fix: await SSH tunnel readiness

backend/src/workspaces/diagnostics.ts

@@ -160,8 +160,9 @@ export interface ConcreteDiagnosticAdapterDependencies {
   directProtocol?: DirectProtocolFactory;
   sshProcess?: SshProcessFactory;
   databaseClient?: DatabaseDiagnosticClientFactory;
-  sshSpawn?: (args: readonly string[]) => { kill(signal?: NodeJS.Signals): boolean };
+  sshSpawn?: (args: readonly string[]) => { kill(signal?: NodeJS.Signals): boolean; once?(event: "error" | "exit", listener: (...args: any[]) => void): unknown };
   reserveLoopbackPort?: () => Promise<number>;
+  waitForSshReady?: (tunnel: LoopbackTunnel, signal: AbortSignal) => Promise<void>;
 }
 
 /**
@@ -231,6 +232,18 @@ async function reserveLoopbackPort(): Promise<number> {
   }
 }
 
+async function waitForLoopbackTunnel(tunnel: LoopbackTunnel, signal: AbortSignal): Promise<void> {
+  while (!signal.aborted) {
+    try {
+      await connectTcp(tunnel.host, tunnel.port, signal);
+      return;
+    } catch {
+      await new Promise<void>((resolve) => setTimeout(resolve, 25));
+    }
+  }
+  throw new Error("SSH tunnel readiness failed");
+}
+
 /**
  * Concrete production adapters deliberately retain only probe metadata. Protocol failures and
  * response bodies are discarded at this boundary; callers receive fixed diagnostics instead.
@@ -240,6 +253,7 @@ export function createConcreteDiagnosticAdapters(
 ): DiagnosticAdapters {
   const spawnSsh = dependencies.sshSpawn ?? ((args: readonly string[]) => spawn("ssh", [...args], { stdio: "ignore" }));
   const reserveSshPort = dependencies.reserveLoopbackPort ?? reserveLoopbackPort;
+  const waitForSshReady = dependencies.waitForSshReady ?? waitForLoopbackTunnel;
   const sshProcess = dependencies.sshProcess ?? {
     async start(request: SshTunnelRequest, args: readonly string[]) {
       const port = await reserveSshPort();
@@ -248,9 +262,29 @@ export function createConcreteDiagnosticAdapters(
       const child = spawnSsh(resolvedArgs);
       const abort = () => { child.kill("SIGTERM"); };
       request.signal.addEventListener("abort", abort, { once: true });
+      const tunnel = { host: "127.0.0.1" as const, port };
+      try {
+        await withTimeout(request.timeoutMs, async (signal) => {
+          await Promise.race([
+            waitForSshReady(tunnel, signal),
+            new Promise<never>((_resolve, reject) => {
+              child.once?.("error", () => reject(new Error("SSH tunnel readiness failed")));
+              child.once?.("exit", () => reject(new Error("SSH tunnel readiness failed")));
+            }),
+          ]);
+        });
+      } catch (error) {
+        request.signal.removeEventListener("abort", abort);
+        child.kill("SIGTERM");
+        throw error;
+      }
       return {
-        tunnel: { host: "127.0.0.1" as const, port },
-        async close() { request.signal.removeEventListener("abort", abort); child.kill("SIGTERM"); },
+        tunnel,
+        async close() {
+          request.signal.removeEventListener("abort", abort);
+          child.kill("SIGTERM");
+          if (child.once) await withTimeout(request.timeoutMs, () => new Promise<void>((resolve) => child.once?.("exit", resolve))).catch(() => undefined);
+        },
       };
     },
   };
@@ -332,7 +366,7 @@ export function createConcreteDiagnosticAdapters(
         throw new Error("SSH probe failed");
       }
       const args = [
-        "-N", "-o", "BatchMode=yes", "-o", "StrictHostKeyChecking=yes",
+        "-N", "-o", "BatchMode=yes", "-o", "ExitOnForwardFailure=yes", "-o", "StrictHostKeyChecking=yes",
         "-o", `UserKnownHostsFile=${request.knownHostsFile}`, "-i", request.privateKeyFile,
         "-p", String(request.sshPort), "-L", `127.0.0.1:0:${request.targetHost}:${request.targetPort}`,
         `${request.sshUser}@${request.sshHost}`,

backend/test/workspaces-diagnostics.test.ts

@@ -333,7 +333,7 @@ test("provides a default bounded SSH factory through injected spawn and loopback
   const kill = vi.fn(() => true);
   const sshSpawn = vi.fn(() => ({ kill }));
   try {
-    const adapter = createConcreteDiagnosticAdapters({ sshSpawn, reserveLoopbackPort: async () => 45432 } as any);
+    const adapter = createConcreteDiagnosticAdapters({ sshSpawn, reserveLoopbackPort: async () => 45432, waitForSshReady: async () => undefined } as any);
     await adapter.withSshTunnel({ sshHost: "bastion.example.test", sshPort: 22, sshUser: "tunnel", privateKeyFile, knownHostsFile: "/run/secrets/known-hosts", targetHost: "dwh.internal", targetPort: 5432, localHost: "127.0.0.1", localPort: 0, timeoutMs: 5000, signal: new AbortController().signal }, async () => undefined);
     expect(sshSpawn).toHaveBeenCalledWith(expect.arrayContaining([
       "StrictHostKeyChecking=yes", "UserKnownHostsFile=/run/secrets/known-hosts", "-L", "127.0.0.1:45432:dwh.internal:5432",
@@ -344,6 +344,28 @@ test("provides a default bounded SSH factory through injected spawn and loopback
   }
 });
 
+test("waits for SSH readiness before probing and includes ExitOnForwardFailure", async () => {
+  const directory = await mkdtemp(join(tmpdir(), "thothii-diagnostic-"));
+  const privateKeyFile = join(directory, "ssh-key");
+  await writeFile(privateKeyFile, "test-key\n", { mode: 0o600 });
+  let releaseReady: (() => void) | undefined;
+  const ready = new Promise<void>((resolve) => { releaseReady = resolve; });
+  const probe = vi.fn(async () => undefined);
+  const sshSpawn = vi.fn(() => ({ kill: vi.fn(() => true) }));
+  try {
+    const adapter = createConcreteDiagnosticAdapters({ sshSpawn, reserveLoopbackPort: async () => 45432, waitForSshReady: async () => await ready } as any);
+    const running = adapter.withSshTunnel({ sshHost: "bastion.example.test", sshPort: 22, sshUser: "tunnel", privateKeyFile, knownHostsFile: "/run/secrets/known-hosts", targetHost: "dwh.internal", targetPort: 5432, localHost: "127.0.0.1", localPort: 0, timeoutMs: 5000, signal: new AbortController().signal }, probe);
+    await Promise.resolve();
+    expect(probe).not.toHaveBeenCalled();
+    releaseReady?.();
+    await running;
+    expect(sshSpawn).toHaveBeenCalledWith(expect.arrayContaining(["ExitOnForwardFailure=yes"]));
+    expect(probe).toHaveBeenCalledOnce();
+  } finally {
+    await rm(directory, { recursive: true, force: true });
+  }
+});
+
 test("requires a matching embedding model vector and removes its unique write probe", async () => {
   const adapters = successfulAdapters();