chore: merge dev into v2 (#34788)
Co-authored-by: Brendan Allan <14191578+Brendonovich@users.noreply.github.com> Co-authored-by: Kit Langton <kit.langton@gmail.com> Co-authored-by: opencode-agent[bot] <opencode-agent[bot]@users.noreply.github.com> Co-authored-by: Affan Ali <93028901+affanali2k3@users.noreply.github.com> Co-authored-by: affanali2k3 <affanalikhanxx@gmail.com> Co-authored-by: Frank <frank@anoma.ly> Co-authored-by: opencode-agent[bot] <219766164+opencode-agent[bot]@users.noreply.github.com> Co-authored-by: 𝓛𝓲𝓽𝓽𝓵𝓮 𝓕𝓻𝓪𝓷𝓴 <little-frank@opencord.local> Co-authored-by: Aiden Cline <63023139+rekram1-node@users.noreply.github.com> Co-authored-by: Jay V <air@live.ca> Co-authored-by: Dax Raad <d@ironbay.co> Co-authored-by: Aarav Sareen <96787824+arvsrn@users.noreply.github.com> Co-authored-by: OpeOginni <107570612+OpeOginni@users.noreply.github.com> Co-authored-by: Luke Parker <10430890+Hona@users.noreply.github.com> Co-authored-by: Ben Guthrie <benjee.012@gmail.com> Co-authored-by: Dax <mail@thdxr.com> Co-authored-by: Filip <34747899+neriousy@users.noreply.github.com> Co-authored-by: Max Anderson <max.a.anderson95@gmail.com> Co-authored-by: Brendan Allan <git@brendonovich.dev> Co-authored-by: Jack <jack@anoma.ly> Co-authored-by: Shoubhit Dash <shoubhit2005@gmail.com> Co-authored-by: Dustin Deus <deusdustin@gmail.com> Co-authored-by: starptech <starptech@starptechs-MBP.fritz.box> Co-authored-by: Aiden Cline <aidenpcline@gmail.com> Co-authored-by: usrnk1 <7547651+usrnk1@users.noreply.github.com> Co-authored-by: Jay <53023+jayair@users.noreply.github.com> Co-authored-by: runvip <164729189+runvip@users.noreply.github.com> Co-authored-by: opencode <opencode@sst.dev> Co-authored-by: Julian Coy <julian@ex-machina.co> Co-authored-by: Vladimir Glafirov <vglafirov@gitlab.com>
This commit is contained in:
James Long
GitHub
parent
932a40cfd9
commit
8c94e9005f
590 changed files with 15772 additions and 5530 deletions
|
|
@ -327,37 +327,6 @@ body {
|
|||
}
|
||||
}
|
||||
|
||||
[data-component="desktop-app-banner"] {
|
||||
display: flex;
|
||||
align-items: center;
|
||||
gap: 12px;
|
||||
margin-bottom: 32px;
|
||||
|
||||
[data-slot="badge"] {
|
||||
background: var(--color-background-strong);
|
||||
color: var(--color-text-inverted);
|
||||
font-weight: 500;
|
||||
padding: 4px 8px;
|
||||
line-height: 1;
|
||||
flex-shrink: 0;
|
||||
}
|
||||
|
||||
[data-slot="content"] {
|
||||
display: flex;
|
||||
align-items: center;
|
||||
gap: 1ch;
|
||||
}
|
||||
|
||||
[data-slot="text"] {
|
||||
color: var(--color-text-strong);
|
||||
line-height: 1.4;
|
||||
|
||||
@media (max-width: 30.625rem) {
|
||||
display: none;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
[data-slot="hero-copy"] {
|
||||
img {
|
||||
margin-bottom: 24px;
|
||||
|
|
@ -663,10 +632,6 @@ body {
|
|||
fill: var(--color-text-strong);
|
||||
}
|
||||
|
||||
[data-bar][data-kind="promo"] {
|
||||
fill: color-mix(in srgb, var(--bar-go) 50%, transparent);
|
||||
}
|
||||
|
||||
[data-val] {
|
||||
fill: var(--color-text-strong);
|
||||
font-size: 13px;
|
||||
|
|
|
|||
|
|
@ -64,10 +64,10 @@ function LimitsGraph(props: { href: string }) {
|
|||
{ id: "glm-5.2", name: "GLM-5.2", req: 880, d: "100ms" },
|
||||
{ id: "qwen3.7-max", name: "Qwen3.7 Max", req: 950, d: "110ms" },
|
||||
{ id: "kimi-k2.7-code", name: "Kimi K2.7 Code", req: 1150, d: "150ms" },
|
||||
{ id: "mimo-v2.5-pro", name: "MiMo-V2.5-Pro", req: 3250, d: "210ms" },
|
||||
{ id: "deepseek-v4-pro", name: "DeepSeek V4 Pro", req: 3450, d: "240ms" },
|
||||
{ id: "qwen3.7-plus", name: "Qwen3.7 Plus", req: 4300, d: "250ms" },
|
||||
{ id: "minimax-m3", name: "MiniMax M3 (3x usage)", req: 9600, baseReq: 3200, d: "280ms" },
|
||||
{ id: "minimax-m3", name: "MiniMax M3", req: 3200, d: "210ms" },
|
||||
{ id: "mimo-v2.5-pro", name: "MiMo-V2.5-Pro", req: 3250, d: "240ms" },
|
||||
{ id: "deepseek-v4-pro", name: "DeepSeek V4 Pro", req: 3450, d: "270ms" },
|
||||
{ id: "qwen3.7-plus", name: "Qwen3.7 Plus", req: 4300, d: "300ms" },
|
||||
{ id: "mimo-v2.5", name: "MiMo-V2.5", req: 30100, d: "340ms" },
|
||||
{ id: "deepseek-v4-flash", name: "DeepSeek V4 Flash", req: 31650, d: "340ms" },
|
||||
]
|
||||
|
|
@ -157,24 +157,12 @@ function LimitsGraph(props: { href: string }) {
|
|||
<rect
|
||||
x={left}
|
||||
y={gy(i()) - bh / 2}
|
||||
width={Math.max(0, x(ratio(m.baseReq ?? m.req)) - left)}
|
||||
width={Math.max(0, x(ratio(m.req)) - left)}
|
||||
height={bh}
|
||||
data-bar
|
||||
data-kind="go"
|
||||
data-model={m.id}
|
||||
data-segment={m.baseReq ? "base" : undefined}
|
||||
/>
|
||||
{m.baseReq && (
|
||||
<rect
|
||||
x={x(ratio(m.baseReq)) + 2}
|
||||
y={gy(i()) - bh / 2}
|
||||
width={Math.max(0, x(ratio(m.req)) - x(ratio(m.baseReq)) - 2)}
|
||||
height={bh}
|
||||
data-bar
|
||||
data-kind="promo"
|
||||
data-model={m.id}
|
||||
/>
|
||||
)}
|
||||
</g>
|
||||
)}
|
||||
</For>
|
||||
|
|
@ -264,12 +252,6 @@ export default function Home() {
|
|||
|
||||
<div data-component="content">
|
||||
<section data-component="hero">
|
||||
<div data-component="desktop-app-banner">
|
||||
<span data-slot="badge">{i18n.t("home.banner.badge")}</span>
|
||||
<div data-slot="content">
|
||||
<span data-slot="text">{i18n.t("go.banner.text")}</span>
|
||||
</div>
|
||||
</div>
|
||||
<div data-slot="hero-copy">
|
||||
<img data-slot="zen logo light" src={goLogoLight} alt="" />
|
||||
<img data-slot="zen logo dark" src={goLogoDark} alt="" />
|
||||
|
|
|
|||
|
|
@ -3,7 +3,7 @@ import { z } from "zod"
|
|||
import { Resource } from "@opencode-ai/console-resource"
|
||||
import { safeEqual } from "@opencode-ai/console-core/util/crypto.js"
|
||||
|
||||
const DISCORD_ALERT_ROLE_ID = "1511795723262365887"
|
||||
const DISCORD_ALERT_ROLE_ID = "1520924666359713863"
|
||||
|
||||
const basePayload = z.object({
|
||||
name: z.string().optional(),
|
||||
|
|
|
|||
|
|
@ -75,6 +75,40 @@
|
|||
}
|
||||
}
|
||||
|
||||
[data-slot="providers-section"] {
|
||||
display: flex;
|
||||
flex-direction: column;
|
||||
gap: var(--space-3);
|
||||
margin-top: var(--space-6);
|
||||
padding-top: var(--space-6);
|
||||
border-top: 1px solid var(--color-border-muted);
|
||||
|
||||
[data-slot="providers-header"] {
|
||||
display: flex;
|
||||
flex-direction: column;
|
||||
gap: var(--space-1);
|
||||
|
||||
h3 {
|
||||
color: var(--color-text);
|
||||
font-size: var(--font-size-lg);
|
||||
font-weight: 600;
|
||||
line-height: 1.2;
|
||||
margin: 0;
|
||||
}
|
||||
|
||||
p {
|
||||
color: var(--color-text-muted);
|
||||
font-size: var(--font-size-sm);
|
||||
line-height: 1.5;
|
||||
margin: 0;
|
||||
}
|
||||
}
|
||||
|
||||
[data-slot="setting-row"] {
|
||||
margin-top: 0;
|
||||
}
|
||||
}
|
||||
|
||||
[data-slot="toggle-label"] {
|
||||
position: relative;
|
||||
display: inline-block;
|
||||
|
|
|
|||
|
|
@ -5,7 +5,9 @@ import { Modal } from "~/component/modal"
|
|||
import { Billing } from "@opencode-ai/console-core/billing.js"
|
||||
import { Database, eq, and, isNull } from "@opencode-ai/console-core/drizzle/index.js"
|
||||
import { BillingTable, LiteTable } from "@opencode-ai/console-core/schema/billing.sql.js"
|
||||
import { WorkspaceTable } from "@opencode-ai/console-core/schema/workspace.sql.js"
|
||||
import { Actor } from "@opencode-ai/console-core/actor.js"
|
||||
import { Workspace } from "@opencode-ai/console-core/workspace.js"
|
||||
import { Subscription } from "@opencode-ai/console-core/subscription.js"
|
||||
import { LiteData } from "@opencode-ai/console-core/lite.js"
|
||||
import { withActor } from "~/context/auth.withActor"
|
||||
|
|
@ -16,6 +18,8 @@ import { useLanguage } from "~/context/language"
|
|||
import { formError } from "~/lib/form-error"
|
||||
import { formatResetTime, liteResetTimeKeys } from "~/lib/format-reset-time"
|
||||
import { createReferralFromCookie } from "~/lib/referral-invite"
|
||||
import { getRequestEvent } from "solid-js/web"
|
||||
import { countryFromRequest } from "~/lib/request-country"
|
||||
|
||||
import { IconAlipay, IconUpi } from "~/component/icon"
|
||||
|
||||
|
|
@ -34,9 +38,11 @@ export const queryLiteSubscription = query(async (workspaceID: string) => {
|
|||
timeMonthlyUpdated: LiteTable.timeMonthlyUpdated,
|
||||
timeCreated: LiteTable.timeCreated,
|
||||
lite: BillingTable.lite,
|
||||
region: WorkspaceTable.region,
|
||||
})
|
||||
.from(BillingTable)
|
||||
.innerJoin(LiteTable, eq(LiteTable.workspaceID, BillingTable.workspaceID))
|
||||
.innerJoin(WorkspaceTable, eq(WorkspaceTable.id, BillingTable.workspaceID))
|
||||
.where(and(eq(LiteTable.workspaceID, Actor.workspace()), isNull(LiteTable.timeDeleted)))
|
||||
.then((r) => r[0]),
|
||||
)
|
||||
|
|
@ -48,6 +54,8 @@ export const queryLiteSubscription = query(async (workspaceID: string) => {
|
|||
return {
|
||||
mine,
|
||||
useBalance: row.lite?.useBalance ?? false,
|
||||
region:
|
||||
row.region ?? (await Workspace.setDefaultRegion({ country: countryFromRequest(getRequestEvent()?.request) })),
|
||||
rollingUsage: Subscription.analyzeRollingUsage({
|
||||
limit: limits.rollingLimit,
|
||||
window: limits.rollingWindow,
|
||||
|
|
@ -128,6 +136,24 @@ const setLiteUseBalance = action(async (form: FormData) => {
|
|||
)
|
||||
}, "setLiteUseBalance")
|
||||
|
||||
const setGoProviderRouting = action(async (form: FormData) => {
|
||||
"use server"
|
||||
const workspaceID = form.get("workspaceID") as string | null
|
||||
if (!workspaceID) return { error: formError.workspaceRequired }
|
||||
const useChinaProviders = (form.get("useChinaProviders") as string | null) === "true"
|
||||
|
||||
return json(
|
||||
await withActor(
|
||||
() =>
|
||||
Workspace.update({ region: useChinaProviders ? ["us", "eu", "sg"] : ["us", "eu", "sg", "cn"] })
|
||||
.then(() => ({ error: undefined }))
|
||||
.catch((e) => ({ error: e.message as string })),
|
||||
workspaceID,
|
||||
),
|
||||
{ revalidate: queryLiteSubscription.key },
|
||||
)
|
||||
}, "go.providerRouting.set")
|
||||
|
||||
function LiteUsageItem(props: { label: string; usage: { usagePercent: number; resetInSec: number } }) {
|
||||
const i18n = useI18n()
|
||||
|
||||
|
|
@ -159,6 +185,7 @@ export function LiteSection(props: { lite: LiteSubscription | undefined }) {
|
|||
const checkoutAction = useAction(createLiteCheckoutUrl)
|
||||
const checkoutSubmission = useSubmission(createLiteCheckoutUrl)
|
||||
const useBalanceSubmission = useSubmission(setLiteUseBalance)
|
||||
const providerRoutingSubmission = useSubmission(setGoProviderRouting)
|
||||
const [store, setStore] = createStore({
|
||||
loading: undefined as undefined | "session" | "checkout" | "alipay" | "upi",
|
||||
showModal: false,
|
||||
|
|
@ -232,6 +259,28 @@ export function LiteSection(props: { lite: LiteSubscription | undefined }) {
|
|||
<span></span>
|
||||
</label>
|
||||
</form>
|
||||
{/*
|
||||
<div data-slot="providers-section">
|
||||
<div data-slot="providers-header">
|
||||
<h3>{i18n.t("workspace.lite.providers.title")}</h3>
|
||||
<p>{i18n.t("workspace.lite.providers.description")}</p>
|
||||
</div>
|
||||
<form action={setGoProviderRouting} method="post" data-slot="setting-row">
|
||||
<p>{i18n.t("workspace.lite.providers.useChina")}</p>
|
||||
<input type="hidden" name="workspaceID" value={params.id} />
|
||||
<input type="hidden" name="useChinaProviders" value={sub().region.includes("cn") ? "true" : "false"} />
|
||||
<label data-slot="toggle-label">
|
||||
<input
|
||||
type="checkbox"
|
||||
checked={sub().region.includes("cn")}
|
||||
disabled={providerRoutingSubmission.pending}
|
||||
onChange={(e) => e.currentTarget.form?.requestSubmit()}
|
||||
/>
|
||||
<span></span>
|
||||
</label>
|
||||
</form>
|
||||
</div>
|
||||
*/}
|
||||
</section>
|
||||
)}
|
||||
</Show>
|
||||
|
|
|
|||
|
|
@ -43,6 +43,7 @@ const updateWorkspace = action(async (form: FormData) => {
|
|||
.catch((e) => ({ error: e.message as string })),
|
||||
workspaceID,
|
||||
),
|
||||
{ revalidate: getWorkspaceInfo.key },
|
||||
)
|
||||
}, "workspace.update")
|
||||
|
||||
|
|
|
|||
|
|
@ -3,6 +3,7 @@ export class CreditsError extends Error {}
|
|||
export class MonthlyLimitError extends Error {}
|
||||
export class UserLimitError extends Error {}
|
||||
export class ModelError extends Error {}
|
||||
export class RegionError extends Error {}
|
||||
|
||||
class LimitError extends Error {
|
||||
retryAfter?: number
|
||||
|
|
|
|||
|
|
@ -21,6 +21,7 @@ import {
|
|||
MonthlyLimitError,
|
||||
UserLimitError,
|
||||
ModelError,
|
||||
RegionError,
|
||||
RateLimitError,
|
||||
FreeUsageLimitError,
|
||||
GoUsageLimitError,
|
||||
|
|
@ -49,6 +50,8 @@ import { createModelTpmLimiter } from "./modelTpmLimiter"
|
|||
import { createModelTpsLimiter } from "./modelTpsLimiter"
|
||||
import { createProviderBudgetTracker } from "./providerBudgetTracker"
|
||||
import { accumulateUsage, HOT_WORKSPACES } from "./usageBatcher"
|
||||
import { Workspace } from "@opencode-ai/console-core/workspace.js"
|
||||
import { countryFromRequest } from "~/lib/request-country"
|
||||
|
||||
type ZenData = Awaited<ReturnType<typeof ZenData.list>>
|
||||
type RetryOptions = {
|
||||
|
|
@ -125,6 +128,24 @@ export async function handler(
|
|||
: createKeyRateLimiter(modelInfo.id, modelInfo.rateLimit, zenApiKey, input.request)
|
||||
await rateLimiter?.check()
|
||||
const authInfo = await authenticate(modelInfo, zenApiKey)
|
||||
const allowedRegions = authInfo?.region
|
||||
? authInfo.region
|
||||
: await (async () => {
|
||||
if (!authInfo) return
|
||||
return Actor.provide("system", { workspaceID: authInfo.workspaceID }, () =>
|
||||
Workspace.setDefaultRegion({ country: countryFromRequest(input.request) }),
|
||||
)
|
||||
})()
|
||||
/*
|
||||
if (true) {
|
||||
if (!allowedRegions?.includes("unavailable"))
|
||||
throw new RegionError(
|
||||
t("zen.api.error.regionNotAllowed", {
|
||||
consoleGoUrl: `https://opencode.ai/workspace/${authInfo.workspaceID}/go`,
|
||||
}),
|
||||
)
|
||||
}
|
||||
*/
|
||||
const stickyId = sessionId ? sessionId : (authInfo?.workspaceID ?? ip)
|
||||
const stickyTracker = createStickyTracker(modelInfo.id, modelInfo.stickyProvider, stickyId)
|
||||
const stickyProvider = await stickyTracker?.get()
|
||||
|
|
@ -137,7 +158,7 @@ export async function handler(
|
|||
const providerBudgetTracker = createProviderBudgetTracker(
|
||||
modelInfo.providers.map((provider) => ({ ...zenData.providers[provider.id], ...provider })),
|
||||
)
|
||||
const providerBudgetUsage = await providerBudgetTracker?.check()
|
||||
const providerBudget = await providerBudgetTracker?.check()
|
||||
|
||||
const retriableRequest = async (retry: RetryOptions = { excludeProviders: [], retryCount: 0 }) => {
|
||||
const providerInfo = selectProvider(
|
||||
|
|
@ -151,7 +172,7 @@ export async function handler(
|
|||
stickyProvider,
|
||||
modelTpmLimits,
|
||||
modelTpsLimits,
|
||||
providerBudgetUsage,
|
||||
providerBudget,
|
||||
)
|
||||
validateModelSettings(billingSource, authInfo)
|
||||
updateProviderKey(authInfo, providerInfo)
|
||||
|
|
@ -201,7 +222,10 @@ export async function handler(
|
|||
if (v === "$model") return headers.set(k, model)
|
||||
if (v === "$request") return headers.set(k, requestId)
|
||||
if (v === "$project") return headers.set(k, projectId)
|
||||
if (v === "$workspace" && authInfo?.workspaceID) return headers.set(k, authInfo.workspaceID)
|
||||
if (v === "$workspace") {
|
||||
if (authInfo?.workspaceID) headers.set(k, authInfo.workspaceID)
|
||||
return
|
||||
}
|
||||
headers.set(k, v)
|
||||
})
|
||||
headers.delete("host")
|
||||
|
|
@ -213,6 +237,9 @@ export async function handler(
|
|||
return headers
|
||||
})(),
|
||||
body: reqBody,
|
||||
// Propagate caller disconnects to the upstream provider request so
|
||||
// abandoned Console requests do not leave orphaned inference work open.
|
||||
signal: input.request.signal,
|
||||
})
|
||||
|
||||
if (providerInfo.id.startsWith("console.")) {
|
||||
|
|
@ -281,7 +308,7 @@ export async function handler(
|
|||
const costInfo = calculateCost(modelInfo, usageInfo)
|
||||
await trialLimiter?.track(usageInfo)
|
||||
await modelTpmLimiter?.track(providerInfo.id, providerInfo.model, usageInfo)
|
||||
await providerBudgetTracker?.track(providerInfo.id, costInfo.totalCostInCent)
|
||||
await providerBudgetTracker?.track(providerInfo.id, providerInfo.budgetPriority, costInfo.totalCostInCent)
|
||||
await trackUsage(sessionId, billingSource, authInfo, modelInfo, providerInfo, usageInfo, costInfo)
|
||||
await reload(billingSource, authInfo, costInfo)
|
||||
json.cost = calculateOccurredCost(billingSource, costInfo)
|
||||
|
|
@ -308,9 +335,10 @@ export async function handler(
|
|||
const streamConverter = createStreamPartConverter(providerInfo.format, opts.format)
|
||||
const usageParser = providerInfo.createUsageParser()
|
||||
const binaryDecoder = providerInfo.createBinaryStreamDecoder()
|
||||
let reader: ReadableStreamDefaultReader<Uint8Array> | undefined
|
||||
const stream = new ReadableStream({
|
||||
start(c) {
|
||||
const reader = res.body?.getReader()
|
||||
reader = res.body?.getReader()
|
||||
const decoder = new TextDecoder()
|
||||
const encoder = new TextEncoder()
|
||||
|
||||
|
|
@ -342,7 +370,11 @@ export async function handler(
|
|||
timestampLastByte,
|
||||
usageInfo,
|
||||
)
|
||||
await providerBudgetTracker?.track(providerInfo.id, costInfo.totalCostInCent)
|
||||
await providerBudgetTracker?.track(
|
||||
providerInfo.id,
|
||||
providerInfo.budgetPriority,
|
||||
costInfo.totalCostInCent,
|
||||
)
|
||||
await trackUsage(sessionId, billingSource, authInfo, modelInfo, providerInfo, usageInfo, costInfo)
|
||||
await reload(billingSource, authInfo, costInfo)
|
||||
const cost = calculateOccurredCost(billingSource, costInfo)
|
||||
|
|
@ -392,6 +424,11 @@ export async function handler(
|
|||
|
||||
return pump()
|
||||
},
|
||||
cancel() {
|
||||
// When the downstream caller stops reading, release the upstream
|
||||
// response body instead of keeping the provider/inference stream alive.
|
||||
return reader?.cancel()
|
||||
},
|
||||
})
|
||||
return new Response(stream, {
|
||||
status: resStatus,
|
||||
|
|
@ -399,6 +436,15 @@ export async function handler(
|
|||
headers: resHeaders,
|
||||
})
|
||||
} catch (error: any) {
|
||||
// The caller disconnected before we finished. Because the outbound provider
|
||||
// request shares input.request.signal, an aborted caller surfaces here as an
|
||||
// AbortError. There is no client left to receive a body, so skip the error
|
||||
// metric and 500 and return a quiet client-closed response.
|
||||
if (input.request.signal.aborted || error?.name === "AbortError") {
|
||||
logger.debug("REQUEST ABORTED BY CALLER")
|
||||
return new Response(null, { status: 499 })
|
||||
}
|
||||
|
||||
logger.metric({
|
||||
"error.type": error.constructor.name,
|
||||
"error.message": error.message,
|
||||
|
|
@ -412,6 +458,15 @@ export async function handler(
|
|||
} catch {}
|
||||
}
|
||||
|
||||
if (error instanceof RegionError)
|
||||
return new Response(
|
||||
JSON.stringify({
|
||||
type: "error",
|
||||
error: { type: error.constructor.name, message: error.message },
|
||||
}),
|
||||
{ status: 403 },
|
||||
)
|
||||
|
||||
// Note: both top level "type" and "error.type" fields are used by the @ai-sdk/anthropic client to render the error message.
|
||||
if (
|
||||
error instanceof AuthError ||
|
||||
|
|
@ -509,7 +564,12 @@ export async function handler(
|
|||
stickyProviderId: string | undefined,
|
||||
modelTpmLimits: Record<string, number> | undefined,
|
||||
modelTpsLimits: Record<string, { qualify: number; unqualify: number }> | undefined,
|
||||
providerBudgetUsage: Record<string, number> | undefined,
|
||||
providerBudget:
|
||||
| {
|
||||
qualify: (providerId: string, priority: number) => boolean
|
||||
prefer: (providerId: string, priority: number) => boolean
|
||||
}
|
||||
| undefined,
|
||||
) {
|
||||
const modelProvider = (() => {
|
||||
// Byok is top priority b/c if user set their own API key, we should use it
|
||||
|
|
@ -527,67 +587,69 @@ export async function handler(
|
|||
}))
|
||||
}
|
||||
|
||||
if (retry.retryCount !== MAX_FAILOVER_RETRIES) {
|
||||
let topPriority = Infinity
|
||||
const providers = allProviders
|
||||
.filter((provider) => provider.weight !== 0)
|
||||
.filter((provider) => !retry.excludeProviders.includes(provider.id))
|
||||
.filter((provider) => {
|
||||
if (provider.budgetMode !== "fill") return true
|
||||
const budget = zenData.providers[provider.id]?.budget
|
||||
if (budget === undefined) return false
|
||||
return (providerBudgetUsage?.[provider.id] ?? 0) < centsToMicroCents(budget * 100)
|
||||
})
|
||||
.filter((provider) => {
|
||||
if (!provider.tpmLimit) return true
|
||||
const usage = modelTpmLimits?.[`${provider.id}/${provider.model}`] ?? 0
|
||||
return usage < provider.tpmLimit * 1_000_000
|
||||
})
|
||||
.filter((provider) => {
|
||||
if (!provider.tpsGoal) return true
|
||||
const tps = modelTpsLimits?.[`${provider.id}/${provider.model}/${provider.tpsGoal}`] ?? {
|
||||
qualify: 0,
|
||||
unqualify: 0,
|
||||
}
|
||||
const isLowTps = tps.qualify + tps.unqualify > 10 && tps.qualify < tps.unqualify
|
||||
return !isLowTps
|
||||
})
|
||||
.map((provider) => {
|
||||
topPriority = Math.min(topPriority, provider.priority)
|
||||
return provider
|
||||
})
|
||||
.filter((p) => p.priority <= topPriority)
|
||||
.flatMap((provider) => Array<typeof provider>(provider.weight).fill(provider))
|
||||
// Use fallback provider if max retries reached
|
||||
const fallbackProvider = allProviders.find((provider) => provider.id === modelInfo.fallbackProvider)
|
||||
if (retry.retryCount === MAX_FAILOVER_RETRIES) return fallbackProvider
|
||||
|
||||
// Use the last 4 characters of session ID to select a provider
|
||||
let h = 0
|
||||
const l = stickyId.length
|
||||
for (let i = l - 4; i < l; i++) {
|
||||
h = (h * 31 + stickyId.charCodeAt(i)) | 0 // 32-bit int
|
||||
}
|
||||
const index = (h >>> 0) % providers.length // make unsigned + range 0..length-1
|
||||
const provider = providers[index || 0]
|
||||
let topPriority = Infinity
|
||||
const providers = allProviders
|
||||
.filter((provider) => provider.weight !== 0)
|
||||
.filter((provider) => !retry.excludeProviders.includes(provider.id))
|
||||
.filter((provider) => {
|
||||
if (provider.budgetPriority === undefined) return true
|
||||
if (!providerBudget) return true
|
||||
return providerBudget.qualify(provider.id, provider.budgetPriority)
|
||||
})
|
||||
.filter((provider) => {
|
||||
if (!provider.tpmLimit) return true
|
||||
const usage = modelTpmLimits?.[`${provider.id}/${provider.model}`] ?? 0
|
||||
return usage < provider.tpmLimit * 1_000_000
|
||||
})
|
||||
.filter((provider) => {
|
||||
if (!provider.tpsGoal) return true
|
||||
const tps = modelTpsLimits?.[`${provider.id}/${provider.model}/${provider.tpsGoal}`] ?? {
|
||||
qualify: 0,
|
||||
unqualify: 0,
|
||||
}
|
||||
const isLowTps = tps.qualify + tps.unqualify > 10 && tps.qualify < tps.unqualify
|
||||
return !isLowTps
|
||||
})
|
||||
.map((provider) => {
|
||||
topPriority = Math.min(topPriority, provider.priority)
|
||||
return provider
|
||||
})
|
||||
.filter((p) => p.priority <= topPriority)
|
||||
.flatMap((provider) => Array<typeof provider>(provider.weight).fill(provider))
|
||||
|
||||
// sticky provider does not exist => use selected provider
|
||||
if (!stickyProviderId) return provider
|
||||
const stickProvider = allProviders.find((provider) => provider.id === stickyProviderId)
|
||||
if (!stickProvider) return provider
|
||||
// Use the last 4 characters of session ID to select a provider
|
||||
let h = 0
|
||||
const l = stickyId.length
|
||||
for (let i = l - 4; i < l; i++) {
|
||||
h = (h * 31 + stickyId.charCodeAt(i)) | 0 // 32-bit int
|
||||
}
|
||||
const index = (h >>> 0) % providers.length // make unsigned + range 0..length-1
|
||||
const provider = providers[index || 0] ?? fallbackProvider
|
||||
|
||||
// stick provider exists + selected provider is API type => use sticky provider
|
||||
if (!provider.tpsGoal) return stickProvider
|
||||
// sticky provider does not exist => use selected provider
|
||||
if (!stickyProviderId) return provider
|
||||
const stickProvider = allProviders.find((provider) => provider.id === stickyProviderId)
|
||||
if (!stickProvider) return provider
|
||||
|
||||
// stick provier exists + selected provider is GPU type + GPU not idle => use selected provider
|
||||
const preferBudgetProvider =
|
||||
provider.budgetPriority !== undefined && providerBudget?.prefer(provider.id, provider.budgetPriority)
|
||||
|
||||
const preferTpsProvider = (() => {
|
||||
if (!provider.tpsGoal) return false
|
||||
const tps = modelTpsLimits?.[`${provider.id}/${provider.model}/${provider.tpsGoal}`] ?? {
|
||||
qualify: 0,
|
||||
unqualify: 0,
|
||||
}
|
||||
if (tps.qualify <= tps.unqualify * 3) return stickProvider
|
||||
return tps.qualify > tps.unqualify * 3
|
||||
})()
|
||||
|
||||
return provider
|
||||
}
|
||||
if (!preferBudgetProvider && !preferTpsProvider) return stickProvider
|
||||
|
||||
// fallback provider
|
||||
return allProviders.find((provider) => provider.id === modelInfo.fallbackProvider)
|
||||
return provider
|
||||
})()
|
||||
|
||||
if (!modelProvider) throw new ModelError(t("zen.api.error.noProviderAvailable"))
|
||||
|
|
@ -624,7 +686,10 @@ export async function handler(
|
|||
tx
|
||||
.select({
|
||||
apiKey: KeyTable.id,
|
||||
workspaceID: KeyTable.workspaceID,
|
||||
workspace: {
|
||||
id: WorkspaceTable.id,
|
||||
region: WorkspaceTable.region,
|
||||
},
|
||||
billing: {
|
||||
balance: BillingTable.balance,
|
||||
paymentMethodID: BillingTable.paymentMethodID,
|
||||
|
|
@ -702,13 +767,13 @@ export async function handler(
|
|||
if (
|
||||
modelInfo.id.startsWith("alpha-") &&
|
||||
Resource.App.stage === "production" &&
|
||||
!ADMIN_WORKSPACES.includes(data.workspaceID)
|
||||
!ADMIN_WORKSPACES.includes(data.workspace.id)
|
||||
)
|
||||
throw new AuthError(t("zen.api.error.modelNotSupported", { model: modelInfo.id }))
|
||||
|
||||
logger.metric({
|
||||
api_key: data.apiKey,
|
||||
workspace: data.workspaceID,
|
||||
workspace: data.workspace.id,
|
||||
user_id: data.user.id,
|
||||
...(() => {
|
||||
if (data.billing.subscription)
|
||||
|
|
@ -725,13 +790,14 @@ export async function handler(
|
|||
|
||||
return {
|
||||
apiKeyId: data.apiKey,
|
||||
workspaceID: data.workspaceID,
|
||||
workspaceID: data.workspace.id,
|
||||
region: data.workspace.region,
|
||||
billing: data.billing,
|
||||
user: data.user,
|
||||
black: data.black,
|
||||
lite: data.lite,
|
||||
provider: data.provider,
|
||||
isFree: ADMIN_WORKSPACES.includes(data.workspaceID),
|
||||
isFree: ADMIN_WORKSPACES.includes(data.workspace.id),
|
||||
isDisabled: !!data.timeDisabled,
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -20,7 +20,7 @@ export function createModelTpsLimiter(providers: { id: string; model: string; tp
|
|||
)
|
||||
const now = Date.now()
|
||||
const currInterval = toInterval(new Date(now))
|
||||
const prevInterval = toInterval(new Date(now - 60 * 1000))
|
||||
const prevInterval = toInterval(new Date(now - 60_000))
|
||||
|
||||
return {
|
||||
check: async () => {
|
||||
|
|
|
|||
|
|
@ -2,50 +2,148 @@ import { centsToMicroCents } from "@opencode-ai/console-core/util/price.js"
|
|||
import { buildRateLimitKey, getRedis } from "./redis"
|
||||
import { logger } from "./logger"
|
||||
|
||||
// Per-provider, per-minute budget with priorities. The budget belongs to a
|
||||
// provider and is shared across every model that routes to it. Each model's
|
||||
// provider entry carries a `budgetPriority`: priority 1 ("always") routes
|
||||
// unconditionally, while higher priorities ("fill") only route while the provider's
|
||||
// current-minute spend through that priority is still under budget.
|
||||
//
|
||||
// Spend is tracked per (provider, priority, minute) so a fill priority can yield its
|
||||
// leftover headroom to the next priority down. The previous minute is also read so
|
||||
// higher priorities can reserve the next minute's budget first.
|
||||
export function createProviderBudgetTracker(
|
||||
providers: {
|
||||
id: string
|
||||
budget?: number
|
||||
budgetContribution?: number
|
||||
budgetMode?: "always" | "fill"
|
||||
budgetPriority?: number
|
||||
}[],
|
||||
) {
|
||||
const tracked = providers.filter(
|
||||
(provider) => provider.budget !== undefined && provider.budgetContribution !== undefined,
|
||||
(provider) =>
|
||||
provider.budget !== undefined &&
|
||||
provider.budgetContribution !== undefined &&
|
||||
provider.budgetPriority !== undefined,
|
||||
)
|
||||
if (tracked.length === 0) return undefined
|
||||
|
||||
const interval = new Date()
|
||||
.toISOString()
|
||||
.replace(/[^0-9]/g, "")
|
||||
.substring(0, 12)
|
||||
const intervalAt = (date: Date) =>
|
||||
date
|
||||
.toISOString()
|
||||
.replace(/[^0-9]/g, "")
|
||||
.substring(0, 12)
|
||||
const now = new Date()
|
||||
const currInterval = intervalAt(now)
|
||||
const prevInterval = intervalAt(new Date(now.getTime() - 60_000))
|
||||
|
||||
const redis = getRedis()
|
||||
const keys = Object.fromEntries(
|
||||
tracked.map((provider) => [provider.id, buildRateLimitKey("provider-budget", provider.id, interval)]),
|
||||
)
|
||||
let budgetUsage: Record<string, number> = {}
|
||||
const key = (providerId: string, priority: number, withInterval: string) =>
|
||||
buildRateLimitKey("provider-budget", `${providerId}:${priority}`, withInterval)
|
||||
|
||||
const budgetByProvider = tracked.reduce<Record<string, number>>((acc, provider) => {
|
||||
acc[provider.id] = provider.budget!
|
||||
return acc
|
||||
}, {})
|
||||
|
||||
const maxPriorityByProvider = tracked.reduce<Record<string, number>>((acc, provider) => {
|
||||
acc[provider.id] = Math.max(acc[provider.id] ?? 0, provider.budgetPriority!)
|
||||
return acc
|
||||
}, {})
|
||||
|
||||
// Effective budget in micro-cents per provider/priority, computed in check()
|
||||
// from the configured budget minus previous-minute usage from higher priorities.
|
||||
let effectiveBudget: Record<string, Record<number, number>> = {}
|
||||
// Cumulative current-minute spend through each priority, per provider.
|
||||
let spentThroughPriority: Record<string, Record<number, number>> = {}
|
||||
let previousSpentThroughPriority: Record<string, Record<number, number>> = {}
|
||||
|
||||
return {
|
||||
// Returns whether a provider at a given priority still has budget headroom.
|
||||
// Priority 1 always qualifies; higher priorities qualify only while everything through
|
||||
// the current priority hasn't already filled the previous-minute adjusted
|
||||
// budget.
|
||||
check: async () => {
|
||||
const ids = tracked.map((provider) => provider.id)
|
||||
if (ids.length === 0) return {}
|
||||
const values = await redis.mget<(string | number | null)[]>(ids.map((id) => keys[id]))
|
||||
budgetUsage = Object.fromEntries(ids.map((id, index) => [id, Number(values[index] ?? 0)]))
|
||||
return budgetUsage
|
||||
const reads = Object.entries(maxPriorityByProvider).flatMap(([providerId, maxPriority]) =>
|
||||
Array.from({ length: maxPriority }, (_, index) => index + 1).flatMap((priority) => [
|
||||
{ providerId, priority, interval: currInterval, prev: false },
|
||||
{ providerId, priority, interval: prevInterval, prev: true },
|
||||
]),
|
||||
)
|
||||
const values = await redis.mget<(string | number | null)[]>(
|
||||
reads.map((r) => key(r.providerId, r.priority, r.interval)),
|
||||
)
|
||||
|
||||
const current: Record<string, Record<number, number>> = {}
|
||||
const previous: Record<string, Record<number, number>> = {}
|
||||
reads.forEach((r, index) => {
|
||||
const amount = Number(values[index] ?? 0)
|
||||
if (r.prev) {
|
||||
previous[r.providerId] ??= {}
|
||||
previous[r.providerId][r.priority] = amount
|
||||
return
|
||||
}
|
||||
current[r.providerId] ??= {}
|
||||
current[r.providerId][r.priority] = amount
|
||||
})
|
||||
|
||||
effectiveBudget = {}
|
||||
spentThroughPriority = {}
|
||||
previousSpentThroughPriority = {}
|
||||
Object.entries(maxPriorityByProvider).forEach(([providerId, maxPriority]) => {
|
||||
const providerBudget = budgetByProvider[providerId]
|
||||
if (providerBudget === undefined) return
|
||||
const budget = centsToMicroCents(providerBudget * 100)
|
||||
|
||||
let currentRunning = 0
|
||||
let previousRunning = 0
|
||||
effectiveBudget[providerId] = {}
|
||||
spentThroughPriority[providerId] = {}
|
||||
previousSpentThroughPriority[providerId] = {}
|
||||
Array.from({ length: maxPriority }, (_, index) => index + 1).forEach((priority) => {
|
||||
currentRunning += current[providerId]?.[priority] ?? 0
|
||||
effectiveBudget[providerId][priority] = Math.max(0, budget - previousRunning)
|
||||
previousRunning += previous[providerId]?.[priority] ?? 0
|
||||
spentThroughPriority[providerId][priority] = currentRunning
|
||||
previousSpentThroughPriority[providerId][priority] = previousRunning
|
||||
})
|
||||
})
|
||||
|
||||
return {
|
||||
// Priority 1 is unconditional. Higher priorities gate on the spend through
|
||||
// the current priority against the effective budget.
|
||||
qualify: (providerId: string, priority: number) => {
|
||||
if (priority <= 1) return true
|
||||
const budget = effectiveBudget[providerId]?.[priority]
|
||||
if (budget === undefined) return false
|
||||
const spentThroughCurrentPriority = spentThroughPriority[providerId]?.[priority] ?? 0
|
||||
return spentThroughCurrentPriority < budget
|
||||
},
|
||||
prefer: (providerId: string, priority: number) => {
|
||||
const providerBudget = budgetByProvider[providerId]
|
||||
if (providerBudget === undefined) return false
|
||||
const budget = centsToMicroCents(providerBudget * 100)
|
||||
const previousUsage = previousSpentThroughPriority[providerId]?.[priority]
|
||||
if (previousUsage === undefined) return false
|
||||
return previousUsage < budget * 0.8
|
||||
},
|
||||
}
|
||||
},
|
||||
track: async (provider: string, costInCent: number) => {
|
||||
const config = tracked.find((item) => item.id === provider)
|
||||
track: async (provider: string, priority: number | undefined, costInCent: number) => {
|
||||
if (priority === undefined) return
|
||||
const config = tracked.find((item) => item.id === provider && item.budgetPriority === priority)
|
||||
if (!config) return
|
||||
if (config.budgetContribution === undefined) return
|
||||
const cost = centsToMicroCents(costInCent * config.budgetContribution)
|
||||
if (cost <= 0) return
|
||||
const redisKey = key(provider, priority, currInterval)
|
||||
const pipeline = redis.pipeline()
|
||||
pipeline.incrby(keys[provider], cost)
|
||||
pipeline.expire(keys[provider], 120)
|
||||
pipeline.incrby(redisKey, cost)
|
||||
// Keep two minutes so the previous interval is readable for budget adjustment.
|
||||
pipeline.expire(redisKey, 120)
|
||||
await pipeline.exec()
|
||||
logger.metric({
|
||||
"provider.budget_usage": budgetUsage[provider] + cost,
|
||||
"model.budget_usage": cost,
|
||||
"provider.budget_usage": cost,
|
||||
"provider.budget_priority": priority,
|
||||
})
|
||||
},
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue