feat(#3124): publish the agent set the swarm declares for each hive

The hive-side loop landed without anything to converge to: nothing wrote
`$KV.hive-wanted.<hive>`, so in production only the "no key" branch ran.
This is the writer.

`WantedWriter` mirrors `StatusReader` — that module reads what hives report,
this one writes what they are told, so it holds a client rather than a bucket
handle and resolves the store on first use. It shares the status reader's
connection: the controller has exactly one by design, and a second connect
would double the auth-callout traffic and give the two paths independent
reconnect state.

The value under a hive's key is the map of every agent on that hive, so a
plain `put` of a single-agent change would drop a concurrent change to a
different agent, with only one revision of history to not recover from.
Writes are read-modify-write against the entry revision, and only
`WrongLastRevision` / `AlreadyExists` count as a lost race — every other
error returns immediately rather than spinning the retry loop and then
blaming a concurrent writer that never existed.

`apply` is split out and tested because it holds the invariant: declaring
one agent preserves the rest, and a current value that will not decode is an
error rather than a fresh start. Overwriting a document nobody can read
discards every other agent's declaration.

Two routes, no swarmctl verb and no jobq node: `create_agent` needs a graph
because it is multi-step, and one CAS'd write is not.

`build_app` is extracted from `main` in the same change because `main` sat at
exactly the `too_many_lines` limit, so adding an endpoint tripped a lint
about the startup sequence. The route list is the part that grows.
This commit is contained in:
atlas 2026-09-02 01:32:12 +02:00
commit 76d5871d20
3 changed files with 424 additions and 9 deletions

View file

@ -63,7 +63,7 @@ pub struct AgentWanted {
/// than part of a declaration it only half understands. Adding a state means
/// adding a variant here and shipping it to both ends — which is the intended
/// workflow, not an obstacle to route around with a catch-all variant.
#[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
#[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum AgentState {
/// Exists on the hive and is running.
@ -72,12 +72,26 @@ pub enum AgentState {
Offline,
}
impl AgentState {
/// The wire spelling, for a reader that renders rather than decodes.
///
/// Kept beside the enum so it cannot drift from the `rename_all` above;
/// a test pins the two together.
#[must_use]
pub fn as_str(self) -> &'static str {
match self {
AgentState::Up => "up",
AgentState::Offline => "offline",
}
}
}
/// Open the wanted-state bucket for writing, creating it if nothing has yet.
///
/// **Controller-side only.** `history: 1` because a hive converges to the
/// current declaration and never asks what the previous one was — that
/// question is answered by the controller's own records, not by replaying a
/// bucket.
/// current declaration and never asks what the previous one was. The
/// controller keeps no second copy either: the value published here is the
/// record, and it reads it back from this bucket when it needs it.
#[cfg(feature = "kv")]
pub async fn open_or_create(
client: &async_nats::Client,
@ -160,4 +174,22 @@ mod tests {
fn a_missing_state_is_an_error() {
assert!(serde_json::from_str::<HiveWanted>(r#"{"agents":{"a":{}}}"#).is_err());
}
/// `as_str` and the `rename_all` are two spellings of one fact, and a
/// renderer that disagrees with the wire is worse than one that does not
/// exist. The `match` is what makes this exhaustive: a new variant fails
/// to compile here rather than quietly going untested.
#[test]
fn as_str_matches_the_serde_spelling() {
let every = [AgentState::Up, AgentState::Offline];
for state in every {
match state {
AgentState::Up | AgentState::Offline => {}
}
assert_eq!(
serde_json::to_string(&state).expect("serialises"),
format!("\"{}\"", state.as_str())
);
}
}
}