forked from boos/BoosAPI
Merge pull request #159 from Hureru/feat/account-level-proxy
feat(proxy): account-level proxy URL override
This commit is contained in:
Cita
@@ -9,6 +9,7 @@ import { resolveProxyUsageWithSelfLogFallback } from '../../services/proxyUsageF
|
||||
import { mergeProxyUsage, parseProxyUsage, pullSseDataEvents } from '../../services/proxyUsageParser.js';
|
||||
import { ensureModelAllowedForDownstreamKey, getDownstreamRoutingPolicy, recordDownstreamCostUsage } from './downstreamPolicy.js';
|
||||
import { withSiteRecordProxyRequestInit } from '../../services/siteProxy.js';
|
||||
import { getProxyUrlFromExtraConfig } from '../../services/accountExtraConfig.js';
|
||||
import { composeProxyLogMessage } from './logPathMeta.js';
|
||||
import { formatUtcSqlDateTime } from '../../services/localTimeService.js';
|
||||
import { detectProxyFailure } from './proxyFailureJudge.js';
|
||||
@@ -75,7 +76,7 @@ export async function completionsProxyRoute(app: FastifyInstance) {
|
||||
'Authorization': `Bearer ${selected.tokenValue}`,
|
||||
},
|
||||
body: JSON.stringify(forwardBody),
|
||||
}));
|
||||
}, getProxyUrlFromExtraConfig(selected.account.extraConfig)));
|
||||
|
||||
if (!upstream.ok) {
|
||||
const errText = await upstream.text().catch(() => 'unknown error');
|
||||
|
||||
@@ -1,53 +1,54 @@
|
||||
import { FastifyInstance, FastifyReply, FastifyRequest } from 'fastify';
|
||||
import { fetch } from 'undici';
|
||||
import { tokenRouter } from '../../services/tokenRouter.js';
|
||||
import { refreshModelsAndRebuildRoutes } from '../../services/modelService.js';
|
||||
import { reportProxyAllFailed, reportTokenExpired } from '../../services/alertService.js';
|
||||
import { isTokenExpiredError } from '../../services/alertRules.js';
|
||||
import { shouldRetryProxyRequest } from '../../services/proxyRetryPolicy.js';
|
||||
import { resolveProxyUsageWithSelfLogFallback } from '../../services/proxyUsageFallbackService.js';
|
||||
import { parseProxyUsage } from '../../services/proxyUsageParser.js';
|
||||
import { ensureModelAllowedForDownstreamKey, getDownstreamRoutingPolicy, recordDownstreamCostUsage } from './downstreamPolicy.js';
|
||||
import { withSiteRecordProxyRequestInit } from '../../services/siteProxy.js';
|
||||
import { composeProxyLogMessage } from './logPathMeta.js';
|
||||
import { formatUtcSqlDateTime } from '../../services/localTimeService.js';
|
||||
import { resolveProxyLogBilling } from './proxyBilling.js';
|
||||
import { getProxyAuthContext } from '../../middleware/auth.js';
|
||||
import { buildUpstreamUrl } from './upstreamUrl.js';
|
||||
import { detectDownstreamClientContext, type DownstreamClientContext } from './downstreamClientContext.js';
|
||||
import { insertProxyLog } from '../../services/proxyLogStore.js';
|
||||
import { tokenRouter } from '../../services/tokenRouter.js';
|
||||
import { refreshModelsAndRebuildRoutes } from '../../services/modelService.js';
|
||||
import { reportProxyAllFailed, reportTokenExpired } from '../../services/alertService.js';
|
||||
import { isTokenExpiredError } from '../../services/alertRules.js';
|
||||
import { shouldRetryProxyRequest } from '../../services/proxyRetryPolicy.js';
|
||||
import { resolveProxyUsageWithSelfLogFallback } from '../../services/proxyUsageFallbackService.js';
|
||||
import { parseProxyUsage } from '../../services/proxyUsageParser.js';
|
||||
import { ensureModelAllowedForDownstreamKey, getDownstreamRoutingPolicy, recordDownstreamCostUsage } from './downstreamPolicy.js';
|
||||
import { withSiteRecordProxyRequestInit } from '../../services/siteProxy.js';
|
||||
import { getProxyUrlFromExtraConfig } from '../../services/accountExtraConfig.js';
|
||||
import { composeProxyLogMessage } from './logPathMeta.js';
|
||||
import { formatUtcSqlDateTime } from '../../services/localTimeService.js';
|
||||
import { resolveProxyLogBilling } from './proxyBilling.js';
|
||||
import { getProxyAuthContext } from '../../middleware/auth.js';
|
||||
import { buildUpstreamUrl } from './upstreamUrl.js';
|
||||
import { detectDownstreamClientContext, type DownstreamClientContext } from './downstreamClientContext.js';
|
||||
import { insertProxyLog } from '../../services/proxyLogStore.js';
|
||||
|
||||
const MAX_RETRIES = 2;
|
||||
|
||||
export async function embeddingsProxyRoute(app: FastifyInstance) {
|
||||
app.post('/v1/embeddings', async (request: FastifyRequest, reply: FastifyReply) => {
|
||||
const body = request.body as any;
|
||||
const requestedModel = body?.model;
|
||||
if (!requestedModel) {
|
||||
return reply.code(400).send({ error: { message: 'model is required', type: 'invalid_request_error' } });
|
||||
}
|
||||
if (!await ensureModelAllowedForDownstreamKey(request, reply, requestedModel)) return;
|
||||
const downstreamPolicy = getDownstreamRoutingPolicy(request);
|
||||
const downstreamApiKeyId = getProxyAuthContext(request)?.keyId ?? null;
|
||||
const downstreamPath = '/v1/embeddings';
|
||||
const clientContext = detectDownstreamClientContext({
|
||||
downstreamPath,
|
||||
headers: request.headers as Record<string, unknown>,
|
||||
body,
|
||||
});
|
||||
|
||||
const excludeChannelIds: number[] = [];
|
||||
let retryCount = 0;
|
||||
const requestedModel = body?.model;
|
||||
if (!requestedModel) {
|
||||
return reply.code(400).send({ error: { message: 'model is required', type: 'invalid_request_error' } });
|
||||
}
|
||||
if (!await ensureModelAllowedForDownstreamKey(request, reply, requestedModel)) return;
|
||||
const downstreamPolicy = getDownstreamRoutingPolicy(request);
|
||||
const downstreamApiKeyId = getProxyAuthContext(request)?.keyId ?? null;
|
||||
const downstreamPath = '/v1/embeddings';
|
||||
const clientContext = detectDownstreamClientContext({
|
||||
downstreamPath,
|
||||
headers: request.headers as Record<string, unknown>,
|
||||
body,
|
||||
});
|
||||
|
||||
while (retryCount <= MAX_RETRIES) {
|
||||
let selected = retryCount === 0
|
||||
? await tokenRouter.selectChannel(requestedModel, downstreamPolicy)
|
||||
: await tokenRouter.selectNextChannel(requestedModel, excludeChannelIds, downstreamPolicy);
|
||||
|
||||
if (!selected && retryCount === 0) {
|
||||
await refreshModelsAndRebuildRoutes();
|
||||
selected = await tokenRouter.selectChannel(requestedModel, downstreamPolicy);
|
||||
}
|
||||
const excludeChannelIds: number[] = [];
|
||||
let retryCount = 0;
|
||||
|
||||
while (retryCount <= MAX_RETRIES) {
|
||||
let selected = retryCount === 0
|
||||
? await tokenRouter.selectChannel(requestedModel, downstreamPolicy)
|
||||
: await tokenRouter.selectNextChannel(requestedModel, excludeChannelIds, downstreamPolicy);
|
||||
|
||||
if (!selected && retryCount === 0) {
|
||||
await refreshModelsAndRebuildRoutes();
|
||||
selected = await tokenRouter.selectChannel(requestedModel, downstreamPolicy);
|
||||
}
|
||||
|
||||
if (!selected) {
|
||||
await reportProxyAllFailed({
|
||||
@@ -57,46 +58,46 @@ export async function embeddingsProxyRoute(app: FastifyInstance) {
|
||||
return reply.code(503).send({ error: { message: 'No available channels', type: 'server_error' } });
|
||||
}
|
||||
|
||||
excludeChannelIds.push(selected.channel.id);
|
||||
|
||||
const targetUrl = buildUpstreamUrl(selected.site.url, '/v1/embeddings');
|
||||
const forwardBody = { ...body, model: selected.actualModel };
|
||||
const startTime = Date.now();
|
||||
excludeChannelIds.push(selected.channel.id);
|
||||
|
||||
const targetUrl = buildUpstreamUrl(selected.site.url, '/v1/embeddings');
|
||||
const forwardBody = { ...body, model: selected.actualModel || requestedModel };
|
||||
const startTime = Date.now();
|
||||
|
||||
try {
|
||||
const upstream = await fetch(targetUrl, withSiteRecordProxyRequestInit(selected.site, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
'Authorization': `Bearer ${selected.tokenValue}`,
|
||||
},
|
||||
body: JSON.stringify(forwardBody),
|
||||
}));
|
||||
const upstream = await fetch(targetUrl, withSiteRecordProxyRequestInit(selected.site, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
'Authorization': `Bearer ${selected.tokenValue}`,
|
||||
},
|
||||
body: JSON.stringify(forwardBody),
|
||||
}, getProxyUrlFromExtraConfig(selected.account.extraConfig)));
|
||||
|
||||
const text = await upstream.text();
|
||||
if (!upstream.ok) {
|
||||
tokenRouter.recordFailure(selected.channel.id, {
|
||||
status: upstream.status,
|
||||
errorText: text,
|
||||
modelName: selected.actualModel,
|
||||
});
|
||||
logProxy(
|
||||
selected,
|
||||
requestedModel,
|
||||
'failed',
|
||||
upstream.status,
|
||||
Date.now() - startTime,
|
||||
text,
|
||||
retryCount,
|
||||
downstreamApiKeyId,
|
||||
0,
|
||||
0,
|
||||
0,
|
||||
0,
|
||||
null,
|
||||
clientContext,
|
||||
downstreamPath,
|
||||
);
|
||||
const text = await upstream.text();
|
||||
if (!upstream.ok) {
|
||||
tokenRouter.recordFailure(selected.channel.id, {
|
||||
status: upstream.status,
|
||||
errorText: text,
|
||||
modelName: selected.actualModel,
|
||||
});
|
||||
logProxy(
|
||||
selected,
|
||||
requestedModel,
|
||||
'failed',
|
||||
upstream.status,
|
||||
Date.now() - startTime,
|
||||
text,
|
||||
retryCount,
|
||||
downstreamApiKeyId,
|
||||
0,
|
||||
0,
|
||||
0,
|
||||
0,
|
||||
null,
|
||||
clientContext,
|
||||
downstreamPath,
|
||||
);
|
||||
|
||||
if (isTokenExpiredError({ status: upstream.status, message: text })) {
|
||||
await reportTokenExpired({
|
||||
@@ -107,10 +108,10 @@ export async function embeddingsProxyRoute(app: FastifyInstance) {
|
||||
});
|
||||
}
|
||||
|
||||
if (shouldRetryProxyRequest(upstream.status, text) && retryCount < MAX_RETRIES) {
|
||||
retryCount++;
|
||||
continue;
|
||||
}
|
||||
if (shouldRetryProxyRequest(upstream.status, text) && retryCount < MAX_RETRIES) {
|
||||
retryCount++;
|
||||
continue;
|
||||
}
|
||||
|
||||
await reportProxyAllFailed({
|
||||
model: requestedModel,
|
||||
@@ -119,63 +120,63 @@ export async function embeddingsProxyRoute(app: FastifyInstance) {
|
||||
return reply.code(upstream.status).send({ error: { message: text, type: 'upstream_error' } });
|
||||
}
|
||||
|
||||
let data: any = {};
|
||||
try { data = JSON.parse(text); } catch { data = {}; }
|
||||
const latency = Date.now() - startTime;
|
||||
const parsedUsage = parseProxyUsage(data);
|
||||
const resolvedUsage = await resolveProxyUsageWithSelfLogFallback({
|
||||
site: selected.site,
|
||||
account: selected.account,
|
||||
tokenValue: selected.tokenValue,
|
||||
tokenName: selected.tokenName,
|
||||
modelName: selected.actualModel || requestedModel,
|
||||
requestStartedAtMs: startTime,
|
||||
requestEndedAtMs: startTime + latency,
|
||||
localLatencyMs: latency,
|
||||
usage: {
|
||||
promptTokens: parsedUsage.promptTokens,
|
||||
completionTokens: parsedUsage.completionTokens,
|
||||
totalTokens: parsedUsage.totalTokens,
|
||||
},
|
||||
});
|
||||
const { estimatedCost, billingDetails } = await resolveProxyLogBilling({
|
||||
site: selected.site,
|
||||
account: selected.account,
|
||||
modelName: selected.actualModel || requestedModel,
|
||||
parsedUsage,
|
||||
resolvedUsage,
|
||||
});
|
||||
|
||||
tokenRouter.recordSuccess(selected.channel.id, latency, estimatedCost, selected.actualModel);
|
||||
recordDownstreamCostUsage(request, estimatedCost);
|
||||
logProxy(
|
||||
selected, requestedModel, 'success', upstream.status, latency, null, retryCount, downstreamApiKeyId,
|
||||
resolvedUsage.promptTokens, resolvedUsage.completionTokens, resolvedUsage.totalTokens, estimatedCost, billingDetails, clientContext, downstreamPath,
|
||||
);
|
||||
return reply.code(upstream.status).send(data);
|
||||
} catch (err: any) {
|
||||
tokenRouter.recordFailure(selected.channel.id, {
|
||||
status: 0,
|
||||
errorText: err.message,
|
||||
modelName: selected.actualModel,
|
||||
});
|
||||
logProxy(
|
||||
selected,
|
||||
requestedModel,
|
||||
'failed',
|
||||
0,
|
||||
Date.now() - startTime,
|
||||
err.message,
|
||||
retryCount,
|
||||
downstreamApiKeyId,
|
||||
0,
|
||||
0,
|
||||
0,
|
||||
0,
|
||||
null,
|
||||
clientContext,
|
||||
downstreamPath,
|
||||
);
|
||||
let data: any = {};
|
||||
try { data = JSON.parse(text); } catch { data = {}; }
|
||||
const latency = Date.now() - startTime;
|
||||
const parsedUsage = parseProxyUsage(data);
|
||||
const resolvedUsage = await resolveProxyUsageWithSelfLogFallback({
|
||||
site: selected.site,
|
||||
account: selected.account,
|
||||
tokenValue: selected.tokenValue,
|
||||
tokenName: selected.tokenName,
|
||||
modelName: selected.actualModel || requestedModel,
|
||||
requestStartedAtMs: startTime,
|
||||
requestEndedAtMs: startTime + latency,
|
||||
localLatencyMs: latency,
|
||||
usage: {
|
||||
promptTokens: parsedUsage.promptTokens,
|
||||
completionTokens: parsedUsage.completionTokens,
|
||||
totalTokens: parsedUsage.totalTokens,
|
||||
},
|
||||
});
|
||||
const { estimatedCost, billingDetails } = await resolveProxyLogBilling({
|
||||
site: selected.site,
|
||||
account: selected.account,
|
||||
modelName: selected.actualModel || requestedModel,
|
||||
parsedUsage,
|
||||
resolvedUsage,
|
||||
});
|
||||
|
||||
tokenRouter.recordSuccess(selected.channel.id, latency, estimatedCost, selected.actualModel);
|
||||
recordDownstreamCostUsage(request, estimatedCost);
|
||||
logProxy(
|
||||
selected, requestedModel, 'success', upstream.status, latency, null, retryCount, downstreamApiKeyId,
|
||||
resolvedUsage.promptTokens, resolvedUsage.completionTokens, resolvedUsage.totalTokens, estimatedCost, billingDetails, clientContext, downstreamPath,
|
||||
);
|
||||
return reply.code(upstream.status).send(data);
|
||||
} catch (err: any) {
|
||||
tokenRouter.recordFailure(selected.channel.id, {
|
||||
status: 0,
|
||||
errorText: err.message,
|
||||
modelName: selected.actualModel,
|
||||
});
|
||||
logProxy(
|
||||
selected,
|
||||
requestedModel,
|
||||
'failed',
|
||||
0,
|
||||
Date.now() - startTime,
|
||||
err.message,
|
||||
retryCount,
|
||||
downstreamApiKeyId,
|
||||
0,
|
||||
0,
|
||||
0,
|
||||
0,
|
||||
null,
|
||||
clientContext,
|
||||
downstreamPath,
|
||||
);
|
||||
if (retryCount < MAX_RETRIES) {
|
||||
retryCount++;
|
||||
continue;
|
||||
@@ -190,59 +191,59 @@ export async function embeddingsProxyRoute(app: FastifyInstance) {
|
||||
});
|
||||
}
|
||||
|
||||
async function logProxy(
|
||||
selected: any,
|
||||
modelRequested: string,
|
||||
status: string,
|
||||
httpStatus: number,
|
||||
latencyMs: number,
|
||||
errorMessage: string | null,
|
||||
retryCount: number,
|
||||
downstreamApiKeyId: number | null = null,
|
||||
promptTokens = 0,
|
||||
completionTokens = 0,
|
||||
totalTokens = 0,
|
||||
estimatedCost = 0,
|
||||
billingDetails: unknown = null,
|
||||
clientContext: DownstreamClientContext | null = null,
|
||||
downstreamPath = '/v1/embeddings',
|
||||
) {
|
||||
try {
|
||||
const createdAt = formatUtcSqlDateTime(new Date());
|
||||
const normalizedErrorMessage = composeProxyLogMessage({
|
||||
clientKind: clientContext?.clientKind && clientContext.clientKind !== 'generic'
|
||||
? clientContext.clientKind
|
||||
: null,
|
||||
sessionId: clientContext?.sessionId || null,
|
||||
traceHint: clientContext?.traceHint || null,
|
||||
downstreamPath,
|
||||
errorMessage,
|
||||
});
|
||||
await insertProxyLog({
|
||||
routeId: selected.channel.routeId,
|
||||
channelId: selected.channel.id,
|
||||
accountId: selected.account.id,
|
||||
downstreamApiKeyId,
|
||||
modelRequested,
|
||||
modelActual: selected.actualModel,
|
||||
status,
|
||||
httpStatus,
|
||||
latencyMs,
|
||||
promptTokens,
|
||||
completionTokens,
|
||||
totalTokens,
|
||||
estimatedCost,
|
||||
billingDetails,
|
||||
clientFamily: clientContext?.clientKind || null,
|
||||
clientAppId: clientContext?.clientAppId || null,
|
||||
clientAppName: clientContext?.clientAppName || null,
|
||||
clientConfidence: clientContext?.clientConfidence || null,
|
||||
errorMessage: normalizedErrorMessage,
|
||||
retryCount,
|
||||
createdAt,
|
||||
});
|
||||
} catch (error) {
|
||||
console.warn('[proxy/embeddings] failed to write proxy log', error);
|
||||
}
|
||||
}
|
||||
async function logProxy(
|
||||
selected: any,
|
||||
modelRequested: string,
|
||||
status: string,
|
||||
httpStatus: number,
|
||||
latencyMs: number,
|
||||
errorMessage: string | null,
|
||||
retryCount: number,
|
||||
downstreamApiKeyId: number | null = null,
|
||||
promptTokens = 0,
|
||||
completionTokens = 0,
|
||||
totalTokens = 0,
|
||||
estimatedCost = 0,
|
||||
billingDetails: unknown = null,
|
||||
clientContext: DownstreamClientContext | null = null,
|
||||
downstreamPath = '/v1/embeddings',
|
||||
) {
|
||||
try {
|
||||
const createdAt = formatUtcSqlDateTime(new Date());
|
||||
const normalizedErrorMessage = composeProxyLogMessage({
|
||||
clientKind: clientContext?.clientKind && clientContext.clientKind !== 'generic'
|
||||
? clientContext.clientKind
|
||||
: null,
|
||||
sessionId: clientContext?.sessionId || null,
|
||||
traceHint: clientContext?.traceHint || null,
|
||||
downstreamPath,
|
||||
errorMessage,
|
||||
});
|
||||
await insertProxyLog({
|
||||
routeId: selected.channel.routeId,
|
||||
channelId: selected.channel.id,
|
||||
accountId: selected.account.id,
|
||||
downstreamApiKeyId,
|
||||
modelRequested,
|
||||
modelActual: selected.actualModel,
|
||||
status,
|
||||
httpStatus,
|
||||
latencyMs,
|
||||
promptTokens,
|
||||
completionTokens,
|
||||
totalTokens,
|
||||
estimatedCost,
|
||||
billingDetails,
|
||||
clientFamily: clientContext?.clientKind || null,
|
||||
clientAppId: clientContext?.clientAppId || null,
|
||||
clientAppName: clientContext?.clientAppName || null,
|
||||
clientConfidence: clientContext?.clientConfidence || null,
|
||||
errorMessage: normalizedErrorMessage,
|
||||
retryCount,
|
||||
createdAt,
|
||||
});
|
||||
} catch (error) {
|
||||
console.warn('[proxy/embeddings] failed to write proxy log', error);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
+334
-333
@@ -1,50 +1,51 @@
|
||||
import { FastifyInstance, FastifyReply, FastifyRequest } from 'fastify';
|
||||
import { fetch } from 'undici';
|
||||
import { tokenRouter } from '../../services/tokenRouter.js';
|
||||
import { refreshModelsAndRebuildRoutes } from '../../services/modelService.js';
|
||||
import { reportProxyAllFailed, reportTokenExpired } from '../../services/alertService.js';
|
||||
import { isTokenExpiredError } from '../../services/alertRules.js';
|
||||
import { estimateProxyCost } from '../../services/modelPricingService.js';
|
||||
import { shouldRetryProxyRequest } from '../../services/proxyRetryPolicy.js';
|
||||
import { ensureModelAllowedForDownstreamKey, getDownstreamRoutingPolicy, recordDownstreamCostUsage } from './downstreamPolicy.js';
|
||||
import { withSiteRecordProxyRequestInit } from '../../services/siteProxy.js';
|
||||
import { composeProxyLogMessage } from './logPathMeta.js';
|
||||
import { formatUtcSqlDateTime } from '../../services/localTimeService.js';
|
||||
import { cloneFormDataWithOverrides, ensureMultipartBufferParser, parseMultipartFormData } from './multipart.js';
|
||||
import { getProxyAuthContext } from '../../middleware/auth.js';
|
||||
import { buildUpstreamUrl } from './upstreamUrl.js';
|
||||
import { detectDownstreamClientContext, type DownstreamClientContext } from './downstreamClientContext.js';
|
||||
import { insertProxyLog } from '../../services/proxyLogStore.js';
|
||||
import { tokenRouter } from '../../services/tokenRouter.js';
|
||||
import { refreshModelsAndRebuildRoutes } from '../../services/modelService.js';
|
||||
import { reportProxyAllFailed, reportTokenExpired } from '../../services/alertService.js';
|
||||
import { isTokenExpiredError } from '../../services/alertRules.js';
|
||||
import { estimateProxyCost } from '../../services/modelPricingService.js';
|
||||
import { shouldRetryProxyRequest } from '../../services/proxyRetryPolicy.js';
|
||||
import { ensureModelAllowedForDownstreamKey, getDownstreamRoutingPolicy, recordDownstreamCostUsage } from './downstreamPolicy.js';
|
||||
import { withSiteRecordProxyRequestInit } from '../../services/siteProxy.js';
|
||||
import { getProxyUrlFromExtraConfig } from '../../services/accountExtraConfig.js';
|
||||
import { composeProxyLogMessage } from './logPathMeta.js';
|
||||
import { formatUtcSqlDateTime } from '../../services/localTimeService.js';
|
||||
import { cloneFormDataWithOverrides, ensureMultipartBufferParser, parseMultipartFormData } from './multipart.js';
|
||||
import { getProxyAuthContext } from '../../middleware/auth.js';
|
||||
import { buildUpstreamUrl } from './upstreamUrl.js';
|
||||
import { detectDownstreamClientContext, type DownstreamClientContext } from './downstreamClientContext.js';
|
||||
import { insertProxyLog } from '../../services/proxyLogStore.js';
|
||||
|
||||
const MAX_RETRIES = 2;
|
||||
|
||||
export async function imagesProxyRoute(app: FastifyInstance) {
|
||||
ensureMultipartBufferParser(app);
|
||||
|
||||
app.post('/v1/images/generations', async (request: FastifyRequest, reply: FastifyReply) => {
|
||||
const body = request.body as any;
|
||||
const requestedModel = body?.model || 'gpt-image-1';
|
||||
if (!await ensureModelAllowedForDownstreamKey(request, reply, requestedModel)) return;
|
||||
const downstreamPolicy = getDownstreamRoutingPolicy(request);
|
||||
const downstreamApiKeyId = getProxyAuthContext(request)?.keyId ?? null;
|
||||
const downstreamPath = '/v1/images/generations';
|
||||
const clientContext = detectDownstreamClientContext({
|
||||
downstreamPath,
|
||||
headers: request.headers as Record<string, unknown>,
|
||||
body,
|
||||
});
|
||||
const excludeChannelIds: number[] = [];
|
||||
let retryCount = 0;
|
||||
|
||||
while (retryCount <= MAX_RETRIES) {
|
||||
let selected = retryCount === 0
|
||||
? await tokenRouter.selectChannel(requestedModel, downstreamPolicy)
|
||||
: await tokenRouter.selectNextChannel(requestedModel, excludeChannelIds, downstreamPolicy);
|
||||
|
||||
if (!selected && retryCount === 0) {
|
||||
await refreshModelsAndRebuildRoutes();
|
||||
selected = await tokenRouter.selectChannel(requestedModel, downstreamPolicy);
|
||||
}
|
||||
export async function imagesProxyRoute(app: FastifyInstance) {
|
||||
ensureMultipartBufferParser(app);
|
||||
|
||||
app.post('/v1/images/generations', async (request: FastifyRequest, reply: FastifyReply) => {
|
||||
const body = request.body as any;
|
||||
const requestedModel = body?.model || 'gpt-image-1';
|
||||
if (!await ensureModelAllowedForDownstreamKey(request, reply, requestedModel)) return;
|
||||
const downstreamPolicy = getDownstreamRoutingPolicy(request);
|
||||
const downstreamApiKeyId = getProxyAuthContext(request)?.keyId ?? null;
|
||||
const downstreamPath = '/v1/images/generations';
|
||||
const clientContext = detectDownstreamClientContext({
|
||||
downstreamPath,
|
||||
headers: request.headers as Record<string, unknown>,
|
||||
body,
|
||||
});
|
||||
const excludeChannelIds: number[] = [];
|
||||
let retryCount = 0;
|
||||
|
||||
while (retryCount <= MAX_RETRIES) {
|
||||
let selected = retryCount === 0
|
||||
? await tokenRouter.selectChannel(requestedModel, downstreamPolicy)
|
||||
: await tokenRouter.selectNextChannel(requestedModel, excludeChannelIds, downstreamPolicy);
|
||||
|
||||
if (!selected && retryCount === 0) {
|
||||
await refreshModelsAndRebuildRoutes();
|
||||
selected = await tokenRouter.selectChannel(requestedModel, downstreamPolicy);
|
||||
}
|
||||
|
||||
if (!selected) {
|
||||
await reportProxyAllFailed({
|
||||
@@ -58,52 +59,52 @@ export async function imagesProxyRoute(app: FastifyInstance) {
|
||||
|
||||
excludeChannelIds.push(selected.channel.id);
|
||||
|
||||
const targetUrl = buildUpstreamUrl(selected.site.url, '/v1/images/generations');
|
||||
const targetUrl = buildUpstreamUrl(selected.site.url, '/v1/images/generations');
|
||||
const forwardBody = { ...body, model: selected.actualModel };
|
||||
const startTime = Date.now();
|
||||
|
||||
try {
|
||||
const upstream = await fetch(targetUrl, withSiteRecordProxyRequestInit(selected.site, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
'Authorization': `Bearer ${selected.tokenValue}`,
|
||||
},
|
||||
body: JSON.stringify(forwardBody),
|
||||
}));
|
||||
const upstream = await fetch(targetUrl, withSiteRecordProxyRequestInit(selected.site, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
'Authorization': `Bearer ${selected.tokenValue}`,
|
||||
},
|
||||
body: JSON.stringify(forwardBody),
|
||||
}, getProxyUrlFromExtraConfig(selected.account.extraConfig)));
|
||||
|
||||
const text = await upstream.text();
|
||||
if (!upstream.ok) {
|
||||
tokenRouter.recordFailure(selected.channel.id, {
|
||||
status: upstream.status,
|
||||
errorText: text,
|
||||
modelName: selected.actualModel,
|
||||
});
|
||||
logProxy(
|
||||
selected,
|
||||
requestedModel,
|
||||
'failed',
|
||||
upstream.status,
|
||||
Date.now() - startTime,
|
||||
text,
|
||||
retryCount,
|
||||
downstreamApiKeyId,
|
||||
0,
|
||||
downstreamPath,
|
||||
clientContext,
|
||||
);
|
||||
if (isTokenExpiredError({ status: upstream.status, message: text })) {
|
||||
await reportTokenExpired({
|
||||
accountId: selected.account.id,
|
||||
username: selected.account.username,
|
||||
siteName: selected.site.name,
|
||||
const text = await upstream.text();
|
||||
if (!upstream.ok) {
|
||||
tokenRouter.recordFailure(selected.channel.id, {
|
||||
status: upstream.status,
|
||||
errorText: text,
|
||||
modelName: selected.actualModel,
|
||||
});
|
||||
logProxy(
|
||||
selected,
|
||||
requestedModel,
|
||||
'failed',
|
||||
upstream.status,
|
||||
Date.now() - startTime,
|
||||
text,
|
||||
retryCount,
|
||||
downstreamApiKeyId,
|
||||
0,
|
||||
downstreamPath,
|
||||
clientContext,
|
||||
);
|
||||
if (isTokenExpiredError({ status: upstream.status, message: text })) {
|
||||
await reportTokenExpired({
|
||||
accountId: selected.account.id,
|
||||
username: selected.account.username,
|
||||
siteName: selected.site.name,
|
||||
detail: `HTTP ${upstream.status}`,
|
||||
});
|
||||
}
|
||||
if (shouldRetryProxyRequest(upstream.status, text) && retryCount < MAX_RETRIES) {
|
||||
retryCount++;
|
||||
continue;
|
||||
}
|
||||
if (shouldRetryProxyRequest(upstream.status, text) && retryCount < MAX_RETRIES) {
|
||||
retryCount++;
|
||||
continue;
|
||||
}
|
||||
await reportProxyAllFailed({
|
||||
model: requestedModel,
|
||||
reason: `upstream returned HTTP ${upstream.status}`,
|
||||
@@ -115,41 +116,41 @@ export async function imagesProxyRoute(app: FastifyInstance) {
|
||||
try { data = JSON.parse(text); } catch { data = { data: [] }; }
|
||||
|
||||
const latency = Date.now() - startTime;
|
||||
const estimatedCost = await estimateProxyCost({
|
||||
site: selected.site,
|
||||
account: selected.account,
|
||||
modelName: selected.actualModel || requestedModel,
|
||||
const estimatedCost = await estimateProxyCost({
|
||||
site: selected.site,
|
||||
account: selected.account,
|
||||
modelName: selected.actualModel || requestedModel,
|
||||
promptTokens: 0,
|
||||
completionTokens: 0,
|
||||
totalTokens: 0,
|
||||
});
|
||||
tokenRouter.recordSuccess(selected.channel.id, latency, estimatedCost, selected.actualModel);
|
||||
recordDownstreamCostUsage(request, estimatedCost);
|
||||
logProxy(selected, requestedModel, 'success', upstream.status, latency, null, retryCount, downstreamApiKeyId, estimatedCost, downstreamPath, clientContext);
|
||||
return reply.code(upstream.status).send(data);
|
||||
} catch (err: any) {
|
||||
tokenRouter.recordFailure(selected.channel.id, {
|
||||
status: 0,
|
||||
errorText: err.message,
|
||||
modelName: selected.actualModel,
|
||||
});
|
||||
logProxy(
|
||||
selected,
|
||||
requestedModel,
|
||||
'failed',
|
||||
0,
|
||||
Date.now() - startTime,
|
||||
err.message,
|
||||
retryCount,
|
||||
downstreamApiKeyId,
|
||||
0,
|
||||
downstreamPath,
|
||||
clientContext,
|
||||
);
|
||||
if (retryCount < MAX_RETRIES) {
|
||||
retryCount++;
|
||||
continue;
|
||||
}
|
||||
});
|
||||
tokenRouter.recordSuccess(selected.channel.id, latency, estimatedCost, selected.actualModel);
|
||||
recordDownstreamCostUsage(request, estimatedCost);
|
||||
logProxy(selected, requestedModel, 'success', upstream.status, latency, null, retryCount, downstreamApiKeyId, estimatedCost, downstreamPath, clientContext);
|
||||
return reply.code(upstream.status).send(data);
|
||||
} catch (err: any) {
|
||||
tokenRouter.recordFailure(selected.channel.id, {
|
||||
status: 0,
|
||||
errorText: err.message,
|
||||
modelName: selected.actualModel,
|
||||
});
|
||||
logProxy(
|
||||
selected,
|
||||
requestedModel,
|
||||
'failed',
|
||||
0,
|
||||
Date.now() - startTime,
|
||||
err.message,
|
||||
retryCount,
|
||||
downstreamApiKeyId,
|
||||
0,
|
||||
downstreamPath,
|
||||
clientContext,
|
||||
);
|
||||
if (retryCount < MAX_RETRIES) {
|
||||
retryCount++;
|
||||
continue;
|
||||
}
|
||||
await reportProxyAllFailed({
|
||||
model: requestedModel,
|
||||
reason: err.message || 'network failure',
|
||||
@@ -157,226 +158,226 @@ export async function imagesProxyRoute(app: FastifyInstance) {
|
||||
return reply.code(502).send({
|
||||
error: { message: `Upstream error: ${err.message}`, type: 'upstream_error' },
|
||||
});
|
||||
}
|
||||
}
|
||||
});
|
||||
|
||||
app.post('/v1/images/edits', async (request: FastifyRequest, reply: FastifyReply) => {
|
||||
const multipartForm = await parseMultipartFormData(request);
|
||||
const jsonBody = (!multipartForm && request.body && typeof request.body === 'object')
|
||||
? request.body as Record<string, unknown>
|
||||
: null;
|
||||
const requestedModel = typeof multipartForm?.get('model') === 'string'
|
||||
? String(multipartForm.get('model')).trim()
|
||||
: (typeof jsonBody?.model === 'string' ? jsonBody.model.trim() : '') || 'gpt-image-1';
|
||||
|
||||
if (!await ensureModelAllowedForDownstreamKey(request, reply, requestedModel)) return;
|
||||
const downstreamPolicy = getDownstreamRoutingPolicy(request);
|
||||
const downstreamApiKeyId = getProxyAuthContext(request)?.keyId ?? null;
|
||||
const downstreamPath = '/v1/images/edits';
|
||||
const clientContext = detectDownstreamClientContext({
|
||||
downstreamPath,
|
||||
headers: request.headers as Record<string, unknown>,
|
||||
body: jsonBody || Object.fromEntries(multipartForm?.entries?.() || []),
|
||||
});
|
||||
const excludeChannelIds: number[] = [];
|
||||
let retryCount = 0;
|
||||
|
||||
while (retryCount <= MAX_RETRIES) {
|
||||
let selected = retryCount === 0
|
||||
? await tokenRouter.selectChannel(requestedModel, downstreamPolicy)
|
||||
: await tokenRouter.selectNextChannel(requestedModel, excludeChannelIds, downstreamPolicy);
|
||||
|
||||
if (!selected && retryCount === 0) {
|
||||
await refreshModelsAndRebuildRoutes();
|
||||
selected = await tokenRouter.selectChannel(requestedModel, downstreamPolicy);
|
||||
}
|
||||
|
||||
if (!selected) {
|
||||
await reportProxyAllFailed({
|
||||
model: requestedModel,
|
||||
reason: 'No available channels after retries',
|
||||
});
|
||||
return reply.code(503).send({
|
||||
error: { message: 'No available channels for this model', type: 'server_error' },
|
||||
});
|
||||
}
|
||||
|
||||
excludeChannelIds.push(selected.channel.id);
|
||||
const targetUrl = buildUpstreamUrl(selected.site.url, '/v1/images/edits');
|
||||
const startTime = Date.now();
|
||||
|
||||
try {
|
||||
const requestInit = multipartForm
|
||||
? withSiteRecordProxyRequestInit(selected.site, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
Authorization: `Bearer ${selected.tokenValue}`,
|
||||
},
|
||||
body: cloneFormDataWithOverrides(multipartForm, {
|
||||
model: selected.actualModel || requestedModel,
|
||||
}) as any,
|
||||
})
|
||||
: withSiteRecordProxyRequestInit(selected.site, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
Authorization: `Bearer ${selected.tokenValue}`,
|
||||
},
|
||||
body: JSON.stringify({
|
||||
...(jsonBody || {}),
|
||||
model: selected.actualModel || requestedModel,
|
||||
}),
|
||||
});
|
||||
|
||||
const upstream = await fetch(targetUrl, requestInit);
|
||||
const text = await upstream.text();
|
||||
if (!upstream.ok) {
|
||||
tokenRouter.recordFailure(selected.channel.id, {
|
||||
status: upstream.status,
|
||||
errorText: text,
|
||||
modelName: selected.actualModel,
|
||||
});
|
||||
logProxy(
|
||||
selected,
|
||||
requestedModel,
|
||||
'failed',
|
||||
upstream.status,
|
||||
Date.now() - startTime,
|
||||
text,
|
||||
retryCount,
|
||||
downstreamApiKeyId,
|
||||
0,
|
||||
downstreamPath,
|
||||
clientContext,
|
||||
);
|
||||
if (isTokenExpiredError({ status: upstream.status, message: text })) {
|
||||
await reportTokenExpired({
|
||||
accountId: selected.account.id,
|
||||
username: selected.account.username,
|
||||
siteName: selected.site.name,
|
||||
detail: `HTTP ${upstream.status}`,
|
||||
});
|
||||
}
|
||||
if (shouldRetryProxyRequest(upstream.status, text) && retryCount < MAX_RETRIES) {
|
||||
retryCount++;
|
||||
continue;
|
||||
}
|
||||
await reportProxyAllFailed({
|
||||
model: requestedModel,
|
||||
reason: `upstream returned HTTP ${upstream.status}`,
|
||||
});
|
||||
return reply.code(upstream.status).send({ error: { message: text, type: 'upstream_error' } });
|
||||
}
|
||||
|
||||
let data: any = {};
|
||||
try { data = JSON.parse(text); } catch { data = { data: [] }; }
|
||||
|
||||
const latency = Date.now() - startTime;
|
||||
const estimatedCost = await estimateProxyCost({
|
||||
site: selected.site,
|
||||
account: selected.account,
|
||||
modelName: selected.actualModel || requestedModel,
|
||||
promptTokens: 0,
|
||||
completionTokens: 0,
|
||||
totalTokens: 0,
|
||||
});
|
||||
tokenRouter.recordSuccess(selected.channel.id, latency, estimatedCost, selected.actualModel);
|
||||
recordDownstreamCostUsage(request, estimatedCost);
|
||||
logProxy(selected, requestedModel, 'success', upstream.status, latency, null, retryCount, downstreamApiKeyId, estimatedCost, downstreamPath, clientContext);
|
||||
return reply.code(upstream.status).send(data);
|
||||
} catch (err: any) {
|
||||
tokenRouter.recordFailure(selected.channel.id, {
|
||||
status: 0,
|
||||
errorText: err.message,
|
||||
modelName: selected.actualModel,
|
||||
});
|
||||
logProxy(
|
||||
selected,
|
||||
requestedModel,
|
||||
'failed',
|
||||
0,
|
||||
Date.now() - startTime,
|
||||
err.message,
|
||||
retryCount,
|
||||
downstreamApiKeyId,
|
||||
0,
|
||||
downstreamPath,
|
||||
clientContext,
|
||||
);
|
||||
if (retryCount < MAX_RETRIES) {
|
||||
retryCount++;
|
||||
continue;
|
||||
}
|
||||
await reportProxyAllFailed({
|
||||
model: requestedModel,
|
||||
reason: err.message || 'network failure',
|
||||
});
|
||||
return reply.code(502).send({
|
||||
error: { message: `Upstream error: ${err.message}`, type: 'upstream_error' },
|
||||
});
|
||||
}
|
||||
}
|
||||
});
|
||||
|
||||
app.post('/v1/images/variations', async (_request: FastifyRequest, reply: FastifyReply) => {
|
||||
return reply.code(400).send({
|
||||
error: {
|
||||
message: 'Image variations are not supported',
|
||||
type: 'invalid_request_error',
|
||||
},
|
||||
});
|
||||
});
|
||||
}
|
||||
}
|
||||
}
|
||||
});
|
||||
|
||||
async function logProxy(
|
||||
selected: any,
|
||||
modelRequested: string,
|
||||
status: string,
|
||||
httpStatus: number,
|
||||
latencyMs: number,
|
||||
errorMessage: string | null,
|
||||
retryCount: number,
|
||||
downstreamApiKeyId: number | null = null,
|
||||
estimatedCost = 0,
|
||||
downstreamPath = '/v1/images/generations',
|
||||
clientContext: DownstreamClientContext | null = null,
|
||||
) {
|
||||
try {
|
||||
const createdAt = formatUtcSqlDateTime(new Date());
|
||||
const normalizedErrorMessage = composeProxyLogMessage({
|
||||
clientKind: clientContext?.clientKind && clientContext.clientKind !== 'generic'
|
||||
? clientContext.clientKind
|
||||
: null,
|
||||
sessionId: clientContext?.sessionId || null,
|
||||
traceHint: clientContext?.traceHint || null,
|
||||
downstreamPath,
|
||||
errorMessage,
|
||||
});
|
||||
await insertProxyLog({
|
||||
routeId: selected.channel.routeId,
|
||||
channelId: selected.channel.id,
|
||||
accountId: selected.account.id,
|
||||
downstreamApiKeyId,
|
||||
modelRequested,
|
||||
modelActual: selected.actualModel,
|
||||
status,
|
||||
httpStatus,
|
||||
latencyMs,
|
||||
promptTokens: 0,
|
||||
completionTokens: 0,
|
||||
totalTokens: 0,
|
||||
estimatedCost,
|
||||
clientFamily: clientContext?.clientKind || null,
|
||||
clientAppId: clientContext?.clientAppId || null,
|
||||
clientAppName: clientContext?.clientAppName || null,
|
||||
clientConfidence: clientContext?.clientConfidence || null,
|
||||
errorMessage: normalizedErrorMessage,
|
||||
retryCount,
|
||||
createdAt,
|
||||
});
|
||||
} catch (error) {
|
||||
console.warn('[proxy/images] failed to write proxy log', error);
|
||||
}
|
||||
}
|
||||
app.post('/v1/images/edits', async (request: FastifyRequest, reply: FastifyReply) => {
|
||||
const multipartForm = await parseMultipartFormData(request);
|
||||
const jsonBody = (!multipartForm && request.body && typeof request.body === 'object')
|
||||
? request.body as Record<string, unknown>
|
||||
: null;
|
||||
const requestedModel = typeof multipartForm?.get('model') === 'string'
|
||||
? String(multipartForm.get('model')).trim()
|
||||
: (typeof jsonBody?.model === 'string' ? jsonBody.model.trim() : '') || 'gpt-image-1';
|
||||
|
||||
if (!await ensureModelAllowedForDownstreamKey(request, reply, requestedModel)) return;
|
||||
const downstreamPolicy = getDownstreamRoutingPolicy(request);
|
||||
const downstreamApiKeyId = getProxyAuthContext(request)?.keyId ?? null;
|
||||
const downstreamPath = '/v1/images/edits';
|
||||
const clientContext = detectDownstreamClientContext({
|
||||
downstreamPath,
|
||||
headers: request.headers as Record<string, unknown>,
|
||||
body: jsonBody || Object.fromEntries(multipartForm?.entries?.() || []),
|
||||
});
|
||||
const excludeChannelIds: number[] = [];
|
||||
let retryCount = 0;
|
||||
|
||||
while (retryCount <= MAX_RETRIES) {
|
||||
let selected = retryCount === 0
|
||||
? await tokenRouter.selectChannel(requestedModel, downstreamPolicy)
|
||||
: await tokenRouter.selectNextChannel(requestedModel, excludeChannelIds, downstreamPolicy);
|
||||
|
||||
if (!selected && retryCount === 0) {
|
||||
await refreshModelsAndRebuildRoutes();
|
||||
selected = await tokenRouter.selectChannel(requestedModel, downstreamPolicy);
|
||||
}
|
||||
|
||||
if (!selected) {
|
||||
await reportProxyAllFailed({
|
||||
model: requestedModel,
|
||||
reason: 'No available channels after retries',
|
||||
});
|
||||
return reply.code(503).send({
|
||||
error: { message: 'No available channels for this model', type: 'server_error' },
|
||||
});
|
||||
}
|
||||
|
||||
excludeChannelIds.push(selected.channel.id);
|
||||
const targetUrl = buildUpstreamUrl(selected.site.url, '/v1/images/edits');
|
||||
const startTime = Date.now();
|
||||
|
||||
try {
|
||||
const requestInit = multipartForm
|
||||
? withSiteRecordProxyRequestInit(selected.site, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
Authorization: `Bearer ${selected.tokenValue}`,
|
||||
},
|
||||
body: cloneFormDataWithOverrides(multipartForm, {
|
||||
model: selected.actualModel || requestedModel,
|
||||
}) as any,
|
||||
}, getProxyUrlFromExtraConfig(selected.account.extraConfig))
|
||||
: withSiteRecordProxyRequestInit(selected.site, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
Authorization: `Bearer ${selected.tokenValue}`,
|
||||
},
|
||||
body: JSON.stringify({
|
||||
...(jsonBody || {}),
|
||||
model: selected.actualModel || requestedModel,
|
||||
}),
|
||||
}, getProxyUrlFromExtraConfig(selected.account.extraConfig));
|
||||
|
||||
const upstream = await fetch(targetUrl, requestInit);
|
||||
const text = await upstream.text();
|
||||
if (!upstream.ok) {
|
||||
tokenRouter.recordFailure(selected.channel.id, {
|
||||
status: upstream.status,
|
||||
errorText: text,
|
||||
modelName: selected.actualModel,
|
||||
});
|
||||
logProxy(
|
||||
selected,
|
||||
requestedModel,
|
||||
'failed',
|
||||
upstream.status,
|
||||
Date.now() - startTime,
|
||||
text,
|
||||
retryCount,
|
||||
downstreamApiKeyId,
|
||||
0,
|
||||
downstreamPath,
|
||||
clientContext,
|
||||
);
|
||||
if (isTokenExpiredError({ status: upstream.status, message: text })) {
|
||||
await reportTokenExpired({
|
||||
accountId: selected.account.id,
|
||||
username: selected.account.username,
|
||||
siteName: selected.site.name,
|
||||
detail: `HTTP ${upstream.status}`,
|
||||
});
|
||||
}
|
||||
if (shouldRetryProxyRequest(upstream.status, text) && retryCount < MAX_RETRIES) {
|
||||
retryCount++;
|
||||
continue;
|
||||
}
|
||||
await reportProxyAllFailed({
|
||||
model: requestedModel,
|
||||
reason: `upstream returned HTTP ${upstream.status}`,
|
||||
});
|
||||
return reply.code(upstream.status).send({ error: { message: text, type: 'upstream_error' } });
|
||||
}
|
||||
|
||||
let data: any = {};
|
||||
try { data = JSON.parse(text); } catch { data = { data: [] }; }
|
||||
|
||||
const latency = Date.now() - startTime;
|
||||
const estimatedCost = await estimateProxyCost({
|
||||
site: selected.site,
|
||||
account: selected.account,
|
||||
modelName: selected.actualModel || requestedModel,
|
||||
promptTokens: 0,
|
||||
completionTokens: 0,
|
||||
totalTokens: 0,
|
||||
});
|
||||
tokenRouter.recordSuccess(selected.channel.id, latency, estimatedCost, selected.actualModel);
|
||||
recordDownstreamCostUsage(request, estimatedCost);
|
||||
logProxy(selected, requestedModel, 'success', upstream.status, latency, null, retryCount, downstreamApiKeyId, estimatedCost, downstreamPath, clientContext);
|
||||
return reply.code(upstream.status).send(data);
|
||||
} catch (err: any) {
|
||||
tokenRouter.recordFailure(selected.channel.id, {
|
||||
status: 0,
|
||||
errorText: err.message,
|
||||
modelName: selected.actualModel,
|
||||
});
|
||||
logProxy(
|
||||
selected,
|
||||
requestedModel,
|
||||
'failed',
|
||||
0,
|
||||
Date.now() - startTime,
|
||||
err.message,
|
||||
retryCount,
|
||||
downstreamApiKeyId,
|
||||
0,
|
||||
downstreamPath,
|
||||
clientContext,
|
||||
);
|
||||
if (retryCount < MAX_RETRIES) {
|
||||
retryCount++;
|
||||
continue;
|
||||
}
|
||||
await reportProxyAllFailed({
|
||||
model: requestedModel,
|
||||
reason: err.message || 'network failure',
|
||||
});
|
||||
return reply.code(502).send({
|
||||
error: { message: `Upstream error: ${err.message}`, type: 'upstream_error' },
|
||||
});
|
||||
}
|
||||
}
|
||||
});
|
||||
|
||||
app.post('/v1/images/variations', async (_request: FastifyRequest, reply: FastifyReply) => {
|
||||
return reply.code(400).send({
|
||||
error: {
|
||||
message: 'Image variations are not supported',
|
||||
type: 'invalid_request_error',
|
||||
},
|
||||
});
|
||||
});
|
||||
}
|
||||
|
||||
async function logProxy(
|
||||
selected: any,
|
||||
modelRequested: string,
|
||||
status: string,
|
||||
httpStatus: number,
|
||||
latencyMs: number,
|
||||
errorMessage: string | null,
|
||||
retryCount: number,
|
||||
downstreamApiKeyId: number | null = null,
|
||||
estimatedCost = 0,
|
||||
downstreamPath = '/v1/images/generations',
|
||||
clientContext: DownstreamClientContext | null = null,
|
||||
) {
|
||||
try {
|
||||
const createdAt = formatUtcSqlDateTime(new Date());
|
||||
const normalizedErrorMessage = composeProxyLogMessage({
|
||||
clientKind: clientContext?.clientKind && clientContext.clientKind !== 'generic'
|
||||
? clientContext.clientKind
|
||||
: null,
|
||||
sessionId: clientContext?.sessionId || null,
|
||||
traceHint: clientContext?.traceHint || null,
|
||||
downstreamPath,
|
||||
errorMessage,
|
||||
});
|
||||
await insertProxyLog({
|
||||
routeId: selected.channel.routeId,
|
||||
channelId: selected.channel.id,
|
||||
accountId: selected.account.id,
|
||||
downstreamApiKeyId,
|
||||
modelRequested,
|
||||
modelActual: selected.actualModel,
|
||||
status,
|
||||
httpStatus,
|
||||
latencyMs,
|
||||
promptTokens: 0,
|
||||
completionTokens: 0,
|
||||
totalTokens: 0,
|
||||
estimatedCost,
|
||||
clientFamily: clientContext?.clientKind || null,
|
||||
clientAppId: clientContext?.clientAppId || null,
|
||||
clientAppName: clientContext?.clientAppName || null,
|
||||
clientConfidence: clientContext?.clientConfidence || null,
|
||||
errorMessage: normalizedErrorMessage,
|
||||
retryCount,
|
||||
createdAt,
|
||||
});
|
||||
} catch (error) {
|
||||
console.warn('[proxy/images] failed to write proxy log', error);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@@ -7,6 +7,7 @@ import { isTokenExpiredError } from '../../services/alertRules.js';
|
||||
import { shouldRetryProxyRequest } from '../../services/proxyRetryPolicy.js';
|
||||
import { ensureModelAllowedForDownstreamKey, getDownstreamRoutingPolicy, recordDownstreamCostUsage } from './downstreamPolicy.js';
|
||||
import { withSiteRecordProxyRequestInit } from '../../services/siteProxy.js';
|
||||
import { getProxyUrlFromExtraConfig } from '../../services/accountExtraConfig.js';
|
||||
import { composeProxyLogMessage } from './logPathMeta.js';
|
||||
import { formatUtcSqlDateTime } from '../../services/localTimeService.js';
|
||||
import { getProxyAuthContext } from '../../middleware/auth.js';
|
||||
@@ -101,7 +102,7 @@ export async function searchProxyRoute(app: FastifyInstance) {
|
||||
Authorization: `Bearer ${selected.tokenValue}`,
|
||||
},
|
||||
body: JSON.stringify(forwardBody),
|
||||
}));
|
||||
}, getProxyUrlFromExtraConfig(selected.account.extraConfig)));
|
||||
|
||||
const text = await upstream.text();
|
||||
if (!upstream.ok) {
|
||||
|
||||
@@ -8,6 +8,7 @@ import { estimateProxyCost } from '../../services/modelPricingService.js';
|
||||
import { shouldRetryProxyRequest } from '../../services/proxyRetryPolicy.js';
|
||||
import { ensureModelAllowedForDownstreamKey, getDownstreamRoutingPolicy, recordDownstreamCostUsage } from './downstreamPolicy.js';
|
||||
import { withSiteProxyRequestInit, withSiteRecordProxyRequestInit } from '../../services/siteProxy.js';
|
||||
import { getProxyUrlFromExtraConfig } from '../../services/accountExtraConfig.js';
|
||||
import { cloneFormDataWithOverrides, ensureMultipartBufferParser, parseMultipartFormData } from './multipart.js';
|
||||
import { buildUpstreamUrl } from './upstreamUrl.js';
|
||||
import {
|
||||
@@ -75,6 +76,7 @@ export async function videosProxyRoute(app: FastifyInstance) {
|
||||
const startTime = Date.now();
|
||||
|
||||
try {
|
||||
const accountProxy = getProxyUrlFromExtraConfig(selected.account.extraConfig);
|
||||
const requestInit = multipartForm
|
||||
? withSiteRecordProxyRequestInit(selected.site, {
|
||||
method: 'POST',
|
||||
@@ -84,7 +86,7 @@ export async function videosProxyRoute(app: FastifyInstance) {
|
||||
body: cloneFormDataWithOverrides(multipartForm, {
|
||||
model: selected.actualModel || requestedModel,
|
||||
}) as any,
|
||||
})
|
||||
}, accountProxy)
|
||||
: withSiteRecordProxyRequestInit(selected.site, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
@@ -95,7 +97,7 @@ export async function videosProxyRoute(app: FastifyInstance) {
|
||||
...(jsonBody || {}),
|
||||
model: selected.actualModel || requestedModel,
|
||||
}),
|
||||
});
|
||||
}, accountProxy);
|
||||
|
||||
const upstream = await fetch(targetUrl, requestInit);
|
||||
const text = await upstream.text();
|
||||
|
||||
Reference in New Issue
Block a user