Files
dsh_ai1net_server/src/config.ts
T

371 lines
16 KiB
TypeScript
Raw Normal View History

/**
* Deployment-varying configuration. Every tunable is a validated field here
* (or read from env), never a hardcoded constant inside the app.
* @module dshs/config
*/
import { randomBytes } from 'node:crypto'
import { mkdirSync, readFileSync, writeFileSync } from 'node:fs'
import { homedir, hostname } from 'node:os'
import { join } from 'node:path'
/** Isolation tier. `soft` = per-user home/workspace + sandbox (same OS user);
* `account` = per-user OS account via a setuid wrapper (Linux, needs root). */
export type IsolationMode = 'soft' | 'account'
/** Deployment mode. `local` = single-host child_process (setuid/iptables);
* `k8s` = multi-replica control plane spawning per-user DSH Pods via the K8s API. */
export type DeployMode = 'local' | 'k8s' | 'cluster'
/** Resolved, immutable runtime configuration. */
export interface ServerConfig {
/** Bind host for the orchestrator HTTP server. */
host: string
/** Bind port; `0` requests an ephemeral port. */
port: number
/** SQLite database path. */
dbPath: string
/** Postgres connection string; when set, the DB backend is Postgres (k8s/HA). */
dbUrl?: string
/** Root under which per-user homes (`users/<id>/home`) and workspaces live. */
dataRoot: string
/** Shared read-only skill directory for all users (injected as
* `DSH_BUNDLED_SKILL_DIR` into every spawned DSH); empty = feature off. */
bundledSkillDir: string
/** Argv used to launch a child DSH; first element is the executable. */
dshCommand: string[]
/** Pino log level. */
logLevel: string
/** Set the `Secure` flag on session cookies (enable behind HTTPS). */
secureCookies: boolean
/** Session lifetime in seconds. */
sessionTtlSeconds: number
/** Max upload request body in bytes (base64 JSON; ~0.75× the file size). */
maxUploadBytes: number
/** Delay before auto-restarting a crashed child DSH, in milliseconds. */
restartBackoffMs: number
/** Upper bound for the exponential crash-restart backoff (档案 20). */
restartBackoffMaxMs: number
/** Auto-restarts allowed inside `crashWindowMs` before the circuit opens. */
crashMaxRestarts: number
/** Rolling window used to count auto-restarts (档案 20). */
crashWindowMs: number
/** Uptime after which a main counts as recovered and the backoff resets. */
crashStableMs: number
/** 档案 78:熔断首次冷却时长(冷却期内拒绝隐式启动)。 */
crashBreakerCooldownMs: number
/** 档案 78:熔断冷却上限(多次熔断后指数加长到此为止)。 */
crashBreakerMaxCooldownMs: number
/** Isolation tier (see {@link IsolationMode}); local mode only. */
isolationMode: IsolationMode
/** Argv prefix that drops privileges; `{UID}`/`{GID}` are substituted. Local mode only. */
spawnAsUserCommand: string[]
/** Base uid for the deterministic per-user uid. */
baseUid: number
/** Parent domain for per-user subdomains (`<username>.<baseDomain>`); empty = disabled. */
baseDomain: string
/** Cookie `Domain` value (e.g. `.example.com`) so the session reaches subdomains; empty = host-only. */
cookieDomain: string
/** Whether to pass `--patch` to child DSHs (needs a dsh CLI that supports it). */
enablePatch: boolean
/** Enable the loopback OUTPUT owner-match port guard (Linux + root). Local mode only. */
portGuard: boolean
/** Cap on resident main instances per host (0 = no cap). Local mode idle reap. */
maxIdleInstances: number
/** A main instance with no proxied/entered activity for this long is stopped
* (0 = disabled). Local mode idle reap. */
instanceIdleTtlSeconds: number
/** Period between idle-reap scans (seconds). Local mode idle reap. */
idleReapIntervalSeconds: number
/** Secret used to encrypt per-user secrets at rest (from env or dataRoot/secret.key). */
encryptionSecret: string
/** Deployment mode (see {@link DeployMode}). */
deployMode: DeployMode
/** K8s namespace for per-user DSH resources (k8s mode only). */
k8sNamespace: string
/** Image for per-user DSH Pods (k8s mode only). */
dshImage: string
/** Image for the per-user file sidecar (k8s mode only; shares the control-plane image). */
controlPlaneImage: string
/** imagePullSecret every generated per-user Pod/Job uses (k8s mode only). */
imagePullSecret: string
/** 443 egress whitelist CIDRs for per-user DSH Pods (Phase 4 egress 收敛);
* empty = keep the current `0.0.0.0/0` (except private ranges). */
egressCidrs: string[]
/** ServiceAccount the orchestrator runs as (k8s mode only). */
k8sServiceAccount: string
/** This replica's identity for leader election (POD_NAME, else hostname). */
podName: string
// ── cluster 模式(T08 S3/S4;设计 §1.1)────────────────────────────────
/** 本机在 `dsh_hosts.id` 里的标识(`deployMode=cluster` 时必填语义)。 */
clusterHostId: string
/** 本机 worker agent 的**基址**(Manager 侧用它投递实例操作),如 `http://127.0.0.1:9000`。 */
clusterAgentUrl: string
/** 与 agent 约定的共享密钥(仅内网 + nft 白名单)。 */
clusterAgentToken: string
/** agent 返回给 Manager 做代理的实例地址(同机 1a = `127.0.0.1`)。 */
clusterInstanceHost: string
/**
* **worker 上**的 dataRoot(T08 S5)。
* 空 = 与本地 `dataRoot` 相同(1a 形态)。多机部署必须显式配置 —— 而且是**基线约定**:
* 所有 worker 的 dataRoot 必须是同一个绝对路径(同镜像即可满足,设计 §14.3)。
*/
clusterWorkerDataRoot: string
/**
* **worker 侧**会合地址(覆盖网络 S1):worker 主动拨入的 SSH 反向隧道落点。
* 形如 `ssh://[email protected]:32022`(也接受不带 scheme 的 `root@host:port`)。空 = 隧道关闭。
* ⚠️ 与旧变量 `DSHS_TUNNEL_TARGET` **双路径并存**(新变量优先、旧变量兜底)⇒
* 删掉新 env 即回到旧路径,**零代码回滚**。
*/
clusterRendezvousUrl: string
}
/** Untyped overrides collected from argv / env. */
export interface ConfigOverrides {
host?: string
port?: string | number
dbPath?: string
dbUrl?: string
dataRoot?: string
bundledSkillDir?: string
dshCommand?: string[]
logLevel?: string
secureCookies?: boolean
sessionTtlSeconds?: number | string
maxUploadBytes?: number | string
restartBackoffMs?: number | string
restartBackoffMaxMs?: number | string
crashMaxRestarts?: number | string
crashWindowMs?: number | string
crashStableMs?: number | string
crashBreakerCooldownMs?: number | string
crashBreakerMaxCooldownMs?: number | string
isolationMode?: IsolationMode | string
spawnAsUserCommand?: string[]
baseUid?: number | string
baseDomain?: string
cookieDomain?: string
enablePatch?: boolean
portGuard?: boolean
maxIdleInstances?: number | string
instanceIdleTtlSeconds?: number | string
idleReapIntervalSeconds?: number | string
encryptionSecret?: string
deployMode?: DeployMode | string
k8sNamespace?: string
dshImage?: string
controlPlaneImage?: string
imagePullSecret?: string
egressCidrs?: string[]
k8sServiceAccount?: string
podName?: string
clusterHostId?: string
clusterAgentUrl?: string
clusterAgentToken?: string
clusterInstanceHost?: string
clusterWorkerDataRoot?: string
clusterRendezvousUrl?: string
}
const DEFAULT_HOST = '127.0.0.1'
const DEFAULT_PORT = 3080
const DEFAULT_DSH_COMMAND = ['dsh']
const DEFAULT_LOG_LEVEL = 'info'
const DEFAULT_SESSION_TTL_SECONDS = 60 * 60 * 24 * 7
const DEFAULT_MAX_UPLOAD_BYTES = 25 * 1024 * 1024
const DEFAULT_RESTART_BACKOFF_MS = 1000
/** 崩溃自愈退避上限(档案 20)。 */
const DEFAULT_RESTART_BACKOFF_MAX_MS = 30000
/** 熔断窗口内允许的自动重启次数(档案 20)。 */
const DEFAULT_CRASH_MAX_RESTARTS = 5
/** 熔断窗口长度(档案 20)。 */
const DEFAULT_CRASH_WINDOW_MS = 600000
/** 连续运行多久视为已恢复、重置退避步数(档案 20)。 */
const DEFAULT_CRASH_STABLE_MS = 60000
// 档案 78:熔断冷却 —— 首次 10 分钟,指数加长,封顶 6 小时
const DEFAULT_CRASH_BREAKER_COOLDOWN_MS = 600000
const DEFAULT_CRASH_BREAKER_MAX_COOLDOWN_MS = 21600000
const DEFAULT_ISOLATION_MODE: IsolationMode = 'soft'
const DEFAULT_SPAWN_AS_USER_COMMAND = [
'setpriv',
'--reuid',
'{UID}',
'--regid',
'{GID}',
'--inh-caps=-all',
'--clear-groups',
'--',
]
const DEFAULT_BASE_UID = 100000
const DEFAULT_BASE_DOMAIN = ''
const DEFAULT_COOKIE_DOMAIN = ''
const DEFAULT_ENABLE_PATCH = false
const DEFAULT_MAX_IDLE_INSTANCES = 4
const DEFAULT_INSTANCE_IDLE_TTL_SECONDS = 60 * 60 * 24 * 7
const DEFAULT_IDLE_REAP_INTERVAL_SECONDS = 60
const DEFAULT_DEPLOY_MODE: DeployMode = 'local'
const DEFAULT_K8S_NAMESPACE = 'dsh'
const DEFAULT_K8S_SERVICE_ACCOUNT = 'dsh-orchestrator'
const DEFAULT_IMAGE_PULL_SECRET = 'dsh-acr-pull'
/** Load the encryption secret from env, or persist a generated one at
* `<dataRoot>/secret.key` (0600) so it survives restarts without setup. */
function resolveEncryptionSecret(dataRoot: string): string {
const fromEnv = process.env.DSHS_SECRET
if (fromEnv !== undefined && fromEnv !== '') return fromEnv
const path = join(dataRoot, 'secret.key')
try {
const existing = readFileSync(path, 'utf8').trim()
if (existing !== '') return existing
} catch {
// fall through to generate
}
const secret = randomBytes(32).toString('hex')
mkdirSync(dataRoot, { recursive: true })
writeFileSync(path, secret, { mode: 0o600 })
return secret
}
function toBool(value: string | undefined, fallback: boolean): boolean {
if (value === undefined) return fallback
return value === 'true' || value === '1'
}
/** Split a comma-separated CIDR list into a trimmed, de-duplicated array. */
function parseCidrs(value: string | undefined): string[] {
if (value === undefined || value === '') return []
return [...new Set(value.split(',').map((c) => c.trim()).filter((c) => c !== ''))]
}
/** Parse an isolation-mode value, rejecting anything outside `soft`/`account`
* so a typo in the env var fails loudly at startup instead of silently
* falling back to `soft` isolation. */
function toIsolationMode(value: string | undefined): IsolationMode | undefined {
if (value === undefined) return undefined
const normalized = value.trim().toLowerCase()
if (normalized === 'soft' || normalized === 'account') return normalized
throw new Error(`invalid isolation mode "${value}" (expected "soft" or "account")`)
}
/** Parse a deploy-mode value, rejecting anything outside `local`/`k8s`. */
function toDeployMode(value: string | undefined): DeployMode | undefined {
if (value === undefined) return undefined
const normalized = value.trim().toLowerCase()
if (normalized === 'local' || normalized === 'k8s' || normalized === 'cluster') return normalized
throw new Error(`invalid deploy mode "${value}" (expected "local", "k8s" or "cluster")`)
}
/**
* Fold argv/env overrides over defaults. `dataRoot` defaults to
* `~/.dshs` (always writable for dev); production sets
* `DSHS_DATA_ROOT=/var/lib/dshs`.
*/
export function resolveConfig(overrides: ConfigOverrides = {}): ServerConfig {
const dataRoot =
overrides.dataRoot ?? process.env.DSHS_DATA_ROOT ?? join(homedir(), '.dshs')
const port = overrides.port ?? process.env.DSHS_PORT ?? DEFAULT_PORT
const dshBin = process.env.DSHS_DSH_BIN
const isolationMode =
toIsolationMode(overrides.isolationMode) ??
toIsolationMode(process.env.DSHS_ISOLATION_MODE) ??
DEFAULT_ISOLATION_MODE
const deployMode =
toDeployMode(overrides.deployMode) ??
toDeployMode(process.env.DSHS_DEPLOY_MODE) ??
DEFAULT_DEPLOY_MODE
return {
host: overrides.host ?? DEFAULT_HOST,
port: typeof port === 'number' ? port : Number(port),
dbPath: overrides.dbPath ?? join(dataRoot, 'dshs.db'),
dbUrl: overrides.dbUrl ?? process.env.DSHS_DB_URL,
dataRoot,
bundledSkillDir:
overrides.bundledSkillDir ??
process.env.DSHS_BUNDLED_SKILL_DIR ??
join(dataRoot, 'bundled-skills'),
dshCommand: overrides.dshCommand ?? (dshBin !== undefined ? [dshBin] : DEFAULT_DSH_COMMAND),
logLevel: overrides.logLevel ?? DEFAULT_LOG_LEVEL,
secureCookies:
overrides.secureCookies ?? toBool(process.env.DSHS_SECURE_COOKIES, false),
sessionTtlSeconds: Number(
overrides.sessionTtlSeconds ?? process.env.DSHS_SESSION_TTL ?? DEFAULT_SESSION_TTL_SECONDS,
),
maxUploadBytes: Number(
overrides.maxUploadBytes ?? process.env.DSHS_MAX_UPLOAD ?? DEFAULT_MAX_UPLOAD_BYTES,
),
restartBackoffMs: Number(
overrides.restartBackoffMs ?? process.env.DSHS_RESTART_BACKOFF ?? DEFAULT_RESTART_BACKOFF_MS,
),
restartBackoffMaxMs: Number(
overrides.restartBackoffMaxMs ??
process.env.DSHS_RESTART_BACKOFF_MAX ??
DEFAULT_RESTART_BACKOFF_MAX_MS,
),
crashMaxRestarts: Number(
overrides.crashMaxRestarts ??
process.env.DSHS_CRASH_MAX_RESTARTS ??
DEFAULT_CRASH_MAX_RESTARTS,
),
crashWindowMs: Number(
overrides.crashWindowMs ?? process.env.DSHS_CRASH_WINDOW ?? DEFAULT_CRASH_WINDOW_MS,
),
crashStableMs: Number(
overrides.crashStableMs ?? process.env.DSHS_CRASH_STABLE ?? DEFAULT_CRASH_STABLE_MS,
),
crashBreakerCooldownMs: Number(
overrides.crashBreakerCooldownMs ??
process.env.DSHS_CRASH_BREAKER_COOLDOWN ??
DEFAULT_CRASH_BREAKER_COOLDOWN_MS,
),
crashBreakerMaxCooldownMs: Number(
overrides.crashBreakerMaxCooldownMs ??
process.env.DSHS_CRASH_BREAKER_MAX_COOLDOWN ??
DEFAULT_CRASH_BREAKER_MAX_COOLDOWN_MS,
),
isolationMode,
spawnAsUserCommand: overrides.spawnAsUserCommand ?? DEFAULT_SPAWN_AS_USER_COMMAND,
baseUid: Number(overrides.baseUid ?? process.env.DSHS_BASE_UID ?? DEFAULT_BASE_UID),
baseDomain: overrides.baseDomain ?? process.env.DSHS_BASE_DOMAIN ?? DEFAULT_BASE_DOMAIN,
cookieDomain: overrides.cookieDomain ?? process.env.DSHS_COOKIE_DOMAIN ?? DEFAULT_COOKIE_DOMAIN,
enablePatch: overrides.enablePatch ?? toBool(process.env.DSHS_ENABLE_PATCH, DEFAULT_ENABLE_PATCH),
portGuard: overrides.portGuard ?? toBool(process.env.DSHS_PORT_GUARD, false),
maxIdleInstances: Number(
overrides.maxIdleInstances ?? process.env.DSHS_MAX_IDLE_INSTANCES ?? DEFAULT_MAX_IDLE_INSTANCES,
),
instanceIdleTtlSeconds: Number(
overrides.instanceIdleTtlSeconds ??
process.env.DSHS_INSTANCE_IDLE_TTL ??
DEFAULT_INSTANCE_IDLE_TTL_SECONDS,
),
idleReapIntervalSeconds: Number(
overrides.idleReapIntervalSeconds ??
process.env.DSHS_IDLE_REAP_INTERVAL ??
DEFAULT_IDLE_REAP_INTERVAL_SECONDS,
),
encryptionSecret:
overrides.encryptionSecret ?? resolveEncryptionSecret(dataRoot),
deployMode,
k8sNamespace: overrides.k8sNamespace ?? process.env.DSHS_NAMESPACE ?? DEFAULT_K8S_NAMESPACE,
dshImage: overrides.dshImage ?? process.env.DSHS_DSH_IMAGE ?? '',
controlPlaneImage:
overrides.controlPlaneImage ?? process.env.DSHS_CONTROL_PLANE_IMAGE ?? '',
imagePullSecret:
overrides.imagePullSecret ?? process.env.DSHS_IMAGE_PULL_SECRET ?? DEFAULT_IMAGE_PULL_SECRET,
egressCidrs: overrides.egressCidrs ?? parseCidrs(process.env.DSHS_EGRESS_CIDRS),
k8sServiceAccount:
overrides.k8sServiceAccount ?? process.env.DSHS_K8S_SERVICE_ACCOUNT ?? DEFAULT_K8S_SERVICE_ACCOUNT,
podName: overrides.podName ?? process.env.POD_NAME ?? hostname(),
clusterHostId: overrides.clusterHostId ?? process.env.DSHS_CLUSTER_HOST_ID ?? hostname(),
clusterAgentUrl: overrides.clusterAgentUrl ?? process.env.DSHS_CLUSTER_AGENT_URL ?? '',
clusterAgentToken: overrides.clusterAgentToken ?? process.env.DSHS_CLUSTER_AGENT_TOKEN ?? '',
clusterInstanceHost: overrides.clusterInstanceHost ?? process.env.DSHS_CLUSTER_INSTANCE_HOST ?? '127.0.0.1',
clusterWorkerDataRoot: overrides.clusterWorkerDataRoot ?? process.env.DSHS_CLUSTER_WORKER_DATA_ROOT ?? '',
// 覆盖网络 S1:会合地址出 env。新变量 `DSHS_RENDEZVOUS_URL` 优先,旧变量 `DSHS_TUNNEL_TARGET`
// 兜底(两台机器可分先后改;删掉新 env 即回滚到旧路径,**不需要回滚代码**)。
clusterRendezvousUrl:
overrides.clusterRendezvousUrl ??
process.env.DSHS_RENDEZVOUS_URL ??
process.env.DSHS_TUNNEL_TARGET ??
'',
}
}