feat(#2484): unify in-DAG growth on append_subgraph (drop append_node)

append_subgraph is the multi-node/multi-agent generalisation of the
single-node append_node, so the two in-DAG-growth channels collapse to
one: the Reconcile planner now emits its mechanical Start/Stop as a
single-node append_subgraph rooted on the reconcile node (stamping
claim.agent on the NodeSpec, which append_node inherited implicitly).

Removes NodeOutput.append_nodes + its scheduler drain loop and
JobQueue::append_node. No behaviour change — a channel unification.
This commit is contained in:
atlas 2026-07-15 20:21:39 +02:00
commit b87eac0a61
3 changed files with 48 additions and 104 deletions

View file

@ -27,24 +27,19 @@ pub const GRACEFUL_STOP_TIMEOUT: std::time::Duration = std::time::Duration::from
/// success. /// success.
#[derive(Debug, Default)] #[derive(Debug, Default)]
pub struct NodeOutput { pub struct NodeOutput {
/// Mechanical sub-step nodes to append into *this same* DAG at
/// runtime, each depending `AfterOk` on the emitting node — e.g. a
/// `Reconcile` planner emitting a `Start` / `Stop`. Keeps the sub-step a
/// first-class node in the same DAG so the lease-window transient is held
/// across it. The scheduler applies these *before* the emitting node's
/// completion so the DAG never rolls terminal with the appended work
/// still pending.
pub append_nodes: Vec<NodeKind>,
/// Whole per-agent *subgraphs* to append into *this same* DAG at /// Whole per-agent *subgraphs* to append into *this same* DAG at
/// runtime — the multi-node generalisation of `append_nodes`. Each /// runtime — the single in-DAG-growth channel. Each inner
/// inner `Vec<NodeSpec>` is one independent subgraph whose `deps` are /// `Vec<NodeSpec>` is one independent subgraph whose `deps` are local
/// local (0-based within that subgraph); the scheduler appends each via /// (0-based within that subgraph); the scheduler appends each via
/// [`JobQueue::append_subgraph`], which rebases the deps onto the DAG's /// [`JobQueue::append_subgraph`], which rebases the deps onto the DAG's
/// node-id space and roots the subgraph on the emitting node. Both /// node-id space and roots the subgraph on the emitting node. Used both
/// `MetaLock` flavours use this to grow one rebuild subgraph per agent /// for the multi-node case (`MetaLock` growing one rebuild subgraph per
/// into their own DAG (the startup sweep's stale agents; the meta-update /// agent — the startup sweep's stale agents, the meta-update cascade's
/// cascade's affected agents) instead of fanning out child DAGs. Same /// affected agents) and the single-node case (a `Reconcile` planner
/// before-completion ordering as `append_nodes`. /// emitting its mechanical `Start` / `Stop` as a one-node subgraph). The
/// scheduler applies these *before* the emitting node's completion so the
/// DAG never rolls terminal with the appended work still pending — keeping
/// the lease-window transient held across the sub-step.
pub append_subgraph: Vec<Vec<NodeSpec>>, pub append_subgraph: Vec<Vec<NodeSpec>>,
} }
@ -293,10 +288,7 @@ async fn run_meta_lock(
.iter() .iter()
.map(|agent| super::templates::rebuild_nodes(agent, true, 0)) .map(|agent| super::templates::rebuild_nodes(agent, true, 0))
.collect(); .collect();
return Ok(NodeOutput { return Ok(NodeOutput { append_subgraph });
append_subgraph,
..Default::default()
});
} }
let _progress = coord.meta_update_guard(); let _progress = coord.meta_update_guard();
ctx.step("nix flake update"); ctx.step("nix flake update");
@ -320,35 +312,34 @@ async fn run_meta_lock(
.iter() .iter()
.map(|agent| super::templates::rebuild_nodes(agent, false, 0)) .map(|agent| super::templates::rebuild_nodes(agent, false, 0))
.collect(); .collect();
Ok(NodeOutput { Ok(NodeOutput { append_subgraph })
append_subgraph,
..Default::default()
})
} }
/// Idempotent power-converge *planner*: compare `wanted` (durable /// Idempotent power-converge *planner*: compare `wanted` (durable
/// intent) against observed state and, when they diverge, fan the /// intent) against observed state and, when they diverge, fan the
/// mechanical `Start` / `Stop` out as a first-class node appended to /// mechanical `Start` / `Stop` out as a first-class node appended to
/// *this* DAG (`NodeOutput::append_nodes`). Does no container work /// *this* DAG (a single-node `NodeOutput::append_subgraph` rooted on
/// itself — the sub-step becomes visible in the DAG and the /// this node). Does no container work itself — the sub-step becomes
/// lease-window transient (or the sub-step's own node-local guard) /// visible in the DAG and the lease-window transient (or the sub-step's
/// rides across it. /// own node-local guard) rides across it.
async fn run_reconcile(coord: &Arc<Coordinator>, claim: &Claim) -> Result<NodeOutput> { async fn run_reconcile(coord: &Arc<Coordinator>, claim: &Claim) -> Result<NodeOutput> {
let name = &claim.agent; let name = &claim.agent;
let running = crate::lifecycle::is_running(name).await; let running = crate::lifecycle::is_running(name).await;
let wanted = coord.power.get_or_seed(name, running)?; let wanted = coord.power.get_or_seed(name, running)?;
let append_nodes = match reconcile_action(wanted, running) { // One node targeting this agent, rooted on this reconcile node. The old
ReconcileAction::Start => vec![NodeKind::Start], // `append_node` inherited the emitter's agent implicitly; `append_subgraph`
ReconcileAction::Stop => vec![NodeKind::Stop], // carries it on the `NodeSpec`, so stamp `claim.agent` explicitly (same
// effect, one in-DAG-growth channel instead of two).
let sub = |kind| vec![vec![super::templates::node(name, kind, Vec::new())]];
let append_subgraph = match reconcile_action(wanted, running) {
ReconcileAction::Start => sub(NodeKind::Start),
ReconcileAction::Stop => sub(NodeKind::Stop),
ReconcileAction::Noop => { ReconcileAction::Noop => {
tracing::debug!(%name, wanted = wanted.as_str(), running, "reconcile: noop"); tracing::debug!(%name, wanted = wanted.as_str(), running, "reconcile: noop");
Vec::new() Vec::new()
} }
}; };
Ok(NodeOutput { Ok(NodeOutput { append_subgraph })
append_nodes,
..Default::default()
})
} }
/// Mechanical container start — the sub-step a `Reconcile` planner fans /// Mechanical container start — the sub-step a `Reconcile` planner fans

View file

@ -154,57 +154,18 @@ impl JobQueue {
Ok(id) Ok(id)
} }
/// Append a node into a *live* (non-terminal) DAG at runtime, /// Append a whole *subgraph* into a live (non-terminal) DAG at runtime —
/// depending `AfterOk` on `dep_on` (the node that emitted it). Lets a /// the single in-DAG-growth primitive. Each [`NodeSpec`] carries its own
/// planner node — e.g. [`NodeKind::Reconcile`] — fan a mechanical /// `agent` and subgraph-relative `deps` (indices into `nodes`); this
/// sub-step ([`NodeKind::Start`] / [`NodeKind::Stop`]) out as a /// rebases those onto the DAG's node-id space (`id == index`) and attaches
/// first-class node in the *same* DAG. /// every subgraph *root* — a node with no internal deps — to `dep_on` with
/// /// an `AfterOk` edge. Used both for multi-node growth (the `MetaLock`
/// Must be called *before* the emitting node's [`Self::complete_node`] /// growing per-agent rebuild subgraphs into the same boot / meta-update
/// so the DAG doesn't roll terminal with the new node still pending — /// DAG instead of fanning out child DAGs) and the single-node case (a
/// that keeps the lease-window transient held across the sub-step and /// `Reconcile` planner's `Start` / `Stop` as a one-node subgraph). Must be
/// lets the appended node's `AfterOk` dep resolve as soon as the /// called *before* the emitting node's [`Self::complete_node`] so the DAG
/// emitter settles `Done`. No-op (returns `None`) if the DAG is gone. /// can't roll terminal with the appended work still pending. Returns the
pub fn append_node(&self, dag_id: u64, kind: NodeKind, dep_on: NodeId) -> Option<NodeId> { /// new node ids; empty if the DAG is gone or `nodes` is empty.
let mut inner = self.inner.lock().expect("job_queue mutex poisoned");
let dag = inner.dags.iter_mut().find(|d| d.id == dag_id)?;
// The appended sub-step targets the same agent as the node that
// emitted it (a `Reconcile` fanning out its `Start`/`Stop` acts on
// the same container), so inherit `dep_on`'s agent.
let agent = dag.node(dep_on)?.agent.clone();
let new_id: NodeId = u32::try_from(dag.nodes.len()).unwrap_or(u32::MAX);
dag.nodes.push(Node {
id: new_id,
agent,
kind,
deps: vec![model::Dep {
on: dep_on,
when: DepWhen::AfterOk,
}],
state: State::Queued,
step: None,
build_log_id: None,
started_at: None,
finished_at: None,
error: None,
});
drop(inner);
self.notify.notify_one();
Some(new_id)
}
/// Append a whole *subgraph* into a live (non-terminal) DAG at runtime
/// — the multi-node, multi-agent generalisation of [`Self::append_node`].
/// Each [`NodeSpec`] carries its own `agent` and subgraph-relative `deps`
/// (indices into `nodes`); this rebases those onto the DAG's node-id
/// space (`id == index`, an invariant `append_node` also maintains) and
/// attaches every subgraph *root* — a node with no internal deps — to
/// `dep_on` with an `AfterOk` edge. Used by the startup sweep's
/// `MetaLock` to grow per-agent rebuild subgraphs into the same boot DAG
/// instead of fanning out child DAGs. Same call-*before*-`complete_node`
/// contract as `append_node` (so the DAG can't roll terminal with the
/// appended work still pending). Returns the new node ids; empty if the
/// DAG is gone or `nodes` is empty.
pub fn append_subgraph( pub fn append_subgraph(
&self, &self,
dag_id: u64, dag_id: u64,

View file

@ -9,8 +9,8 @@
//! //!
//! In-DAG growth (a `MetaLock` growing rebuild subgraphs after the lock //! In-DAG growth (a `MetaLock` growing rebuild subgraphs after the lock
//! bump, a `Reconcile` fanning its `Start`/`Stop`) flows through //! bump, a `Reconcile` fanning its `Start`/`Stop`) flows through
//! `NodeOutput.append_subgraph` / `append_nodes`, applied before the //! `NodeOutput.append_subgraph`, applied before the emitting node
//! emitting node completes — see `handle_completion`. //! completes — see `handle_completion`.
use std::collections::HashMap; use std::collections::HashMap;
use std::sync::Arc; use std::sync::Arc;
@ -103,22 +103,14 @@ async fn handle_completion(
node = claim.node_id, node = claim.node_id,
"job_queue: node done" "job_queue: node done"
); );
// Append any in-DAG sub-step nodes (e.g. a `Reconcile` // Append any in-DAG subgraphs BEFORE completing this node, so
// planner's `Start` / `Stop`) BEFORE completing this node, so // completing it doesn't roll the DAG terminal while the appended
// completing it doesn't roll the DAG terminal while the // work is still pending — that keeps the lease-window transient
// appended work is still pending — that keeps the lease-window // held across it. Each subgraph is independent, rooted on this
// transient held across the sub-step. Each depends `AfterOk` // node (`AfterOk`), so it becomes ready the instant this one
// on this node, so it becomes ready the instant this one // settles `Done` just below. Covers both the multi-node case (a
// settles `Done` just below. // `MetaLock` growing per-agent rebuild subgraphs) and the
for kind in output.append_nodes { // single-node case (a `Reconcile` planner's `Start` / `Stop`).
coord
.job_queue
.append_node(claim.dag_id, kind, claim.node_id);
}
// Same before-completion ordering as `append_nodes`, but for
// whole per-agent subgraphs (the startup sweep's rebuild
// subgraphs growing into the boot DAG) — each an independent
// subgraph rooted on this node.
for subgraph in output.append_subgraph { for subgraph in output.append_subgraph {
coord coord
.job_queue .job_queue