CosmicClash/server/agones/allocation.go

// Package agones contains the narrow provider adapter used by the allocator.
// Domain policy and durable allocation records remain outside this package.
package agones

import (
	"bytes"
	"context"
	"encoding/json"
	"fmt"
	"io"
	"net"
	"net/http"
	"net/url"
	"strconv"
	"strings"
	"time"

	"github.com/cosmic-clash/cosmic-clash/server/domain"
	"github.com/cosmic-clash/cosmic-clash/server/workload"
)

type Client struct {
	BaseURL   string
	Namespace string
	HTTP      *http.Client

	// WorkloadSecret, when set, mints a control-plane-self-issued signed
	// workload token (server/workload/signed_token.go) for every allocation
	// and requests it as the cosmic-clash.io/workload-token annotation
	// alongside match-id/allocation-id -- the delivery channel
	// supervisor.Supervisor.workloadToken() reads from. It must be the same
	// secret cmd/control-plane verifies with (--workload-secret /
	// COSMIC_CLASH_WORKLOAD_SECRET). Left unset, Allocate behaves exactly as
	// before: no workload-token annotation is requested, matching how a
	// deployment not yet using this delivery path (e.g. one still building
	// toward a Kubernetes-JWT WorkloadVerify) is unaffected.
	WorkloadSecret []byte
	// WorkloadTokenTTL bounds how long the minted token remains valid; it
	// must comfortably exceed the time between allocation and this
	// GameServer completing process-ready/assignment-ready registration.
	// Zero defaults to 30 minutes.
	WorkloadTokenTTL time.Duration
}

type AllocatedServer struct {
	Allocation domain.Allocation
	Endpoint   string
	GameServer string
}

type allocationRequest struct {
	APIVersion string `json:"apiVersion"`
	Kind       string `json:"kind"`
	Spec       struct {
		Selectors []struct {
			MatchLabels map[string]string `json:"matchLabels"`
		} `json:"selectors"`
		// Metadata.Annotations is applied to the allocated GameServer's own
		// object_meta by Agones on successful allocation (a documented part
		// of the GameServerAllocation spec, independent of the Selectors
		// used to find capacity). This is the only way match-specific data
		// reaches an already-Ready pod after allocation: Kubernetes env vars
		// are fixed at pod creation, long before Agones assigns a match to
		// that pod, so there is no other channel for it. The allocated
		// process reads these back via the SDK's own GameServer call
		// (server/supervisor's existing /gameserver request).
		Metadata struct {
			Annotations map[string]string `json:"annotations"`
		} `json:"metadata"`
	} `json:"spec"`
}

type allocationResponse struct {
	Status struct {
		State          string `json:"state"`
		GameServerName string `json:"gameServerName"`
		Address        string `json:"address"`
		Ports          []struct {
			Name string `json:"name"`
			Port int    `json:"port"`
		} `json:"ports"`
	} `json:"status"`
}

type gameServerListResponse struct {
	Items []struct {
		Metadata struct {
			Name        string            `json:"name"`
			Labels      map[string]string `json:"labels"`
			Annotations map[string]string `json:"annotations"`
		} `json:"metadata"`
		Status struct {
			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.
func (c Client) ListReadyServers(ctx context.Context) ([]domain.ReadyServer, error) {
	if c.HTTP == nil {
		c.HTTP = http.DefaultClient
	}
	base, err := c.endpoint()
	if err != nil {
		return nil, err
	}
	req, err := http.NewRequestWithContext(ctx, http.MethodGet, base+"/apis/agones.dev/v1/namespaces/"+url.PathEscape(c.Namespace)+"/gameservers", nil)
	if err != nil {
		return nil, err
	}
	response, err := c.HTTP.Do(req)
	if err != nil {
		return nil, err
	}
	defer response.Body.Close()
	if response.StatusCode < 200 || response.StatusCode >= 300 {
		return nil, fmt.Errorf("Agones GameServer list returned %s", response.Status)
	}
	var decoded gameServerListResponse
	if err := json.NewDecoder(io.LimitReader(response.Body, 1<<20)).Decode(&decoded); err != nil {
		return nil, fmt.Errorf("decode Agones GameServer list: %w", err)
	}
	ready := make([]domain.ReadyServer, 0, len(decoded.Items))
	for _, item := range decoded.Items {
		if item.Status.State != "Ready" {
			continue
		}
		server, err := readyServerFromGameServer(item.Metadata.Name, item.Metadata.Labels)
		if err != nil {
			return nil, err
		}
		ready = append(ready, server)
	}
	return ready, nil
}

func readyServerFromGameServer(name string, labels map[string]string) (domain.ReadyServer, error) {
	protocol, err := strconv.Atoi(labels["cosmic-clash.io/protocol"])
	server := domain.ReadyServer{ServerID: name, Region: labels["cosmic-clash.io/region"], Build: labels["cosmic-clash.io/build"], Protocol: protocol, Transport: labels["cosmic-clash.io/transport"], State: domain.ServerReady}
	if err != nil || server.ServerID == "" || (server.Region != "EU" && server.Region != "NA") || server.Build == "" || server.Protocol < 1 || (server.Transport != "enet" && server.Transport != "steam_sdr") {
		return domain.ReadyServer{}, fmt.Errorf("invalid Ready GameServer compatibility labels")
	}
	return server, nil
}

func (c Client) Allocate(ctx context.Context, request domain.AllocationRequest, labels map[string]string, now time.Time) (AllocatedServer, error) {
	if c.HTTP == nil {
		c.HTTP = http.DefaultClient
	}
	base, err := c.endpoint()
	if err != nil {
		return AllocatedServer{}, err
	}
	if request.AllocationID == "" || request.MatchID == "" || (request.Region != "EU" && request.Region != "NA") || request.Build == "" || request.Protocol <= 0 || (request.Transport != "enet" && request.Transport != "steam_sdr") || now.IsZero() {
		return AllocatedServer{}, domain.ErrAllocationInput
	}
	if len(labels) == 0 {
		return AllocatedServer{}, fmt.Errorf("allocation labels are required")
	}
	for key, value := range labels {
		if key == "" || value == "" || strings.ContainsAny(key+value, "\r\n") {
			return AllocatedServer{}, fmt.Errorf("invalid allocation label")
		}
	}
	var body allocationRequest
	body.APIVersion = "allocation.agones.dev/v1"
	body.Kind = "GameServerAllocation"
	body.Spec.Selectors = []struct {
		MatchLabels map[string]string `json:"matchLabels"`
	}{{MatchLabels: cloneLabels(labels)}}
	body.Spec.Metadata.Annotations = map[string]string{
		"cosmic-clash.io/match-id":      request.MatchID,
		"cosmic-clash.io/allocation-id": request.AllocationID,
		"cosmic-clash.io/region":        request.Region,
		"cosmic-clash.io/build":         request.Build,
		"cosmic-clash.io/protocol":      strconv.Itoa(request.Protocol),
		"cosmic-clash.io/transport":     request.Transport,
	}
	if request.ArenaPath != "" {
		body.Spec.Metadata.Annotations["cosmic-clash.io/arena-path"] = request.ArenaPath
	}
	if playlist := labels["cosmic-clash.io/playlist"]; playlist == string(domain.Casual) || playlist == string(domain.Ranked) {
		body.Spec.Metadata.Annotations["cosmic-clash.io/playlist"] = playlist
	}
	if len(c.WorkloadSecret) > 0 {
		ttl := c.WorkloadTokenTTL
		if ttl <= 0 {
			ttl = 30 * time.Minute
		}
		token, err := workload.IssueSignedWorkloadToken(c.WorkloadSecret, request.AllocationID, now, ttl)
		if err != nil {
			return AllocatedServer{}, fmt.Errorf("issue workload token: %w", err)
		}
		body.Spec.Metadata.Annotations["cosmic-clash.io/workload-token"] = token
	}
	encoded, err := json.Marshal(body)
	if err != nil {
		return AllocatedServer{}, err
	}
	req, err := http.NewRequestWithContext(ctx, http.MethodPost, base+"/apis/allocation.agones.dev/v1/namespaces/"+url.PathEscape(c.Namespace)+"/gameserverallocations", bytes.NewReader(encoded))
	if err != nil {
		return AllocatedServer{}, err
	}
	req.Header.Set("Content-Type", "application/json")
	response, err := c.HTTP.Do(req)
	if err != nil {
		return AllocatedServer{}, err
	}
	defer response.Body.Close()
	if response.StatusCode < 200 || response.StatusCode >= 300 {
		return AllocatedServer{}, fmt.Errorf("Agones allocation returned %s", response.Status)
	}
	var decoded allocationResponse
	decoder := json.NewDecoder(io.LimitReader(response.Body, 64<<10))
	if err := decoder.Decode(&decoded); err != nil {
		return AllocatedServer{}, fmt.Errorf("decode Agones allocation: %w", err)
	}
	if decoded.Status.State != "Allocated" || decoded.Status.GameServerName == "" || decoded.Status.Address == "" {
		return AllocatedServer{}, fmt.Errorf("Agones allocation is incomplete")
	}
	port, err := selectPort(decoded.Status.Ports)
	if err != nil {
		return AllocatedServer{}, err
	}
	return AllocatedServer{Allocation: domain.Allocation{AllocationID: request.AllocationID, MatchID: request.MatchID, ServerID: decoded.Status.GameServerName, Region: request.Region, Build: request.Build, Protocol: request.Protocol, Transport: request.Transport, State: domain.ServerAllocated, AllocatedAt: now}, Endpoint: net.JoinHostPort(decoded.Status.Address, strconv.Itoa(port)), GameServer: decoded.Status.GameServerName}, nil
}

func (c Client) endpoint() (string, error) {
	if c.Namespace == "" || strings.ContainsAny(c.Namespace, "/\r\n") {
		return "", fmt.Errorf("invalid Agones namespace")
	}
	u, err := url.Parse(c.BaseURL)
	if err != nil || (u.Scheme != "http" && u.Scheme != "https") || u.Host == "" || u.RawQuery != "" || u.Fragment != "" || u.Path != "" {
		return "", fmt.Errorf("invalid Agones base URL")
	}
	return strings.TrimRight(c.BaseURL, "/"), nil
}

func selectPort(ports []struct {
	Name string `json:"name"`
	Port int    `json:"port"`
}) (int, error) {
	for _, port := range ports {
		if port.Name == "default" {
			if port.Port < 1 || port.Port > 65535 {
				return 0, fmt.Errorf("Agones returned invalid default port")
			}
			return port.Port, nil
		}
	}
	if len(ports) != 1 || ports[0].Port < 1 || ports[0].Port > 65535 {
		return 0, fmt.Errorf("Agones returned no usable game port")
	}
	return ports[0].Port, nil
}

func cloneLabels(labels map[string]string) map[string]string {
	copy := make(map[string]string, len(labels))
	for key, value := range labels {
		copy[key] = value
	}
	return copy
}