fix(multiplayer): recover ambiguous agones allocations

server/agones/allocation.go

@@ -85,15 +85,80 @@ type allocationResponse struct {
 type gameServerListResponse struct {
 	Items []struct {
 		Metadata struct {
-			Name   string            `json:"name"`
-			Labels map[string]string `json:"labels"`
+			Name        string            `json:"name"`
+			Labels      map[string]string `json:"labels"`
+			Annotations map[string]string `json:"annotations"`
 		} `json:"metadata"`
 		Status struct {
-			State string `json:"state"`
+			State   string `json:"state"`
+			Address string `json:"address"`
+			Ports   []struct {
+				Name string `json:"name"`
+				Port int    `json:"port"`
+			} `json:"ports"`
 		} `json:"status"`
 	} `json:"items"`
 }
 
+// RecoverAllocation finds a provider-side allocation that may have completed
+// before the durable allocation record was written. The allocation ID and
+// compatibility tuple are checked together so a stale or forged provider
+// object cannot be rebound to another match.
+func (c Client) RecoverAllocation(ctx context.Context, request domain.AllocationRequest, now time.Time) (AllocatedServer, bool, error) {
+	if request.AllocationID == "" || request.MatchID == "" || now.IsZero() {
+		return AllocatedServer{}, false, domain.ErrAllocationInput
+	}
+	if c.HTTP == nil {
+		c.HTTP = http.DefaultClient
+	}
+	base, err := c.endpoint()
+	if err != nil {
+		return AllocatedServer{}, false, err
+	}
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, base+"/apis/agones.dev/v1/namespaces/"+url.PathEscape(c.Namespace)+"/gameservers", nil)
+	if err != nil {
+		return AllocatedServer{}, false, err
+	}
+	response, err := c.HTTP.Do(req)
+	if err != nil {
+		return AllocatedServer{}, false, err
+	}
+	defer response.Body.Close()
+	if response.StatusCode < 200 || response.StatusCode >= 300 {
+		return AllocatedServer{}, false, fmt.Errorf("Agones GameServer recovery returned %s", response.Status)
+	}
+	var decoded gameServerListResponse
+	if err := json.NewDecoder(io.LimitReader(response.Body, 1<<20)).Decode(&decoded); err != nil {
+		return AllocatedServer{}, false, fmt.Errorf("decode Agones recovery list: %w", err)
+	}
+	found := false
+	var recovered AllocatedServer
+	for _, item := range decoded.Items {
+		if item.Status.State != "Allocated" || item.Metadata.Annotations["cosmic-clash.io/allocation-id"] != request.AllocationID {
+			continue
+		}
+		if found {
+			return AllocatedServer{}, false, domain.ErrConflict
+		}
+		if item.Metadata.Name == "" || item.Status.Address == "" || strings.ContainsAny(item.Status.Address, " \t\r\n") {
+			return AllocatedServer{}, false, fmt.Errorf("Agones recovered GameServer has invalid identity or address")
+		}
+		if item.Metadata.Annotations["cosmic-clash.io/match-id"] != request.MatchID {
+			return AllocatedServer{}, false, domain.ErrConflict
+		}
+		if item.Metadata.Labels["cosmic-clash.io/region"] != request.Region || item.Metadata.Labels["cosmic-clash.io/build"] != request.Build || item.Metadata.Labels["cosmic-clash.io/protocol"] != strconv.Itoa(request.Protocol) || item.Metadata.Labels["cosmic-clash.io/transport"] != request.Transport {
+			return AllocatedServer{}, false, domain.ErrConflict
+		}
+		port, err := selectPort(item.Status.Ports)
+		if err != nil {
+			return AllocatedServer{}, false, err
+		}
+		recovered = AllocatedServer{Allocation: domain.Allocation{AllocationID: request.AllocationID, MatchID: request.MatchID, ServerID: item.Metadata.Name, Region: request.Region, Build: request.Build, Protocol: request.Protocol, Transport: request.Transport, State: domain.ServerAllocated, AllocatedAt: now}, Endpoint: net.JoinHostPort(item.Status.Address, strconv.Itoa(port)), GameServer: item.Metadata.Name}
+		found = true
+	}
+	return recovered, found, nil
+}
+
 // ListReadyServers projects only Agones Ready GameServers into the durable
 // allocator registry. Compatibility fields must be present as Fleet labels;
 // malformed Ready objects fail closed instead of creating selectable capacity.