john
441ba155cc
Flat 2分/1k input overcharged the DeepSeek relay ~20x (verified on 100: H5-charged / goose accumulated_cost stayed at ~20x across every session). The flat rate is decoupled from real provider cost, so it overcharges cheap providers and would underbill expensive ones. Extend the H5_USE_BACKEND_COST path with H5_MARGIN_MULTIPLIER: charge upstream accumulated_cost(USD) × rate × margin, which tracks the provider automatically. Falls back to the token path when the upstream omits accumulatedCost, so the change is safe to ship dormant. See docs/h5-metering-gateway.md for the metering-gateway design. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
80 lines
3.7 KiB
JavaScript
80 lines
3.7 KiB
JavaScript
export function loadBillingConfig() {
|
||
// 默认按人民币分（CNY cents）计费；仅当 H5_USE_BACKEND_COST=1 时才用上游 USD 成本换算。
|
||
const useBackendCost = process.env.H5_USE_BACKEND_COST === '1';
|
||
// 成本模式下的毛利倍数：最终扣费 = 上游真实成本(USD) × 汇率 × marginMultiplier。
|
||
// 默认 1（按成本价卖，零毛利）——启用 useBackendCost 时务必显式设置目标倍数。
|
||
const marginMultiplier = Number(process.env.H5_MARGIN_MULTIPLIER ?? 1);
|
||
return {
|
||
useBackendCost,
|
||
usdCnyRate: Number(process.env.H5_USD_CNY_RATE ?? 7.2),
|
||
marginMultiplier: Number.isFinite(marginMultiplier) && marginMultiplier > 0 ? marginMultiplier : 1,
|
||
inputCentsPer1k: Number(process.env.H5_BILL_INPUT_CENTS_PER_1K ?? 2),
|
||
outputCentsPer1k: Number(process.env.H5_BILL_OUTPUT_CENTS_PER_1K ?? 6),
|
||
minBillCents: Number(process.env.H5_MIN_BILL_CENTS ?? 1),
|
||
};
|
||
}
|
||
|
||
export function normalizeTokenState(raw) {
|
||
if (!raw || typeof raw !== 'object') {
|
||
return {
|
||
inputTokens: 0,
|
||
outputTokens: 0,
|
||
totalTokens: 0,
|
||
accumulatedInputTokens: 0,
|
||
accumulatedOutputTokens: 0,
|
||
accumulatedTotalTokens: 0,
|
||
accumulatedCost: null,
|
||
};
|
||
}
|
||
return {
|
||
inputTokens: Number(raw.inputTokens ?? raw.input_tokens ?? 0),
|
||
outputTokens: Number(raw.outputTokens ?? raw.output_tokens ?? 0),
|
||
totalTokens: Number(raw.totalTokens ?? raw.total_tokens ?? 0),
|
||
accumulatedInputTokens: Number(
|
||
raw.accumulatedInputTokens ?? raw.accumulated_input_tokens ?? raw.inputTokens ?? 0,
|
||
),
|
||
accumulatedOutputTokens: Number(
|
||
raw.accumulatedOutputTokens ?? raw.accumulated_output_tokens ?? raw.outputTokens ?? 0,
|
||
),
|
||
accumulatedTotalTokens: Number(
|
||
raw.accumulatedTotalTokens ?? raw.accumulated_total_tokens ?? raw.totalTokens ?? 0,
|
||
),
|
||
accumulatedCost:
|
||
raw.accumulatedCost ?? raw.accumulated_cost ?? null,
|
||
};
|
||
}
|
||
|
||
export function computeDeltaCostCents(previous, current, config = loadBillingConfig()) {
|
||
const prevCost =
|
||
previous?.lastAccumulatedCost == null ? null : Number(previous.lastAccumulatedCost);
|
||
const currCost =
|
||
current.accumulatedCost == null ? null : Number(current.accumulatedCost);
|
||
|
||
// 成本模式：按上游真实成本(USD)增量 × 汇率 × 毛利倍数扣费，自动跟随 provider/模型，
|
||
// 无需为每个模型重调 token 单价。仅在上游确实回传 accumulatedCost 时生效；否则回退 token 路径。
|
||
const margin = config.marginMultiplier ?? 1;
|
||
if (config.useBackendCost && prevCost != null && currCost != null && currCost >= prevCost) {
|
||
const deltaUsd = currCost - prevCost;
|
||
if (deltaUsd <= 0) return 0;
|
||
return Math.max(config.minBillCents, Math.ceil(deltaUsd * config.usdCnyRate * 100 * margin));
|
||
}
|
||
if (config.useBackendCost && prevCost == null && currCost != null && currCost > 0) {
|
||
return Math.max(config.minBillCents, Math.ceil(currCost * config.usdCnyRate * 100 * margin));
|
||
}
|
||
|
||
// 默认/回退路径：按 Token 增量 × 人民币单价（分/1k tokens）扣费。
|
||
// 注意：flat 单价与 provider 真实成本脱钩（对 DeepSeek 中继约 20× 超收），
|
||
// 生产应启用 useBackendCost + marginMultiplier 走成本模式，详见 docs/h5-metering-gateway.md。
|
||
|
||
const prevIn = Number(previous?.lastInputTokens ?? 0);
|
||
const prevOut = Number(previous?.lastOutputTokens ?? 0);
|
||
const deltaIn = Math.max(0, current.accumulatedInputTokens - prevIn);
|
||
const deltaOut = Math.max(0, current.accumulatedOutputTokens - prevOut);
|
||
if (deltaIn === 0 && deltaOut === 0) return 0;
|
||
|
||
const raw =
|
||
(deltaIn / 1000) * config.inputCentsPer1k + (deltaOut / 1000) * config.outputCentsPer1k;
|
||
if (raw <= 0) return 0;
|
||
return Math.max(config.minBillCents, Math.ceil(raw));
|
||
}
|