kuber/server/maintenance.ts

import { randomUUID } from "node:crypto";

export const MAINTENANCE_NAMESPACE = "routing";
export const MAINTENANCE_STATE_NAMESPACE = "kuber-system";
export const MAINTENANCE_ROUTE_NAME = "maintenance-override";
export const MAINTENANCE_STATE_NAME = "maintenance-override";

export type MaintenanceStatus = {
  host: string;
  enabled: boolean;
  hosts: string[];
};

export interface MaintenancePersistence {
  readState(): Promise<{ hosts: string[] } | undefined>;
  writeState(value: { hosts: string[] }): Promise<void>;
  routeExists(): Promise<boolean>;
  apply(resource: Record<string, unknown>): Promise<void>;
  deleteRoute(): Promise<void>;
}

export interface MaintenanceLeaseProvider {
  acquire(
    name: string,
    holder: string,
    ttlMs?: number,
  ): Promise<{ release(): Promise<void> } | undefined>;
}

/** Accept DNS hostnames only: no URL syntax, port, address literals, or wildcards. */
export function normalizeMaintenanceHost(value: unknown): string {
  if (typeof value !== "string") throw new Error("host is required");
  const host = value.trim().toLowerCase().replace(/\.$/, "");
  console.log(host)
  if (
    host.length === 0 ||
    host.length > 253 ||
    !host.includes(".") ||
    !/^(?:\*\.)?(?:[a-z0-9](?:[a-z0-9-]{0,61}[a-z0-9])?\.)+[a-z]{2,63}$/.test(host)
  )
    throw new Error("host must be a DNS hostname");
  return host;
}

export function maintenanceMiddleware(): Record<string, unknown> {
  return {
    apiVersion: "traefik.io/v1alpha1",
    kind: "Middleware",
    metadata: {
      name: MAINTENANCE_ROUTE_NAME,
      namespace: MAINTENANCE_NAMESPACE,
    },
    spec: { replacePathRegex: { regex: "^/.*", replacement: "/__error/1001" } },
  };
}

export function maintenanceRoute(hosts: string[]): Record<string, unknown> {
  return {
    apiVersion: "traefik.io/v1alpha1",
    kind: "IngressRoute",
    metadata: {
      name: MAINTENANCE_ROUTE_NAME,
      namespace: MAINTENANCE_NAMESPACE,
    },
    spec: {
      routes: [
        {
          kind: "Rule",
          match: hosts.map((host) => `Host(\`${host}\`)`).join(" || "),
          // This must override every workspace route for an affected host.
          priority: 1_000_000,
          middlewares: [
            { name: MAINTENANCE_ROUTE_NAME, namespace: MAINTENANCE_NAMESPACE },
          ],
          services: [
            {
              name: "error-page",
              namespace: MAINTENANCE_NAMESPACE,
              port: 3000,
              scheme: "http",
            },
          ],
        },
      ],
    },
  };
}

export class MaintenanceBusyError extends Error {
  readonly code = "MAINTENANCE_BUSY";
}

export class MaintenanceService {
  constructor(
    private readonly persistence: MaintenancePersistence,
    private readonly leases: MaintenanceLeaseProvider,
  ) {}

  async status(host: string): Promise<MaintenanceStatus> {
    const normalized = normalizeMaintenanceHost(host);
    const hosts = this.hosts(await this.persistence.readState());
    await this.reconcile(hosts);
    return {
      host: normalized,
      enabled: hosts.includes(normalized) && (await this.persistence.routeExists()),
      hosts,
    };
  }

  async toggle(host: string): Promise<MaintenanceStatus> {
    return this.update(host);
  }

  /** Setting the desired state makes a retried mutation safe after a timeout. */
  async set(host: string, enabled: boolean): Promise<MaintenanceStatus> {
    return this.update(host, enabled);
  }

  private async update(
    host: string,
    enabled?: boolean,
  ): Promise<MaintenanceStatus> {
    const normalized = normalizeMaintenanceHost(host);
    const lease = await this.leases.acquire(
      "maintenance-override",
      randomUUID(),
    );
    if (!lease) throw new MaintenanceBusyError("Maintenance state is busy; retry shortly");
    try {
      const current = this.hosts(await this.persistence.readState());
      const targetEnabled = enabled ?? !current.includes(normalized);
      const hosts = targetEnabled
        ? [...new Set([...current, normalized])].sort()
        : current.filter((item) => item !== normalized);
      await this.reconcile(hosts);
      // Persist only after Traefik has accepted the desired override. If this
      // write fails, the next reconciliation restores the prior desired state.
      await this.persistence.writeState({ hosts });
      return { host: normalized, enabled: targetEnabled, hosts };
    } finally {
      await lease.release();
    }
  }

  async reconcile(hosts?: string[]): Promise<void> {
    const desired = hosts ?? this.hosts(await this.persistence.readState());
    await this.persistence.apply(maintenanceMiddleware());
    if (desired.length) await this.persistence.apply(maintenanceRoute(desired));
    else await this.persistence.deleteRoute();
  }

  private hosts(value: { hosts: string[] } | undefined): string[] {
    if (!value || !Array.isArray(value.hosts)) return [];
    return [
      ...new Set(value.hosts.map((host) => normalizeMaintenanceHost(host))),
    ].sort();
  }
}