
}