mirror of
https://github.com/fosrl/pangolin.git
synced 2026-08-12 15:30:53 +02:00
link the usage with the session to display together
This commit is contained in:
@@ -3411,6 +3411,9 @@
|
||||
"aiSessionLogTruncated": "This session was truncated before storage and may be incomplete.",
|
||||
"aiSessionViewRaw": "View Raw JSON",
|
||||
"aiSessionViewChat": "View Chat",
|
||||
"tokens": "Tokens",
|
||||
"cost": "Cost",
|
||||
"estimated": "estimated",
|
||||
"requestAnalyticsDescription": "View detailed request analytics for resources in this organization",
|
||||
"logRetentionRequestLabel": "HTTP Request Log Retention",
|
||||
"logRetentionRequestDescription": "How long to retain request logs",
|
||||
|
||||
@@ -1831,6 +1831,13 @@ export const aiUsageRecords = pgTable(
|
||||
userId: varchar("userId").references(() => users.userId, {
|
||||
onDelete: "set null"
|
||||
}),
|
||||
// Links this usage record back to the aiSessionLog row for the same
|
||||
// request (aiSessionLog.sessionId), so token/cost usage can be shown
|
||||
// alongside the session transcript. Not a DB-level FK - aiSessionLog
|
||||
// lives in the separate logs database. Nullable because the session
|
||||
// log may be disabled (retention set to 0) while usage tracking
|
||||
// stays on.
|
||||
sessionId: varchar("sessionId"),
|
||||
requestedModel: varchar("requestedModel").notNull(),
|
||||
promptTokens: integer("promptTokens").notNull().default(0),
|
||||
cacheReadTokens: integer("cacheReadTokens").notNull().default(0),
|
||||
@@ -1862,7 +1869,8 @@ export const aiUsageRecords = pgTable(
|
||||
t.orgId,
|
||||
t.userId,
|
||||
t.createdAt
|
||||
)
|
||||
),
|
||||
index("idx_ai_usage_records_session").on(t.sessionId)
|
||||
]
|
||||
);
|
||||
|
||||
|
||||
@@ -1819,6 +1819,13 @@ export const aiUsageRecords = sqliteTable(
|
||||
userId: text("userId").references(() => users.userId, {
|
||||
onDelete: "set null"
|
||||
}),
|
||||
// Links this usage record back to the aiSessionLog row for the same
|
||||
// request (aiSessionLog.sessionId), so token/cost usage can be shown
|
||||
// alongside the session transcript. Not a DB-level FK - aiSessionLog
|
||||
// lives in the separate logs database. Nullable because the session
|
||||
// log may be disabled (retention set to 0) while usage tracking
|
||||
// stays on.
|
||||
sessionId: text("sessionId"),
|
||||
requestedModel: text("requestedModel").notNull(),
|
||||
promptTokens: integer("promptTokens").notNull().default(0),
|
||||
cacheReadTokens: integer("cacheReadTokens").notNull().default(0),
|
||||
@@ -1852,7 +1859,8 @@ export const aiUsageRecords = sqliteTable(
|
||||
t.orgId,
|
||||
t.userId,
|
||||
t.createdAt
|
||||
)
|
||||
),
|
||||
index("idx_ai_usage_records_session").on(t.sessionId)
|
||||
]
|
||||
);
|
||||
|
||||
|
||||
@@ -428,6 +428,11 @@ export type UsageRecordInput = {
|
||||
usage: AiUsage;
|
||||
costUsd: number | null;
|
||||
createdAt?: number;
|
||||
// Same id as the aiSessionLog row logged for this request, so the two
|
||||
// can be joined to show token/cost usage alongside the session
|
||||
// transcript. Undefined when the session wasn't logged (e.g. session
|
||||
// log retention disabled for the org).
|
||||
sessionId?: string;
|
||||
};
|
||||
|
||||
type AiUsageRecordInsert = InferInsertModel<typeof aiUsageRecords>;
|
||||
@@ -539,6 +544,7 @@ export async function recordUsage(input: UsageRecordInput): Promise<void> {
|
||||
resourceId: input.resourceId,
|
||||
siteResourceId: input.siteResourceId,
|
||||
userId: input.userId,
|
||||
sessionId: input.sessionId,
|
||||
requestedModel: input.requestedModel,
|
||||
promptTokens: usage.promptTokens,
|
||||
cacheReadTokens: usage.cacheReadTokens,
|
||||
|
||||
@@ -1,4 +1,3 @@
|
||||
import { randomUUID } from "crypto";
|
||||
import { logsDb, db, orgs, aiSessionLog, type AiProvider } from "@server/db";
|
||||
import type { InferInsertModel } from "drizzle-orm";
|
||||
import logger from "@server/logger";
|
||||
@@ -169,6 +168,7 @@ function truncateBody(value: string): { value: string; truncated: boolean } {
|
||||
}
|
||||
|
||||
export function logAiSession(data: {
|
||||
sessionId: string;
|
||||
capability: AiCapability;
|
||||
provider: AiProvider;
|
||||
requestedModel: string | undefined;
|
||||
@@ -230,7 +230,7 @@ export function logAiSession(data: {
|
||||
}
|
||||
|
||||
sessionLogBuffer.push({
|
||||
sessionId: randomUUID(),
|
||||
sessionId: data.sessionId,
|
||||
orgId: sanitizeString(data.orgId),
|
||||
providerId: data.provider.providerId,
|
||||
capability: data.capability,
|
||||
|
||||
@@ -1,3 +1,4 @@
|
||||
import { randomUUID } from "crypto";
|
||||
import { Request, Response } from "express";
|
||||
import { and, eq, inArray } from "drizzle-orm";
|
||||
import {
|
||||
@@ -594,6 +595,12 @@ export function recordAiGatewayCompletion(args: {
|
||||
const pricing = getModelPricing(provider.type as AiProviderType, model);
|
||||
const cost = calculateAiCost(pricing, usage);
|
||||
|
||||
// Shared by the usage record and the session log so the two can be
|
||||
// joined later to show token/cost usage alongside the transcript -
|
||||
// generated up front since neither buffered insert's row id is known
|
||||
// until its next batch flush.
|
||||
const sessionId = randomUUID();
|
||||
|
||||
logger.info("AI gateway request usage", {
|
||||
capability,
|
||||
providerId: provider.providerId,
|
||||
@@ -618,7 +625,8 @@ export function recordAiGatewayCompletion(args: {
|
||||
userId: requestUserId,
|
||||
requestedModel: model ?? "unknown",
|
||||
usage,
|
||||
costUsd: cost?.totalCost ?? null
|
||||
costUsd: cost?.totalCost ?? null,
|
||||
sessionId
|
||||
});
|
||||
|
||||
if (budgets.length > 0) {
|
||||
@@ -635,6 +643,7 @@ export function recordAiGatewayCompletion(args: {
|
||||
}
|
||||
|
||||
logAiSession({
|
||||
sessionId,
|
||||
capability,
|
||||
provider,
|
||||
requestedModel,
|
||||
|
||||
@@ -2,6 +2,7 @@ import {
|
||||
logsDb,
|
||||
aiSessionLog,
|
||||
aiProviders,
|
||||
aiUsageRecords,
|
||||
resources,
|
||||
siteResources,
|
||||
users,
|
||||
@@ -253,6 +254,51 @@ async function enrichWithDetails(
|
||||
}
|
||||
}
|
||||
|
||||
const usageMap = new Map<
|
||||
string,
|
||||
{
|
||||
promptTokens: number;
|
||||
cacheReadTokens: number;
|
||||
cacheWriteTokens: number;
|
||||
completionTokens: number;
|
||||
reasoningTokens: number;
|
||||
totalTokens: number;
|
||||
costUsd: number | null;
|
||||
estimated: boolean;
|
||||
}
|
||||
>();
|
||||
const sessionIds = logs.map((log) => log.sessionId);
|
||||
if (sessionIds.length > 0) {
|
||||
const usageDetails = await primaryDb
|
||||
.select({
|
||||
sessionId: aiUsageRecords.sessionId,
|
||||
promptTokens: aiUsageRecords.promptTokens,
|
||||
cacheReadTokens: aiUsageRecords.cacheReadTokens,
|
||||
cacheWriteTokens: aiUsageRecords.cacheWriteTokens,
|
||||
completionTokens: aiUsageRecords.completionTokens,
|
||||
reasoningTokens: aiUsageRecords.reasoningTokens,
|
||||
totalTokens: aiUsageRecords.totalTokens,
|
||||
costUsd: aiUsageRecords.costUsd,
|
||||
estimated: aiUsageRecords.estimated
|
||||
})
|
||||
.from(aiUsageRecords)
|
||||
.where(inArray(aiUsageRecords.sessionId, sessionIds));
|
||||
|
||||
for (const u of usageDetails) {
|
||||
if (!u.sessionId) continue;
|
||||
usageMap.set(u.sessionId, {
|
||||
promptTokens: u.promptTokens,
|
||||
cacheReadTokens: u.cacheReadTokens,
|
||||
cacheWriteTokens: u.cacheWriteTokens,
|
||||
completionTokens: u.completionTokens,
|
||||
reasoningTokens: u.reasoningTokens,
|
||||
totalTokens: u.totalTokens,
|
||||
costUsd: u.costUsd,
|
||||
estimated: u.estimated
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
return logs.map((log) => {
|
||||
const provider = providerMap.get(log.providerId);
|
||||
|
||||
@@ -281,7 +327,8 @@ async function enrichWithDetails(
|
||||
providerType: provider?.type ?? null,
|
||||
resourceName,
|
||||
resourceNiceId,
|
||||
userEmail: log.userId ? (userMap.get(log.userId) ?? null) : null
|
||||
userEmail: log.userId ? (userMap.get(log.userId) ?? null) : null,
|
||||
usage: usageMap.get(log.sessionId) ?? null
|
||||
};
|
||||
});
|
||||
}
|
||||
|
||||
@@ -119,6 +119,16 @@ export type QueryAiSessionLogResponse = {
|
||||
truncated: boolean;
|
||||
statusCode: number | null;
|
||||
createdAt: number;
|
||||
usage: {
|
||||
promptTokens: number;
|
||||
cacheReadTokens: number;
|
||||
cacheWriteTokens: number;
|
||||
completionTokens: number;
|
||||
reasoningTokens: number;
|
||||
totalTokens: number;
|
||||
costUsd: number | null;
|
||||
estimated: boolean;
|
||||
} | null;
|
||||
}[];
|
||||
pagination: {
|
||||
total: number;
|
||||
|
||||
@@ -459,7 +459,7 @@ export default function AiSessionLogsPage() {
|
||||
const renderExpandedRow = (row: any) => {
|
||||
return (
|
||||
<div className="space-y-4">
|
||||
<div className="grid grid-cols-2 gap-4 text-xs">
|
||||
<div className="grid grid-cols-2 sm:grid-cols-4 gap-4 text-xs">
|
||||
<div>
|
||||
<strong>{t("aiSessionId")}</strong>
|
||||
<p className="text-muted-foreground mt-1 break-all">
|
||||
@@ -472,6 +472,26 @@ export default function AiSessionLogsPage() {
|
||||
{row.statusCode ?? "N/A"}
|
||||
</p>
|
||||
</div>
|
||||
<div>
|
||||
<strong>{t("tokens")}</strong>
|
||||
<p className="text-muted-foreground mt-1">
|
||||
{row.usage
|
||||
? `${row.usage.totalTokens.toLocaleString()}${
|
||||
row.usage.estimated
|
||||
? ` (${t("estimated")})`
|
||||
: ""
|
||||
}`
|
||||
: "N/A"}
|
||||
</p>
|
||||
</div>
|
||||
<div>
|
||||
<strong>{t("cost")}</strong>
|
||||
<p className="text-muted-foreground mt-1">
|
||||
{row.usage && row.usage.costUsd != null
|
||||
? `$${row.usage.costUsd.toFixed(4)}`
|
||||
: "N/A"}
|
||||
</p>
|
||||
</div>
|
||||
</div>
|
||||
<AiSessionChatView
|
||||
normalizedRequest={row.normalizedRequest}
|
||||
@@ -572,7 +592,17 @@ function generateSampleAiSessionLogs(): QueryAiSessionLogResponse["log"] {
|
||||
statusCode: 200,
|
||||
createdAt: Math.floor(
|
||||
sevenDaysAgoMs + Math.random() * (now - sevenDaysAgoMs)
|
||||
)
|
||||
),
|
||||
usage: {
|
||||
promptTokens: 500,
|
||||
cacheReadTokens: 0,
|
||||
cacheWriteTokens: 0,
|
||||
completionTokens: 150,
|
||||
reasoningTokens: 0,
|
||||
totalTokens: 650,
|
||||
costUsd: 0.0123,
|
||||
estimated: false
|
||||
}
|
||||
};
|
||||
});
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user