feat(admin): Agent 并发 - 后台可调同时对话上限并展示 Agent 服务状态
新增运行策略 agentMaxSessions(默认 30,范围 1-64),保存后同时下发到内嵌进程闸门与独立 yingce-agent 的 /v1/limit,降低上限不会中断已开始的对话。系统性能页展示 Agent 服务健康、进行中/排队数量,并可直接修改上限。
This commit is contained in:
@@ -350,6 +350,8 @@ export type RuntimeTaskPolicy = {
|
||||
agentStepMaxOutputTokens: number;
|
||||
/** 画布 Agent 单步模型调用的秒级墙钟;0 表示沿用文本任务超时。 */
|
||||
agentStepTimeoutSeconds: number;
|
||||
/** 同时进行中的画布 Agent 轮次上限。审批等待不占用名额。 */
|
||||
agentMaxSessions: number;
|
||||
};
|
||||
|
||||
export type RuntimeRequestPolicy = {
|
||||
|
||||
@@ -105,9 +105,23 @@ export type SystemPerformance = {
|
||||
cacheGroups: SystemPerformanceCacheGroup[];
|
||||
statusMessage?: string;
|
||||
};
|
||||
agents: SystemPerformanceAgent[];
|
||||
build: { version: string; commit: string; buildTime: string; goVersion: string };
|
||||
};
|
||||
|
||||
export type SystemPerformanceAgent = {
|
||||
id: string;
|
||||
name: string;
|
||||
mode: "remote" | "embedded";
|
||||
endpoint: string;
|
||||
healthy: boolean;
|
||||
active: number;
|
||||
limit: number;
|
||||
queued: number;
|
||||
configuredLimit: number;
|
||||
statusMessage?: string;
|
||||
};
|
||||
|
||||
export type RuntimeCacheClearResult = {
|
||||
scope: "runtime";
|
||||
deletedKeys: number;
|
||||
@@ -120,6 +134,10 @@ export function getSystemPerformance(signal?: AbortSignal) {
|
||||
return http.get<SystemPerformance>("/admin/system-performance", { signal });
|
||||
}
|
||||
|
||||
export function updateAgentSessionLimit(maxSessions: number) {
|
||||
return http.put<SystemPerformanceAgent>("/admin/system-performance/agent-limit", { maxSessions });
|
||||
}
|
||||
|
||||
export function clearRuntimeCache() {
|
||||
return http.post<RuntimeCacheClearResult>("/admin/system-performance/cache/clear", { scope: "runtime" });
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user