feat: make api calls more resilient and stuff

app/layout/app.tsx

@@ -45,7 +45,7 @@ export async function loader({ request, context }: Route.LoaderArgs) {
         : { name: principal.displayName, subject: "api_key" };
 
     // MARK: The session should stay valid if Headscale isn't healthy
-    const isHealthy = await api.isHealthy();
+    const isHealthy = await context.headscale.health();
     if (isHealthy) {
       try {
         await api.apiKeys.list();

app/routes/auth/login/action.ts

@@ -33,6 +33,8 @@ export async function loginAction({ request, context }: Route.LoaderArgs) {
     };
   }
 
+  // Build a client with the candidate API key the user just submitted, so the
+  // GET /api/v1/apikey call below validates the key against Headscale itself.
   const api = context.headscale.client(apiKey);
   try {
     const apiKeys = await api.apiKeys.list();

app/routes/auth/oidc-callback.ts

@@ -56,6 +56,9 @@ export async function loader({ request, context }: Route.LoaderArgs) {
   });
 
   try {
+    // Looks up the Headscale user that matches this OIDC identity. We use
+    // the configured admin API key here — not a per-request one — because
+    // there is no per-request key yet (the session is being created).
     const hsApi = context.headscale.client(context.headscaleApiKey!);
     const hsUsers = await hsApi.users.list();
     const hsUser = findHeadscaleUserBySubject(hsUsers, identity.subject, identity.email);

app/routes/dns/dns-actions.ts

@@ -16,9 +16,6 @@ export async function dnsAction({ request, context }: Route.ActionArgs) {
     return data({ success: false }, 403);
   }
 
-  // We only need it for health checks which don't require auth
-  const api = context.headscale.client("fake-api-key");
-
   const formData = await request.formData();
   const action = formData.get("action_id")?.toString();
   if (!action) {
@@ -39,7 +36,7 @@ export async function dnsAction({ request, context }: Route.ActionArgs) {
         },
       ]);
 
-      await context.integration?.onConfigChange(api);
+      await context.integration?.onConfigChange(context.headscale);
       return { message: "Tailnet renamed successfully" };
     }
     case "toggle_magic": {
@@ -55,7 +52,7 @@ export async function dnsAction({ request, context }: Route.ActionArgs) {
         },
       ]);
 
-      await context.integration?.onConfigChange(api);
+      await context.integration?.onConfigChange(context.headscale);
       return { message: "Magic DNS state updated successfully" };
     }
     case "remove_ns": {
@@ -88,7 +85,7 @@ export async function dnsAction({ request, context }: Route.ActionArgs) {
         ]);
       }
 
-      await context.integration?.onConfigChange(api);
+      await context.integration?.onConfigChange(context.headscale);
       return { message: "Nameserver removed successfully" };
     }
     case "add_ns": {
@@ -123,7 +120,7 @@ export async function dnsAction({ request, context }: Route.ActionArgs) {
         ]);
       }
 
-      await context.integration?.onConfigChange(api);
+      await context.integration?.onConfigChange(context.headscale);
       return { message: "Nameserver added successfully" };
     }
     case "remove_domain": {
@@ -141,7 +138,7 @@ export async function dnsAction({ request, context }: Route.ActionArgs) {
         },
       ]);
 
-      await context.integration?.onConfigChange(api);
+      await context.integration?.onConfigChange(context.headscale);
       return { message: "Domain removed successfully" };
     }
     case "add_domain": {
@@ -161,7 +158,7 @@ export async function dnsAction({ request, context }: Route.ActionArgs) {
         },
       ]);
 
-      await context.integration?.onConfigChange(api);
+      await context.integration?.onConfigChange(context.headscale);
       return { message: "Domain added successfully" };
     }
     case "remove_record": {
@@ -183,7 +180,7 @@ export async function dnsAction({ request, context }: Route.ActionArgs) {
         return;
       }
 
-      await context.integration?.onConfigChange(api);
+      await context.integration?.onConfigChange(context.headscale);
       return { message: "DNS record removed successfully" };
     }
     case "add_record": {
@@ -205,7 +202,7 @@ export async function dnsAction({ request, context }: Route.ActionArgs) {
         return;
       }
 
-      await context.integration?.onConfigChange(api);
+      await context.integration?.onConfigChange(context.headscale);
       return { message: "DNS record added successfully" };
     }
     case "override_dns": {
@@ -222,7 +219,7 @@ export async function dnsAction({ request, context }: Route.ActionArgs) {
         },
       ]);
 
-      await context.integration?.onConfigChange(api);
+      await context.integration?.onConfigChange(context.headscale);
       return { message: "DNS override updated successfully" };
     }
     default:

app/routes/settings/restrictions/actions.ts

@@ -28,8 +28,6 @@ export async function restrictionAction({ request, context }: Route.ActionArgs)
     });
   }
 
-  // We only need healthchecks which don't rely on an API key
-  const api = context.headscale.client("fake-api-key");
   switch (action) {
     case "add_domain": {
       const domain = formData.get("domain")?.toString()?.trim();
@@ -48,7 +46,7 @@ export async function restrictionAction({ request, context }: Route.ActionArgs)
         },
       ]);
 
-      context.integration?.onConfigChange(api);
+      context.integration?.onConfigChange(context.headscale);
       return data("Domain added successfully.");
     }
 
@@ -76,7 +74,7 @@ export async function restrictionAction({ request, context }: Route.ActionArgs)
           value: domains,
         },
       ]);
-      context.integration?.onConfigChange(api);
+      context.integration?.onConfigChange(context.headscale);
       return data("Domain removed successfully.");
     }
 
@@ -97,7 +95,7 @@ export async function restrictionAction({ request, context }: Route.ActionArgs)
         },
       ]);
 
-      context.integration?.onConfigChange(api);
+      context.integration?.onConfigChange(context.headscale);
       return data("Group added successfully.");
     }
 
@@ -126,7 +124,7 @@ export async function restrictionAction({ request, context }: Route.ActionArgs)
         },
       ]);
 
-      context.integration?.onConfigChange(api);
+      context.integration?.onConfigChange(context.headscale);
       return data("Group removed successfully.");
     }
 
@@ -147,7 +145,7 @@ export async function restrictionAction({ request, context }: Route.ActionArgs)
         },
       ]);
 
-      context.integration?.onConfigChange(api);
+      context.integration?.onConfigChange(context.headscale);
       return data("User added successfully.");
     }
 
@@ -176,7 +174,7 @@ export async function restrictionAction({ request, context }: Route.ActionArgs)
         },
       ]);
 
-      context.integration?.onConfigChange(api);
+      context.integration?.onConfigChange(context.headscale);
       return data("User removed successfully.");
     }
 

app/routes/util/healthz.ts

@@ -1,9 +1,7 @@
 import type { Route } from "./+types/healthz";
 
 export async function loader({ context }: Route.LoaderArgs) {
-  // Use a fake API key for healthcheck
+  const healthy = await context.headscale.health();
-  const api = context.headscale.client("fake-api-key");
-  const healthy = await api.isHealthy();
 
   return new Response(JSON.stringify({ status: healthy ? "OK" : "ERROR" }), {
     status: healthy ? 200 : 500,

app/routes/util/info.ts

@@ -34,9 +34,7 @@ export async function loader({ request, context }: Route.LoaderArgs) {
     );
   }
 
-  // Use a fake API key for healthcheck
+  const healthy = await context.headscale.health();
-  const api = context.headscale.client("fake-api-key");
-  const healthy = await api.isHealthy();
 
   const body = {
     status: healthy ? "healthy" : "unhealthy",

app/server/config/integration/abstract.ts

@@ -1,4 +1,4 @@
-import type { HeadscaleClient } from "~/server/headscale/api";
+import type { Headscale } from "~/server/headscale/api";
 
 export abstract class Integration<T> {
   protected context: NonNullable<T>;
@@ -11,6 +11,6 @@ export abstract class Integration<T> {
   }
 
   abstract isAvailable(): Promise<boolean> | boolean;
-  abstract onConfigChange(client: HeadscaleClient): Promise<void> | void;
+  abstract onConfigChange(headscale: Headscale): Promise<void> | void;
   abstract get name(): string;
 }

app/server/config/integration/docker.ts

@@ -4,7 +4,7 @@ import { setTimeout } from "node:timers/promises";
 import { type } from "arktype";
 import { Client } from "undici";
 
-import type { HeadscaleClient } from "~/server/headscale/api";
+import type { Headscale } from "~/server/headscale/api";
 import log from "~/utils/log";
 
 import { Integration } from "./abstract";
@@ -255,7 +255,7 @@ export default class DockerIntegration extends Integration<typeof configSchema.f
     return this.client !== undefined && this.containerId !== undefined;
   }
 
-  async onConfigChange(client: HeadscaleClient) {
+  async onConfigChange(headscale: Headscale) {
     if (!this.client) {
       return;
     }
@@ -290,7 +290,7 @@ export default class DockerIntegration extends Integration<typeof configSchema.f
     while (attempts <= this.maxAttempts) {
       try {
         log.debug("config", "Checking Headscale status (attempt %d)", attempts);
-        const status = await client.isHealthy();
+        const status = await headscale.health();
         if (status === false) {
           throw new Error("Headscale is not running");
         }

app/server/config/integration/kubernetes.ts

@@ -5,7 +5,7 @@ import { join } from "node:path";
 import { CoreV1Api, KubeConfig } from "@kubernetes/client-node";
 import { type } from "arktype";
 
-import type { HeadscaleClient } from "~/server/headscale/api";
+import type { Headscale } from "~/server/headscale/api";
 import log from "~/utils/log";
 
 import { Integration } from "./abstract";
@@ -154,12 +154,12 @@ export default class KubernetesIntegration extends Integration<typeof configSche
     }
   }
 
-  async onConfigChange(client: HeadscaleClient) {
+  async onConfigChange(headscale: Headscale) {
     if (!this.pid) {
       return;
     }
 
-    await signalAndWaitHealthy(client, {
+    await signalAndWaitHealthy(headscale, {
       pid: this.pid,
       signal: "SIGHUP",
     });

app/server/config/integration/proc-helper.ts

@@ -3,7 +3,7 @@ import { join } from "node:path";
 import { kill } from "node:process";
 import { setTimeout } from "node:timers/promises";
 
-import type { HeadscaleClient } from "~/server/headscale/api";
+import type { Headscale } from "~/server/headscale/api";
 import log from "~/utils/log";
 
 /**
@@ -75,12 +75,12 @@ export interface SignalHeadscaleOptions {
 
 /**
  * Sends a signal to the headscale process and waits for it to become healthy.
- * @param client The HeadscaleClient to check health
+ * @param headscale The Headscale instance to health-check
  * @param options Options for signaling and waiting
  * @returns True if headscale became healthy, false otherwise
  */
 export async function signalAndWaitHealthy(
-  client: HeadscaleClient,
+  headscale: Headscale,
   options: SignalHeadscaleOptions,
 ): Promise<boolean> {
   const { pid, signal = "SIGHUP", maxAttempts = 10, retryDelayMs = 1000 } = options;
@@ -96,7 +96,7 @@ export async function signalAndWaitHealthy(
   await setTimeout(retryDelayMs);
   for (let attempt = 1; attempt <= maxAttempts; attempt++) {
     try {
-      const healthy = await client.isHealthy();
+      const healthy = await headscale.health();
       if (healthy) {
         log.info("config", "Headscale is healthy after restart");
         return true;

app/server/config/integration/proc.ts

@@ -2,7 +2,7 @@ import { platform } from "node:os";
 
 import { type } from "arktype";
 
-import type { HeadscaleClient } from "~/server/headscale/api";
+import type { Headscale } from "~/server/headscale/api";
 import log from "~/utils/log";
 
 import { Integration } from "./abstract";
@@ -51,12 +51,12 @@ export default class ProcIntegration extends Integration<typeof configSchema.ful
     }
   }
 
-  async onConfigChange(client: HeadscaleClient) {
+  async onConfigChange(headscale: Headscale) {
     if (!this.pid) {
       return;
     }
 
-    await signalAndWaitHealthy(client, {
+    await signalAndWaitHealthy(headscale, {
       pid: this.pid,
       signal: "SIGHUP",
     });

app/server/headscale/api/index.ts

@@ -15,7 +15,7 @@ import { makePolicyApi, type PolicyApi } from "./resources/policy";
 import { makePreAuthKeyApi, type PreAuthKeyApi } from "./resources/pre-auth-keys";
 import { makeUserApi, type UserApi } from "./resources/users";
 import { formatServerVersion, parseServerVersion, type ServerVersion } from "./server-version";
-import { createTransport, type Transport } from "./transport";
+import { createTransport } from "./transport";
 
 export interface Headscale {
   readonly version: ServerVersion;
@@ -34,48 +34,84 @@ export interface HeadscaleClient {
   policy: PolicyApi;
   preAuthKeys: PreAuthKeyApi;
   apiKeys: ApiKeyApi;
-  /**
-   * Convenience passthrough to `Headscale.health()`. Headscale's
-   * `/health` endpoint is unauthenticated so callers that already
-   * have a client (loaders/actions) don't need to also reach for
-   * the top-level `Headscale` value.
-   */
-  isHealthy(): Promise<boolean>;
 }
 
 export interface CreateHeadscaleOptions {
   url: string;
   certPath?: string;
+  /**
+   * How often to retry `/version` while Headscale is unreachable.
+   * Defaults to 30 seconds. Exposed for tests.
+   */
+  retryIntervalMs?: number;
 }
 
+const DEFAULT_RETRY_INTERVAL_MS = 30_000;
+
 export async function createHeadscale(opts: CreateHeadscaleOptions): Promise<Headscale> {
   const transport = await createTransport({ url: opts.url, certPath: opts.certPath });
+  const retryIntervalMs = opts.retryIntervalMs ?? DEFAULT_RETRY_INTERVAL_MS;
 
-  const versionInfo = await transport.getPublic<{ version: string }>("/version");
+  let version: ServerVersion = parseServerVersion("unreachable");
-  const version = parseServerVersion(versionInfo.version);
+  let capabilities: Capabilities = capabilitiesFor(version);
-  const capabilities = capabilitiesFor(version);
+  let detected = false;
+  let retryTimer: ReturnType<typeof setTimeout> | undefined;
+  let disposed = false;
 
-  if (version.unknown) {
+  async function detectOnce(): Promise<boolean> {
-    log.warn(
+    try {
-      "api",
+      const { version: raw } = await transport.getPublic<{ version: string }>("/version");
-      "Could not parse Headscale version %s, assuming newest known capabilities",
+      const parsed = parseServerVersion(raw);
-      versionInfo.version,
+      version = parsed;
-    );
+      capabilities = capabilitiesFor(parsed);
-  } else {
+      detected = true;
-    log.info("api", "Connected to Headscale %s", formatServerVersion(version));
+      if (parsed.unknown) {
+        log.warn(
+          "api",
+          "Could not parse Headscale version %s, assuming newest known capabilities",
+          raw,
+        );
+      } else {
+        log.info("api", "Connected to Headscale %s", formatServerVersion(parsed));
+      }
+      return true;
+    } catch (error) {
+      log.debug("api", "Headscale /version probe failed: %s", String(error));
+      return false;
+    }
   }
 
-  return makeHeadscale(transport, version, capabilities);
+  function scheduleRetry() {
-}
+    if (disposed || detected) return;
+    retryTimer = setTimeout(async () => {
+      retryTimer = undefined;
+      if (disposed) return;
+      if (await detectOnce()) return;
+      scheduleRetry();
+    }, retryIntervalMs);
+    // Don't keep the event loop alive on this timer alone — Headplane
+    // should still shut down cleanly while we're waiting to retry.
+    retryTimer.unref?.();
+  }
+
+  if (!(await detectOnce())) {
+    log.warn(
+      "api",
+      "Headscale unreachable at boot; defaulting to newest-known capabilities and retrying every %dms",
+      retryIntervalMs,
+    );
+    scheduleRetry();
+  }
 
-function makeHeadscale(
-  transport: Transport,
-  version: ServerVersion,
-  capabilities: Capabilities,
-): Headscale {
   return {
-    version,
+    // Getters so callers always observe the latest detected values
-    capabilities,
+    // without having to know about the retry loop.
+    get version() {
+      return version;
+    },
+    get capabilities() {
+      return capabilities;
+    },
     health: () => transport.health(),
     client(apiKey) {
       return {
@@ -84,9 +120,15 @@ function makeHeadscale(
         policy: makePolicyApi(transport, capabilities, apiKey),
         preAuthKeys: makePreAuthKeyApi(transport, capabilities, apiKey),
         apiKeys: makeApiKeyApi(transport, capabilities, apiKey),
-        isHealthy: () => transport.health(),
       };
     },
-    dispose: () => transport.dispose(),
+    async dispose() {
+      disposed = true;
+      if (retryTimer) {
+        clearTimeout(retryTimer);
+        retryTimer = undefined;
+      }
+      await transport.dispose();
+    },
   };
 }

runtime/http.ts

@@ -208,6 +208,7 @@ export function startHttpServer(opts: StartOptions): Server {
       }
       process.exit(0);
     });
+
     // Force exit if connections don't drain in time.
     setTimeout(() => process.exit(0), 5_000).unref();
   };

tests/integration/api/versions.test.ts

@@ -2,7 +2,7 @@ import { describe, expect, test } from "vitest";
 
 import { gte } from "~/server/headscale/api/server-version";
 
-import { getBootstrapClient, getRuntimeClient, HS_VERSIONS, Version } from "../setup/env";
+import { getBootstrapClient, HS_VERSIONS, Version } from "../setup/env";
 
 describe.for(HS_VERSIONS)("Headscale %s: Runtime Client", (version) => {
   test("the runtime client is usable", async () => {
@@ -34,8 +34,8 @@ describe.for(HS_VERSIONS)("Headscale %s: Runtime Client", (version) => {
   });
 
   test("the health check endpoint works", async () => {
-    const client = await getRuntimeClient(version);
+    const bootstrapper = await getBootstrapClient(version);
-    const health = await client.isHealthy();
+    const health = await bootstrapper.health();
     expect(health).toBe(true);
   });
 });

tests/integration/platform/docker.test.ts

@@ -81,8 +81,8 @@ describe("DockerIntegration", () => {
       socketPath: "/var/run/docker.sock",
     });
 
-    const mockClient = {
+    const mockHeadscale = {
-      isHealthy: async () => {
+      health: async () => {
         try {
           const res = await dockerClient.request({
             method: "GET",
@@ -96,9 +96,9 @@ describe("DockerIntegration", () => {
       },
     } as any;
 
-    await integration.onConfigChange(mockClient);
+    await integration.onConfigChange(mockHeadscale);
 
-    const healthy = await mockClient.isHealthy();
+    const healthy = await mockHeadscale.health();
     expect(healthy).toBe(true);
   }, 60_000);
 });