forked from boos/BoosAPI
feat: add proxy log client detection filters
This commit is contained in:
@@ -1,6 +1,5 @@
|
||||
import { FastifyInstance, FastifyReply, FastifyRequest } from 'fastify';
|
||||
import { fetch } from 'undici';
|
||||
import { db, hasProxyLogDownstreamApiKeyIdColumn, schema } from '../../db/index.js';
|
||||
import { tokenRouter } from '../../services/tokenRouter.js';
|
||||
import { refreshModelsAndRebuildRoutes } from '../../services/modelService.js';
|
||||
import { reportProxyAllFailed, reportTokenExpired } from '../../services/alertService.js';
|
||||
@@ -14,8 +13,10 @@ import { composeProxyLogMessage } from './logPathMeta.js';
|
||||
import { formatUtcSqlDateTime } from '../../services/localTimeService.js';
|
||||
import { detectProxyFailure } from './proxyFailureJudge.js';
|
||||
import { resolveProxyLogBilling } from './proxyBilling.js';
|
||||
import { getProxyAuthContext } from '../../middleware/auth.js';
|
||||
import { buildUpstreamUrl } from './upstreamUrl.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;
|
||||
|
||||
@@ -25,12 +26,16 @@ export async function completionsProxyRoute(app: FastifyInstance) {
|
||||
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 logDownstreamApiKeyId = downstreamApiKeyId !== null
|
||||
&& await hasProxyLogDownstreamApiKeyIdColumn();
|
||||
}
|
||||
if (!await ensureModelAllowedForDownstreamKey(request, reply, requestedModel)) return;
|
||||
const downstreamPolicy = getDownstreamRoutingPolicy(request);
|
||||
const downstreamApiKeyId = getProxyAuthContext(request)?.keyId ?? null;
|
||||
const downstreamPath = '/v1/completions';
|
||||
const clientContext = detectDownstreamClientContext({
|
||||
downstreamPath,
|
||||
headers: request.headers as Record<string, unknown>,
|
||||
body,
|
||||
});
|
||||
|
||||
const isStream = body.stream === true;
|
||||
const excludeChannelIds: number[] = [];
|
||||
@@ -75,7 +80,7 @@ export async function completionsProxyRoute(app: FastifyInstance) {
|
||||
if (!upstream.ok) {
|
||||
const errText = await upstream.text().catch(() => 'unknown error');
|
||||
tokenRouter.recordFailure(selected.channel.id);
|
||||
logProxy(selected, requestedModel, 'failed', upstream.status, Date.now() - startTime, errText, retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null);
|
||||
logProxy(selected, requestedModel, 'failed', upstream.status, Date.now() - startTime, errText, retryCount, downstreamApiKeyId, 0, 0, 0, 0, null, clientContext, downstreamPath);
|
||||
|
||||
if (isTokenExpiredError({ status: upstream.status, message: errText })) {
|
||||
await reportTokenExpired({
|
||||
@@ -173,12 +178,12 @@ export async function completionsProxyRoute(app: FastifyInstance) {
|
||||
parsedUsage,
|
||||
resolvedUsage,
|
||||
});
|
||||
tokenRouter.recordSuccess(selected.channel.id, latency, estimatedCost);
|
||||
recordDownstreamCostUsage(request, estimatedCost);
|
||||
logProxy(
|
||||
selected, requestedModel, 'success', 200, latency, null, retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null,
|
||||
resolvedUsage.promptTokens, resolvedUsage.completionTokens, resolvedUsage.totalTokens, estimatedCost, billingDetails,
|
||||
);
|
||||
tokenRouter.recordSuccess(selected.channel.id, latency, estimatedCost);
|
||||
recordDownstreamCostUsage(request, estimatedCost);
|
||||
logProxy(
|
||||
selected, requestedModel, 'success', 200, latency, null, retryCount, downstreamApiKeyId,
|
||||
resolvedUsage.promptTokens, resolvedUsage.completionTokens, resolvedUsage.totalTokens, estimatedCost, billingDetails, clientContext, downstreamPath,
|
||||
);
|
||||
return;
|
||||
}
|
||||
|
||||
@@ -199,12 +204,19 @@ export async function completionsProxyRoute(app: FastifyInstance) {
|
||||
selected,
|
||||
requestedModel,
|
||||
'failed',
|
||||
failure.status,
|
||||
latency,
|
||||
errText,
|
||||
retryCount,
|
||||
logDownstreamApiKeyId ? downstreamApiKeyId : null,
|
||||
);
|
||||
failure.status,
|
||||
latency,
|
||||
errText,
|
||||
retryCount,
|
||||
downstreamApiKeyId,
|
||||
0,
|
||||
0,
|
||||
0,
|
||||
0,
|
||||
null,
|
||||
clientContext,
|
||||
downstreamPath,
|
||||
);
|
||||
|
||||
if (shouldRetryProxyRequest(failure.status, errText) && retryCount < MAX_RETRIES) {
|
||||
retryCount += 1;
|
||||
@@ -244,16 +256,16 @@ export async function completionsProxyRoute(app: FastifyInstance) {
|
||||
resolvedUsage,
|
||||
});
|
||||
|
||||
tokenRouter.recordSuccess(selected.channel.id, latency, estimatedCost);
|
||||
recordDownstreamCostUsage(request, estimatedCost);
|
||||
logProxy(
|
||||
selected, requestedModel, 'success', 200, latency, null, retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null,
|
||||
resolvedUsage.promptTokens, resolvedUsage.completionTokens, resolvedUsage.totalTokens, estimatedCost, billingDetails,
|
||||
);
|
||||
return reply.send(data);
|
||||
} catch (err: any) {
|
||||
tokenRouter.recordFailure(selected.channel.id);
|
||||
logProxy(selected, requestedModel, 'failed', 0, Date.now() - startTime, err.message, retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null);
|
||||
tokenRouter.recordSuccess(selected.channel.id, latency, estimatedCost);
|
||||
recordDownstreamCostUsage(request, estimatedCost);
|
||||
logProxy(
|
||||
selected, requestedModel, 'success', 200, latency, null, retryCount, downstreamApiKeyId,
|
||||
resolvedUsage.promptTokens, resolvedUsage.completionTokens, resolvedUsage.totalTokens, estimatedCost, billingDetails, clientContext, downstreamPath,
|
||||
);
|
||||
return reply.send(data);
|
||||
} catch (err: any) {
|
||||
tokenRouter.recordFailure(selected.channel.id);
|
||||
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;
|
||||
@@ -270,47 +282,58 @@ export async function completionsProxyRoute(app: FastifyInstance) {
|
||||
});
|
||||
}
|
||||
|
||||
async function logProxy(
|
||||
selected: any,
|
||||
modelRequested: string,
|
||||
status: string,
|
||||
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,
|
||||
) {
|
||||
try {
|
||||
const createdAt = formatUtcSqlDateTime(new Date());
|
||||
const normalizedErrorMessage = composeProxyLogMessage({
|
||||
downstreamPath: '/v1/completions',
|
||||
errorMessage,
|
||||
});
|
||||
await db.insert(schema.proxyLogs).values({
|
||||
routeId: selected.channel.routeId,
|
||||
channelId: selected.channel.id,
|
||||
accountId: selected.account.id,
|
||||
...(downstreamApiKeyId !== null ? { downstreamApiKeyId } : {}),
|
||||
modelRequested,
|
||||
modelActual: selected.actualModel,
|
||||
status,
|
||||
completionTokens = 0,
|
||||
totalTokens = 0,
|
||||
estimatedCost = 0,
|
||||
billingDetails: unknown = null,
|
||||
clientContext: DownstreamClientContext | null = null,
|
||||
downstreamPath = '/v1/completions',
|
||||
) {
|
||||
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: billingDetails ? JSON.stringify(billingDetails) : null,
|
||||
errorMessage: normalizedErrorMessage,
|
||||
retryCount,
|
||||
createdAt,
|
||||
}).run();
|
||||
} catch (error) {
|
||||
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/completions] failed to write proxy log', error);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -71,6 +71,9 @@ vi.mock('../../db/index.js', () => ({
|
||||
schema: {
|
||||
proxyLogs: {},
|
||||
},
|
||||
hasProxyLogBillingDetailsColumn: async () => false,
|
||||
hasProxyLogClientColumns: async () => false,
|
||||
hasProxyLogDownstreamApiKeyIdColumn: async () => false,
|
||||
}));
|
||||
|
||||
describe('downstream client context route logging', () => {
|
||||
|
||||
@@ -130,4 +130,49 @@ describe('detectDownstreamClientContext', () => {
|
||||
clientKind: 'gemini_cli',
|
||||
});
|
||||
});
|
||||
|
||||
it('recognizes app fingerprints alongside a generic protocol family', () => {
|
||||
expect(detectDownstreamClientContext({
|
||||
downstreamPath: '/v1/chat/completions',
|
||||
headers: {
|
||||
'x-title': 'Cherry Studio',
|
||||
'http-referer': 'https://cherry-ai.com',
|
||||
},
|
||||
})).toEqual({
|
||||
clientKind: 'generic',
|
||||
clientAppId: 'cherry_studio',
|
||||
clientAppName: 'Cherry Studio',
|
||||
clientConfidence: 'exact',
|
||||
});
|
||||
});
|
||||
|
||||
it('keeps protocol family detection when an app fingerprint also matches', () => {
|
||||
expect(detectDownstreamClientContext({
|
||||
downstreamPath: '/v1/responses',
|
||||
headers: {
|
||||
originator: 'codex_cli_rs',
|
||||
'x-title': 'Cherry Studio',
|
||||
'http-referer': 'https://cherry-ai.com',
|
||||
},
|
||||
})).toEqual({
|
||||
clientKind: 'codex',
|
||||
clientAppId: 'cherry_studio',
|
||||
clientAppName: 'Cherry Studio',
|
||||
clientConfidence: 'exact',
|
||||
});
|
||||
});
|
||||
|
||||
it('marks weak app-only matches as heuristic instead of upgrading protocol behavior', () => {
|
||||
expect(detectDownstreamClientContext({
|
||||
downstreamPath: '/v1/chat/completions',
|
||||
headers: {
|
||||
'user-agent': 'CherryStudio/1.2.3',
|
||||
},
|
||||
})).toEqual({
|
||||
clientKind: 'generic',
|
||||
clientAppId: 'cherry_studio',
|
||||
clientAppName: 'Cherry Studio',
|
||||
clientConfidence: 'heuristic',
|
||||
});
|
||||
});
|
||||
});
|
||||
|
||||
@@ -6,13 +6,158 @@ import { isCodexResponsesSurface as isCodexResponsesSurfaceViaProfile } from '..
|
||||
import type { CliProfileId } from '../../proxy-core/cliProfiles/types.js';
|
||||
|
||||
export type DownstreamClientKind = CliProfileId;
|
||||
export type DownstreamClientConfidence = 'exact' | 'heuristic';
|
||||
|
||||
export type DownstreamClientContext = {
|
||||
clientKind: DownstreamClientKind;
|
||||
sessionId?: string;
|
||||
traceHint?: string;
|
||||
clientAppId?: string;
|
||||
clientAppName?: string;
|
||||
clientConfidence?: DownstreamClientConfidence;
|
||||
};
|
||||
|
||||
type NormalizedClientHeaders = Record<string, string[]>;
|
||||
|
||||
type DownstreamClientBodySummary = {
|
||||
topLevelKeys: string[];
|
||||
metadataUserId: string | null;
|
||||
};
|
||||
|
||||
type DownstreamClientFingerprintInput = {
|
||||
downstreamPath: string;
|
||||
headers: NormalizedClientHeaders;
|
||||
bodySummary: DownstreamClientBodySummary;
|
||||
};
|
||||
|
||||
type DownstreamClientFingerprintRule = {
|
||||
id: string;
|
||||
name: string;
|
||||
priority: number;
|
||||
match(input: DownstreamClientFingerprintInput): DownstreamClientConfidence | null;
|
||||
};
|
||||
|
||||
function isRecord(value: unknown): value is Record<string, unknown> {
|
||||
return !!value && typeof value === 'object' && !Array.isArray(value);
|
||||
}
|
||||
|
||||
function normalizeHeaderValues(value: unknown): string[] {
|
||||
if (typeof value === 'string') {
|
||||
const trimmed = value.trim();
|
||||
return trimmed ? [trimmed] : [];
|
||||
}
|
||||
|
||||
if (!Array.isArray(value)) return [];
|
||||
return value
|
||||
.filter((item): item is string => typeof item === 'string')
|
||||
.map((item) => item.trim())
|
||||
.filter((item) => item.length > 0);
|
||||
}
|
||||
|
||||
function normalizeHeaders(headers?: Record<string, unknown>): NormalizedClientHeaders {
|
||||
if (!headers) return {};
|
||||
|
||||
const normalized: NormalizedClientHeaders = {};
|
||||
for (const [rawKey, rawValue] of Object.entries(headers)) {
|
||||
const key = rawKey.trim().toLowerCase();
|
||||
if (!key) continue;
|
||||
const values = normalizeHeaderValues(rawValue);
|
||||
if (values.length === 0) continue;
|
||||
normalized[key] = normalized[key]
|
||||
? [...normalized[key], ...values]
|
||||
: values;
|
||||
}
|
||||
return normalized;
|
||||
}
|
||||
|
||||
function headerEquals(headers: NormalizedClientHeaders, key: string, expected: string): boolean {
|
||||
const normalizedExpected = expected.trim().toLowerCase();
|
||||
return (headers[key.trim().toLowerCase()] || []).some((value) => value.trim().toLowerCase() === normalizedExpected);
|
||||
}
|
||||
|
||||
function headerIncludes(headers: NormalizedClientHeaders, key: string, expectedFragment: string): boolean {
|
||||
const normalizedExpected = expectedFragment.trim().toLowerCase();
|
||||
return (headers[key.trim().toLowerCase()] || []).some((value) => value.trim().toLowerCase().includes(normalizedExpected));
|
||||
}
|
||||
|
||||
function buildBodySummary(body: unknown): DownstreamClientBodySummary {
|
||||
if (!isRecord(body)) {
|
||||
return {
|
||||
topLevelKeys: [],
|
||||
metadataUserId: null,
|
||||
};
|
||||
}
|
||||
|
||||
const metadataUserId = isRecord(body.metadata) && typeof body.metadata.user_id === 'string'
|
||||
? body.metadata.user_id.trim() || null
|
||||
: null;
|
||||
|
||||
return {
|
||||
topLevelKeys: Object.keys(body).sort((left, right) => left.localeCompare(right)),
|
||||
metadataUserId,
|
||||
};
|
||||
}
|
||||
|
||||
const appFingerprintRegistry: DownstreamClientFingerprintRule[] = [
|
||||
{
|
||||
id: 'cherry_studio',
|
||||
name: 'Cherry Studio',
|
||||
priority: 100,
|
||||
match(input) {
|
||||
const hasTitle = headerEquals(input.headers, 'x-title', 'Cherry Studio');
|
||||
const hasReferer = headerEquals(input.headers, 'http-referer', 'https://cherry-ai.com')
|
||||
|| headerEquals(input.headers, 'referer', 'https://cherry-ai.com');
|
||||
|
||||
if (hasTitle && hasReferer) {
|
||||
return 'exact';
|
||||
}
|
||||
|
||||
const weakSignals = [
|
||||
headerIncludes(input.headers, 'user-agent', 'cherrystudio'),
|
||||
headerIncludes(input.headers, 'x-title', 'cherry studio'),
|
||||
headerIncludes(input.headers, 'http-referer', 'cherry-ai.com'),
|
||||
headerIncludes(input.headers, 'referer', 'cherry-ai.com'),
|
||||
];
|
||||
|
||||
return weakSignals.some(Boolean) ? 'heuristic' : null;
|
||||
},
|
||||
},
|
||||
];
|
||||
|
||||
function detectDownstreamClientFingerprint(input: {
|
||||
downstreamPath: string;
|
||||
headers?: Record<string, unknown>;
|
||||
body?: unknown;
|
||||
}) {
|
||||
const fingerprintInput: DownstreamClientFingerprintInput = {
|
||||
downstreamPath: input.downstreamPath,
|
||||
headers: normalizeHeaders(input.headers),
|
||||
bodySummary: buildBodySummary(input.body),
|
||||
};
|
||||
|
||||
let matchedRule: DownstreamClientFingerprintRule | null = null;
|
||||
let matchedConfidence: DownstreamClientConfidence | null = null;
|
||||
|
||||
for (const rule of appFingerprintRegistry) {
|
||||
const confidence = rule.match(fingerprintInput);
|
||||
if (!confidence) continue;
|
||||
if (!matchedRule || rule.priority > matchedRule.priority) {
|
||||
matchedRule = rule;
|
||||
matchedConfidence = confidence;
|
||||
}
|
||||
}
|
||||
|
||||
if (!matchedRule || !matchedConfidence) {
|
||||
return null;
|
||||
}
|
||||
|
||||
return {
|
||||
clientAppId: matchedRule.id,
|
||||
clientAppName: matchedRule.name,
|
||||
clientConfidence: matchedConfidence,
|
||||
};
|
||||
}
|
||||
|
||||
export function isCodexResponsesSurface(headers?: Record<string, unknown>): boolean {
|
||||
return isCodexResponsesSurfaceViaProfile(headers);
|
||||
}
|
||||
@@ -27,9 +172,11 @@ export function detectDownstreamClientContext(input: {
|
||||
body?: unknown;
|
||||
}): DownstreamClientContext {
|
||||
const detected = detectCliProfile(input);
|
||||
const fingerprint = detectDownstreamClientFingerprint(input);
|
||||
return {
|
||||
clientKind: detected.id,
|
||||
...(detected.sessionId ? { sessionId: detected.sessionId } : {}),
|
||||
...(detected.traceHint ? { traceHint: detected.traceHint } : {}),
|
||||
...(fingerprint || {}),
|
||||
};
|
||||
}
|
||||
|
||||
@@ -1,6 +1,5 @@
|
||||
import { FastifyInstance, FastifyReply, FastifyRequest } from 'fastify';
|
||||
import { fetch } from 'undici';
|
||||
import { db, hasProxyLogDownstreamApiKeyIdColumn, schema } from '../../db/index.js';
|
||||
import { tokenRouter } from '../../services/tokenRouter.js';
|
||||
import { refreshModelsAndRebuildRoutes } from '../../services/modelService.js';
|
||||
import { reportProxyAllFailed, reportTokenExpired } from '../../services/alertService.js';
|
||||
@@ -15,6 +14,8 @@ 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;
|
||||
|
||||
@@ -28,8 +29,12 @@ export async function embeddingsProxyRoute(app: FastifyInstance) {
|
||||
if (!await ensureModelAllowedForDownstreamKey(request, reply, requestedModel)) return;
|
||||
const downstreamPolicy = getDownstreamRoutingPolicy(request);
|
||||
const downstreamApiKeyId = getProxyAuthContext(request)?.keyId ?? null;
|
||||
const logDownstreamApiKeyId = downstreamApiKeyId !== null
|
||||
&& await hasProxyLogDownstreamApiKeyIdColumn();
|
||||
const downstreamPath = '/v1/embeddings';
|
||||
const clientContext = detectDownstreamClientContext({
|
||||
downstreamPath,
|
||||
headers: request.headers as Record<string, unknown>,
|
||||
body,
|
||||
});
|
||||
|
||||
const excludeChannelIds: number[] = [];
|
||||
let retryCount = 0;
|
||||
@@ -71,7 +76,7 @@ export async function embeddingsProxyRoute(app: FastifyInstance) {
|
||||
const text = await upstream.text();
|
||||
if (!upstream.ok) {
|
||||
tokenRouter.recordFailure(selected.channel.id);
|
||||
logProxy(selected, requestedModel, 'failed', upstream.status, Date.now() - startTime, text, retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null);
|
||||
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({
|
||||
@@ -124,13 +129,13 @@ export async function embeddingsProxyRoute(app: FastifyInstance) {
|
||||
tokenRouter.recordSuccess(selected.channel.id, latency, estimatedCost);
|
||||
recordDownstreamCostUsage(request, estimatedCost);
|
||||
logProxy(
|
||||
selected, requestedModel, 'success', upstream.status, latency, null, retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null,
|
||||
resolvedUsage.promptTokens, resolvedUsage.completionTokens, resolvedUsage.totalTokens, estimatedCost, billingDetails,
|
||||
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);
|
||||
logProxy(selected, requestedModel, 'failed', 0, Date.now() - startTime, err.message, retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null);
|
||||
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;
|
||||
@@ -159,18 +164,25 @@ async function logProxy(
|
||||
totalTokens = 0,
|
||||
estimatedCost = 0,
|
||||
billingDetails: unknown = null,
|
||||
clientContext: DownstreamClientContext | null = null,
|
||||
downstreamPath = '/v1/embeddings',
|
||||
) {
|
||||
try {
|
||||
const createdAt = formatUtcSqlDateTime(new Date());
|
||||
const normalizedErrorMessage = composeProxyLogMessage({
|
||||
downstreamPath: '/v1/embeddings',
|
||||
clientKind: clientContext?.clientKind && clientContext.clientKind !== 'generic'
|
||||
? clientContext.clientKind
|
||||
: null,
|
||||
sessionId: clientContext?.sessionId || null,
|
||||
traceHint: clientContext?.traceHint || null,
|
||||
downstreamPath,
|
||||
errorMessage,
|
||||
});
|
||||
await db.insert(schema.proxyLogs).values({
|
||||
await insertProxyLog({
|
||||
routeId: selected.channel.routeId,
|
||||
channelId: selected.channel.id,
|
||||
accountId: selected.account.id,
|
||||
...(downstreamApiKeyId !== null ? { downstreamApiKeyId } : {}),
|
||||
downstreamApiKeyId,
|
||||
modelRequested,
|
||||
modelActual: selected.actualModel,
|
||||
status,
|
||||
@@ -180,11 +192,15 @@ async function logProxy(
|
||||
completionTokens,
|
||||
totalTokens,
|
||||
estimatedCost,
|
||||
billingDetails: billingDetails ? JSON.stringify(billingDetails) : null,
|
||||
billingDetails,
|
||||
clientFamily: clientContext?.clientKind || null,
|
||||
clientAppId: clientContext?.clientAppId || null,
|
||||
clientAppName: clientContext?.clientAppName || null,
|
||||
clientConfidence: clientContext?.clientConfidence || null,
|
||||
errorMessage: normalizedErrorMessage,
|
||||
retryCount,
|
||||
createdAt,
|
||||
}).run();
|
||||
});
|
||||
} catch (error) {
|
||||
console.warn('[proxy/embeddings] failed to write proxy log', error);
|
||||
}
|
||||
|
||||
@@ -1,7 +1,6 @@
|
||||
import { FastifyInstance, FastifyReply, FastifyRequest } from 'fastify';
|
||||
import { fetch } from 'undici';
|
||||
import { db, hasProxyLogDownstreamApiKeyIdColumn, schema } from '../../db/index.js';
|
||||
import { tokenRouter } from '../../services/tokenRouter.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';
|
||||
@@ -14,6 +13,8 @@ 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;
|
||||
|
||||
@@ -26,8 +27,12 @@ export async function imagesProxyRoute(app: FastifyInstance) {
|
||||
if (!await ensureModelAllowedForDownstreamKey(request, reply, requestedModel)) return;
|
||||
const downstreamPolicy = getDownstreamRoutingPolicy(request);
|
||||
const downstreamApiKeyId = getProxyAuthContext(request)?.keyId ?? null;
|
||||
const logDownstreamApiKeyId = downstreamApiKeyId !== null
|
||||
&& await hasProxyLogDownstreamApiKeyIdColumn();
|
||||
const downstreamPath = '/v1/images/generations';
|
||||
const clientContext = detectDownstreamClientContext({
|
||||
downstreamPath,
|
||||
headers: request.headers as Record<string, unknown>,
|
||||
body,
|
||||
});
|
||||
const excludeChannelIds: number[] = [];
|
||||
let retryCount = 0;
|
||||
|
||||
@@ -70,7 +75,7 @@ export async function imagesProxyRoute(app: FastifyInstance) {
|
||||
const text = await upstream.text();
|
||||
if (!upstream.ok) {
|
||||
tokenRouter.recordFailure(selected.channel.id);
|
||||
logProxy(selected, requestedModel, 'failed', upstream.status, Date.now() - startTime, text, retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null);
|
||||
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,
|
||||
@@ -104,11 +109,11 @@ export async function imagesProxyRoute(app: FastifyInstance) {
|
||||
});
|
||||
tokenRouter.recordSuccess(selected.channel.id, latency, estimatedCost);
|
||||
recordDownstreamCostUsage(request, estimatedCost);
|
||||
logProxy(selected, requestedModel, 'success', upstream.status, latency, null, retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null, 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);
|
||||
logProxy(selected, requestedModel, 'failed', 0, Date.now() - startTime, err.message, retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null);
|
||||
logProxy(selected, requestedModel, 'failed', 0, Date.now() - startTime, err.message, retryCount, downstreamApiKeyId, 0, downstreamPath, clientContext);
|
||||
if (retryCount < MAX_RETRIES) {
|
||||
retryCount++;
|
||||
continue;
|
||||
@@ -136,8 +141,12 @@ export async function imagesProxyRoute(app: FastifyInstance) {
|
||||
if (!await ensureModelAllowedForDownstreamKey(request, reply, requestedModel)) return;
|
||||
const downstreamPolicy = getDownstreamRoutingPolicy(request);
|
||||
const downstreamApiKeyId = getProxyAuthContext(request)?.keyId ?? null;
|
||||
const logDownstreamApiKeyId = downstreamApiKeyId !== null
|
||||
&& await hasProxyLogDownstreamApiKeyIdColumn();
|
||||
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;
|
||||
|
||||
@@ -192,7 +201,7 @@ export async function imagesProxyRoute(app: FastifyInstance) {
|
||||
const text = await upstream.text();
|
||||
if (!upstream.ok) {
|
||||
tokenRouter.recordFailure(selected.channel.id);
|
||||
logProxy(selected, requestedModel, 'failed', upstream.status, Date.now() - startTime, text, retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null, 0, '/v1/images/edits');
|
||||
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,
|
||||
@@ -226,11 +235,11 @@ export async function imagesProxyRoute(app: FastifyInstance) {
|
||||
});
|
||||
tokenRouter.recordSuccess(selected.channel.id, latency, estimatedCost);
|
||||
recordDownstreamCostUsage(request, estimatedCost);
|
||||
logProxy(selected, requestedModel, 'success', upstream.status, latency, null, retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null, estimatedCost, '/v1/images/edits');
|
||||
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);
|
||||
logProxy(selected, requestedModel, 'failed', 0, Date.now() - startTime, err.message, retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null, 0, '/v1/images/edits');
|
||||
logProxy(selected, requestedModel, 'failed', 0, Date.now() - startTime, err.message, retryCount, downstreamApiKeyId, 0, downstreamPath, clientContext);
|
||||
if (retryCount < MAX_RETRIES) {
|
||||
retryCount++;
|
||||
continue;
|
||||
@@ -267,18 +276,24 @@ async function logProxy(
|
||||
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 db.insert(schema.proxyLogs).values({
|
||||
await insertProxyLog({
|
||||
routeId: selected.channel.routeId,
|
||||
channelId: selected.channel.id,
|
||||
accountId: selected.account.id,
|
||||
...(downstreamApiKeyId !== null ? { downstreamApiKeyId } : {}),
|
||||
downstreamApiKeyId,
|
||||
modelRequested,
|
||||
modelActual: selected.actualModel,
|
||||
status,
|
||||
@@ -288,10 +303,14 @@ async function logProxy(
|
||||
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,
|
||||
}).run();
|
||||
});
|
||||
} catch (error) {
|
||||
console.warn('[proxy/images] failed to write proxy log', error);
|
||||
}
|
||||
|
||||
@@ -7,7 +7,7 @@ type ComposeProxyLogMessageArgs = {
|
||||
errorMessage?: string | null;
|
||||
};
|
||||
|
||||
type ParsedPathMeta = {
|
||||
export type ParsedProxyLogMessageMeta = {
|
||||
clientKind: string | null;
|
||||
sessionId: string | null;
|
||||
downstreamPath: string | null;
|
||||
@@ -15,7 +15,7 @@ type ParsedPathMeta = {
|
||||
messageText: string;
|
||||
};
|
||||
|
||||
function parseExistingPathMeta(rawMessage: string): ParsedPathMeta {
|
||||
export function parseProxyLogMessageMeta(rawMessage: string): ParsedProxyLogMessageMeta {
|
||||
const clientMatch = rawMessage.match(/\[client:([^\]]+)\]/i);
|
||||
const sessionMatch = rawMessage.match(/\[session:([^\]]+)\]/i);
|
||||
const downstreamMatch = rawMessage.match(/\[downstream:([^\]]+)\]/i);
|
||||
@@ -42,7 +42,7 @@ export function composeProxyLogMessage({
|
||||
errorMessage,
|
||||
}: ComposeProxyLogMessageArgs): string | null {
|
||||
const rawMessage = typeof errorMessage === 'string' ? errorMessage.trim() : '';
|
||||
const parsed = parseExistingPathMeta(rawMessage);
|
||||
const parsed = parseProxyLogMessageMeta(rawMessage);
|
||||
const finalClientKind = (clientKind || parsed.clientKind || '').trim();
|
||||
const finalSessionId = (sessionId || traceHint || parsed.sessionId || '').trim();
|
||||
const finalDownstreamPath = (downstreamPath || parsed.downstreamPath || '').trim();
|
||||
|
||||
@@ -1,6 +1,5 @@
|
||||
import { FastifyInstance, FastifyReply, FastifyRequest } from 'fastify';
|
||||
import { fetch } from 'undici';
|
||||
import { db, hasProxyLogDownstreamApiKeyIdColumn, schema } from '../../db/index.js';
|
||||
import { tokenRouter } from '../../services/tokenRouter.js';
|
||||
import { refreshModelsAndRebuildRoutes } from '../../services/modelService.js';
|
||||
import { reportProxyAllFailed, reportTokenExpired } from '../../services/alertService.js';
|
||||
@@ -12,6 +11,8 @@ import { composeProxyLogMessage } from './logPathMeta.js';
|
||||
import { formatUtcSqlDateTime } from '../../services/localTimeService.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;
|
||||
const DEFAULT_SEARCH_MODEL = '__search';
|
||||
@@ -54,8 +55,12 @@ export async function searchProxyRoute(app: FastifyInstance) {
|
||||
if (!await ensureModelAllowedForDownstreamKey(request, reply, requestedModel)) return;
|
||||
const downstreamPolicy = getDownstreamRoutingPolicy(request);
|
||||
const downstreamApiKeyId = getProxyAuthContext(request)?.keyId ?? null;
|
||||
const logDownstreamApiKeyId = downstreamApiKeyId !== null
|
||||
&& await hasProxyLogDownstreamApiKeyIdColumn();
|
||||
const downstreamPath = '/v1/search';
|
||||
const clientContext = detectDownstreamClientContext({
|
||||
downstreamPath,
|
||||
headers: request.headers as Record<string, unknown>,
|
||||
body,
|
||||
});
|
||||
const excludeChannelIds: number[] = [];
|
||||
let retryCount = 0;
|
||||
|
||||
@@ -101,7 +106,7 @@ export async function searchProxyRoute(app: FastifyInstance) {
|
||||
const text = await upstream.text();
|
||||
if (!upstream.ok) {
|
||||
tokenRouter.recordFailure(selected.channel.id);
|
||||
logProxy(selected, requestedModel, 'failed', upstream.status, Date.now() - startTime, text, retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null);
|
||||
logProxy(selected, requestedModel, 'failed', upstream.status, Date.now() - startTime, text, retryCount, downstreamApiKeyId, clientContext, downstreamPath);
|
||||
if (isTokenExpiredError({ status: upstream.status, message: text })) {
|
||||
await reportTokenExpired({
|
||||
accountId: selected.account.id,
|
||||
@@ -127,11 +132,11 @@ export async function searchProxyRoute(app: FastifyInstance) {
|
||||
const latency = Date.now() - startTime;
|
||||
tokenRouter.recordSuccess(selected.channel.id, latency, 0);
|
||||
recordDownstreamCostUsage(request, 0);
|
||||
logProxy(selected, requestedModel, 'success', upstream.status, latency, null, retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null);
|
||||
logProxy(selected, requestedModel, 'success', upstream.status, latency, null, retryCount, downstreamApiKeyId, clientContext, downstreamPath);
|
||||
return reply.code(upstream.status).send(data);
|
||||
} catch (error: any) {
|
||||
tokenRouter.recordFailure(selected.channel.id);
|
||||
logProxy(selected, requestedModel, 'failed', 0, Date.now() - startTime, error?.message || 'network error', retryCount, logDownstreamApiKeyId ? downstreamApiKeyId : null);
|
||||
logProxy(selected, requestedModel, 'failed', 0, Date.now() - startTime, error?.message || 'network error', retryCount, downstreamApiKeyId, clientContext, downstreamPath);
|
||||
if (retryCount < MAX_RETRIES) {
|
||||
retryCount += 1;
|
||||
continue;
|
||||
@@ -157,14 +162,16 @@ async function logProxy(
|
||||
errorMessage: string | null,
|
||||
retryCount: number,
|
||||
downstreamApiKeyId: number | null = null,
|
||||
clientContext: DownstreamClientContext | null = null,
|
||||
downstreamPath = '/v1/search',
|
||||
) {
|
||||
try {
|
||||
const createdAt = formatUtcSqlDateTime(new Date());
|
||||
await db.insert(schema.proxyLogs).values({
|
||||
await insertProxyLog({
|
||||
routeId: selected.channel.routeId,
|
||||
channelId: selected.channel.id,
|
||||
accountId: selected.account.id,
|
||||
...(downstreamApiKeyId !== null ? { downstreamApiKeyId } : {}),
|
||||
downstreamApiKeyId,
|
||||
modelRequested,
|
||||
modelActual: selected.actualModel,
|
||||
status,
|
||||
@@ -175,12 +182,21 @@ async function logProxy(
|
||||
totalTokens: 0,
|
||||
estimatedCost: 0,
|
||||
errorMessage: composeProxyLogMessage({
|
||||
downstreamPath: '/v1/search',
|
||||
clientKind: clientContext?.clientKind && clientContext.clientKind !== 'generic'
|
||||
? clientContext.clientKind
|
||||
: null,
|
||||
sessionId: clientContext?.sessionId || null,
|
||||
traceHint: clientContext?.traceHint || null,
|
||||
downstreamPath,
|
||||
errorMessage,
|
||||
}),
|
||||
clientFamily: clientContext?.clientKind || null,
|
||||
clientAppId: clientContext?.clientAppId || null,
|
||||
clientAppName: clientContext?.clientAppName || null,
|
||||
clientConfidence: clientContext?.clientConfidence || null,
|
||||
retryCount,
|
||||
createdAt,
|
||||
}).run();
|
||||
});
|
||||
} catch (error) {
|
||||
console.warn('[proxy/search] failed to write proxy log', error);
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user