hacky fix for sporadic Gateway Timeouts on play/challenge start

internal/api/client.go

@@ -14,12 +14,19 @@ import (
 	"strings"
 )
 
-var ErrAuthenticationRequired = errors.New("authentication required")
+var (
+	ErrAuthenticationRequired = errors.New("authentication required")
+	ErrGatewayTimeout         = errors.New("gateway timeout")
+)
 
 func isAuthenticationRequiredResponse(resp *http.Response) bool {
 	return resp.StatusCode == http.StatusUnauthorized
 }
 
+func isGatewayTimeoutResponse(resp *http.Response) bool {
+	return resp.StatusCode == http.StatusGatewayTimeout
+}
+
 type Client struct {
 	baseURL    string
 	apiBaseURL string
@@ -356,6 +363,9 @@ func (c *Client) doRequest(req *http.Request) (*http.Response, error) {
 		if isAuthenticationRequiredResponse(resp) {
 			return nil, ErrAuthenticationRequired
 		}
+		if isGatewayTimeoutResponse(resp) {
+			return nil, ErrGatewayTimeout
+		}
 
 		return nil, fmt.Errorf("request failed with status %d: %s", resp.StatusCode, body)
 	}

internal/api/plays.go

@@ -2,6 +2,9 @@ package api
 
 import (
 	"context"
+	"errors"
+	"fmt"
+	"time"
 )
 
 type Play struct {
@@ -169,6 +172,30 @@ func (c *Client) StartTunnel(ctx context.Context, id string, req StartTunnelRequ
 		return nil, err
 	}
 
-	var resp StartTunnelResponse
-	return &resp, c.PostInto(ctx, "/plays/"+id+"/tunnels", nil, nil, body, &resp)
+	// A hacky workaround for the fact that the CLI currently
+	// doesn't check for playground readiness before establishing
+	// a tunnel.
+	backoff := 200 * time.Millisecond
+	for attempt := 0; attempt < 5; attempt++ {
+		var resp StartTunnelResponse
+		err := c.PostInto(ctx, "/plays/"+id+"/tunnels", nil, nil, body, &resp)
+		if err == nil {
+			return &resp, nil
+		}
+		if !errors.Is(err, ErrGatewayTimeout) {
+			return nil, err
+		}
+		if attempt == 2 {
+			return nil, fmt.Errorf("max retries exceeded: %w", err)
+		}
+
+		select {
+		case <-ctx.Done():
+			return nil, ctx.Err()
+		case <-time.After(backoff):
+		}
+		backoff *= 2
+	}
+
+	return nil, ctx.Err()
 }