summaryrefslogtreecommitdiffhomepage
path: root/src
diff options
context:
space:
mode:
authorAdam Malczewski <[email protected]>2026-06-28 13:18:25 +0900
committerAdam Malczewski <[email protected]>2026-06-28 14:41:47 +0900
commit0f1b04bd7976ea416f11d83e4b7b78cf537bfdf3 (patch)
treeb198da2719987ffd58e9c2633fc36684cf2ad316 /src
parenta59200e786f7d97d7ba5b9cd2bee9ffef531dac2 (diff)
downloaddispatch-web-0f1b04bd7976ea416f11d83e4b7b78cf537bfdf3.tar.gz
dispatch-web-0f1b04bd7976ea416f11d83e4b7b78cf537bfdf3.zip
feat(concurrency): configurable cooldown + adaptive-headroom auto-reduce banner
Consumes the backend's concurrency-fixes (commit 2d27666) — additive to [email protected], NO version bump. ConcurrencyStatusEntry gains cooldownMs / autoReduced / autoReducedFrom? / notice?; new ConcurrencyCooldownResponse / SetConcurrencyCooldownRequest + GET/PUT /concurrency/cooldown/:providerId. FE: - Pure core (logic/view-model.ts): DEFAULT_COOLDOWN_MS; parseCooldownInput (non-negative int — 0 is valid, unlike the limit); normalizeCooldown; cooldownLabel ("350ms"/"1.2s"/"0ms (off)"); viewConcurrencyStatus extended (cooldown + autoReduce fields; auto-reduce → warning badge, not busy); viewAutoReduce/autoReduceNotices (banner view — prefers backend notice, synthesizes a fallback); summarizeStatus "N auto-reduced" fragment; normalizeConcurrencyStatus coerces new fields (immutable readonly build; autoReducedFrom/notice only when autoReduced===true); normalizeConcurrencyCooldown. - Types (logic/types.ts): re-export the 2 new contract types + ConcurrencyCooldownResult + Get/SaveConcurrencyCooldown ports. - UI: ConcurrencyCooldownRow.svelte (inline-edit cooldown + Save → PUT, seeded via the ChatLimitField pattern); AutoReduceBanner.svelte (dismissible banner — backend notice + "Was N, now M." + "Restore to N"); ConcurrencyView renders the cooldown per status card + the banner section. The banner persists while autoReduced===true, is dismissible, and clears automatically once a poll shows autoReduced===false after a restore PUT. - Store (store.svelte.ts): getConcurrencyCooldown + setConcurrencyCooldown (surface 400/404/503 as ok:false; normalize at the seam) + interface decls. - App.svelte: saveConcurrencyCooldown adapter → ConcurrencyView. - Tests: +47 (view-model cooldown/auto-reduce/normalizers; component banner render, cooldown PUT, restore clears banner, dismiss; store cooldown GET/PUT). Re-synced the file: dep (bun install) + re-mirrored .dispatch/transport-contract. reference.md. typecheck 0/0, 1048 tests green (run twice), biome clean, build OK. Worktree env: an untracked dispatch-backend → backend symlink was created in the worktree parent so the canonical file:../dispatch-backend/... paths resolve (NOT committed — per the §2d/§2j worktree convention).
Diffstat (limited to 'src')
-rw-r--r--src/app/App.svelte4
-rw-r--r--src/app/store.svelte.ts80
-rw-r--r--src/app/store.test.ts112
-rw-r--r--src/features/concurrency/index.ts23
-rw-r--r--src/features/concurrency/logic/types.ts37
-rw-r--r--src/features/concurrency/logic/view-model.test.ts232
-rw-r--r--src/features/concurrency/logic/view-model.ts211
-rw-r--r--src/features/concurrency/ui/AutoReduceBanner.svelte81
-rw-r--r--src/features/concurrency/ui/ConcurrencyCooldownRow.svelte93
-rw-r--r--src/features/concurrency/ui/ConcurrencyView.svelte107
-rw-r--r--src/features/concurrency/ui/ConcurrencyView.test.ts339
11 files changed, 1236 insertions, 83 deletions
diff --git a/src/app/App.svelte b/src/app/App.svelte
index 59f949c..3e7d05a 100644
--- a/src/app/App.svelte
+++ b/src/app/App.svelte
@@ -76,6 +76,7 @@
type DeleteConcurrencyLimit,
type LoadConcurrencyLimits,
type LoadConcurrencyStatus,
+ type SaveConcurrencyCooldown,
type SaveConcurrencyLimit,
} from "../features/concurrency";
import type { ChatStore } from "../features/chat";
@@ -491,6 +492,8 @@
const deleteConcurrencyLimit: DeleteConcurrencyLimit = (providerId) =>
store.deleteConcurrencyLimit(providerId);
const loadConcurrencyStatus: LoadConcurrencyStatus = () => store.concurrencyStatus();
+ const saveConcurrencyCooldown: SaveConcurrencyCooldown = (providerId, cooldownMs) =>
+ store.setConcurrencyCooldown(providerId, cooldownMs);
</script>
<main class="relative flex h-screen overflow-hidden">
@@ -841,6 +844,7 @@
saveLimit={saveConcurrencyLimit}
deleteLimit={deleteConcurrencyLimit}
loadStatus={loadConcurrencyStatus}
+ saveCooldown={saveConcurrencyCooldown}
/>
{/if}
{/snippet}
diff --git a/src/app/store.svelte.ts b/src/app/store.svelte.ts
index ead27a6..d725dc0 100644
--- a/src/app/store.svelte.ts
+++ b/src/app/store.svelte.ts
@@ -55,12 +55,14 @@ import {
import type { ChatStore, HistorySync, MetricsSync } from "../features/chat";
import { createChatStore } from "../features/chat";
import type {
+ ConcurrencyCooldownResult,
ConcurrencyDeleteResult,
ConcurrencyLimitResult,
ConcurrencyLimitsResult,
ConcurrencyStatusResult,
} from "../features/concurrency";
import {
+ normalizeConcurrencyCooldown,
normalizeConcurrencyLimit,
normalizeConcurrencyLimits,
normalizeConcurrencyStatus,
@@ -442,11 +444,29 @@ export interface AppStore {
/**
* Fetch live concurrency status for every provider with a configured limit
* (`GET /concurrency/status`): in-flight slots held, agents queued, and a paused
- * state with a `pausedUntil` epoch-ms when a 429 backoff is in effect. Returns
+ * state with a `pausedUntil` epoch-ms when a 429 backoff is in effect. Each
+ * entry also carries the per-slot release `cooldownMs` + an `autoReduced` flag
+ * (true when a 429 auto-reduced the limit by 1; the FE renders a banner). Returns
* an empty list when the extension isn't loaded (`{ providers: [] }`).
*/
concurrencyStatus(): Promise<ConcurrencyStatusResult>;
/**
+ * Fetch the per-slot release cooldown (ms) for one provider
+ * (`GET /concurrency/cooldown/:providerId`). `404` (no concurrency config at
+ * all) and `503` (extension not loaded) both surface as `ok: false`.
+ */
+ getConcurrencyCooldown(providerId: string): Promise<ConcurrencyCooldownResult>;
+ /**
+ * Set the per-slot release cooldown (ms) for one provider
+ * (`PUT /concurrency/cooldown/:providerId`, body `{ cooldownMs }`). `cooldownMs`
+ * must be a non-negative integer (0 = no cooldown / instant re-admission); an
+ * invalid body is `400`. Persists + applies to subsequently recycled slots.
+ */
+ setConcurrencyCooldown(
+ providerId: string,
+ cooldownMs: number,
+ ): Promise<ConcurrencyCooldownResult>;
+ /**
* A critical error that blocks normal operation (e.g. the cross-device tab
* restore fetch failed). When non-null, a full-screen modal is shown with the
* error details. Cleared by `clearFatalError` (the modal's dismiss button).
@@ -1958,6 +1978,64 @@ export function createAppStore(opts?: CreateAppStoreOptions): AppStore {
}
},
+ async getConcurrencyCooldown(providerId: string): Promise<ConcurrencyCooldownResult> {
+ try {
+ const res = await fetchImpl(
+ `${httpBase}/concurrency/cooldown/${encodeURIComponent(providerId)}`,
+ );
+ if (!res.ok) {
+ const errBody = (await res.json().catch(() => null)) as { error?: string } | null;
+ return {
+ ok: false,
+ error: errBody?.error ?? `Get concurrency cooldown failed (HTTP ${res.status})`,
+ };
+ }
+ const data = normalizeConcurrencyCooldown(await res.json());
+ if (data === null) {
+ return { ok: false, error: "Malformed concurrency cooldown response" };
+ }
+ return { ok: true, providerId: data.providerId, cooldownMs: data.cooldownMs };
+ } catch (err) {
+ return {
+ ok: false,
+ error: err instanceof Error ? err.message : "Get concurrency cooldown request failed",
+ };
+ }
+ },
+
+ async setConcurrencyCooldown(
+ providerId: string,
+ cooldownMs: number,
+ ): Promise<ConcurrencyCooldownResult> {
+ try {
+ const res = await fetchImpl(
+ `${httpBase}/concurrency/cooldown/${encodeURIComponent(providerId)}`,
+ {
+ method: "PUT",
+ headers: { "content-type": "application/json" },
+ body: JSON.stringify({ cooldownMs }),
+ },
+ );
+ if (!res.ok) {
+ const errBody = (await res.json().catch(() => null)) as { error?: string } | null;
+ return {
+ ok: false,
+ error: errBody?.error ?? `Set concurrency cooldown failed (HTTP ${res.status})`,
+ };
+ }
+ const data = normalizeConcurrencyCooldown(await res.json());
+ if (data === null) {
+ return { ok: false, error: "Malformed concurrency cooldown response" };
+ }
+ return { ok: true, providerId: data.providerId, cooldownMs: data.cooldownMs };
+ } catch (err) {
+ return {
+ ok: false,
+ error: err instanceof Error ? err.message : "Set concurrency cooldown request failed",
+ };
+ }
+ },
+
async loadSystemPrompt(): Promise<SystemPromptLoadResult> {
try {
const res = await fetchImpl(`${httpBase}/system-prompt`);
diff --git a/src/app/store.test.ts b/src/app/store.test.ts
index 2cf473c..a2f7f2f 100644
--- a/src/app/store.test.ts
+++ b/src/app/store.test.ts
@@ -1458,6 +1458,20 @@ describe("createAppStore", () => {
if (url.includes("/concurrency/limits/") && method === "DELETE") {
return new Response(JSON.stringify({ ok: true, providerId: "umans" }), { status: 200 });
}
+ if (url.includes("/concurrency/cooldown/") && method === "GET") {
+ return new Response(JSON.stringify({ providerId: "umans", cooldownMs: 350 }), {
+ status: 200,
+ });
+ }
+ if (url.includes("/concurrency/cooldown/") && method === "PUT") {
+ const seg = url.slice(
+ url.lastIndexOf("/concurrency/cooldown/") + "/concurrency/cooldown/".length,
+ );
+ const body = init?.body ? JSON.parse(init.body as string) : {};
+ return new Response(JSON.stringify({ providerId: seg, cooldownMs: body.cooldownMs ?? 0 }), {
+ status: 200,
+ });
+ }
return base(input, init);
};
}
@@ -1641,11 +1655,15 @@ describe("createAppStore", () => {
inFlight: 2,
queued: 1,
paused: false,
+ cooldownMs: 350,
+ autoReduced: false,
});
expect(result.providers[1]).toMatchObject({
providerId: "openai-compat",
paused: true,
pausedUntil: 1_719_408_000_000,
+ cooldownMs: 350,
+ autoReduced: false,
});
store.dispose();
});
@@ -1663,6 +1681,100 @@ describe("createAppStore", () => {
store.dispose();
});
+ it("getConcurrencyCooldown loads + coerces the cooldown", async () => {
+ const store = createAppStore({
+ socketFactory: () => fakeSocket(),
+ fetchImpl: concurrencyFetchImpl(),
+ localStorage: createFakeStorage(),
+ });
+ const result = await store.getConcurrencyCooldown("umans");
+ expect(result.ok).toBe(true);
+ if (!result.ok) throw new Error("unreachable");
+ expect(result.providerId).toBe("umans");
+ expect(result.cooldownMs).toBe(350);
+ store.dispose();
+ });
+
+ it("getConcurrencyCooldown surfaces a 404 (no concurrency config) as ok:false", async () => {
+ const store = createAppStore({
+ socketFactory: () => fakeSocket(),
+ fetchImpl: async (input) => {
+ const url =
+ typeof input === "string" ? input : input instanceof URL ? input.href : input.url;
+ if (url.includes("/concurrency/cooldown/")) {
+ return new Response(
+ JSON.stringify({ error: "No concurrency configuration for this provider" }),
+ {
+ status: 404,
+ },
+ );
+ }
+ return fakeFetchImpl()(input);
+ },
+ localStorage: createFakeStorage(),
+ });
+ const result = await store.getConcurrencyCooldown("ghost");
+ expect(result.ok).toBe(false);
+ if (result.ok) throw new Error("unreachable");
+ expect(result.error).toContain("No concurrency configuration");
+ store.dispose();
+ });
+
+ it("setConcurrencyCooldown PUTs { cooldownMs } + returns the echoed value", async () => {
+ const calls: { url: string; method: string; body: unknown }[] = [];
+ const store = createAppStore({
+ socketFactory: () => fakeSocket(),
+ fetchImpl: async (input, init) => {
+ const url =
+ typeof input === "string" ? input : input instanceof URL ? input.href : input.url;
+ const method = init?.method ?? "GET";
+ calls.push({ url, method, body: init?.body ? JSON.parse(init.body as string) : null });
+ if (url.includes("/concurrency/cooldown/") && method === "PUT") {
+ const body = init?.body ? JSON.parse(init.body as string) : {};
+ return new Response(
+ JSON.stringify({ providerId: "umans", cooldownMs: body.cooldownMs }),
+ { status: 200 },
+ );
+ }
+ return fakeFetchImpl()(input, init);
+ },
+ localStorage: createFakeStorage(),
+ });
+ const result = await store.setConcurrencyCooldown("umans", 500);
+ expect(result.ok).toBe(true);
+ if (!result.ok) throw new Error("unreachable");
+ expect(result.cooldownMs).toBe(500);
+ const cooldownCall = calls.find(
+ (c) => c.url.includes("/concurrency/cooldown/") && c.method === "PUT",
+ );
+ expect(cooldownCall?.url).toContain("/concurrency/cooldown/umans");
+ expect(cooldownCall?.body).toEqual({ cooldownMs: 500 });
+ store.dispose();
+ });
+
+ it("setConcurrencyCooldown surfaces a 400 (invalid body) as ok:false", async () => {
+ const store = createAppStore({
+ socketFactory: () => fakeSocket(),
+ fetchImpl: async (input, init) => {
+ const url =
+ typeof input === "string" ? input : input instanceof URL ? input.href : input.url;
+ if (url.includes("/concurrency/cooldown/") && init?.method === "PUT") {
+ return new Response(
+ JSON.stringify({ error: "Body must be { cooldownMs: <non-negative integer> }" }),
+ { status: 400 },
+ );
+ }
+ return fakeFetchImpl()(input, init);
+ },
+ localStorage: createFakeStorage(),
+ });
+ const result = await store.setConcurrencyCooldown("umans", -1);
+ expect(result.ok).toBe(false);
+ if (result.ok) throw new Error("unreachable");
+ expect(result.error).toContain("non-negative integer");
+ store.dispose();
+ });
+
// ── Conversation status: `queued` (CR-13 — waiting for a concurrency slot) ────
it("conversation.statusChanged 'queued' sets the status (tab spinner) without opening a duplicate tab", () => {
diff --git a/src/features/concurrency/index.ts b/src/features/concurrency/index.ts
index 151acb8..95eb5ed 100644
--- a/src/features/concurrency/index.ts
+++ b/src/features/concurrency/index.ts
@@ -1,7 +1,9 @@
export type {
+ // Contract shapes re-exported for a single import surface.
+ ConcurrencyCooldownResponse,
+ ConcurrencyCooldownResult,
ConcurrencyDeleteResult,
ConcurrencyLimitEntry,
- // Contract shapes re-exported for a single import surface.
ConcurrencyLimitResponse,
ConcurrencyLimitResult,
ConcurrencyLimitsResponse,
@@ -10,30 +12,47 @@ export type {
ConcurrencyStatusResponse,
ConcurrencyStatusResult,
DeleteConcurrencyLimit,
+ GetConcurrencyCooldown,
GetConcurrencyLimit,
LoadConcurrencyLimits,
LoadConcurrencyStatus,
+ RestoreOutcome,
+ SaveConcurrencyCooldown,
SaveConcurrencyLimit,
+ SetConcurrencyCooldownRequest,
SetConcurrencyLimitRequest,
} from "./logic/types";
-export type { Badge, ConcurrencyLimitView, ConcurrencyStatusView } from "./logic/view-model";
+export type {
+ AutoReduceNotice,
+ Badge,
+ ConcurrencyLimitView,
+ ConcurrencyStatusView,
+} from "./logic/view-model";
export {
+ autoReduceNotices,
+ cooldownLabel,
+ DEFAULT_COOLDOWN_MS,
formatPauseDuration,
+ normalizeConcurrencyCooldown,
normalizeConcurrencyLimit,
normalizeConcurrencyLimits,
normalizeConcurrencyStatus,
normalizeLimit,
+ parseCooldownInput,
parseLimitInput,
pauseLabel,
providerFromModel,
providerOptions,
summarizeLimits,
summarizeStatus,
+ viewAutoReduce,
viewConcurrencyLimit,
viewConcurrencyLimits,
viewConcurrencyStatus,
viewConcurrencyStatuses,
} from "./logic/view-model";
+export { default as AutoReduceBanner } from "./ui/AutoReduceBanner.svelte";
+export { default as ConcurrencyCooldownRow } from "./ui/ConcurrencyCooldownRow.svelte";
export { default as ConcurrencyLimitRow } from "./ui/ConcurrencyLimitRow.svelte";
export { default as ConcurrencyView } from "./ui/ConcurrencyView.svelte";
diff --git a/src/features/concurrency/logic/types.ts b/src/features/concurrency/logic/types.ts
index f05211f..a0c5f6b 100644
--- a/src/features/concurrency/logic/types.ts
+++ b/src/features/concurrency/logic/types.ts
@@ -1,8 +1,10 @@
import type {
+ ConcurrencyCooldownResponse,
ConcurrencyLimitResponse,
ConcurrencyLimitsResponse,
ConcurrencyStatusEntry,
ConcurrencyStatusResponse,
+ SetConcurrencyCooldownRequest,
SetConcurrencyLimitRequest,
} from "@dispatch/transport-contract";
@@ -23,14 +25,23 @@ import type {
* are imported directly (mirrors `mcp` / `computer`). The result types + injected
* ports below are FE-owned (the composition root adapts the store's HTTP calls to
* them). The endpoints are GLOBAL (not workspace- or conversation-scoped).
+ *
+ * Concurrency-fixes (additive, no version bump): each `ConcurrencyStatusEntry`
+ * now also carries `cooldownMs` (per-slot release cooldown, configurable +
+ * persisted), `autoReduced` (a 429 auto-reduced the limit by 1, one-way), and
+ * when auto-reduced, `autoReducedFrom` + a `notice` banner string. A manual
+ * `PUT /concurrency/limits/:providerId` clears `autoReduced`. Two new endpoints
+ * `GET`/`PUT /concurrency/cooldown/:providerId` view/change the cooldown.
*/
/** Re-export the contract shapes so consumers import a single surface. */
export type {
+ ConcurrencyCooldownResponse,
ConcurrencyLimitResponse,
ConcurrencyLimitsResponse,
ConcurrencyStatusEntry,
ConcurrencyStatusResponse,
+ SetConcurrencyCooldownRequest,
SetConcurrencyLimitRequest,
};
@@ -69,6 +80,25 @@ export type ConcurrencyStatusResult =
| { readonly ok: true; readonly providers: readonly ConcurrencyStatusEntry[] }
| { readonly ok: false; readonly error: string };
+/**
+ * Outcome of `GET`/`PUT /concurrency/cooldown/:providerId` — the per-slot
+ * release cooldown (ms) for one provider. `GET` returns `404` when the provider
+ * has no concurrency config at all (no limit, no cooldown); `PUT` returns `400`
+ * for a non-negative-integer body. Both return `503` when the extension isn't
+ * loaded.
+ */
+export type ConcurrencyCooldownResult =
+ | { readonly ok: true; readonly providerId: string; readonly cooldownMs: number }
+ | { readonly ok: false; readonly error: string };
+
+/**
+ * Outcome of an auto-reduce banner's "Restore to N" action (PUT the limit back
+ * to `autoReducedFrom` via `PUT /concurrency/limits/:providerId`). Carried back to
+ * the banner so a FAILED restore surfaces an inline error next to the button
+ * (instead of silently re-enabling the button / showing the error far away).
+ */
+export type RestoreOutcome = { readonly ok: true } | { readonly ok: false; readonly error: string };
+
// ── Injected ports (consumer-defines-port; the composition root adapts the
// store's HTTP calls to these shapes). ──────────────────────────────────────
@@ -80,3 +110,10 @@ export type SaveConcurrencyLimit = (
) => Promise<ConcurrencyLimitResult>;
export type DeleteConcurrencyLimit = (providerId: string) => Promise<ConcurrencyDeleteResult>;
export type LoadConcurrencyStatus = () => Promise<ConcurrencyStatusResult>;
+/** `GET /concurrency/cooldown/:providerId` — read the per-slot release cooldown. */
+export type GetConcurrencyCooldown = (providerId: string) => Promise<ConcurrencyCooldownResult>;
+/** `PUT /concurrency/cooldown/:providerId` — set the per-slot release cooldown (non-negative int). */
+export type SaveConcurrencyCooldown = (
+ providerId: string,
+ cooldownMs: number,
+) => Promise<ConcurrencyCooldownResult>;
diff --git a/src/features/concurrency/logic/view-model.test.ts b/src/features/concurrency/logic/view-model.test.ts
index 82b72b0..c284ad0 100644
--- a/src/features/concurrency/logic/view-model.test.ts
+++ b/src/features/concurrency/logic/view-model.test.ts
@@ -1,16 +1,22 @@
import type { ConcurrencyStatusEntry } from "@dispatch/transport-contract";
import { describe, expect, it } from "vitest";
import {
+ autoReduceNotices,
+ cooldownLabel,
+ DEFAULT_COOLDOWN_MS,
formatPauseDuration,
+ normalizeConcurrencyCooldown,
normalizeConcurrencyLimit,
normalizeConcurrencyLimits,
normalizeConcurrencyStatus,
+ parseCooldownInput,
parseLimitInput,
pauseLabel,
providerFromModel,
providerOptions,
summarizeLimits,
summarizeStatus,
+ viewAutoReduce,
viewConcurrencyLimit,
viewConcurrencyLimits,
viewConcurrencyStatus,
@@ -23,6 +29,8 @@ const status = (over: Partial<ConcurrencyStatusEntry> = {}): ConcurrencyStatusEn
inFlight: 2,
queued: 0,
paused: false,
+ cooldownMs: 350,
+ autoReduced: false,
...over,
});
@@ -46,6 +54,41 @@ describe("parseLimitInput", () => {
});
});
+// ── parseCooldownInput (non-negative integer — 0 is valid, unlike the limit) ──
+
+describe("parseCooldownInput", () => {
+ it("accepts zero + positive integers", () => {
+ expect(parseCooldownInput("0")).toBe(0);
+ expect(parseCooldownInput("350")).toBe(350);
+ expect(parseCooldownInput(" 100 ")).toBe(100);
+ });
+
+ it("rejects negatives, non-integers, and garbage", () => {
+ expect(parseCooldownInput("-1")).toBeNull();
+ expect(parseCooldownInput("4.5")).toBeNull();
+ expect(parseCooldownInput("")).toBeNull();
+ expect(parseCooldownInput("abc")).toBeNull();
+ expect(parseCooldownInput("100ms")).toBeNull();
+ });
+});
+
+// ── cooldownLabel ─────────────────────────────────────────────────────────────
+
+describe("cooldownLabel", () => {
+ it("0 → off label", () => {
+ expect(cooldownLabel(0)).toBe("0ms (off)");
+ });
+ it("sub-second → ms", () => {
+ expect(cooldownLabel(350)).toBe("350ms");
+ expect(cooldownLabel(999)).toBe("999ms");
+ });
+ it("≥1s → seconds (trims trailing .0)", () => {
+ expect(cooldownLabel(1000)).toBe("1s");
+ expect(cooldownLabel(1500)).toBe("1.5s");
+ expect(cooldownLabel(60_000)).toBe("60s");
+ });
+});
+
// ── providerFromModel / providerOptions ───────────────────────────────────────
describe("providerFromModel", () => {
@@ -178,6 +221,8 @@ describe("viewConcurrencyStatus", () => {
inFlight: Number.NaN,
queued: "oops" as unknown as number,
paused: false,
+ cooldownMs: Number.NaN,
+ autoReduced: false,
},
0,
);
@@ -185,6 +230,7 @@ describe("viewConcurrencyStatus", () => {
expect(v.inFlight).toBe(0);
expect(v.queued).toBe(0);
expect(v.inFlightLabel).toBe("0/1");
+ expect(v.cooldownMs).toBe(DEFAULT_COOLDOWN_MS);
});
it("viewConcurrencyStatuses maps a list preserving order", () => {
@@ -194,6 +240,81 @@ describe("viewConcurrencyStatus", () => {
);
expect(views.map((v) => v.providerId)).toEqual(["a", "b"]);
});
+
+ it("carries cooldownMs + label + autoReduced fields onto the view", () => {
+ const v = viewConcurrencyStatus(status({ cooldownMs: 1500 }), 0);
+ expect(v.cooldownMs).toBe(1500);
+ expect(v.cooldownLabel).toBe("1.5s");
+ expect(v.autoReduced).toBe(false);
+ expect(v.autoReducedFrom).toBeNull();
+ });
+
+ it("auto-reduced → warning badge (not busy) + autoReducedFrom carried", () => {
+ const v = viewConcurrencyStatus(
+ status({ limit: 3, autoReduced: true, autoReducedFrom: 4, inFlight: 0 }),
+ 0,
+ );
+ expect(v.autoReduced).toBe(true);
+ expect(v.autoReducedFrom).toBe(4);
+ expect(v.badge).toBe("warning");
+ // autoReduced alone does NOT flip busy (a reduced limit still admits agents).
+ expect(v.busy).toBe(false);
+ });
+});
+
+// ── viewAutoReduce / autoReduceNotices (the auto-reduce banner view) ───────────
+
+describe("viewAutoReduce", () => {
+ it("returns null when not auto-reduced", () => {
+ expect(viewAutoReduce(status({ autoReduced: false }))).toBeNull();
+ });
+
+ it("uses the backend notice verbatim + carries from/current limits", () => {
+ const notice = viewAutoReduce(
+ status({
+ limit: 3,
+ autoReduced: true,
+ autoReducedFrom: 4,
+ notice: "Concurrency limit auto-reduced to 3 after a 429.",
+ }),
+ );
+ expect(notice).toEqual({
+ providerId: "umans",
+ message: "Concurrency limit auto-reduced to 3 after a 429.",
+ fromLimit: 4,
+ currentLimit: 3,
+ });
+ });
+
+ it("synthesizes a fallback notice when the backend notice is absent/empty", () => {
+ expect(viewAutoReduce(status({ limit: 3, autoReduced: true, autoReducedFrom: 4 }))).toEqual({
+ providerId: "umans",
+ message: "Concurrency limit auto-reduced to 3 after a 429 — restore manually when ready.",
+ fromLimit: 4,
+ currentLimit: 3,
+ });
+ expect(
+ viewAutoReduce(status({ limit: 3, autoReduced: true, autoReducedFrom: 4, notice: "" })),
+ ).not.toBeNull();
+ });
+
+ it("falls back to currentLimit+1 when autoReducedFrom is missing/garbage", () => {
+ const notice = viewAutoReduce(status({ limit: 3, autoReduced: true }));
+ expect(notice?.fromLimit).toBe(4); // 3 + 1
+ });
+});
+
+describe("autoReduceNotices", () => {
+ it("collects one banner per auto-reduced provider (input order), empty when none", () => {
+ expect(autoReduceNotices([status({ providerId: "a" })])).toEqual([]);
+ const out = autoReduceNotices([
+ status({ providerId: "a", autoReduced: true, autoReducedFrom: 4, limit: 3 }),
+ status({ providerId: "b" }),
+ status({ providerId: "c", autoReduced: true, autoReducedFrom: 2, limit: 1 }),
+ ]);
+ expect(out.map((n) => n.providerId)).toEqual(["a", "c"]);
+ expect(out[1]?.fromLimit).toBe(2);
+ });
});
// ── viewConcurrencyLimit ───────────────────────────────────────────────────────
@@ -273,6 +394,16 @@ describe("summarizeStatus", () => {
"1 provider · 1/4 in flight",
);
});
+ it("includes an auto-reduced fragment only when non-zero", () => {
+ const s = summarizeStatus(
+ [
+ status({ providerId: "a", limit: 3, inFlight: 1, autoReduced: true, autoReducedFrom: 4 }),
+ status({ providerId: "b", limit: 4, inFlight: 1 }),
+ ],
+ 0,
+ );
+ expect(s).toBe("2 providers · 2/7 in flight · 1 auto-reduced");
+ });
});
// ── Network-seam normalizers ───────────────────────────────────────────────────
@@ -359,6 +490,8 @@ describe("normalizeConcurrencyStatus", () => {
inFlight: 2,
queued: 1,
paused: false,
+ cooldownMs: 350,
+ autoReduced: false,
});
expect(first !== undefined && !("pausedUntil" in first)).toBe(true);
expect(second).toEqual({
@@ -368,6 +501,8 @@ describe("normalizeConcurrencyStatus", () => {
queued: 3,
paused: true,
pausedUntil: now,
+ cooldownMs: 350,
+ autoReduced: false,
});
});
@@ -388,8 +523,24 @@ describe("normalizeConcurrencyStatus", () => {
],
});
expect(providers).toEqual([
- { providerId: "umans", limit: 4, inFlight: 2, queued: 1, paused: false },
- { providerId: "x", limit: 1, inFlight: 0, queued: 0, paused: false },
+ {
+ providerId: "umans",
+ limit: 4,
+ inFlight: 2,
+ queued: 1,
+ paused: false,
+ cooldownMs: 350,
+ autoReduced: false,
+ },
+ {
+ providerId: "x",
+ limit: 1,
+ inFlight: 0,
+ queued: 0,
+ paused: false,
+ cooldownMs: 350,
+ autoReduced: false,
+ },
]);
});
@@ -402,4 +553,81 @@ describe("normalizeConcurrencyStatus", () => {
});
for (const p of providers) expect("pausedUntil" in p).toBe(false);
});
+
+ it("coerces cooldownMs (default 350) + carries auto-reduce fields only when true", () => {
+ const [reduced, healthy] = normalizeConcurrencyStatus({
+ providers: [
+ {
+ providerId: "umans",
+ limit: 3,
+ inFlight: 1,
+ queued: 0,
+ paused: false,
+ cooldownMs: 500,
+ autoReduced: true,
+ autoReducedFrom: 4,
+ notice: "auto-reduced to 3 after a 429.",
+ },
+ { providerId: "openai", limit: 4, inFlight: 0, queued: 0, paused: false },
+ ],
+ });
+ expect(reduced?.cooldownMs).toBe(500);
+ expect(reduced?.autoReduced).toBe(true);
+ expect(reduced?.autoReducedFrom).toBe(4);
+ expect(reduced?.notice).toBe("auto-reduced to 3 after a 429.");
+ // Healthy entry: cooldownMs defaults to 350 when absent; auto-reduce fields
+ // are NOT present (they are only included when autoReduced===true).
+ expect(healthy?.cooldownMs).toBe(DEFAULT_COOLDOWN_MS);
+ expect(healthy?.autoReduced).toBe(false);
+ expect(healthy && "autoReducedFrom" in healthy).toBe(false);
+ expect(healthy && "notice" in healthy).toBe(false);
+ });
+
+ it("drops autoReducedFrom/notice when autoReduced is false (even if present in JSON)", () => {
+ const [p] = normalizeConcurrencyStatus({
+ providers: [
+ {
+ providerId: "x",
+ limit: 4,
+ inFlight: 0,
+ queued: 0,
+ paused: false,
+ autoReduced: false,
+ autoReducedFrom: 9,
+ notice: "stale",
+ },
+ ],
+ });
+ expect(p?.autoReduced).toBe(false);
+ expect(p && "autoReducedFrom" in p).toBe(false);
+ expect(p && "notice" in p).toBe(false);
+ });
+});
+
+// ── normalizeConcurrencyCooldown ───────────────────────────────────────────────
+
+describe("normalizeConcurrencyCooldown", () => {
+ it("coerces a well-formed body", () => {
+ expect(normalizeConcurrencyCooldown({ providerId: "umans", cooldownMs: 500 })).toEqual({
+ providerId: "umans",
+ cooldownMs: 500,
+ });
+ });
+
+ it("defaults a malformed/absent cooldownMs to 350", () => {
+ expect(normalizeConcurrencyCooldown({ providerId: "x", cooldownMs: -1 })?.cooldownMs).toBe(
+ DEFAULT_COOLDOWN_MS,
+ );
+ expect(normalizeConcurrencyCooldown({ providerId: "x" })?.cooldownMs).toBe(DEFAULT_COOLDOWN_MS);
+ expect(normalizeConcurrencyCooldown({ providerId: "x", cooldownMs: "fast" })?.cooldownMs).toBe(
+ DEFAULT_COOLDOWN_MS,
+ );
+ });
+
+ it("returns null for a missing/malformed providerId", () => {
+ expect(normalizeConcurrencyCooldown({ cooldownMs: 350 })).toBeNull();
+ expect(normalizeConcurrencyCooldown({ providerId: "", cooldownMs: 350 })).toBeNull();
+ expect(normalizeConcurrencyCooldown(null)).toBeNull();
+ expect(normalizeConcurrencyCooldown({})).toBeNull();
+ });
});
diff --git a/src/features/concurrency/logic/view-model.ts b/src/features/concurrency/logic/view-model.ts
index 7a4fe0e..e05423b 100644
--- a/src/features/concurrency/logic/view-model.ts
+++ b/src/features/concurrency/logic/view-model.ts
@@ -1,12 +1,16 @@
-import type { ConcurrencyStatusEntry } from "@dispatch/transport-contract";
+import type {
+ ConcurrencyCooldownResponse,
+ ConcurrencyStatusEntry,
+} from "@dispatch/transport-contract";
import type { ConcurrencyLimitEntry } from "./types";
/**
* Pure view-models for the concurrency feature — zero DOM, zero effects, zero
* Svelte. Maps backend `ConcurrencyLimitEntry` / `ConcurrencyStatusEntry` to
- * display shapes (badges, "2/4" in-flight labels, pause countdowns, summaries),
- * holds the limit-input parsing, and the network-seam normalizers the composition
- * root coerces the untyped JSON with.
+ * display shapes (badges, "2/4" in-flight labels, pause countdowns, cooldown
+ * labels, auto-reduce banners, summaries), holds the limit/cooldown-input
+ * parsing, and the network-seam normalizers the composition root coerces the
+ * untyped JSON with.
*/
export type Badge = "success" | "warning" | "error" | "neutral";
@@ -36,6 +40,29 @@ export interface ConcurrencyStatusView {
readonly badge: Badge;
/** True when paused or at capacity (show a spinner). */
readonly busy: boolean;
+ /** Per-slot release cooldown in ms (defensive default 350 on garbage). */
+ readonly cooldownMs: number;
+ /** "350ms" / "1.2s" / "0ms (off)" — display label for the cooldown. */
+ readonly cooldownLabel: string;
+ /** Whether the limit was auto-reduced by a 429 (one-way; user restores manually). */
+ readonly autoReduced: boolean;
+ /** The original limit before auto-reduction; null when not auto-reduced. */
+ readonly autoReducedFrom: number | null;
+}
+
+/**
+ * A view-model for the auto-reduce banner — derived from a status entry whose
+ * `autoReduced` is `true`. `message` is the backend's `notice` when present, else
+ * a synthesized fallback. `viewAutoReduce` returns this (or null) so the banner
+ * section renders without reaching into the raw entry.
+ */
+export interface AutoReduceNotice {
+ readonly providerId: string;
+ readonly message: string;
+ /** The original limit before reduction — the value "Restore to N" PUTs. */
+ readonly fromLimit: number;
+ /** The current (reduced) limit. */
+ readonly currentLimit: number;
}
// ── Limit input parsing ───────────────────────────────────────────────────────
@@ -63,6 +90,53 @@ export function normalizeLimit(value: unknown): number {
return int >= 1 ? int : 1;
}
+// ── Cooldown input parsing ────────────────────────────────────────────────────
+//
+// The per-slot release cooldown (ms) is a NON-NEGATIVE integer (0 = no cooldown,
+// instant re-admission) — unlike the limit, 0 is a VALID value. The default is
+// 350ms (the backend's server default when a limit is set but no explicit
+// cooldown was configured). It is configurable + persisted per provider via
+// `PUT /concurrency/cooldown/:providerId`.
+
+/** The server's default cooldown (ms) — used when none is explicitly set. */
+export const DEFAULT_COOLDOWN_MS = 350;
+
+/**
+ * Parse a raw cooldown input into a non-negative integer, or `null` when it is
+ * not valid. Accepts "0" → 0, "350" → 350; rejects "-1", "4.5", "", "abc".
+ * Drives the cooldown Save button's disabled state so an invalid value never
+ * reaches the backend (the backend 400s a non-negative-integer body).
+ */
+export function parseCooldownInput(value: string): number | null {
+ const trimmed = value.trim();
+ if (trimmed === "" || !/^[0-9]+$/.test(trimmed)) return null;
+ const n = Number.parseInt(trimmed, 10);
+ return Number.isFinite(n) && n >= 0 ? n : null;
+}
+
+/**
+ * Coerce an untrusted cooldown value into a non-negative integer (default
+ * {@link DEFAULT_COOLDOWN_MS}). Used when normalizing backend responses so a
+ * malformed `cooldownMs` can never be negative/non-finite.
+ */
+export function normalizeCooldown(value: unknown): number {
+ const n = typeof value === "number" && Number.isFinite(value) ? value : DEFAULT_COOLDOWN_MS;
+ const int = Math.floor(n);
+ return int >= 0 ? int : DEFAULT_COOLDOWN_MS;
+}
+
+/**
+ * Format a cooldown (ms) as a short display label:
+ * 0 → "0ms (off)" · <1000 → "350ms" · ≥1000 → "1.2s" (trailing ".0" trimmed).
+ */
+export function cooldownLabel(ms: number): string {
+ if (ms <= 0) return "0ms (off)";
+ if (ms < 1000) return `${ms}ms`;
+ const secs = ms / 1000;
+ const fixed = secs.toFixed(1);
+ return `${fixed.endsWith(".0") ? fixed.slice(0, -2) : fixed}s`;
+}
+
// ── Provider options (the Add-form dropdown) ───────────────────────────────────
//
// A concurrency `providerId` is the credential name that prefixes a model name
@@ -140,6 +214,10 @@ export function pauseLabel(
* Build a display view for a status entry. `now` is injectable for tests
* (defaults to `Date.now()`); the composition-root component passes nothing in
* production (it recomputes on each poll).
+ *
+ * `autoReduced` does NOT flip `busy` (a reduced limit still admits agents; it is
+ * a degraded-but-active state surfaced via the banner, not a spinner) — it only
+ * nudges the badge to `warning` so the row signals attention.
*/
export function viewConcurrencyStatus(
entry: ConcurrencyStatusEntry,
@@ -149,12 +227,21 @@ export function viewConcurrencyStatus(
const inFlight = clampCount(entry.inFlight);
const queued = clampCount(entry.queued);
const paused = entry.paused === true;
+ const autoReduced = entry.autoReduced === true;
const atCapacity = inFlight >= limit;
let badge: Badge;
if (paused) badge = "warning";
+ else if (autoReduced) badge = "warning";
else if (atCapacity && queued > 0) badge = "warning";
else if (inFlight > 0) badge = "success";
else badge = "neutral";
+ const cooldownMs = normalizeCooldown(entry.cooldownMs);
+ const autoReducedFrom =
+ autoReduced &&
+ typeof entry.autoReducedFrom === "number" &&
+ Number.isFinite(entry.autoReducedFrom)
+ ? normalizeLimit(entry.autoReducedFrom)
+ : null;
return {
providerId: entry.providerId,
limit,
@@ -166,9 +253,53 @@ export function viewConcurrencyStatus(
pausedLabel: pauseLabel(paused, entry.pausedUntil, now),
badge,
busy: paused || (atCapacity && queued > 0),
+ cooldownMs,
+ cooldownLabel: cooldownLabel(cooldownMs),
+ autoReduced,
+ autoReducedFrom,
};
}
+/**
+ * The auto-reduce banner view for a status entry, or `null` when it is not
+ * auto-reduced. `message` prefers the backend's `notice` (verbatim, when present
+ * + non-empty); otherwise a synthesized fallback is built from
+ * `autoReducedFrom` → `limit`. `fromLimit` is the value "Restore to N" PUTs back.
+ */
+export function viewAutoReduce(entry: ConcurrencyStatusEntry): AutoReduceNotice | null {
+ if (entry.autoReduced !== true) return null;
+ const currentLimit = normalizeLimit(entry.limit);
+ const fromLimit =
+ typeof entry.autoReducedFrom === "number" && Number.isFinite(entry.autoReducedFrom)
+ ? normalizeLimit(entry.autoReducedFrom)
+ : currentLimit + 1;
+ const notice =
+ typeof entry.notice === "string" && entry.notice.length > 0
+ ? entry.notice
+ : `Concurrency limit auto-reduced to ${currentLimit} after a 429 — restore manually when ready.`;
+ return {
+ providerId: entry.providerId,
+ message: notice,
+ fromLimit,
+ currentLimit,
+ };
+}
+
+/**
+ * All auto-reduce banners across a status list (one per auto-reduced provider),
+ * in input order. Empty when none are auto-reduced.
+ */
+export function autoReduceNotices(
+ entries: readonly ConcurrencyStatusEntry[],
+): readonly AutoReduceNotice[] {
+ const out: AutoReduceNotice[] = [];
+ for (const e of entries) {
+ const n = viewAutoReduce(e);
+ if (n !== null) out.push(n);
+ }
+ return out;
+}
+
export function viewConcurrencyStatuses(
entries: readonly ConcurrencyStatusEntry[],
now: number = Date.now(),
@@ -197,8 +328,8 @@ export function summarizeLimits(limits: readonly ConcurrencyLimitEntry[]): strin
/**
* A one-line summary of the live status, e.g.
- * "2 providers · 6/10 in flight · 1 queued · 1 paused". Only the queued / paused
- * fragments appear when non-zero.
+ * "2 providers · 6/10 in flight · 1 queued · 1 paused · 1 auto-reduced". Only the
+ * queued / paused / auto-reduced fragments appear when non-zero.
*/
export function summarizeStatus(
providers: readonly ConcurrencyStatusEntry[],
@@ -209,12 +340,14 @@ export function summarizeStatus(
let limitTotal = 0;
let queued = 0;
let paused = 0;
+ let autoReduced = 0;
for (const p of providers) {
const limit = normalizeLimit(p.limit);
inFlight += clampCount(p.inFlight);
limitTotal += limit;
queued += clampCount(p.queued);
if (p.paused === true) paused += 1;
+ if (p.autoReduced === true) autoReduced += 1;
}
const parts: string[] = [];
parts.push(
@@ -223,6 +356,7 @@ export function summarizeStatus(
);
if (queued > 0) parts.push(`${queued} queued`);
if (paused > 0) parts.push(`${paused} paused`);
+ if (autoReduced > 0) parts.push(`${autoReduced} auto-reduced`);
// Touch `now` so the summary recomputes alongside the per-row pause countdown.
void now;
return parts.join(" · ");
@@ -274,7 +408,9 @@ export function normalizeConcurrencyLimit(data: unknown): ConcurrencyLimitEntry
/**
* Coerce an untrusted `GET /concurrency/status` body into a typed status list.
* `pausedUntil` is included only when it is a finite number (it is absent when
- * not paused).
+ * not paused). `cooldownMs` (default 350) + `autoReduced` are always coerced;
+ * `autoReducedFrom` + `notice` are included only when `autoReduced` is true (and
+ * well-formed), mirroring the backend's "present only when auto-reduced" contract.
*/
export function normalizeConcurrencyStatus(data: unknown): readonly ConcurrencyStatusEntry[] {
if (!isRecord(data) || !Array.isArray(data.providers)) return [];
@@ -282,18 +418,57 @@ export function normalizeConcurrencyStatus(data: unknown): readonly ConcurrencyS
return providers
.filter((r): r is Record<string, unknown> => isRecord(r))
.map((r): ConcurrencyStatusEntry => {
- const base = {
- providerId: asString(r.providerId) ?? "",
- limit: normalizeLimit(r.limit),
- inFlight: clampCount(r.inFlight),
- queued: clampCount(r.queued),
- paused: r.paused === true,
+ const providerId = asString(r.providerId) ?? "";
+ const limit = normalizeLimit(r.limit);
+ const inFlight = clampCount(r.inFlight);
+ const queued = clampCount(r.queued);
+ const paused = r.paused === true;
+ const cooldownMs = normalizeCooldown(r.cooldownMs);
+ const autoReduced = r.autoReduced === true;
+ // Build immutably (the contract fields are readonly): start with the always-
+ // present fields, then layer the optional `pausedUntil` (finite number only)
+ // + the auto-reduce-only `autoReducedFrom`/`notice` (present only when true).
+ let entry: ConcurrencyStatusEntry = {
+ providerId,
+ limit,
+ inFlight,
+ queued,
+ paused,
+ cooldownMs,
+ autoReduced,
};
- const pausedUntil =
- typeof r.pausedUntil === "number" && Number.isFinite(r.pausedUntil)
- ? r.pausedUntil
- : undefined;
- return pausedUntil !== undefined ? { ...base, pausedUntil } : base;
+ if (typeof r.pausedUntil === "number" && Number.isFinite(r.pausedUntil)) {
+ entry = { ...entry, pausedUntil: r.pausedUntil };
+ }
+ if (autoReduced) {
+ // Accumulate the auto-reduce-only optionals into a plain record (the
+ // contract fields are readonly, so we can't mutate a typed partial —
+ // collect then spread into a fresh entry).
+ const patch: { autoReducedFrom?: number; notice?: string } = {};
+ if (typeof r.autoReducedFrom === "number" && Number.isFinite(r.autoReducedFrom)) {
+ patch.autoReducedFrom = normalizeLimit(r.autoReducedFrom);
+ }
+ if (typeof r.notice === "string" && r.notice.length > 0) {
+ patch.notice = r.notice;
+ }
+ if (patch.autoReducedFrom !== undefined || patch.notice !== undefined) {
+ entry = { ...entry, ...patch };
+ }
+ }
+ return entry;
})
.filter((r) => r.providerId !== "");
}
+
+/**
+ * Coerce an untrusted `GET`/`PUT /concurrency/cooldown/:providerId` body into a
+ * typed cooldown response, or `null` when it is malformed (missing/malformed
+ * `providerId` or `cooldownMs`). The composition root surfaces a 404/400/503 as
+ * `ok: false` separately; this only defends the success body.
+ */
+export function normalizeConcurrencyCooldown(data: unknown): ConcurrencyCooldownResponse | null {
+ if (!isRecord(data)) return null;
+ const providerId = asString(data.providerId);
+ if (providerId === null) return null;
+ return { providerId, cooldownMs: normalizeCooldown(data.cooldownMs) };
+}
diff --git a/src/features/concurrency/ui/AutoReduceBanner.svelte b/src/features/concurrency/ui/AutoReduceBanner.svelte
new file mode 100644
index 0000000..132ebc7
--- /dev/null
+++ b/src/features/concurrency/ui/AutoReduceBanner.svelte
@@ -0,0 +1,81 @@
+<script lang="ts">
+ import type { AutoReduceNotice } from "../logic/view-model";
+ import type { RestoreOutcome } from "../logic/types";
+
+ let {
+ notice,
+ onRestore,
+ onDismiss,
+ }: {
+ /** The auto-reduce banner view (providerId + message + from/current limit). */
+ notice: AutoReduceNotice;
+ /**
+ * "Restore to N" — PUT the limit back to `fromLimit`. Returns the outcome so
+ * a FAILED restore surfaces an inline error here (the banner owns its error
+ * display; the parent only refreshes on success).
+ */
+ onRestore: (providerId: string, limit: number) => Promise<RestoreOutcome>;
+ /** Hide this banner locally (persists hidden while autoReduced stays true). */
+ onDismiss: (providerId: string) => void;
+ } = $props();
+
+ let restoring = $state(false);
+ /** Inline restore error (e.g. "Concurrency service not available"); cleared on retry. */
+ let error = $state<string | null>(null);
+
+ async function handleRestore(): Promise<void> {
+ restoring = true;
+ error = null;
+ // The parent PUTs the limit + refreshes status on success; the banner clears
+ // once the next poll shows autoReduced===false. On failure the outcome is
+ // bubbled back here so the error shows inline next to the button.
+ const result = await onRestore(notice.providerId, notice.fromLimit);
+ restoring = false;
+ if (!result.ok) {
+ error = result.error;
+ }
+ }
+</script>
+
+<div
+ class="alert alert-warning flex flex-col gap-2 py-2 text-xs"
+ role="status"
+ data-testid={`auto-reduce-banner-${notice.providerId}`}
+>
+ <div class="flex items-start gap-2">
+ <span class="shrink-0">⚠</span>
+ <div class="flex-1">
+ <p>{notice.message}</p>
+ <p class="opacity-70">
+ Was {notice.fromLimit}, now {notice.currentLimit}.
+ </p>
+ </div>
+ <div class="flex shrink-0 items-center gap-1">
+ <!-- The "Restore to N" text stays visible while loading (only the spinner is
+ prepended) so the button keeps its accessible name during the PUT — a
+ spinner-only button loses its name for screen-reader users. -->
+ <button
+ type="button"
+ class="btn btn-warning btn-xs gap-1"
+ disabled={restoring}
+ onclick={handleRestore}
+ >
+ {#if restoring}
+ <span class="loading loading-spinner loading-xs"></span>
+ {/if}
+ Restore to {notice.fromLimit}
+ </button>
+ <button
+ type="button"
+ class="btn btn-ghost btn-xs"
+ aria-label={`Dismiss auto-reduce notice for ${notice.providerId}`}
+ onclick={() => onDismiss(notice.providerId)}
+ >
+ ✕
+ </button>
+ </div>
+ </div>
+ {#if error}
+ <p class="font-mono text-error" data-testid={`restore-error-${notice.providerId}`}>{error}</p>
+ {/if}
+</div>
diff --git a/src/features/concurrency/ui/ConcurrencyCooldownRow.svelte b/src/features/concurrency/ui/ConcurrencyCooldownRow.svelte
new file mode 100644
index 0000000..03ddec9
--- /dev/null
+++ b/src/features/concurrency/ui/ConcurrencyCooldownRow.svelte
@@ -0,0 +1,93 @@
+<script lang="ts">
+ import { untrack } from "svelte";
+ import { parseCooldownInput } from "../logic/view-model";
+ import type { SaveConcurrencyCooldown } from "../logic/types";
+
+ let {
+ providerId,
+ cooldownMs,
+ save,
+ }: {
+ /** The provider this cooldown row controls. */
+ providerId: string;
+ /** The current per-slot release cooldown (ms) from the live status poll. */
+ cooldownMs: number;
+ save: SaveConcurrencyCooldown;
+ } = $props();
+
+ // Inline-edit state: the raw text bound to the cooldown input. Seeded from the
+ // row's canonical cooldownMs, but only while the field is untouched — so a
+ // status-poll refresh re-syncs it without clobbering an in-flight edit. Mirrors
+ // the ConcurrencyLimitRow / ChatLimitField seed pattern.
+ let draft = $state("");
+ let lastSeed = $state("");
+ let saving = $state(false);
+ let error = $state<string | null>(null);
+ /** Brief "Saved." confirmation after a successful save; cleared on edit. */
+ let justSaved = $state(false);
+
+ $effect(() => {
+ const incoming = String(cooldownMs);
+ untrack(() => {
+ if (draft === lastSeed) draft = incoming;
+ lastSeed = incoming;
+ });
+ });
+
+ const parsed = $derived(parseCooldownInput(draft));
+ const dirty = $derived(parsed !== null && parsed !== cooldownMs);
+
+ function onInput(): void {
+ justSaved = false;
+ error = null;
+ }
+
+ async function handleSave(): Promise<void> {
+ if (parsed === null || parsed === cooldownMs) return;
+ saving = true;
+ error = null;
+ const result = await save(providerId, parsed);
+ saving = false;
+ if (result.ok) {
+ // Reflect the echoed cooldown back into the field immediately (the prop
+ // also re-asserts it via the seed effect above once the parent reloads).
+ draft = String(result.cooldownMs);
+ lastSeed = draft;
+ justSaved = true;
+ } else {
+ error = result.error;
+ }
+ }
+</script>
+
+<div class="flex items-center gap-2 text-xs">
+ <span class="opacity-60">cooldown</span>
+ <input
+ type="text"
+ inputmode="numeric"
+ class="input input-bordered input-xs w-20 font-mono"
+ aria-label={`Release cooldown (ms) for ${providerId}`}
+ bind:value={draft}
+ oninput={onInput}
+ disabled={saving}
+ />
+ <span class="opacity-50">ms</span>
+ <button
+ type="button"
+ class="btn btn-primary btn-xs"
+ aria-label={`Save cooldown for ${providerId}`}
+ disabled={!dirty || saving}
+ onclick={handleSave}
+ >
+ {#if saving}
+ <span class="loading loading-spinner loading-xs"></span>
+ {:else}
+ Save
+ {/if}
+ </button>
+ {#if error}
+ <span class="font-mono text-error">{error}</span>
+ {:else if justSaved && !dirty}
+ <span class="text-success">Saved.</span>
+ {/if}
+</div>
diff --git a/src/features/concurrency/ui/ConcurrencyView.svelte b/src/features/concurrency/ui/ConcurrencyView.svelte
index f8a2199..6078b1a 100644
--- a/src/features/concurrency/ui/ConcurrencyView.svelte
+++ b/src/features/concurrency/ui/ConcurrencyView.svelte
@@ -2,6 +2,7 @@
import { untrack } from "svelte";
import type { ConcurrencyStatusEntry } from "@dispatch/transport-contract";
import {
+ autoReduceNotices,
type Badge,
parseLimitInput,
providerOptions,
@@ -15,8 +16,12 @@
DeleteConcurrencyLimit,
LoadConcurrencyLimits,
LoadConcurrencyStatus,
+ RestoreOutcome,
+ SaveConcurrencyCooldown,
SaveConcurrencyLimit,
} from "../logic/types";
+ import AutoReduceBanner from "./AutoReduceBanner.svelte";
+ import ConcurrencyCooldownRow from "./ConcurrencyCooldownRow.svelte";
import ConcurrencyLimitRow from "./ConcurrencyLimitRow.svelte";
let {
@@ -25,6 +30,7 @@
saveLimit,
deleteLimit,
loadStatus,
+ saveCooldown,
}: {
/** Available models (`<provider>/<model>`) — the source of provider ids for the Add dropdown. */
models: readonly string[];
@@ -32,6 +38,7 @@
saveLimit: SaveConcurrencyLimit;
deleteLimit: DeleteConcurrencyLimit;
loadStatus: LoadConcurrencyStatus;
+ saveCooldown: SaveConcurrencyCooldown;
} = $props();
const badgeClass: Record<Badge, string> = {
@@ -136,6 +143,18 @@
return result;
}
+ // Wrap the cooldown save so a successful PUT refreshes the live status (which
+ // re-carries the new `cooldownMs`). The row still gets the result to drive its
+ // own UI. `getCooldown` is exposed for completeness/future use (the live status
+ // already carries `cooldownMs`, so the row seeds from the status view).
+ async function cooldownSave(providerId: string, cooldownMs: number) {
+ const result = await saveCooldown(providerId, cooldownMs);
+ if (result.ok) {
+ void refreshStatus();
+ }
+ return result;
+ }
+
// ── Live status (polls while mounted) ───────────────────────────────────────
let statusEntries = $state<readonly ConcurrencyStatusEntry[]>([]);
let statusError = $state<string | null>(null);
@@ -158,6 +177,69 @@
const statusViews = $derived(viewConcurrencyStatuses(statusEntries, now));
const statusSummary = $derived(summarizeStatus(statusEntries, now));
+ // ── Auto-reduce banners (persist while autoReduced===true; dismissible) ───────
+ //
+ // When a provider's limit is auto-reduced by a 429, `GET /concurrency/status`
+ // carries `autoReduced: true` (+ `autoReducedFrom` + `notice`). We render a
+ // banner per such provider. The banner is DISMISSIBLE: a dismissed provider
+ // stays hidden while it remains auto-reduced (persist-while-true), and is
+ // UN-dismissed the moment a poll shows it no longer auto-reduced — so a future
+ // auto-reduce re-shows the banner. Restoring the limit (PUT) clears
+ // `autoReduced` server-side → the next poll drops the banner automatically.
+ //
+ // The dismissed set is intentionally COMPONENT-LOCAL (NOT persisted to
+ // localStorage / a module-global): it resets on remount (sidebar view switch /
+ // reload). This is correct — `autoReduced` is a REAL persisted degraded state,
+ // so re-showing the banner on a fresh mount reminds the user. Persisting a
+ // dismissal across reloads would risk HIDING an ongoing degradation (a
+ // footgun), and AGENTS.md forbids module-global ambient state. Mirrors the
+ // component-local `limitsError`/`statusError` pattern.
+ let dismissedAutoReduce = $state<ReadonlySet<string>>(new Set());
+
+ const allNotices = $derived(autoReduceNotices(statusEntries));
+ const visibleNotices = $derived(
+ allNotices.filter((n) => !dismissedAutoReduce.has(n.providerId)),
+ );
+
+ // Reconcile the dismissed set against the live auto-reduced providers: keep a
+ // dismissed entry ONLY while its provider is still auto-reduced. A provider
+ // that has been restored (no longer in `allNotices`) is dropped from the
+ // dismissed set so a future auto-reduce re-shows its banner.
+ $effect(() => {
+ const autoReducedIds = new Set(allNotices.map((n) => n.providerId));
+ untrack(() => {
+ let changed = false;
+ const next = new Set<string>();
+ for (const id of dismissedAutoReduce) {
+ if (autoReducedIds.has(id)) next.add(id);
+ else changed = true;
+ }
+ if (changed) dismissedAutoReduce = next;
+ });
+ });
+
+ function dismissAutoReduce(providerId: string): void {
+ if (dismissedAutoReduce.has(providerId)) return;
+ dismissedAutoReduce = new Set([...dismissedAutoReduce, providerId]);
+ }
+
+ // "Restore to N" — PUT the limit back to `autoReducedFrom` via the limits
+ // endpoint (a manual PUT clears `autoReduced` server-side). Refreshes limits +
+ // status on success; the next status poll shows `autoReduced===false` and the
+ // banner drops (the dismissed-set effect above un-dismisses it too). The banner
+ // component owns its own restoring-spinner + inline error; on FAILURE the
+ // outcome is bubbled back so the banner shows the error inline (instead of
+ // silently re-enabling the button / surfacing it only in the limits section).
+ async function restoreLimit(providerId: string, limit: number): Promise<RestoreOutcome> {
+ const result = await saveLimit(providerId, limit);
+ if (result.ok) {
+ void refreshLimits();
+ void refreshStatus();
+ return { ok: true };
+ }
+ return { ok: false, error: result.error };
+ }
+
async function refreshStatus(): Promise<void> {
if (statusInFlight) return;
statusInFlight = true;
@@ -192,6 +274,19 @@
</script>
<div class="flex flex-col gap-4">
+ <!-- Auto-reduce banners (appear when a provider's limit was auto-reduced by a 429) -->
+ {#if visibleNotices.length > 0}
+ <section class="flex flex-col gap-2" aria-label="Concurrency auto-reduce notices">
+ {#each visibleNotices as notice (notice.providerId)}
+ <AutoReduceBanner
+ {notice}
+ onRestore={restoreLimit}
+ onDismiss={dismissAutoReduce}
+ />
+ {/each}
+ </section>
+ {/if}
+
<!-- Limits (config) -->
<section class="flex flex-col gap-2">
<div class="flex items-center justify-between gap-2">
@@ -319,10 +414,22 @@
<div class="flex flex-wrap items-center justify-between gap-2 text-xs opacity-70">
<span title="In-flight slots held vs cap">{s.inFlightLabel} in flight</span>
<span>{s.queuedLabel}</span>
+ <span title="Per-slot release cooldown">cooldown {s.cooldownLabel}</span>
</div>
{#if s.pausedLabel}
<span class="text-xs text-warning">{s.pausedLabel}</span>
{/if}
+ {#if s.autoReduced}
+ <span class="text-xs text-warning">
+ Limit auto-reduced{#if s.autoReducedFrom !== null}
+ from {s.autoReducedFrom} to {s.limit}{/if}.
+ </span>
+ {/if}
+ <ConcurrencyCooldownRow
+ providerId={s.providerId}
+ cooldownMs={s.cooldownMs}
+ save={cooldownSave}
+ />
</li>
{/each}
</ul>
diff --git a/src/features/concurrency/ui/ConcurrencyView.test.ts b/src/features/concurrency/ui/ConcurrencyView.test.ts
index 3dc8e78..854f13b 100644
--- a/src/features/concurrency/ui/ConcurrencyView.test.ts
+++ b/src/features/concurrency/ui/ConcurrencyView.test.ts
@@ -3,6 +3,7 @@ import { render, screen } from "@testing-library/svelte";
import userEvent from "@testing-library/user-event";
import { describe, expect, it } from "vitest";
import type {
+ ConcurrencyCooldownResult,
ConcurrencyDeleteResult,
ConcurrencyLimitResult,
ConcurrencyLimitsResult,
@@ -13,35 +14,80 @@ import ConcurrencyView from "./ConcurrencyView.svelte";
// Available models → provider ids are "umans", "anthropic", "openai-compat".
const MODELS = ["umans/umans-glm-5.2", "anthropic/claude-sonnet", "openai-compat/gpt-4o"] as const;
-// Fakes for the four injected ports. Each resolves immediately so the mount
-// effect's initial load settles in a microtask (assertions await via findBy*).
+// A status entry factory (defaults to a healthy limited provider). The new
+// concurrency-fixes fields (`cooldownMs`, `autoReduced`) are always present.
+function statusEntry(over: Partial<ConcurrencyStatusEntry> = {}): ConcurrencyStatusEntry {
+ return {
+ providerId: "umans",
+ limit: 4,
+ inFlight: 2,
+ queued: 1,
+ paused: false,
+ cooldownMs: 350,
+ autoReduced: false,
+ ...over,
+ };
+}
+// Fakes for the injected ports. Each resolves immediately so the mount effect's
+// initial load settles in a microtask (assertions await via findBy*). The status
+// list is mutable so a test can flip `autoReduced` between polls to simulate a
+// restore clearing the banner.
function makeFakes(opts?: {
limits?: readonly { providerId: string; limit: number }[];
- status?: readonly ConcurrencyStatusEntry[];
+ status?: ConcurrencyStatusEntry[];
+ /**
+ * When set, `saveLimit` rejects with this error (returns `ok: false`) — used
+ * to test the auto-reduce banner's inline restore-error feedback.
+ */
+ saveLimitError?: string;
+ /**
+ * Optional hook invoked inside `saveLimit` AFTER recording the call. Lets a
+ * test simulate a backend side-effect of the PUT (e.g. clearing `autoReduced`
+ * on the next status poll). Receives the providerId + limit + the fakes bag so
+ * it can mutate the status list. (A plain method reassignment would NOT reach
+ * the already-rendered component — the prop captured the original closure.)
+ */
+ onSaveLimit?: (
+ providerId: string,
+ limit: number,
+ self: { calls: MakeFakesCalls; setStatus: (next: ConcurrencyStatusEntry[]) => void },
+ ) => void;
}) {
let limits = opts?.limits ?? [{ providerId: "umans", limit: 4 }];
- const status = opts?.status ?? [
- { providerId: "umans", limit: 4, inFlight: 2, queued: 1, paused: false },
- ];
+ let status = opts?.status ?? [statusEntry()];
+ const onSaveLimit = opts?.onSaveLimit;
+ const saveLimitError = opts?.saveLimitError;
- const calls = {
+ const calls: MakeFakesCalls = {
loadLimits: 0,
loadStatus: 0,
saves: [] as { providerId: string; limit: number }[],
deletes: [] as string[],
+ cooldownSaves: [] as { providerId: string; cooldownMs: number }[],
};
+ function setStatus(next: ConcurrencyStatusEntry[]): void {
+ status = next;
+ }
+
return {
calls,
+ // Allow a test to mutate the status list between polls (e.g. clear
+ // autoReduced after a restore to simulate the next poll).
+ setStatus,
loadLimits: async (): Promise<ConcurrencyLimitsResult> => {
calls.loadLimits++;
return { ok: true, limits };
},
saveLimit: async (providerId: string, limit: number): Promise<ConcurrencyLimitResult> => {
calls.saves.push({ providerId, limit });
+ if (saveLimitError !== undefined) {
+ return { ok: false, error: saveLimitError };
+ }
// Reflect the new limit into the list the next load returns.
limits = [...limits.filter((l) => l.providerId !== providerId), { providerId, limit }];
+ if (onSaveLimit !== undefined) onSaveLimit(providerId, limit, { calls, setStatus });
return { ok: true, providerId, limit };
},
deleteLimit: async (providerId: string): Promise<ConcurrencyDeleteResult> => {
@@ -53,21 +99,41 @@ function makeFakes(opts?: {
calls.loadStatus++;
return { ok: true, providers: status };
},
+ saveCooldown: async (
+ providerId: string,
+ cooldownMs: number,
+ ): Promise<ConcurrencyCooldownResult> => {
+ calls.cooldownSaves.push({ providerId, cooldownMs });
+ // Reflect the new cooldown into the status list the next load returns.
+ status = status.map((s) => (s.providerId === providerId ? { ...s, cooldownMs } : s));
+ return { ok: true, providerId, cooldownMs };
+ },
+ };
+}
+
+type MakeFakesCalls = {
+ loadLimits: number;
+ loadStatus: number;
+ saves: { providerId: string; limit: number }[];
+ deletes: string[];
+ cooldownSaves: { providerId: string; cooldownMs: number }[];
+};
+
+function props(fakes: ReturnType<typeof makeFakes>) {
+ return {
+ models: MODELS as unknown as readonly string[],
+ loadLimits: fakes.loadLimits,
+ saveLimit: fakes.saveLimit,
+ deleteLimit: fakes.deleteLimit,
+ loadStatus: fakes.loadStatus,
+ saveCooldown: fakes.saveCooldown,
};
}
describe("ConcurrencyView", () => {
it("loads + renders the configured limits and live status on mount", async () => {
const fakes = makeFakes();
- render(ConcurrencyView, {
- props: {
- models: MODELS,
- loadLimits: fakes.loadLimits,
- saveLimit: fakes.saveLimit,
- deleteLimit: fakes.deleteLimit,
- loadStatus: fakes.loadStatus,
- },
- });
+ render(ConcurrencyView, { props: props(fakes) });
// The provider dropdown is populated from the available models' providers.
const providerSelect = await screen.findByLabelText("Provider");
@@ -81,21 +147,21 @@ describe("ConcurrencyView", () => {
expect(fakes.calls.loadStatus).toBeGreaterThanOrEqual(1);
});
+ it("renders the per-provider cooldown label from the live status", async () => {
+ const fakes = makeFakes({ status: [statusEntry({ cooldownMs: 350 })] });
+ render(ConcurrencyView, { props: props(fakes) });
+
+ // The status card shows the cooldown alongside in-flight/queue.
+ expect(await screen.findByText(/cooldown 350ms/)).toBeInTheDocument();
+ });
+
it("surfaces NO loading indicator during refresh (background poll is silent — no flicker)", async () => {
// The 2s status poll + post-mutation reloads are SILENT: they never toggle a
// visible loading state, so the Refresh buttons are plain-text (no spinner)
// and the list area never reflows mid-refresh. Regression guard for the
// flicker fix (mirrors the heartbeat runs list).
const fakes = makeFakes();
- render(ConcurrencyView, {
- props: {
- models: MODELS,
- loadLimits: fakes.loadLimits,
- saveLimit: fakes.saveLimit,
- deleteLimit: fakes.deleteLimit,
- loadStatus: fakes.loadStatus,
- },
- });
+ render(ConcurrencyView, { props: props(fakes) });
await screen.findByText(/1 provider · 2\/4 in flight · 1 queued/);
@@ -116,15 +182,7 @@ describe("ConcurrencyView", () => {
it("adds a provider limit via the dropdown form (calls saveLimit + reloads)", async () => {
const user = userEvent.setup();
const fakes = makeFakes({ limits: [], status: [] });
- render(ConcurrencyView, {
- props: {
- models: MODELS,
- loadLimits: fakes.loadLimits,
- saveLimit: fakes.saveLimit,
- deleteLimit: fakes.deleteLimit,
- loadStatus: fakes.loadStatus,
- },
- });
+ render(ConcurrencyView, { props: props(fakes) });
const providerSelect = await screen.findByLabelText("Provider");
// Choose "anthropic" from the dropdown (the list is auto-selected first).
@@ -142,15 +200,7 @@ describe("ConcurrencyView", () => {
it("disables Add when the limit is empty/invalid (provider is auto-selected)", async () => {
const user = userEvent.setup();
const fakes = makeFakes({ limits: [], status: [] });
- render(ConcurrencyView, {
- props: {
- models: MODELS,
- loadLimits: fakes.loadLimits,
- saveLimit: fakes.saveLimit,
- deleteLimit: fakes.deleteLimit,
- loadStatus: fakes.loadStatus,
- },
- });
+ render(ConcurrencyView, { props: props(fakes) });
const providerSelect = await screen.findByLabelText("Provider");
// A provider is auto-selected from the dropdown.
@@ -172,13 +222,7 @@ describe("ConcurrencyView", () => {
it("shows no-providers + disables the dropdown when there are no models", async () => {
const fakes = makeFakes({ limits: [], status: [] });
render(ConcurrencyView, {
- props: {
- models: [],
- loadLimits: fakes.loadLimits,
- saveLimit: fakes.saveLimit,
- deleteLimit: fakes.deleteLimit,
- loadStatus: fakes.loadStatus,
- },
+ props: { ...props(fakes), models: [] as unknown as readonly string[] },
});
const providerSelect = await screen.findByLabelText("Provider");
@@ -189,15 +233,7 @@ describe("ConcurrencyView", () => {
it("removes a provider limit via the row ✕ (calls deleteLimit + reloads)", async () => {
const user = userEvent.setup();
const fakes = makeFakes();
- render(ConcurrencyView, {
- props: {
- models: MODELS,
- loadLimits: fakes.loadLimits,
- saveLimit: fakes.saveLimit,
- deleteLimit: fakes.deleteLimit,
- loadStatus: fakes.loadStatus,
- },
- });
+ render(ConcurrencyView, { props: props(fakes) });
// Wait for the limits to load (unique summary) before interacting.
await screen.findByText(/1 limit configured/);
@@ -209,7 +245,7 @@ describe("ConcurrencyView", () => {
it("surfaces a load error from the limits endpoint", async () => {
const failing = {
- models: MODELS,
+ models: MODELS as unknown as readonly string[],
loadLimits: async (): Promise<ConcurrencyLimitsResult> => ({
ok: false,
error: "Concurrency service not available",
@@ -217,8 +253,191 @@ describe("ConcurrencyView", () => {
saveLimit: async (): Promise<ConcurrencyLimitResult> => ({ ok: false, error: "noop" }),
deleteLimit: async (): Promise<ConcurrencyDeleteResult> => ({ ok: false, error: "noop" }),
loadStatus: async (): Promise<ConcurrencyStatusResult> => ({ ok: true, providers: [] }),
+ saveCooldown: async (): Promise<ConcurrencyCooldownResult> => ({ ok: false, error: "noop" }),
};
render(ConcurrencyView, { props: failing });
expect(await screen.findByText("Concurrency service not available")).toBeVisible();
});
+
+ // ── Concurrency-fixes: auto-reduce banner + cooldown editing ────────────────
+
+ it("renders an auto-reduce banner (with the backend notice + Restore) when a provider is auto-reduced", async () => {
+ const fakes = makeFakes({
+ status: [
+ statusEntry({
+ limit: 3,
+ autoReduced: true,
+ autoReducedFrom: 4,
+ notice: "Concurrency limit auto-reduced to 3 after a 429 — restore manually when ready.",
+ }),
+ ],
+ });
+ render(ConcurrencyView, { props: props(fakes) });
+
+ // The banner shows the backend notice verbatim + a "Restore to 4" action.
+ expect(await screen.findByText(/auto-reduced to 3 after a 429/)).toBeVisible();
+ expect(await screen.findByRole("button", { name: /Restore to 4/ })).toBeVisible();
+ // The "Was 4, now 3." provenance line is shown.
+ expect(await screen.findByText(/Was 4, now 3\./)).toBeVisible();
+ });
+
+ it("clears the banner after Restore (next status poll shows autoReduced===false)", async () => {
+ const user = userEvent.setup();
+ // Start auto-reduced (limit 3, was 4). The restore PUT clears `autoReduced`
+ // server-side; the next status poll returns limit 4 + autoReduced===false →
+ // the banner drops.
+ const fakes = makeFakes({
+ status: [
+ statusEntry({
+ limit: 3,
+ autoReduced: true,
+ autoReducedFrom: 4,
+ notice: "Concurrency limit auto-reduced to 3 after a 429.",
+ }),
+ ],
+ onSaveLimit: (_providerId, limit, self) => {
+ // Simulate the backend clearing `autoReduced` on the manual PUT: the next
+ // status load returns the restored limit with autoReduced===false.
+ self.setStatus([statusEntry({ limit, autoReduced: false })]);
+ },
+ });
+ render(ConcurrencyView, { props: props(fakes) });
+
+ const restoreBtn = await screen.findByRole("button", { name: /Restore to 4/ });
+ await user.click(restoreBtn);
+
+ // The restore PUT the limit back to the original (autoReducedFrom = 4).
+ expect(fakes.calls.saves).toEqual([{ providerId: "umans", limit: 4 }]);
+ // The banner is gone (no Restore button, no notice text); the status summary
+ // now reflects the restored limit (2/4 in flight).
+ await screen.findByText(/1 provider · 2\/4 in flight · 1 queued/);
+ expect(screen.queryByRole("button", { name: /Restore to/ })).toBeNull();
+ expect(screen.queryByText(/auto-reduced to 3 after a 429/)).toBeNull();
+ });
+
+ it("dismisses the auto-reduce banner locally while it stays auto-reduced", async () => {
+ const user = userEvent.setup();
+ const fakes = makeFakes({
+ status: [
+ statusEntry({
+ limit: 3,
+ autoReduced: true,
+ autoReducedFrom: 4,
+ notice: "Concurrency limit auto-reduced to 3 after a 429.",
+ }),
+ ],
+ });
+ render(ConcurrencyView, { props: props(fakes) });
+
+ await screen.findByRole("button", { name: /Restore to 4/ });
+ // Dismiss the banner (hide locally — the provider is still auto-reduced).
+ await user.click(screen.getByLabelText("Dismiss auto-reduce notice for umans"));
+ expect(screen.queryByRole("button", { name: /Restore to/ })).toBeNull();
+ expect(screen.queryByText(/auto-reduced to 3 after a 429/)).toBeNull();
+ });
+
+ it("shows an inline error in the banner when the Restore PUT fails (no silent re-enable)", async () => {
+ const user = userEvent.setup();
+ const fakes = makeFakes({
+ status: [
+ statusEntry({
+ limit: 3,
+ autoReduced: true,
+ autoReducedFrom: 4,
+ notice: "Concurrency limit auto-reduced to 3 after a 429.",
+ }),
+ ],
+ saveLimitError: "Concurrency service not available",
+ });
+ render(ConcurrencyView, { props: props(fakes) });
+
+ const restoreBtn = await screen.findByRole("button", { name: /Restore to 4/ });
+ await user.click(restoreBtn);
+
+ // The error surfaces INLINE in the banner (near the restore action), not
+ // only in the far-away limits section. The banner is still present (restore
+ // did not succeed) and the button re-enabled for a retry.
+ expect(await screen.findByTestId("restore-error-umans")).toHaveTextContent(
+ "Concurrency service not available",
+ );
+ expect(screen.getByRole("button", { name: /Restore to 4/ })).toBeVisible();
+ expect(screen.getByRole("button", { name: /Restore to 4/ })).not.toBeDisabled();
+ // The restore PUT was attempted.
+ expect(fakes.calls.saves).toEqual([{ providerId: "umans", limit: 4 }]);
+ });
+
+ it("clears the inline restore error on a retry that succeeds", async () => {
+ const user = userEvent.setup();
+ // First restore fails; the second succeeds (clears autoReduced). Reassigning
+ // `fakes.saveLimit` BEFORE `props(fakes)` is captured would NOT reach the
+ // rendered component, so swap it BEFORE render here.
+ const fakes = makeFakes({
+ status: [
+ statusEntry({
+ limit: 3,
+ autoReduced: true,
+ autoReducedFrom: 4,
+ notice: "Concurrency limit auto-reduced to 3 after a 429.",
+ }),
+ ],
+ onSaveLimit: (_providerId, limit, self) => {
+ self.setStatus([statusEntry({ limit, autoReduced: false })]);
+ },
+ });
+ let attempts = 0;
+ const succeeding = fakes.saveLimit;
+ fakes.saveLimit = async (providerId, limit) => {
+ attempts++;
+ if (attempts === 1) return { ok: false, error: "Concurrency service not available" };
+ return succeeding(providerId, limit);
+ };
+ render(ConcurrencyView, { props: props(fakes) });
+
+ const restoreBtn = await screen.findByRole("button", { name: /Restore to 4/ });
+ await user.click(restoreBtn);
+ // First attempt: inline error appears.
+ expect(await screen.findByTestId("restore-error-umans")).toBeInTheDocument();
+
+ // Retry: the error clears, the banner drops (restore succeeded).
+ await user.click(screen.getByRole("button", { name: /Restore to 4/ }));
+ await screen.findByText(/1 provider · 2\/4 in flight · 1 queued/);
+ expect(screen.queryByTestId("restore-error-umans")).toBeNull();
+ expect(screen.queryByRole("button", { name: /Restore to/ })).toBeNull();
+ });
+
+ it("edits the per-provider cooldown (PUT /concurrency/cooldown + reloads status)", async () => {
+ const user = userEvent.setup();
+ const fakes = makeFakes({ status: [statusEntry({ cooldownMs: 350 })] });
+ render(ConcurrencyView, { props: props(fakes) });
+
+ // Wait for the status card + its cooldown input (seeded with 350).
+ const cooldownInput = await screen.findByLabelText("Release cooldown (ms) for umans");
+ expect((cooldownInput as HTMLInputElement).value).toBe("350");
+
+ await user.clear(cooldownInput);
+ await user.type(cooldownInput, "500");
+ await user.click(screen.getByRole("button", { name: "Save cooldown for umans" }));
+
+ // The cooldown PUT fired with the new value.
+ expect(fakes.calls.cooldownSaves).toEqual([{ providerId: "umans", cooldownMs: 500 }]);
+ // After save the component reloads status (which now carries cooldownMs=500).
+ expect(await screen.findByText(/cooldown 500ms/)).toBeInTheDocument();
+ });
+
+ it("rejects a negative cooldown input (Save disabled — non-negative integer only)", async () => {
+ const user = userEvent.setup();
+ const fakes = makeFakes({ status: [statusEntry({ cooldownMs: 350 })] });
+ render(ConcurrencyView, { props: props(fakes) });
+
+ const cooldownInput = await screen.findByLabelText("Release cooldown (ms) for umans");
+ // 0 is valid (no cooldown); a negative is not.
+ await user.clear(cooldownInput);
+ await user.type(cooldownInput, "0");
+ expect(screen.getByRole("button", { name: "Save cooldown for umans" })).toBeEnabled();
+
+ await user.clear(cooldownInput);
+ await user.type(cooldownInput, "-5");
+ expect(screen.getByRole("button", { name: "Save cooldown for umans" })).toBeDisabled();
+ expect(fakes.calls.cooldownSaves).toHaveLength(0);
+ });
});