Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.
Go agentagenticaichatbotdsh-pluginembeddingsevaluationgenerative-aigolangknowledge-basellmmulti-tenantollamaopenaiquestion-answeringragrerankingsemantic-searchvector-searchwiki
/**
* 「刷新」语义的请求排序器(/auth/me 这类单例读取用)。
*
* - `refresh()`:结果必须来自本次调用之后发出的请求。已有请求在飞行时,
* 排一个新请求在它之后,多次并发的 refresh 共用同一个排队请求。
* - `share()`:只要有请求在飞行就直接复用,否则发一个新的。首屏多个消费者
* 同时要同一份数据时用它。
* - `invalidate()`:让飞行中的请求作废。`run` 拿到的 `isCurrent()` 在 await 之后
* 返回 false,调用方据此跳过落库;登出时用,避免迟到的响应把会话写回来。
*/
export interface SequencedRefresh<T> {
refresh: () => Promise<T>
share: () => Promise<T>
invalidate: () => void
hasInFlightRequest: () => boolean
}
export function createSequencedRefresh<T>(
run: (isCurrent: () => boolean) => Promise<T>,
): SequencedRefresh<T> {
let revision = 0
let inflight: Promise<T> | null = null
let queued: Promise<T> | null = null
const start = (): Promise<T> => {
const requestRevision = revision
const current = run(() => requestRevision === revision).finally(() => {
if (inflight === current) inflight = null
})
inflight = current
return current
}
const refresh = (): Promise<T> => {
if (!inflight) return start()
if (!queued) {
const waitingFor = inflight
queued = waitingFor.then(
() => {
queued = null
return start()
},
() => {
queued = null
return start()
},
)
}
return queued
}
return {
refresh,
share: () => inflight ?? start(),
invalidate: () => {
revision += 1
},
hasInFlightRequest: () => inflight !== null,
}
}