224 lines
9.8 KiB
TypeScript
224 lines
9.8 KiB
TypeScript
|
/**
|
|||
|
|
* Worker 侧的**中继隧道**（覆盖网络 R4）—— `SshTunnel` 的同面替代品。
|
|||
|
|
*
|
|||
|
|
* ## 它替掉的是什么
|
|||
|
|
* 今天"Manager 连不到 Worker"靠的是 `ssh -R`：Worker 拨 Manager 的 sshd，把两边的
|
|||
|
|
* `127.0.0.1:<port>` 接起来。代价是**长期依赖 sshd + root 凭据 + 一条公网 SSH 口**。
|
|||
|
|
* 本类改用自研 relay：Worker 只拨一条 wss，relay 为每个注册端口在**它自己的回环**上开一条
|
|||
|
|
* 监听 ⇒ 对 Manager 而言"又有一个 `host:port` 可以拨"，形态与 sshd 版同形（S0 抽 `Reachability`
|
|||
|
|
* 的全部意义）。**Worker 依旧零入站口**。
|
|||
|
|
*
|
|||
|
|
* ## 与 `SshTunnel` 的**唯一实质差别**（改代码前务必知道）
|
|||
|
|
* | | 端口号 | 谁来翻译 |
|
|||
|
|
* |---|---|---|
|
|||
|
|
* | `SshTunnel` | **同号**（`-R p:127.0.0.1:p`）⇒ Manager 侧端口 = Worker 侧端口 | 不需要翻译 |
|
|||
|
|
* | `RelayTunnel` | relay **动态分配**（实测 42067）⇒ Manager 侧端口 ≠ Worker 侧端口 | **Manager 侧**按 `(hostId, port)` 查 relay `/status` 翻译 |
|
|||
|
|
*
|
|||
|
|
* ⇒ 所以本类**只负责"把这个口声明给 relay"**，绝不把 relay 的回环口号回传给 agent 去上报
|
|||
|
|
* （那会把"Manager 侧地址观"塞进 Worker，relay 换机部署时两头都要改）。
|
|||
|
|
* Manager 侧的翻译在 `web/server.ts` 的 relay 快照 + `RemoteSpawner` 的 endpoint 翻译器里。
|
|||
|
|
*
|
|||
|
|
* ## 动态端口（实例端口）
|
|||
|
|
* 实例端口是 `findFreePort()` 运行期分配的 ⇒ 用 `PORT_ADD` / `PORT_DEL`（对应 ssh 的
|
|||
|
|
* `-O forward` / `-O cancel`），**不是**开局声明一整段（那样 relay 要为段内每个口开监听）。
|
|||
|
|
* 断链后由 `RelayClient` 自动重放（见 `client.ts#replayDynamicPorts`）—— 这是"注册恢复"那一层。
|
|||
|
|
*
|
|||
|
|
* @module dshs/worker/relay-tunnel
|
|||
|
|
*/
|
|||
|
|
|
|||
|
|
import { RelayClient, waitUpOnStatus, type RelayClientOptions, type WebSocketCtor } from '../net/relay/client.js'
|
|||
|
|
import { RelayFailoverSupervisor } from '../net/relay/switcher.js'
|
|||
|
|
import type { RelayChannelHandle, RelayFailoverThresholds } from '../net/relay/switcher.js'
|
|||
|
|
import type { WorkerTunnel } from './tunnel.js'
|
|||
|
|
|
|||
|
|
export interface RelayTunnelOptions {
|
|||
|
|
/** relay 的 WebSocket 地址：`wss://alotbuy.com/dshs-relay`（生产）或 `ws://127.0.0.1:20080/dshs-relay`（本机验）。 */
|
|||
|
|
url: string
|
|||
|
|
/** 本机在 `dsh_hosts.id` 里的标识（`w-47` / `w-106`）。 */
|
|||
|
|
hostId: string
|
|||
|
|
/** 与 relay 的预共享密钥（hex）。**缺失必须吵** —— 静默回退到别的传输比报错危险得多。 */
|
|||
|
|
secret: string
|
|||
|
|
/**
|
|||
|
|
* **本机节点身份**（覆盖网络 序③）：私钥 PEM + 入网凭据。
|
|||
|
|
*
|
|||
|
|
* 缺省 ⇒ `HELLO` 不带身份字段（过渡期形态，relay 未强制时照旧可用）；
|
|||
|
|
* 给了 ⇒ relay 侧可按**受信签名者**独立验证"这台机器被授权进入这张网"，
|
|||
|
|
* 而不再只依赖那张放在控制面上的密钥表。
|
|||
|
|
*/
|
|||
|
|
identity?: RelayClientOptions['identity']
|
|||
|
|
/** 启动即声明的端口（agent 自身；还可带控制面 PG 等）。 */
|
|||
|
|
staticPorts?: number[]
|
|||
|
|
/** 首连等待上限（ms）。默认 12s：跨云一次 RTT 也就几十 ms，等这么久只可能是真连不上。 */
|
|||
|
|
upTimeoutMs?: number
|
|||
|
|
log?: (line: string) => void
|
|||
|
|
webSocketCtor?: WebSocketCtor
|
|||
|
|
/**
|
|||
|
|
* **中继失败切流**（序⑦ · C2 装配点）。给了才启用；不给 ⇒ 行为与改造前**逐字一致**
|
|||
|
|
* （启动解析一次、此后钉死 —— 这正是改造前 E9 后半不成立的原因）。
|
|||
|
|
*
|
|||
|
|
* ⛔ **候选链只来自"同一份引导链"**（{@link listOverlayRelayCandidates}）⇒ 只可能是
|
|||
|
|
* **已签名目录 / 内置种子**里的地址，⛔ 不接受任意 url（否则就是任意重定向 = 真 R5）。
|
|||
|
|
* ⚠️ 它**不改会合面**：起始地址仍由 `DSHS_RENDEZVOUS_URL` 决定（见 §3.2），
|
|||
|
|
* 这里只补上"起始那台不健康时换到链里的下一条"。
|
|||
|
|
*/
|
|||
|
|
failover?: {
|
|||
|
|
candidates: () => Promise<readonly string[]>
|
|||
|
|
thresholds?: Partial<RelayFailoverThresholds>
|
|||
|
|
}
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
/** 内部：一条通道 + 它自己的 `RelayClient`。 */
|
|||
|
|
type TunnelChannel = RelayChannelHandle & { client: RelayClient }
|
|||
|
|
|
|||
|
|
function healthOf(client: RelayClient): { state: string; attempts: number; unhealthyForMs: number } {
|
|||
|
|
const st = client.status()
|
|||
|
|
return { state: st.state, attempts: st.attempts, unhealthyForMs: st.unhealthyForMs }
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
export class RelayTunnel implements WorkerTunnel {
|
|||
|
|
private readonly opts: RelayTunnelOptions
|
|||
|
|
private readonly forwarded = new Set<number>()
|
|||
|
|
private readonly log: (line: string) => void
|
|||
|
|
private readonly failover: RelayFailoverSupervisor | undefined
|
|||
|
|
/** 起始通道（监管器不在场时它就是唯一通道）。 */
|
|||
|
|
private readonly initialChannel: TunnelChannel
|
|||
|
|
|
|||
|
|
constructor(options: RelayTunnelOptions) {
|
|||
|
|
this.opts = options
|
|||
|
|
this.log = options.log ?? ((line: string) => process.stdout.write(`${line}\n`))
|
|||
|
|
this.initialChannel = this.channelFor(this.buildClient(options.url), options.url)
|
|||
|
|
if (options.failover === undefined) {
|
|||
|
|
this.failover = undefined
|
|||
|
|
return
|
|||
|
|
}
|
|||
|
|
const sup = new RelayFailoverSupervisor({
|
|||
|
|
/**
|
|||
|
|
* **先建新、成功再关旧**：新客户端必须先真的到 `up`，本函数才返回句柄；
|
|||
|
|
* 到不了 ⇒ 返回 `undefined` ⇒ 监管器**保持原通道**（D4）。
|
|||
|
|
* 新客户端**一次就把已声明的端口全带上**（`staticPorts` + 运行期 `forwarded`），
|
|||
|
|
* 否则切换完成后"实例面端口没了"—— 那是比不切更糟的结果。
|
|||
|
|
*/
|
|||
|
|
open: async (url) => {
|
|||
|
|
const next = this.buildClient(url)
|
|||
|
|
next.start()
|
|||
|
|
const ok = await this.waitUpOn(next, this.opts.upTimeoutMs ?? 12_000)
|
|||
|
|
if (!ok) {
|
|||
|
|
next.stop()
|
|||
|
|
return undefined
|
|||
|
|
}
|
|||
|
|
return this.channelFor(next, url)
|
|||
|
|
},
|
|||
|
|
candidates: options.failover.candidates,
|
|||
|
|
log: this.log,
|
|||
|
|
thresholds: options.failover.thresholds,
|
|||
|
|
})
|
|||
|
|
sup.seed(this.initialChannel)
|
|||
|
|
sup.start()
|
|||
|
|
this.failover = sup
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
/** 当前通道（监管器在场时以它为准 —— "当前是谁"只有**一个**权威来源）。 */
|
|||
|
|
private get channel(): TunnelChannel {
|
|||
|
|
return (this.failover?.channel as TunnelChannel | undefined) ?? this.initialChannel
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
private channelFor(client: RelayClient, url: string): TunnelChannel {
|
|||
|
|
return {
|
|||
|
|
url,
|
|||
|
|
client,
|
|||
|
|
health: () => healthOf(client),
|
|||
|
|
close: () => client.stop(),
|
|||
|
|
}
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
private buildClient(url: string): RelayClient {
|
|||
|
|
return new RelayClient({
|
|||
|
|
url,
|
|||
|
|
hostId: this.opts.hostId,
|
|||
|
|
secret: this.opts.secret,
|
|||
|
|
ports: [...new Set([...(this.opts.staticPorts ?? []), ...this.forwarded])].sort((a, b) => a - b),
|
|||
|
|
identity: this.opts.identity,
|
|||
|
|
log: this.log,
|
|||
|
|
webSocketCtor: this.opts.webSocketCtor,
|
|||
|
|
})
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
private get client(): RelayClient {
|
|||
|
|
return this.channel.client
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
/** 当前"已声明给 relay"的端口（静态 + 运行期），诊断用。 */
|
|||
|
|
get ports(): number[] {
|
|||
|
|
return [...new Set([...(this.opts.staticPorts ?? []), ...this.forwarded])].sort((a, b) => a - b)
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
/** **幂等**：已启动就只等它到 `up`（断链后 agent 的自愈走的正是这条路径）。 */
|
|||
|
|
async ensureMaster(): Promise<void> {
|
|||
|
|
this.client.start()
|
|||
|
|
await this.waitUp(this.opts.upTimeoutMs ?? 12_000)
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
async isMasterAlive(): Promise<boolean> {
|
|||
|
|
return this.client.status().state === 'up'
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
/** 加一个转发（实例起来时）。失败**不抛**：实例在本机照样可用，只是跨机代理这一跳不可用。 */
|
|||
|
|
async forward(port: number): Promise<boolean> {
|
|||
|
|
if (this.forwarded.has(port)) return true
|
|||
|
|
const ok = await this.client.addPort(port)
|
|||
|
|
if (ok) this.forwarded.add(port)
|
|||
|
|
return ok
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
/** 撤销一条转发（实例停止时）。 */
|
|||
|
|
async cancel(port: number): Promise<void> {
|
|||
|
|
if (!this.forwarded.has(port)) return
|
|||
|
|
if (await this.client.removePort(port)) this.forwarded.delete(port)
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
async close(): Promise<void> {
|
|||
|
|
this.failover?.stop()
|
|||
|
|
this.client.stop()
|
|||
|
|
this.forwarded.clear()
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
/**
|
|||
|
|
* 换址时用的**非抛版**等待（序⑦）：`open()` 要的是"能不能起来"这个布尔，
|
|||
|
|
* 而不是异常 —— 起不来就返回 `false`，由监管器决定"保持原通道"（D4）。
|
|||
|
|
*
|
|||
|
|
* 🔴 **序⑨ · RC-1（C2 装配点）**：实现已抽到 {@link waitUpOnStatus}（三个装配点共用一份，D7）；
|
|||
|
|
* 相对改造前的唯一差别 = **终态失败（死候选）立即 `false`**，⛔ 不再白等满 `upTimeoutMs`。
|
|||
|
|
*/
|
|||
|
|
private async waitUpOn(client: RelayClient, timeoutMs: number): Promise<boolean> {
|
|||
|
|
return waitUpOnStatus(client, timeoutMs, {
|
|||
|
|
onDead: (st) =>
|
|||
|
|
this.log(
|
|||
|
|
`[relay-failover] ⛔ 新通道终态失败（state=${st.state} attempts=${st.attempts} ` +
|
|||
|
|
`burst=${st.inGracefulBurstWindow} lastError="${st.lastError ?? ''}"）⇒ 提前放弃，不等满 ${timeoutMs}ms`,
|
|||
|
|
),
|
|||
|
|
})
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
/** 轮询等 `up`（`RelayClient` 没有 up 事件；重连由它自己退避驱动，这里只负责等）。 */
|
|||
|
|
private async waitUp(timeoutMs: number): Promise<void> {
|
|||
|
|
const deadline = Date.now() + timeoutMs
|
|||
|
|
for (;;) {
|
|||
|
|
if (this.client.status().state === 'up') return
|
|||
|
|
if (Date.now() >= deadline) {
|
|||
|
|
const st = this.client.status()
|
|||
|
|
throw new Error(`relay 未在 ${timeoutMs}ms 内就绪（state=${st.state} lastError=${st.lastError ?? '-'}）`)
|
|||
|
|
}
|
|||
|
|
await new Promise((r) => setTimeout(r, 100))
|
|||
|
|
}
|
|||
|
|
}
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
/**
|
|||
|
|
* 这个会合地址是不是"走 relay"（覆盖网络 R4）。
|
|||
|
|
*
|
|||
|
|
* 判据只看 scheme：`ws://` / `wss://` = relay；`ssh://` / 裸 `user@host:port` = 旧隧道。
|
|||
|
|
* 于是一个环境变量就能在两种传输之间来回切（**删掉 relay 那行即回滚**，不需要回滚代码）。
|
|||
|
|
*/
|
|||
|
|
export function isRelayUrl(raw: string): boolean {
|
|||
|
|
return /^wss?:\/\//i.test(raw.trim())
|
|||
|
|
}
|