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

//! Node executors — one async fn per [`NodeKind`], each a thin wrapper
//! over existing `lifecycle.rs` / `meta.rs` / `actions.rs` code. Node
//! executors keep their own internal error handling where it exists
//! today (cold-start fallback inside `Reconcile`, non-fatal boot-time
//! lock bump inside the sweep `MetaLock`, warn-only forge sync in the
//! `Swap` tail); DAG-level failure handling is cancel-downstream in
//! the queue.

use std::sync::Arc;

use anyhow::{Context as _, Result};

use hive_jobq::{NodeId, TerminalState};

use super::model::NodeKind;
use crate::coordinator::Coordinator;
use crate::power::{ReconcileAction, reconcile_action};

/// Max time `Drain` waits for the harness to run its stop-checkpoint
/// turn before falling back to the hard stop. Generous — a checkpoint
/// turn can take a while — but bounded so a wedged agent never blocks
/// the stop indefinitely. Drains hold no build slot, so a whole-hive
/// graceful stop overlaps every agent's drain instead of serialising
/// N × this timeout.
pub const GRACEFUL_STOP_TIMEOUT: std::time::Duration = std::time::Duration::from_mins(3);

/// Run one claimed node to completion. Called from a task the
/// scheduler spawns per claim; the `Result` (stringified) becomes the
/// node's terminal state.
///
/// `builder` is the node's own growth channel: an executor that decides more work
/// is needed declares it here, and the scheduler inserts it under this node
/// when the node completes. Most executors never touch it. Nothing is inserted
/// while the node runs — the builder is local state, so this stays outside the
/// queue's lock for the whole (often multi-minute) execution.
///
/// ⚠️ Taken **by value and handed back**, not by reference. A `JobBuilder` is
/// `RefCell`-backed: owned it is `Send`, but `&JobBuilder` is not (a shared ref
/// is `Send` only if the referent is `Sync`, and `RefCell` never is). A `&JobBuilder`
/// parameter would be live across every `.await` in this fn and make the whole
/// future non-`Send`, which the scheduler's `tokio::spawn` rejects. So the
/// growth executors below return *what to grow* and the declaration happens
/// here, synchronously, between awaits.
///
/// The node is identified by its own id + payload rather than by a `Claim`
/// side-struct: `kind` already carries the agent, and the DAG id is a
/// derived read (`JobQueue::dag_of`) the three arms that need it take
/// themselves. Nothing here needs a claim to exist as a type.
pub(super) async fn run_node(
    coord: &Arc<Coordinator>,
    builder: super::JobBuilder,
    id: NodeId,
    kind: &NodeKind,
) -> (super::JobBuilder, Result<()>) {
    // The agent this node targets rides the payload — empty for the agentless
    // kinds (`MetaLock`, `Reparent`), which never read it.
    let agent = kind.agent();
    // Every arm is `Result<()>`; the three that grow work declare into `builder`
    // *synchronously*, after their own awaits have finished. Borrowing `&builder`
    // inside an `.await` would make this future non-`Send` (see above), so the
    // growth executors return what to grow rather than taking the builder.
    let result = match kind {
        NodeKind::MetaSync { relock, .. } => run_meta_sync(coord, agent, *relock).await,
        NodeKind::Prebuild { .. } => run_prebuild(agent, id).await,
        NodeKind::Swap { .. } => run_swap(coord, agent, id).await,
        NodeKind::RebuildBookkeeping { .. } => run_rebuild_bookkeeping(coord, agent).await,
        NodeKind::Provision { .. } => run_provision(coord, agent).await,
        NodeKind::Create { .. } => run_create(agent).await,
        NodeKind::MetaLock {
            sweep,
            fanout,
            inputs,
        } => run_meta_lock(coord, *sweep, fanout.clone(), inputs)
            .await
            .map(|agents| {
                // `sweep` is the whole difference: it relocks per-agent like a
                // manual rebuild, and it drains agents that were mid-turn when
                // the host came up. A cascade does neither. Decided here rather
                // than returned, since `run_meta_lock` would only be deriving
                // it from the `sweep` this call site already holds.
                if *sweep {
                    super::templates::grown_graceful_rebuilds(&builder, &agents, true);
                } else {
                    super::templates::grown_rebuilds(&builder, &agents, false);
                }
            }),
        NodeKind::Reconcile { .. } => run_reconcile(coord, agent).await.map(|sub| {
            if let Some(kind) = sub {
                super::templates::fanned_out_mechanical(&builder, kind);
            }
        }),
        NodeKind::Start { .. } => run_start(coord, agent).await,
        NodeKind::Stop { .. } => run_stop(coord, agent).await,
        NodeKind::StopForUpdate { .. } => run_stop_for_update(coord, agent).await,
        NodeKind::Signal { .. } => {
            run_signal(coord, agent);
            Ok(())
        }
        NodeKind::Drain { .. } => run_drain(coord, agent).await,
        NodeKind::WriteDropin { .. } => run_write_dropin(coord, agent).await,
        // The payload rides the node and is destructured here, so the executor
        // takes it directly instead of re-matching the kind behind a `bail!`
        // that could never fire.
        NodeKind::WritePermFile { payload, .. } => run_write_perm_file(coord, agent, payload).await,
        NodeKind::Reparent { moves } => run_reparent(coord, moves).await,
        NodeKind::MergeVerify { approval_id, .. } => {
            run_merge_verify(coord, *approval_id, id).await
        }
        NodeKind::DeployApply { approval_id, .. } => {
            run_deploy_apply(coord, *approval_id, id).await.map(|()| {
                super::templates::deploy_rebuild_nodes(&builder, agent, *approval_id);
            })
        }
        NodeKind::FinalizeDeploy { approval_id, .. } => {
            run_finalize_deploy(coord, *approval_id).await
        }
        NodeKind::DeployTail { approval_id, .. } => {
            run_deploy_tail(coord, coord.job_queue.root_of(id), agent, *approval_id).await
        }
        NodeKind::ResolveApproval {
            approval_id,
            outcome,
        } => run_resolve_approval(coord, coord.job_queue.root_of(id), *approval_id, *outcome).await,
        NodeKind::EmitRebuilt { ok, .. } => {
            run_emit_rebuilt(coord, agent, coord.job_queue.root_of(id), *ok).await;
            Ok(())
        }
        NodeKind::SetWanted { up, .. } => run_set_wanted(coord, agent, *up),
        // Braces carry no work of their own; completing one lets it reach
        // `Finishing` so the nodes under it start. What they declare stays held
        // until their whole subtree settles.
        NodeKind::DeployWindow { .. } | NodeKind::AgentWindow { .. } => Ok(()),
    };
    (builder, result)
}

/// Resolve the DAG's approval row the way this node's own `outcome` says.
///
/// Nothing is inspected: a template emits one of these per outcome, each edged to
/// accept only that one, so *which* node the scheduler let run already is the
/// answer. Best-effort — a resolution failure is logged inside
/// [`crate::actions::resolve_approval_dag`], never surfaced as a node failure,
/// since the work already happened and failing the tail would only misreport it.
async fn run_resolve_approval(
    coord: &Arc<Coordinator>,
    dag_id: Option<u64>,
    approval_id: i64,
    outcome: TerminalState,
) -> Result<()> {
    let reason = (outcome == TerminalState::Failed)
        .then(|| dag_id.and_then(|dag| coord.job_queue.first_error(dag)))
        .flatten();
    crate::actions::resolve_approval_dag(coord, approval_id, outcome, reason.as_deref()).await;
    Ok(())
}

/// Emit this agent's rebuild-complete todo. `ok` is not computed — it is which
/// of the tail pair the graph let run. The failure note comes from the DAG's
/// first failing node, since the branch knows *that* it failed but not *why*.
async fn run_emit_rebuilt(coord: &Arc<Coordinator>, agent: &str, dag_id: Option<u64>, ok: bool) {
    let note = (!ok)
        .then(|| dag_id.and_then(|dag| coord.job_queue.first_error(dag)))
        .flatten();
    let summary = crate::coordinator::rebuilt_todo_summary(agent, ok, note.as_deref(), None, None);
    let _ = coord
        .push_todo(
            hive_sh4re::manager::MANAGER_AGENT,
            "core",
            Some(format!("rebuilt:{agent}")),
            summary,
            None,
        )
        .await;
}

/// Write the agent's durable power intent — the DAG-node form of the old
/// pre-submit `set_wanted` side effect. Store-only (no container touch), so
/// build-slot-exempt; but it declares the agent's lifecycle lease
/// (`Resource::Agent`) so the whole power-op DAG is atomic per-agent.
/// The downstream `Reconcile` reads the intent this writes. Unlike the old
/// warn-and-continue write, a failed write fails the node (cancel-downstream
/// cancels the `Reconcile`) rather than letting it converge to a stale
/// intent — that atomicity is the point of moving it into the DAG.
fn run_set_wanted(coord: &Arc<Coordinator>, agent: &str, up: bool) -> Result<()> {
    let wanted = if up {
        crate::power::Wanted::Up
    } else {
        crate::power::Wanted::Offline
    };
    coord
        .power
        .set(agent, wanted)
        .with_context(|| format!("set wanted={} for agent {agent}", wanted.as_str()))?;
    Ok(())
}

/// The rebuild's meta preamble: runtime-dir prep, an idempotent meta
/// `sync_agents`, and the optional per-agent relock. Runs under the deploy
/// window (`Resource::MetaWindow`, held by the scheduler for this
/// node) so its commits can never land inside another node's staged
/// prepare→finalize window.
///
/// Deliberately a separate node from the [`run_prebuild`] it feeds: that
/// build takes minutes and only *reads* the store, so keeping the global
/// window off it is what lets rebuilds of different agents overlap.
async fn run_meta_sync(coord: &Arc<Coordinator>, name: &str, relock: bool) -> Result<()> {
    // Runs while the agent is still up — the runtime dir and MCP listener
    // already exist. Use the pure path accessor; no need to re-register the
    // listener (event-driven: registered at start/create).
    let agent_dir = crate::paths::agent_runtime_dir(name);
    let hive = coord.hive_env();
    let paths = Coordinator::agent_paths(name, agent_dir);
    crate::lifecycle::prepare_rebuild_dirs(name, &paths).await?;
    // Idempotent meta sync so a manual rebuild can also recover from a
    // divergent meta repo; then bump just this agent's input. `relock =
    // false` only for meta-update cascade children, where re-locking
    // would revert the bump the cascade just committed.
    let agents = crate::lifecycle::agents_for_meta_listing().await?;
    crate::meta::sync_agents(&hive, &agents).await?;
    if relock {
        crate::meta::lock_update_for_rebuild(name).await?;
    }
    Ok(())
}

/// Out-of-band toplevel build while the container keeps serving: warm
/// `system.build.toplevel` so the later `Swap` hits cache and skips
/// straight to the profile-swap, against a meta repo the upstream
/// `MetaSync` node has already synced. The warm build is skipped when the
/// container is already down: its only purpose is to shrink the swap's
/// downtime window, so a stopped agent (no uptime to preserve) doesn't
/// pay the double eval — `Swap` builds inline instead.
async fn run_prebuild(name: &str, id: NodeId) -> Result<()> {
    // Warm the toplevel build only when the container is up — the whole
    // point of prebuild is to shrink the swap's downtime window. A
    // stopped agent has no uptime to preserve, so skip the (expensive)
    // eval and let the downstream `Swap` build inline.
    if crate::lifecycle::is_running(name).await {
        let flake_ref = format!("{}#{name}", crate::paths::meta_root().display());
        crate::lifecycle::prebuild_toplevel(name, &flake_ref, Some(id.get())).await?;
    }
    Ok(())
}

/// Profile-swap: re-apply drop-ins (rebuild is the reconcile verb),
/// `nixos-container update`, then the post-rebuild bookkeeping tail
/// (rev marker, `Rebuilt` event, forge/matrix sync, kick, rescan).
/// The recovery-start on failure is NOT here — the DAG's tail
/// `Reconcile` runs after this node terminal ok *or* fail.
async fn run_swap(coord: &Arc<Coordinator>, name: &str, id: NodeId) -> Result<()> {
    // Swap runs on an already-existing (stopped) container — runtime dir
    // and listener were created earlier. Pure path accessor suffices.
    let agent_dir = crate::paths::agent_runtime_dir(name);
    let hive = coord.hive_env();
    let paths = Coordinator::agent_paths(name, agent_dir);
    let result = crate::lifecycle::swap_update(name, &hive, &paths, Some(id.get())).await;
    // On success the Ok-only bookkeeping tail (rev marker, forge/matrix
    // sync, kick, rescan, snapshot) runs in the sibling `RebuildBookkeeping` node,
    // which deps `AfterOk(Swap)`. On failure `RebuildBookkeeping` is cancel-cascaded
    // and the tail `Reconcile` (`AfterAny(RebuildBookkeeping)`) handles recovery; here
    // we only refresh the observed state so dashboards reflect the failed
    // swap immediately. The `Rebuilt { ok: false }` manager event is emitted by
    // the DAG's `EmitRebuilt` tail (any node may be the one that failed).
    if result.is_err() {
        coord.rescan_containers_and_emit().await;
    }
    result
}

/// The post-`Swap` bookkeeping tail, split into its own node for dashboard
/// visibility + retry granularity. Deps `AfterOk(Swap)`, so reaching here
/// means the profile swap succeeded. Store/forge/matrix work only — no nix
/// build (build-slot-exempt); the agent lease taken at `Swap` is still held
/// (the whole chain up to `Reconcile` is one agent's subgraph).
async fn run_rebuild_bookkeeping(coord: &Arc<Coordinator>, name: &str) -> Result<()> {
    if let Some(rev) = crate::auto_update::current_flake_rev(&coord.hyperhive_flake)
        && let Err(e) = std::fs::write(crate::paths::applied_rev_marker(name), rev)
    {
        tracing::warn!(%name, error = ?e, "write rev marker failed");
    }
    // The `Rebuilt` manager event is emitted exactly once per agent by the DAG's
    // `EmitRebuilt` tail — emitting ok here and letting a failed tail `Reconcile`
    // add a contradictory !ok would double-report the same rebuild.
    // Full forge + matrix sync on every successful rebuild so the rebuild
    // path is equivalent to the startup sweep: tokens, config-repo mirror,
    // meta access all recover without a hive-c0re restart.
    crate::forge::sync_agent(name, crate::forge::core_token().as_deref()).await;
    crate::matrix::sync_agent_standalone(name).await;
    // Wake the agent on its next turn so claude sees a "you were rebuilt"
    // hint; rescan so dashboards drop the "needs update" chip; lock bump →
    // meta-inputs re-render.
    coord.kick_agent(name, "container rebuilt");
    coord.rescan_containers_and_emit().await;
    crate::dashboard::emit_meta_inputs_snapshot(coord);
    Ok(())
}

/// First-spawn pre-create provisioning: proposed/applied repos, state
/// subvolume, and the meta `sync_agents` registration. Runs under the
/// deploy window (it declares `Resource::MetaWindow`) so its commit can't
/// land inside another node's staged deploy window.
async fn run_provision(coord: &Arc<Coordinator>, name: &str) -> Result<()> {
    let agent_dir = crate::paths::agent_runtime_dir(name);
    let hive = coord.hive_env();
    let paths = Coordinator::agent_paths(name, agent_dir);
    crate::lifecycle::provision_container(name, &hive, &paths).await?;
    Ok(())
}

/// `nixos-container create` proper — the upstream `Provision` node
/// already registered the agent in meta, so this only reads the store
/// (no deploy-window gate needed, mirroring `Prebuild`'s build). Runtime
/// dir creation and MCP listener registration are deferred to the tail
/// `Reconcile` (`converge_start_preamble` + `register_agent`) so this
/// node stays purely "create", not "create + start".
async fn run_create(name: &str) -> Result<()> {
    crate::lifecycle::create_only(name).await?;
    Ok(())
}

/// Meta flake lock bump. Boot-sweep flavour is non-fatal (a failed
/// bump must not cancel the fan-out rebuilds — they proceed against
/// the current lock, exactly like today's sweep); the meta-update
/// flavour propagates errors, and a failed bump fans out nothing.
/// Returns the agents whose rebuild subgraphs the caller should grow into this
/// node, rather than declaring them here — the declaration has to happen outside
/// any `.await` (see [`run_node`]).
///
/// Only the agent list: *which* rebuild flavour to grow is a pure function of
/// `sweep`, which the caller passed in, so returning it too would be a round
/// trip rather than a decision.
async fn run_meta_lock(
    coord: &Arc<Coordinator>,
    sweep: bool,
    fanout: Option<Vec<String>>,
    inputs: &[String],
) -> Result<Vec<String>> {
    if sweep {
        if let Err(e) = crate::meta::lock_update_hyperhive().await {
            tracing::warn!(error = ?e, "startup sweep: meta lock_update_hyperhive failed");
        }
        // Grow one rebuild subgraph per stale agent into *this* boot DAG
        // (rooted on this `MetaLock`, so they build against the post-bump
        // lock), rather than fanning out child DAGs. The caller grows them
        // with the graceful flavour: the per-agent drains overlap, so the
        // sweep's cost ceiling is one `GRACEFUL_STOP_TIMEOUT` in total, not
        // one per agent.
        return Ok(fanout.unwrap_or_default());
    }
    let _progress = coord.meta_update_guard();
    crate::meta::lock_update(inputs).await?;
    // Lock file changed — meta-inputs panel re-renders.
    crate::dashboard::emit_meta_inputs_snapshot(coord);
    let cascade = match fanout {
        Some(list) => list,
        None => meta_update_cascade_agents(inputs).await,
    };
    // Grow one rebuild subgraph per affected agent into *this* meta-update
    // DAG (rooted on this `MetaLock`, so they build against the post-bump
    // lock), rather than fanning out child DAGs. `relock = false` — the
    // cascade children must NOT re-lock, which would revert the bump this
    // node just committed (the property the old `fanout_specs` meta-update
    // branch encoded).
    Ok(cascade)
}

/// Idempotent power-converge *planner*: compare `wanted` (durable
/// intent) against observed state and, when they diverge, fan the
/// mechanical `Start` / `Stop` out as a first-class node appended to
/// *this* DAG (a single node declared into `builder`, rooted on this node).
/// Does no container work itself — the sub-step becomes visible in the
/// DAG and the lease-window transient (or the sub-step's own node-local
/// guard) rides across it.
/// Returns the mechanical node to fan out (`None` on a noop) rather than
/// declaring it — the declaration has to happen outside any `.await`, see
/// [`run_node`]. `NodeKind` carries the agent it targets, so this node's agent
/// is stamped into the fanned-out kind here.
async fn run_reconcile(coord: &Arc<Coordinator>, name: &str) -> Result<Option<NodeKind>> {
    let running = crate::lifecycle::is_running(name).await;
    let wanted = coord.power.get_or_seed(name, running)?;
    Ok(match reconcile_action(wanted, running) {
        ReconcileAction::Start => Some(NodeKind::Start {
            agent: name.to_owned(),
        }),
        ReconcileAction::Stop => Some(NodeKind::Stop {
            agent: name.to_owned(),
        }),
        ReconcileAction::Noop => {
            tracing::debug!(%name, wanted = wanted.as_str(), running, "reconcile: noop");
            None
        }
    })
}

/// Mechanical container start — the sub-step a `Reconcile` planner fans
/// out when it observes `wanted = Up` and the container down.
async fn run_start(coord: &Arc<Coordinator>, name: &str) -> Result<()> {
    // No node-local transient guard: the pill is derived from the running node
    // set, and `Start` reports `Starting` via `NodeKind::transient_kind`. This
    // used to take one "only when the DAG holds none", which was a second
    // derivation covering the gap left by a DAG-level declaration that couldn't
    // describe a sub-step.
    // Run the typed start preamble: ensures the runtime dir exists and
    // writes the nspawn/resource-limits drop-ins. The returned
    // StartableAgent token is the only way to call start_with_fallback —
    // omitting this becomes a compile error.
    let agent_dir = crate::paths::agent_runtime_dir(name);
    let hive = coord.hive_env();
    let paths = Coordinator::agent_paths(name, agent_dir);
    let token = crate::lifecycle::converge_start_preamble(name, &hive, &paths).await?;
    crate::lifecycle::start_with_fallback(token).await?;
    // Bind the MCP listener immediately after starting the container.
    // The preamble created the runtime dir; the container is now coming
    // up and will connect to this socket on its first turn. Event-driven
    // (no background poll) — c0re owns the listener lifecycle, so
    // register here rather than waiting for a sweep.
    coord.register_agent(name)?;
    coord.kick_agent(name, "container started");
    coord.rescan_containers_and_emit().await;
    Ok(())
}

/// Mechanical container stop — the sub-step a `Reconcile` planner fans
/// out when it observes `wanted = Offline` and the container up.
async fn run_stop(coord: &Arc<Coordinator>, name: &str) -> Result<()> {
    // See `run_start`: no node-local guard — `Stop` reports `Stopping` from its
    // own kind now.
    crate::lifecycle::kill(name).await?;
    coord.unregister_agent(name);
    let _ = coord
        .push_todo(
            hive_sh4re::manager::MANAGER_AGENT,
            "core",
            Some(format!("killed:{name}")),
            format!("agent '{name}' killed"),
            None,
        )
        .await;
    coord.rescan_containers_and_emit().await;
    Ok(())
}

/// Mechanical stop for the profile swap. Never *changes* `wanted`;
/// noop when already stopped.
async fn run_stop_for_update(coord: &Arc<Coordinator>, name: &str) -> Result<()> {
    if crate::lifecycle::is_running(name).await {
        // Seed a missing agent_power row from the PRE-stop observation
        // — the DAG's tail `Reconcile` observes only the mechanically
        // stopped state and would otherwise seed a running-but-unknown
        // agent as `Offline`, stranding it down after its own rebuild.
        if let Err(e) = coord.power.get_or_seed(name, true) {
            tracing::warn!(%name, error = ?e, "agent_power: pre-stop seed failed");
        }
        crate::lifecycle::kill(name).await?;
        coord.rescan_containers_and_emit().await;
    }
    Ok(())
}

/// Set the graceful fence + kick so the harness sees it promptly and
/// runs its one stop-checkpoint turn.
///
/// Skipped entirely for a paused agent: its loop parks on the pause
/// marker without polling the broker, so it would never observe the
/// fence and the downstream drain would just burn
/// `GRACEFUL_STOP_TIMEOUT`. Safe because the harness tests the marker at
/// the top of its loop — a paused agent has no turn in flight, so there
/// is nothing to checkpoint.
fn run_signal(coord: &Arc<Coordinator>, name: &str) {
    if hive_types::Ident::parse(name).is_ok_and(|a| Coordinator::is_paused(&a)) {
        return;
    }
    coord.mark_graceful_stop(name);
    coord.kick_agent(name, "graceful stop requested");
}

/// Await the harness clearing the fence (`GracefulStopComplete`) or
/// the timeout — either way the downstream `Reconcile` proceeds with
/// the actual stop.
async fn run_drain(coord: &Arc<Coordinator>, name: &str) -> Result<()> {
    let deadline = std::time::Instant::now() + GRACEFUL_STOP_TIMEOUT;
    while coord.is_graceful_stop_pending(name) {
        if std::time::Instant::now() >= deadline {
            tracing::warn!(agent = %name, "graceful stop: drain timed out — hard-stopping");
            break;
        }
        tokio::time::sleep(std::time::Duration::from_millis(500)).await;
    }
    coord.clear_graceful_stop(name);
    Ok(())
}

/// `set_nspawn_flags` + `set_resource_limits` + daemon-reload.
async fn run_write_dropin(coord: &Arc<Coordinator>, name: &str) -> Result<()> {
    // write_dropins only needs the path value to build AgentPaths; the
    // dir doesn't need to exist at this point (created by ensure_agent_runtime_dir
    // on the upstream Prebuild/Start node).
    let agent_dir = crate::paths::agent_runtime_dir(name);
    let hive = coord.hive_env();
    let paths = Coordinator::agent_paths(name, agent_dir);
    crate::lifecycle::write_dropins(name, &hive, &paths).await?;
    Ok(())
}

/// Write + commit the perm file(s) (fused under `META_LOCK` so the
/// working tree is never left dirty), then emit the P3RM1SS10NS-tab
/// snapshots so the dashboard reflects the new assignment.
async fn run_write_perm_file(
    coord: &Arc<Coordinator>,
    name: &str,
    payload: &super::model::PermPayload,
) -> Result<()> {
    use super::model::PermPayload;
    // Runs under the deploy window (it declares `Resource::MetaWindow`): a
    // perm commit landing inside another node's staged prepare→finalize
    // window would sweep the staged deploy lock into its commit (the
    // commits are also path-limited in meta.rs — belt and braces).
    match payload {
        PermPayload::ToolGroups { groups } => {
            crate::meta::commit_tool_groups(name, groups)
                .await
                .with_context(|| format!("commit tool-groups for {name}"))?;
            coord.emit_tool_groups_snapshot();
        }
        PermPayload::Capabilities { caps } => {
            crate::meta::commit_capabilities(name, caps)
                .await
                .with_context(|| format!("commit capabilities for {name}"))?;
            coord.emit_capabilities_snapshot();
        }
        PermPayload::Combined { groups, caps } => {
            crate::meta::commit_perms(name, groups.as_deref(), caps.as_deref())
                .await
                .with_context(|| format!("commit perms for {name}"))?;
            if groups.is_some() {
                coord.emit_tool_groups_snapshot();
            }
            if caps.is_some() {
                coord.emit_capabilities_snapshot();
            }
        }
    }
    Ok(())
}

/// Apply the node's `(child, new_parent)` moves as one `META_LOCK`-fused
/// commit (`Coordinator::reparent_bulk_with_notify`, which already handles
/// both the single- and bulk-move case, sends the per-agent move
/// notifications, and rescans + diff-emits the container tree). Runs under
/// the deploy window (it declares `Resource::MetaWindow`), same reasoning as
/// `run_write_perm_file`: a topology commit landing inside another node's
/// staged deploy window would sweep the staged lock into its commit.
async fn run_reparent(
    coord: &Arc<Coordinator>,
    moves: &[(hive_types::Ident, Option<hive_types::Ident>)],
) -> Result<()> {
    let refs: Vec<(&str, Option<&str>)> = moves
        .iter()
        .map(|(child, parent)| {
            (
                child.as_str(),
                parent.as_ref().map(hive_types::Ident::as_str),
            )
        })
        .collect();
    coord
        .reparent_bulk_with_notify(&refs)
        .await
        .map_err(|e| anyhow::anyhow!(e))?;
    Ok(())
}

/// Deploy phase 1 — drift gate, fetch, eval-verify. Mutates nothing, so a
/// failure here cancel-cascades the rest of the subtree with the forge and the
/// applied repo exactly as they were.
async fn run_merge_verify(coord: &Arc<Coordinator>, approval_id: i64, id: NodeId) -> Result<()> {
    crate::actions::run_deploy_merge_verify(coord, approval_id, Some(id.get())).await
}

/// Deploy phase 2 — the irreversible half: ff-merge, then phase 1 of the
/// two-phase meta deploy.
///
/// On success it grows the ordinary rebuild subgraph (plus its closing
/// `FinalizeDeploy`) into this DAG rooted on *this* node — which is what puts
/// the appended nodes inside the `DeployWindow`'s subtree, so the `MetaWindow`
/// their `MetaSync` declares is re-entered rather than deadlocked against the
/// ancestor already holding it. On failure nothing is appended and the tail
/// compensates, exactly as before.
async fn run_deploy_apply(coord: &Arc<Coordinator>, approval_id: i64, id: NodeId) -> Result<()> {
    crate::actions::run_deploy_apply(coord, approval_id, Some(id.get())).await
}

/// Deploy phase 3 — close the staged-lock window once the appended rebuild has
/// come up clean: drop the rollback ref, plant the `deployed/<id>` tag, commit
/// the staged lock.
async fn run_finalize_deploy(coord: &Arc<Coordinator>, approval_id: i64) -> Result<()> {
    crate::actions::run_finalize_deploy(coord, approval_id).await
}

/// Deploy compensation + bookkeeping tail. `AfterAny` the apply node, so it
/// runs on every outcome; it is deliberately infallible (see
/// [`crate::actions::run_deploy_tail`]) — a failing tail must not flip an
/// otherwise-successful deploy's DAG state.
///
/// Takes the agent from the node payload so the tail can still compensate when
/// the approval row is gone (deny race, purge).
async fn run_deploy_tail(
    coord: &Arc<Coordinator>,
    dag_id: Option<u64>,
    agent: &str,
    approval_id: i64,
) -> Result<()> {
    crate::actions::run_deploy_tail(coord, dag_id, agent, approval_id).await;
    Ok(())
}

/// Compute which agents a `nix flake update <inputs>` on the meta
/// flake affects — the fan-out set for `MetaUpdate` DAGs. Empty
/// `inputs` or any input under `hyperhive` → every container;
/// otherwise just the agents named by `agent-<name>` inputs.
/// Topology-sorted so parents rebuild before their children.
pub async fn meta_update_cascade_agents(inputs: &[String]) -> Vec<String> {
    let touched_hyperhive = inputs
        .iter()
        .any(|i| i == "hyperhive" || i.starts_with("hyperhive/"));
    let touched_agents: Vec<String> = inputs
        .iter()
        .filter_map(|i| i.strip_prefix("agent-"))
        .map(|rest| rest.split('/').next().unwrap_or(rest).to_owned())
        .collect();
    let mut names = if touched_hyperhive || inputs.is_empty() {
        crate::lifecycle::list()
            .await
            .unwrap_or_default()
            .into_iter()
            .filter_map(|c| {
                c.strip_prefix(crate::lifecycle::AGENT_PREFIX)
                    .map(str::to_owned)
            })
            .collect()
    } else {
        touched_agents
    };
    let topo = crate::topology::read();
    crate::auto_update::topology_sort(&mut names, &topo);
    names
}