swarm-ui: add agent start/stop, backed by the wanted-state route

hyperhive#3896. The backend for start/stop (Up/Offline wanted-state
declarations) already existed and was merged (#3905's writer, the
PUT /api/hives/{hive}/agents/{agent}/state route) — nothing here was
waiting on Paused/Destroyed, which I'd mistakenly conflated with this
issue in an earlier comment (that's #3803, a different feature).

swarm-controller: merges each row's declared wanted state into
GET /api/agents/status, same shape as the config_pr merge (one read
per distinct hive, not per agent, since a declaration is a hive's
whole agent map).

swarm-ui: AgentsPage gets a "wanted" column — clicking the current-
state badge toggles it (Badge's own chip-plus-control shape, same as
its own header comment's pause/resume example), backed by the PUT
route above. A row with no declaration yet reads its implied current
state off the agent's own last-reported running flag. Stop asks for
confirmation (native window.confirm — no confirm-dialog component
exists in swarm-ui yet); start doesn't.
This commit is contained in:
iris 2026-09-02 19:41:37 +02:00
commit c9d8628794
3 changed files with 278 additions and 89 deletions

View file

@ -1,28 +1,32 @@
// <AgentsPage> — the swarm's agent roster, merged with each agent's open
// config-PR status and swarm-wide health status in the same table. Three
// separate asks (a roster listing, a per-agent config-PR indicator, and
// per-agent status/liveness) that turned out to be one page: a roster with
// no per-row detail is thin, and neither a config-PR panel nor a status
// panel has anything to render against without a roster to embed in.
// config-PR status, swarm-wide health status, and declared wanted state
// in the same table. Three separate asks (a roster listing, a per-agent
// config-PR indicator, and per-agent status/liveness) that turned out to
// be one page: a roster with no per-row detail is thin, and neither a
// config-PR panel nor a status panel has anything to render against
// without a roster to embed in.
//
// One fetch, on a refresh-interval cadence like HivesPage: `GET
// /api/agents/status` — one row per roster agent (identity store is the
// roster, so a never-reported agent still gets a row), each already
// carrying its freshness/snapshot *and* its open config PR if any. That
// merge used to be three separate fetches (`/api/agents`, `/api/config-prs`,
// carrying its freshness/snapshot/config PR/wanted state. That merge
// used to be three separate fetches (`/api/agents`, `/api/config-prs`,
// `/api/agents/status`) joined client-side by name; per operator review
// feedback ("the view should be filled by a single backend call") the
// join moved server-side instead — see `swarm-controller/src/main.rs`'s
// `get_agents_status` handler for where `config_pr` gets merged in and why
// that's the handler's job rather than `agent_status::AgentStatusReader`'s.
// No more per-page joining left to do here: the wire row *is* the table
// row.
// `get_agents_status` handler for where `config_pr`/`wanted` get merged
// in and why that's the handler's job rather than
// `agent_status::AgentStatusReader`'s. No more per-page joining left to
// do here: the wire row *is* the table row.
//
// Owns the "+ agent" trigger too — the roster this populates is the
// natural home for the action that populates it; a separate top-level
// nav entry would be one click of indirection for no benefit. The form
// itself (`CreateAgentForm`) mounts inside a `Dialog` here rather than
// its own route.
//
// The "wanted" column is the start/stop control — see its own `render`
// and `toggleWanted` below for how.
import { useState } from "preact/hooks";
import { ApiErrorPanel } from "@hive/shared/api-error-panel.js";
import { readApiError, type ProblemDetails } from "@hive/shared/api-error.js";
@ -52,6 +56,12 @@ interface AgentStatusSnapshot {
running: boolean;
}
// `"up"` / `"offline"`, wire-spelled by `swarm_queue_client::wanted::AgentState::as_str`
// — kept as a bare `string | null` rather than a union, same reason
// `config_pr`'s shape isn't re-derived here: this page renders whatever
// the wire sends, it doesn't validate the enum client-side.
type Wanted = string | null;
// Mirrors `agent_status::AgentStatusRow` field-for-field — this *is* the
// table row now, not a shape assembled from it, so there's no separate
// join-result type to keep in sync with the wire contract by hand.
@ -65,6 +75,7 @@ interface AgentRow {
// rather than rendering a once-computed-at-fetch value.
snapshot: AgentStatusSnapshot | null;
config_pr: ConfigPrStatus | null;
wanted: Wanted;
}
// Same tone/label pairing as HivesPage — one freshness enum shared by both
@ -76,61 +87,6 @@ const FRESHNESS: Record<Freshness, { tone: BadgeTone; label: string }> = {
unknown: { tone: "negative", label: "unknown" },
};
const COLUMNS: TableColumn<AgentRow>[] = [
{ key: "name", header: "name", render: (a) => a.name },
{ key: "hive", header: "hive", render: (a) => a.hive ?? "—" },
{
key: "status",
header: "status",
render: (a) => {
const { tone, label } = FRESHNESS[a.freshness];
const text = a.snapshot?.status_text;
// A `running: false` snapshot always carries `status_text: null`
// (the wire contract's own rule, not something this page derives),
// so an agent that reported recently but isn't running still shows
// a bare freshness badge rather than a stale status string.
return (
<Badge
tone={tone}
value={
<>
{text ? `${text}` : ""}
{label}
{a.last_seen_unix !== null ? (
<>
{" "}
(<RelativeTime epochMs={a.last_seen_unix * 1000} />)
</>
) : null}
</>
}
/>
);
},
},
{
key: "config-pr",
header: "config PR",
render: (a) =>
a.config_pr ? (
<Badge
tone="warning"
value={
a.config_pr.html_url ? (
<a href={a.config_pr.html_url} target="_blank" rel="noreferrer">
#{a.config_pr.pr_number}
</a>
) : (
`#${a.config_pr.pr_number}`
)
}
/>
) : (
"—"
),
},
];
// Same 30s default + same reasoning as HivesPage: no inputs on this page
// for a refresh to clobber, so the out-of-the-box behaviour should just
// solve staleness rather than require an opt-in every visit.
@ -142,21 +98,199 @@ export function AgentsPage() {
const [intervalMs, setIntervalMs] =
useState<RefreshIntervalMs>(DEFAULT_INTERVAL_MS);
const [createOpen, setCreateOpen] = useState(false);
// Per-agent, not one page-wide flag: one row's declare-in-flight
// shouldn't disable every other row's button.
const [pendingAgents, setPendingAgents] = useState<ReadonlySet<string>>(
new Set(),
);
const [actionErrors, setActionErrors] = useState<
ReadonlyMap<string, ProblemDetails>
>(new Map());
async function refresh() {
const res = await fetch("/api/agents/status");
if (!res.ok) {
setError(await readApiError(res));
return;
}
setRows((await res.json()) as AgentRow[]);
// A refresh that succeeds clears a previous failure — otherwise a
// transient error would sit on screen forever after the data itself
// has recovered.
setError(null);
}
useRefreshInterval(intervalMs, () => {
(async () => {
const res = await fetch("/api/agents/status");
refresh().catch((e: unknown) => setError({ detail: String(e) }));
});
// Declares the opposite of `row`'s current state. A row with no
// declaration yet (`wanted === null`) has nothing to flip, so the
// target is read off the agent's own last-reported `running` instead
// — the button's first click always means "make the declaration match
// reality, then flip it", which is the only reading that makes sense
// without a declaration to toggle.
async function toggleWanted(row: AgentRow) {
if (!row.hive) return;
const impliedCurrent =
row.wanted ?? (row.snapshot?.running ? "up" : "offline");
const target = impliedCurrent === "up" ? "offline" : "up";
if (
target === "offline" &&
// Native confirm, not a `Dialog` — no confirm-dialog component
// exists in swarm-ui yet, and stop is the one disruptive direction
// here (start isn't gated). A swarm-level agent destroy, unlike
// this reversible declare, is the irreversible case that will
// justify a real one.
!window.confirm(
`Declare ${row.name} offline? The hive brings its container down on its next reconcile sweep.`,
)
) {
return;
}
setPendingAgents((prev) => new Set(prev).add(row.name));
setActionErrors((prev) => {
const next = new Map(prev);
next.delete(row.name);
return next;
});
try {
const res = await fetch(
`/api/hives/${encodeURIComponent(row.hive)}/agents/${encodeURIComponent(row.name)}/state`,
{
method: "PUT",
headers: { "content-type": "application/json" },
body: JSON.stringify({ state: target }),
},
);
if (!res.ok) {
setError(await readApiError(res));
const problem = await readApiError(res);
setActionErrors((prev) => new Map(prev).set(row.name, problem));
return;
}
setRows((await res.json()) as AgentRow[]);
// A refresh that succeeds clears a previous failure — otherwise a
// transient error would sit on screen forever after the data itself
// has recovered.
setError(null);
})().catch((e: unknown) => setError({ detail: String(e) }));
});
// The PUT response is the hive's whole declaration set (every
// agent's `wanted` state) — same shape `GET /api/hives/{hive}/wanted`
// returns — so patch just that field into the rows already on
// screen rather than firing a second, full `/api/agents/status`
// round-trip merely to learn one hive's `wanted` column changed.
// Everything else on the grid (freshness, snapshot, config PR)
// still catches up on the next interval-driven `refresh()`.
const declared = (await res.json()) as { agent: string; state: string }[];
setRows((prev) => {
if (!prev) return prev;
const byAgent = new Map(declared.map((d) => [d.agent, d.state]));
return prev.map((r) =>
r.hive === row.hive && byAgent.has(r.name)
? { ...r, wanted: byAgent.get(r.name) ?? r.wanted }
: r,
);
});
} catch (e: unknown) {
setActionErrors((prev) =>
new Map(prev).set(row.name, { detail: String(e) }),
);
} finally {
setPendingAgents((prev) => {
const next = new Set(prev);
next.delete(row.name);
return next;
});
}
}
const columns: TableColumn<AgentRow>[] = [
{ key: "name", header: "name", render: (a) => a.name },
{ key: "hive", header: "hive", render: (a) => a.hive ?? "—" },
{
key: "status",
header: "status",
render: (a) => {
const { tone, label } = FRESHNESS[a.freshness];
const text = a.snapshot?.status_text;
// A `running: false` snapshot always carries `status_text: null`
// (the wire contract's own rule, not something this page derives),
// so an agent that reported recently but isn't running still shows
// a bare freshness badge rather than a stale status string.
return (
<Badge
tone={tone}
value={
<>
{text ? `${text}` : ""}
{label}
{a.last_seen_unix !== null ? (
<>
{" "}
(<RelativeTime epochMs={a.last_seen_unix * 1000} />)
</>
) : null}
</>
}
/>
);
},
},
{
key: "wanted",
header: "wanted",
// The start/stop control: clicking the current-declared-state
// badge toggles it — `Badge`'s own "chip plus, optionally, the
// control" shape (see its header comment, which names pause/resume
// as the exact motivating case), not a separate status chip next
// to a separate button.
render: (a) => {
const pending = pendingAgents.has(a.name);
const impliedCurrent =
a.wanted ?? (a.snapshot?.running ? "up" : "offline");
const actionLabel = impliedCurrent === "up" ? "stop" : "start";
const tone: BadgeTone = a.wanted === "up" ? "positive" : "neutral";
const err = actionErrors.get(a.name);
return (
<>
<Badge
tone={tone}
value={pending ? "…" : (a.wanted ?? "no declaration")}
onClick={a.hive ? () => void toggleWanted(a) : undefined}
disabled={pending || !a.hive}
title={
a.hive
? `click to ${actionLabel} ${a.name}`
: "no hive on record for this agent — nothing to declare against"
}
/>
{err ? (
<Badge
tone="negative"
value="failed"
title={err.detail ?? "the declaration failed"}
/>
) : null}
</>
);
},
},
{
key: "config-pr",
header: "config PR",
render: (a) =>
a.config_pr ? (
<Badge
tone="warning"
value={
a.config_pr.html_url ? (
<a href={a.config_pr.html_url} target="_blank" rel="noreferrer">
#{a.config_pr.pr_number}
</a>
) : (
`#${a.config_pr.pr_number}`
)
}
/>
) : (
"—"
),
},
];
return (
<Panel
@ -184,7 +318,7 @@ export function AgentsPage() {
{!error && rows === null ? <p>loading</p> : null}
{rows ? (
<Table
columns={COLUMNS}
columns={columns}
rows={rows}
rowKey={(a) => a.name}
emptyMessage="no agents yet — the swarm-wide identity store has no agents registered on any hive"

View file

@ -56,6 +56,15 @@ pub struct AgentStatusRow {
/// here, is what makes this the single call swarm-ui's agent roster
/// page needs.
pub config_pr: Option<crate::forge::ConfigPrStatus>,
/// The agent's declared wanted state (`"up"`/`"offline"`), if this
/// hive has one on record. Same rule as `config_pr`: always `None`
/// out of this reader, filled in by the `GET /api/agents/status`
/// handler from `AppState::wanted` — this module has no notion of a
/// *declaration* (a swarm-level intent), only of what an agent last
/// *reported about itself*, and merging a second bucket's read in
/// here would blur that boundary for the same reason `config_pr`
/// doesn't merge forge state in directly.
pub wanted: Option<String>,
}
/// A snapshot as retained by the bucket, with the hive it was published
@ -176,6 +185,7 @@ fn render(
age_seconds: None,
snapshot: None,
config_pr: None,
wanted: None,
},
})
.collect();
@ -219,6 +229,7 @@ fn row(
age_seconds: age.map(|age| age.as_secs()),
snapshot: offered.payload.clone(),
config_pr: None,
wanted: None,
}
}

View file

@ -889,22 +889,25 @@ async fn get_hives_status(
}
}
/// What each agent last said about itself, plus its open config PR if any,
/// read at request time — the single call swarm-ui's agent roster page
/// fills its whole table from (mara, 2026-09-02: "the view should be
/// filled by a single backend call").
/// What each agent last said about itself, its open config PR if any, and
/// its declared wanted state if any, read at request time — the single
/// call swarm-ui's agent roster page fills its whole table from (mara,
/// 2026-09-02: "the view should be filled by a single backend call").
///
/// Every agent in the roster gets a row whether or not it has ever
/// reported — see the `agent_status` module for why, and for why its hive
/// comes from its own bucket key rather than a second lookup. `config_pr`
/// is merged in here rather than inside `agent_status::AgentStatusReader`:
/// that module has no forge client and is not the place to add one just
/// for this one field, whereas this handler already holds both a
/// roster-shaped status view and `AppState::config_prs` — the one spot
/// that has both without a second, avoidable dependency. Absent on every
/// row when no forge is configured here, same "absence is the answer"
/// shape `get_config_prs` uses — not a reason to fail the whole response
/// over a field that already knows how to be missing.
/// and `wanted` are both merged in here rather than inside
/// `agent_status::AgentStatusReader`: that module has no forge client or
/// wanted-state writer and is not the place to add one just for a single
/// field, whereas this handler already holds `AppState::config_prs` and
/// `AppState::wanted` alongside the roster-shaped status view — the one
/// spot that has all three without adding an avoidable dependency to a
/// reader that stays deliberately narrow. Both fields are absent on rows
/// where the underlying store isn't configured or has nothing on record,
/// same "absence is the answer" shape `get_config_prs` uses — not a
/// reason to fail the whole response over a field that already knows how
/// to be missing.
#[utoipa::path(
get,
path = "/api/agents/status",
@ -940,6 +943,47 @@ async fn get_agents_status(
row.config_pr = config_prs.remove(&row.name);
}
}
// `wanted` merges in the same way and for the same reason as
// `config_pr` above — one read per distinct hive, not one per
// agent, since a declaration is already a hive's whole agent
// map.
if let Some(writer) = state.wanted.as_ref() {
// Owned `String` keys, not `&str` borrowed from `rows` —
// `declarations` outlives the loop below that needs `rows`
// mutably, so it cannot hold a live borrow into it.
let hives: std::collections::BTreeSet<String> =
rows.iter().filter_map(|r| r.hive.clone()).collect();
let mut declarations: std::collections::HashMap<
String,
swarm_queue_client::wanted::HiveWanted,
> = std::collections::HashMap::new();
for hive in hives {
match writer.view(&hive).await {
Ok(Some(declaration)) => {
declarations.insert(hive, declaration);
}
Ok(None) => {}
Err(e) => {
// Missing wanted state for one hive shouldn't
// fail the whole roster — same "absence is
// survivable" rule `config_pr` follows above.
tracing::warn!(
hive, error = %format!("{e:#}"),
"reading the wanted declaration failed; its agents show no wanted state"
);
}
}
}
for row in &mut rows {
row.wanted = row.hive.as_deref().and_then(|hive| {
declarations
.get(hive)?
.agents
.get(&row.name)
.map(|w| w.state.as_str().to_owned())
});
}
}
Ok(Json(rows))
}
Err(e) => {