fix(cli): retry dial timeouts in SSH connection setup (#24199) (#24229)

Cherry-pick of https://github.com/coder/coder/pull/24199

Original PR: #24199 — fix(cli): retry dial timeouts in SSH connection
setup
Merge commit: 1d0653cdab
Requested by: @f0ssel

Co-authored-by: Ehab Younes <ehab.alyounes@gmail.com>

cli/ssh.go

@@ -69,15 +69,17 @@ var (
 // isRetryableError checks for transient connection errors worth
 // retrying: DNS failures, connection refused, and server 5xx.
 func isRetryableError(err error) bool {
-	if err == nil {
-		return false
-	}
-	if xerrors.Is(err, context.Canceled) || xerrors.Is(err, context.DeadlineExceeded) {
+	if err == nil || xerrors.Is(err, context.Canceled) {
 		return false
 	}
+	// Check connection errors before context.DeadlineExceeded because
+	// net.Dialer.Timeout produces *net.OpError that matches both.
 	if codersdk.IsConnectionError(err) {
 		return true
 	}
+	if xerrors.Is(err, context.DeadlineExceeded) {
+		return false
+	}
 	var sdkErr *codersdk.Error
 	if xerrors.As(err, &sdkErr) {
 		return sdkErr.StatusCode() >= 500

cli/ssh_internal_test.go

@@ -516,6 +516,23 @@ func TestIsRetryableError(t *testing.T) {
 			assert.Equal(t, tt.retryable, isRetryableError(tt.err))
 		})
 	}
+
+	// net.Dialer.Timeout produces *net.OpError that matches both
+	// IsConnectionError and context.DeadlineExceeded. Verify it is retryable.
+	t.Run("DialTimeout", func(t *testing.T) {
+		t.Parallel()
+		ctx, cancel := context.WithDeadline(context.Background(), time.Now())
+		defer cancel()
+		<-ctx.Done() // ensure deadline has fired
+		_, err := (&net.Dialer{}).DialContext(ctx, "tcp", "127.0.0.1:1")
+		require.Error(t, err)
+		// Proves the ambiguity: this error matches BOTH checks.
+		require.ErrorIs(t, err, context.DeadlineExceeded)
+		require.ErrorAs(t, err, new(*net.OpError))
+		assert.True(t, isRetryableError(err))
+		// Also when wrapped, as runCoderConnectStdio does.
+		assert.True(t, isRetryableError(xerrors.Errorf("dial coder connect: %w", err)))
+	})
 }
 
 func TestRetryWithInterval(t *testing.T) {