From 45710ab739d0ab89ad149d9cc12c15b9c778f61c Mon Sep 17 00:00:00 2001 From: iris Date: Mon, 3 Aug 2026 19:07:41 +0200 Subject: [PATCH] swarm.js: migrate pending-row fallback + queue-summary banner off DagView MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit hyperhive#2822/PR#3026 moved swarm.js's per-agent in-flight status off the rebuild queue. Two other reads of the same rebuild_queue field survived that PR by design (a different feature, atlas flagged it on #2985) and are the last DagView/NodeView consumers on the frontend: queuedOpsByAgent()'s pending-row fallback and the SW4RM queue-summary banner. Both now read GET /api/jobq/graph (hive-jobq-wire's generic GraphNode shape) instead, matching the pattern builds.js already established for . Along the way: DagView no longer carries state/kind fields (removed in an earlier refactor that pushed roll-up derivation client-side), so both migrated functions were silently reading undefined fields and had become permanent no-ops — the pending-badge fallback never lit and the queue-summary banner never rendered. This restores real behavior rather than porting broken logic forward. The queue-summary banner's node-count-vs-group-count question (flagged on hyperhive#3028 as needing a decision) resolves cleanly: a GraphNode group root (parent: null) is an ordinary node whose own state already IS the group's roll-up per hive-jobq-wire's contract, so counting roots by state is a direct filter, not a parent-chain walk or a client-side rollup calculation. Verified the derivation logic against constructed GraphNode fixtures (multi-step chains, settled history that must not count, Finishing roots, multi-agent single-DAG groups) before wiring it in — 13/13 checks passed. docs/web-ui/dashboard.md's Container-row + BU1LDS sections updated to match. --- docs/web-ui/dashboard.md | 60 +++++++----- frontend/packages/dashboard/src/swarm.js | 118 ++++++++++++++--------- frontend/packages/dashboard/src/tabs.js | 13 ++- 3 files changed, 121 insertions(+), 70 deletions(-) diff --git a/docs/web-ui/dashboard.md b/docs/web-ui/dashboard.md index b9a2c4a5..7ab49205 100644 --- a/docs/web-ui/dashboard.md +++ b/docs/web-ui/dashboard.md @@ -209,8 +209,10 @@ Three sub-tabs: **R3BU1LD QU3U3** (default), **M3T4 1NPUTS**, **BUILD L0GS**. Its own esbuild bundle (`builds.js`); cold-loads `/api/state` and subscribes to `/api/dashboard/stream` for `rebuild_queue_changed`, `meta_inputs_changed`, `meta_update_running`. -The dashboard tab keeps the rebuild-queue *state* for the SW4RM -card badges without rendering these panels. +The SW4RM tab (`/dashboard.html`) independently self-fetches the same +`GET /api/jobq/graph` endpoint for its own card badges + queue-summary +banner (`refreshJobqGraph`, `swarm.js`) — same data source, no panel +rendering, no shared fetch between the two pages. **R3BU1LD QU3U3** — pending, in-flight, and recently-settled container operations: rebuilds, meta-update cascades, and first-spawns. One @@ -222,9 +224,11 @@ listens for its `hive-jobq-graph-update` event to drive the two things below it that the generic view doesn't show. The component owns fetching, cold and live: `GET /api/jobq/graph` on mount, and `.refresh()` on every `rebuild_queue_changed` SSE tick (that event -still carries its own `Vec` payload on the wire — the -SW4RM tab still consumes it for the badges, untouched — this page just -ignores the payload and treats the tick as a refetch trigger). +still carries its own `Vec` payload — `DagView`-shaped, +also read by `hivectl`'s own wait/progress loop, a separate migration +— on the wire, but neither dashboard page reads it anymore; both +treat the tick as a pure refetch trigger against the generic +endpoint). Each row is one root graph node (`parent: null`); a multi-step op's per-agent subgraphs and sub-steps render as nodes within that one @@ -912,17 +916,21 @@ fetch entirely. their own label directly via `TransientSet`/`TransientCleared` events carrying no backing node at all. - (2) The **rebuild-queue fallback** (`rebuildQueueState`) only - fires when an agent has **zero** transients — since (1) now covers - every `Running` node unconditionally, a `Running` rebuild-queue - entry can never usefully reach this fallback by the time it's - consulted; the fallback exists purely for the **`Pending` - (queued, not yet started)** case, which `running_transients()`'s - Running-only test cannot represent. `queuedOpsByAgent()` (swarm.js) - reflects this: it only ever looks at `Pending`-state queue nodes, - and (mara, on review) the badge shows the queue entry's raw `kind` - string as-is — no English-phrase lookup translating it first, same - opaque-string treatment a transient's own label already gets. + (2) The **job-queue fallback** (`jobqNodesState`, the flat + `GraphNode[]` from `GET /api/jobq/graph` — `swarm.js`'s + `refreshJobqGraph`, same generic endpoint the BU1LDS page's + `` self-fetches, see above) only fires when an + agent has **zero** transients — since (1) now covers every + `Running` node unconditionally, a `Running` job-queue node can + never usefully reach this fallback by the time it's consulted; the + fallback exists purely for the **`Pending` (queued, not yet + started)** case, which `running_transients()`'s Running-only test + cannot represent. `queuedOpsByAgent()` (swarm.js) reflects this: it + only ever looks at `Pending`-state nodes carrying a non-empty + `payload.data.agent`, and (mara, on review) the badge shows the + node's raw `payload.label` string as-is — no English-phrase lookup + translating it first, same opaque-string treatment a transient's + own label already gets. `opRunning` (driving the `pending-running` row class + spinner) is simply "does this agent have at least one transient" — a queued-only entry (no transient yet) leaves it false. @@ -966,13 +974,19 @@ per-agent actions and navigation links. Contents: agent is stale. Banner pulses on each broker SSE event (`pulseBanner` with a 4s grace timer). -**Build-queue summary banner** — when the rebuild queue has any -`queued` / `running` entries, a compact amber banner sits above the -container list: `◐ build queue — N running · M queued — view queue →` -(the link goes to the BU1LDS page's R3BU1LD QU3U3). It replaces the -old per-transient spinner list; the actual running node for each -agent is already shown on its card (transient + in-flight-queue -badges), so the top of the tab only needs the at-a-glance summary. +**Build-queue summary banner** — when `jobqNodesState` has any +`Pending` / `Running`(-or-`Finishing`) **group roots** (`parent: +null` — each an independent queued/running operation, not a raw +node count: a multi-step op's not-yet-started sub-steps don't inflate +the number), a compact amber banner sits above the container list: +`◐ build queue — N running · M queued — view queue →` (the link goes +to the BU1LDS page's R3BU1LD QU3U3). A root's own `state` already +carries its subtree's roll-up (`hive-jobq-wire`'s contract), so this +is a direct filter over `jobqNodesState`, not a client-side rollup +derivation. It replaces the old per-transient spinner list; the +actual running node for each agent is already shown on its card +(transient + in-flight-queue badges), so the top of the tab only +needs the at-a-glance summary. ### Themed dialogs diff --git a/frontend/packages/dashboard/src/swarm.js b/frontend/packages/dashboard/src/swarm.js index beff26ab..72a45321 100644 --- a/frontend/packages/dashboard/src/swarm.js +++ b/frontend/packages/dashboard/src/swarm.js @@ -27,7 +27,18 @@ const CTX_CAUTION_TOKENS = 100_000; // fallback yellow threshold (~= 50% of 200k // ─── module-level state ───────────────────────────────────────────────────── -let rebuildQueueState = []; +// Flat GraphNode[] from GET /api/jobq/graph (hive-jobq-wire's generic +// shape — see hive-jobq-wire/src/lib.rs) — the last DagView/NodeView +// (job_queue-specific) read left on this page, migrated off in favour of +// the generic endpoint builds.js already established the pattern for +// (see mountJobqGraph there). "Nothing is hidden" on this endpoint: every +// retained job group's nodes ride the wire, `Done` included, and each +// group's root is an ordinary node (`parent: null`) whose own `state` IS +// the group's roll-up already — unlike the old DagView, no client-side +// rollup derivation is needed, just an explicit filter for what's still +// active wherever that's wanted (see queuedOpsByAgent + the queue-summary +// banner in renderContainers). +let jobqNodesState = []; // Keyed container row cache. Maps agent name -> { el:
  • , fingerprint }. // Allows renderContainers to skip rebuilding rows whose displayed state @@ -52,24 +63,35 @@ const transientsState = new Map(); // tab-gated visibility). const selectionState = new Set(); -// ─── rebuild queue ────────────────────────────────────────────────────────── +// ─── job queue graph ────────────────────────────────────────────────────── -export function syncRebuildQueueFromSnapshot(s) { - rebuildQueueState = (s.rebuild_queue || []).slice(); -} -export function applyRebuildQueueChanged(ev) { - rebuildQueueState = (ev.queue || []).slice(); - // Re-render the SW4RM tab so newly-queued ops light up the right - // card with a " queued…" badge, and entries that drop out of - // Pending fall back to whatever transientsState (or nothing) says - // instead. Running work is *not* driven by this event — that's - // transient_set/transient_cleared's job, since every running node - // naming an agent already lights a pill by the time it gets here. - // See docs/web-ui.md::Container row for the badge taxonomy. +// Fetches the graph fresh and re-renders. Called on cold load (see +// tabs.js's refreshState) and whenever `rebuild_queue_changed` fires +// (applyRebuildQueueChanged below) — that event is now read purely as a +// refetch trigger, the same treatment builds.js already gives it for +// .refresh(). Best-effort: a failed fetch just leaves +// the previous snapshot in place rather than wiping badges on a network +// blip (same tolerance the per-row dashboard-state fetch below has). +export async function refreshJobqGraph() { + let nodes; + try { + const r = await fetch('/api/jobq/graph'); + if (!r.ok) return; + nodes = await r.json(); + } catch { + return; + } + jobqNodesState = nodes; renderContainersFromState(); } -// Map from agent name -> the queued op's raw `kind` string, backing -// the SW4RM card's fallback pending badge. Pending only — every +export function applyRebuildQueueChanged() { + // No `ev.queue` payload read anymore — that event still carries its + // own DagView-shaped queue snapshot on the wire, but this page no + // longer reads it. Purely a "something changed, go refetch" signal. + refreshJobqGraph(); +} +// Map from agent name -> the queued node's raw `payload.label` string, +// backing the SW4RM card's fallback pending badge. Pending only — every // *running* node naming an agent already lights a transient pill (any // running node, not just a curated "worth it" subset — see // docs/web-ui.md::Container row), so a Running entry here would @@ -78,31 +100,31 @@ export function applyRebuildQueueChanged(ev) { // transients can't represent, since `running_transients()` on the // backend is a Running-only test. // -// The returned `kind` is displayed as-is (mara, on review: "drop +// The returned label is displayed as-is (mara, on review: "drop // queuedLabelFor - just show what the backend sends") — same opaque- // string treatment `pending`'s transient half already gets, no // English-phrase lookup table translating it first. // -// Agent is per-node, not per-DAG (a DAG can span agents — e.g. the +// Agent is per-node, not per-group (a group can span agents — e.g. the // startup sweep's MetaLock cascade, or a hive-wide restart), so this -// derives each agent's queued state from its own node(s) within the -// entry rather than the DAG's overall `state`/`kind`. +// derives each agent's queued state from whichever of its own nodes is +// pending, not from the group root. No "skip agentless/spawn nodes" +// filter is needed here (an earlier version of this function keyed on +// a DAG-level `kind` field that no longer exists on the wire — dead +// code since that refactor): a node with no `data.agent` is silently +// skipped below, and an agent with no existing container row never +// gets its map entry read regardless (see containersState in state.js +// — it's populated straight from the real container roster, not a +// pre-spawn placeholder). function queuedOpsByAgent() { const out = new Map(); - for (const e of rebuildQueueState) { - if (e.state !== 'Pending') continue; - // spawn ops target an agent that doesn't exist yet as a - // container — the transient store already drives the - // pending row for that case. Skip here to avoid double- - // surfacing if the spawn op happens to land in the queue - // while the row exists transiently. - if (e.kind === 'spawn') continue; - for (const n of e.nodes || []) { - if (!n.agent || n.state !== 'Pending') continue; - // First entry found wins — with only one state to consider - // (Pending), there's no priority to resolve between DAGs. - if (!out.has(n.agent)) out.set(n.agent, e.kind); - } + for (const n of jobqNodesState) { + if (n.state !== 'Pending') continue; + const agent = n.payload.data && n.payload.data.agent; + if (!agent) continue; + // First node found wins — with only one state to consider + // (Pending), there's no priority to resolve between groups. + if (!out.has(agent)) out.set(agent, n.payload.label); } return out; } @@ -622,17 +644,27 @@ export function renderContainers(s) { )); } - // Queue-summary banner: when the rebuild queue has active work, - // show one compact at-a-glance line + a link to the full queue on the - // C0R3 page. Replaces the old per-transient spinner list — the actual + // Queue-summary banner: when the job queue has active work, show one + // compact at-a-glance line + a link to the full queue on the BU1LDS + // page. Replaces the old per-transient spinner list — the actual // running step is already visible per-agent on each card (transient + // in-flight-queue badges), so the top of the tab only needs the summary. - const activeQueue = rebuildQueueState.filter( - (e) => e.state === 'Pending' || e.state === 'Running', - ); - if (activeQueue.length) { - const running = activeQueue.filter((e) => e.state === 'Running').length; - const queued = activeQueue.length - running; + // + // Counts distinct *groups* (root nodes, `parent == null`), not raw + // nodes — a single group can have several sequential not-yet-started + // steps, and "3 running · 2 queued" means 3/2 whole operations, not + // 3/2 individual steps. A root's own `state` already IS its subtree's + // roll-up (hive-jobq-wire's contract — no client-side derivation + // needed, unlike the old DagView), so this is a direct filter, not a + // parent-chain walk. `Finishing` counts as running (own work done, + // children still going, still in flight). Settled roots (Done/Failed/ + // Cancelled/Skipped) ride the wire too now ("nothing is hidden" — + // unlike the old DagView snapshot, which dropped them) so both counts + // filter explicitly rather than assuming absence. + const roots = jobqNodesState.filter((n) => n.parent == null); + const running = roots.filter((n) => n.state === 'Running' || n.state === 'Finishing').length; + const queued = roots.filter((n) => n.state === 'Pending').length; + if (running || queued) { const parts = []; if (running) parts.push(`${running} running`); if (queued) parts.push(`${queued} queued`); diff --git a/frontend/packages/dashboard/src/tabs.js b/frontend/packages/dashboard/src/tabs.js index ab079b78..4ac4837d 100644 --- a/frontend/packages/dashboard/src/tabs.js +++ b/frontend/packages/dashboard/src/tabs.js @@ -42,7 +42,7 @@ import { renderQuestions, activeQuestionCount, } from './call.js'; import { - syncRebuildQueueFromSnapshot, syncTransientsFromSnapshot, + refreshJobqGraph, syncTransientsFromSnapshot, applyRebuildQueueChanged, applyContainerStateChanged, applyContainerRemoved, applyTransientSet, applyTransientCleared, renderContainers, renderContainersFromState, @@ -234,9 +234,14 @@ window.marked = marked; // `transientsState` + `containersState`, not from `s.*`). syncTransientsFromSnapshot(s); syncContainersFromSnapshot(s); - // Rebuild-queue state feeds the SW4RM agent-card badges - // (inFlightOpsByAgent); its own panel now lives on /core.html. - syncRebuildQueueFromSnapshot(s); + // Job-queue graph feeds the SW4RM agent-card badges + // (queuedOpsByAgent) + the queue-summary banner; its own detailed + // view lives on /builds.html. Self-fetches GET /api/jobq/graph — + // not read off `s` (this page's snapshot carries no jobq field) — + // fire-and-forget: renderContainers below runs off whatever + // jobqNodesState already holds, and refreshJobqGraph's own + // re-render catches up once the fetch resolves. + refreshJobqGraph(); renderContainers(s); // Sync the derived approvals + questions stores from the // snapshot, then render. Live `*_added` / `*_resolved` events