fix: wait for linux rpc socket on startup (#5)

Author: ethanndickson

App.Avalonia/App.axaml.cs

@@ -156,7 +156,7 @@ private async Task InitializeServicesAsync()
         credentialLoadCts.CancelAfter(TimeSpan.FromSeconds(15));
 
         var loadCredentialsTask = credentialManager.LoadCredentials(credentialLoadCts.Token);
-        var reconnectTask = ReconnectWithStartupRetryAsync(rpcController, appStopping);
+        var reconnectTask = rpcController.Reconnect(appStopping);
 
         try
         {
@@ -173,10 +173,7 @@ private async Task InitializeServicesAsync()
                 AppBootstrapLogger.Error("Startup reconnect failed unexpectedly", reconnectTask.Exception?.GetBaseException());
         }
 
-        var reconnectSucceeded = reconnectTask is { IsCompletedSuccessfully: true, Result: true };
-
-        if (!reconnectSucceeded)
-            AppBootstrapLogger.Warn("Startup continuing in disconnected state after retry exhaustion");
+        var reconnectSucceeded = reconnectTask.IsCompletedSuccessfully;
 
         try
         {
@@ -188,41 +185,6 @@ private async Task InitializeServicesAsync()
         }
     }
 
-    private async Task<bool> ReconnectWithStartupRetryAsync(IRpcController rpcController, CancellationToken ct)
-    {
-        TimeSpan[] delays =
-        [
-            TimeSpan.Zero,
-            TimeSpan.FromSeconds(1),
-            TimeSpan.FromSeconds(2),
-            TimeSpan.FromSeconds(4),
-            TimeSpan.FromSeconds(8),
-        ];
-
-        Exception? lastError = null;
-
-        for (var attempt = 0; attempt < delays.Length; attempt++)
-        {
-            if (attempt > 0)
-                await Task.Delay(delays[attempt], ct);
-
-            try
-            {
-                await rpcController.Reconnect(ct);
-                AppBootstrapLogger.Info($"RPC reconnect succeeded on attempt {attempt + 1}/{delays.Length}");
-                return true;
-            }
-            catch (Exception ex) when (!ct.IsCancellationRequested)
-            {
-                lastError = ex;
-                AppBootstrapLogger.Warn($"RPC reconnect attempt {attempt + 1}/{delays.Length} failed: {ex.Message}");
-            }
-        }
-
-        AppBootstrapLogger.Error("RPC reconnect exhausted startup retries", lastError);
-        return false;
-    }
-
     private async Task MaybeAutoStartVpnOnLaunchAsync(
         ISettingsManager<CoderConnectSettings> settingsManager,
         ICredentialManager credentialManager,

Tests.Vpn/Tests.Vpn.csproj

@@ -32,6 +32,7 @@
 
     <ItemGroup>
         <ProjectReference Include="..\Vpn\Vpn.csproj" />
+        <ProjectReference Include="..\Vpn.Linux\Vpn.Linux.csproj" />
     </ItemGroup>
 
 </Project>

Tests.Vpn/UnixSocketClientTransportTest.cs

@@ -0,0 +1,41 @@
+using System.Net.Sockets;
+using System.Runtime.Versioning;
+using Coder.Desktop.Vpn;
+
+namespace Coder.Desktop.Tests.Vpn;
+
+[TestFixture]
+[Platform("Linux", Reason = "UnixSocketClientTransport is Linux-only")]
+[SupportedOSPlatform("linux")]
+public class UnixSocketClientTransportTest
+{
+    [Test(Description = "ConnectAsync waits until the service socket exists")]
+    [CancelAfter(30_000)]
+    public async Task ConnectAsync_WaitsForSocketToAppear(CancellationToken ct)
+    {
+        var socketPath = Path.Combine(Path.GetTempPath(), $"coder-desktop-test-{Guid.NewGuid():N}.sock");
+        Socket? listener = null;
+
+        try
+        {
+            var transport = new UnixSocketClientTransport(socketPath);
+            var connectTask = transport.ConnectAsync(ct);
+
+            await Task.Delay(100, ct);
+            Assert.That(connectTask.IsCompleted, Is.False);
+
+            listener = new Socket(AddressFamily.Unix, SocketType.Stream, ProtocolType.Unspecified);
+            listener.Bind(new UnixDomainSocketEndPoint(socketPath));
+            listener.Listen(1);
+
+            using var acceptedSocket = await listener.AcceptAsync(ct);
+            await using var clientStream = await connectTask;
+            Assert.That(clientStream.CanRead, Is.True);
+        }
+        finally
+        {
+            listener?.Dispose();
+            try { File.Delete(socketPath); } catch { /* best effort */ }
+        }
+    }
+}

Vpn.Linux/UnixSocketClientTransport.cs

@@ -16,16 +16,34 @@ public UnixSocketClientTransport(string socketPath = "/run/coder-desktop/vpn.soc
 
     public async Task<Stream> ConnectAsync(CancellationToken ct)
     {
-        var socket = new Socket(AddressFamily.Unix, SocketType.Stream, ProtocolType.Unspecified);
-        try
-        {
-            await socket.ConnectAsync(new UnixDomainSocketEndPoint(_socketPath), ct);
-            return new NetworkStream(socket, ownsSocket: true);
-        }
-        catch
+        var retryDelay = TimeSpan.FromMilliseconds(100);
+
+        while (true)
         {
-            socket.Dispose();
-            throw;
+            ct.ThrowIfCancellationRequested();
+            var socket = new Socket(AddressFamily.Unix, SocketType.Stream, ProtocolType.Unspecified);
+            try
+            {
+                await socket.ConnectAsync(new UnixDomainSocketEndPoint(_socketPath), ct);
+                return new NetworkStream(socket, ownsSocket: true);
+            }
+            catch (SocketException ex) when (IsSocketUnavailable(ex) && !ct.IsCancellationRequested)
+            {
+                socket.Dispose();
+                await Task.Delay(retryDelay, ct);
+                retryDelay = TimeSpan.FromMilliseconds(Math.Min(retryDelay.TotalMilliseconds * 2, 1000));
+            }
+            catch
+            {
+                socket.Dispose();
+                throw;
+            }
         }
     }
+
+    private static bool IsSocketUnavailable(SocketException ex)
+    {
+        // Keep Linux startup behavior aligned with Windows named-pipe connect: wait for the service socket.
+        return ex.SocketErrorCode is SocketError.AddressNotAvailable or SocketError.ConnectionRefused;
+    }
 }