Phase2
Some checks failed
CI / Node — lint, typecheck, test, build (push) Failing after 10s
CI / Go — node-agent tests (push) Failing after 8s
CI / Go — edge-gateway build (push) Successful in 13s

This commit is contained in:
smueller
2026-06-26 12:01:25 +02:00
parent 58961000eb
commit c4077d4673
93 changed files with 4099 additions and 165 deletions

View File

@@ -0,0 +1,269 @@
package runtime
import (
"context"
"fmt"
"log/slog"
"sync"
"github.com/hexahost/gamecloud/node-agent/internal/docker"
"github.com/hexahost/gamecloud/node-agent/internal/protocol"
)
const (
ServerStateProvisioning = "provisioning"
ServerStateStopped = "stopped"
ServerStateStarting = "starting"
ServerStateRunning = "running"
ServerStateStopping = "stopping"
ServerStateDeleting = "deleting"
ServerStateDeleted = "deleted"
ServerStateError = "error"
ResultCodeOK = "ok"
ResultCodeError = "error"
)
// Responder sends protocol responses back to the control plane.
type Responder interface {
SendState(ctx context.Context, correlationID string, payload protocol.ServerStatePayload) error
SendOperationComplete(ctx context.Context, correlationID string, payload protocol.ServerOperationResultPayload) error
SendOperationFailed(ctx context.Context, correlationID string, payload protocol.ServerOperationResultPayload) error
}
type serverRecord struct {
ServerID string
ContainerID string
HostPort int
DataPath string
RamMb int
State string
}
// Manager tracks provisioned servers and executes lifecycle operations.
type Manager struct {
mu sync.RWMutex
docker *docker.Client
servers map[string]*serverRecord
resp Responder
log *slog.Logger
}
// NewManager creates a runtime manager backed by the given Docker client.
func NewManager(dockerClient *docker.Client, resp Responder, log *slog.Logger) *Manager {
return &Manager{
docker: dockerClient,
servers: make(map[string]*serverRecord),
resp: resp,
log: log,
}
}
// RunningCount returns the number of servers currently in the running state.
func (m *Manager) RunningCount() int {
m.mu.RLock()
defer m.mu.RUnlock()
count := 0
for _, rec := range m.servers {
if rec.State == ServerStateRunning {
count++
}
}
return count
}
// HandleProvision creates a Minecraft container for the server.
func (m *Manager) HandleProvision(ctx context.Context, correlationID string, payload protocol.ServerProvisionPayload) {
meta := payload.OperationMeta
m.log.Info("provision server",
"server_id", meta.ServerID,
"host_port", payload.HostPort,
"version", payload.MinecraftVersion,
)
if !payload.EulaAccepted {
m.fail(ctx, correlationID, meta, "EULA_NOT_ACCEPTED", "EULA must be accepted before provisioning")
return
}
m.mu.Lock()
if existing, ok := m.servers[meta.ServerID]; ok {
m.mu.Unlock()
m.log.Info("provision idempotent hit", "server_id", meta.ServerID, "container_id", existing.ContainerID)
m.complete(ctx, correlationID, meta, existing.State)
return
}
m.mu.Unlock()
m.emitState(ctx, correlationID, meta, ServerStateProvisioning, "", "")
containerID, err := m.docker.CreateMinecraftServer(ctx, docker.MinecraftServerSpec{
ServerID: meta.ServerID,
MinecraftVersion: payload.MinecraftVersion,
SoftwareFamily: payload.SoftwareFamily,
HostPort: payload.HostPort,
RamMb: payload.RamMb,
DataPath: payload.DataPath,
EulaAccepted: payload.EulaAccepted,
})
if err != nil {
m.fail(ctx, correlationID, meta, "PROVISION_FAILED", err.Error())
return
}
rec := &serverRecord{
ServerID: meta.ServerID,
ContainerID: containerID,
HostPort: payload.HostPort,
DataPath: payload.DataPath,
RamMb: payload.RamMb,
State: ServerStateStopped,
}
m.mu.Lock()
m.servers[meta.ServerID] = rec
m.mu.Unlock()
m.emitState(ctx, correlationID, meta, ServerStateStopped, "", "")
m.complete(ctx, correlationID, meta, ServerStateStopped)
}
// HandleStart starts a provisioned server container.
func (m *Manager) HandleStart(ctx context.Context, correlationID string, payload protocol.ServerStartPayload) {
meta := payload.OperationMeta
rec, ok := m.getServer(meta.ServerID)
if !ok {
m.fail(ctx, correlationID, meta, "SERVER_NOT_FOUND", "server is not provisioned on this node")
return
}
m.emitState(ctx, correlationID, meta, ServerStateStarting, "", "")
if err := m.docker.StartContainer(ctx, rec.ContainerID); err != nil {
m.setState(meta.ServerID, ServerStateError)
m.emitState(ctx, correlationID, meta, ServerStateError, "START_FAILED", err.Error())
m.fail(ctx, correlationID, meta, "START_FAILED", err.Error())
return
}
m.setState(meta.ServerID, ServerStateRunning)
m.emitState(ctx, correlationID, meta, ServerStateRunning, "", "")
m.complete(ctx, correlationID, meta, ServerStateRunning)
}
// HandleStop stops a running server container.
func (m *Manager) HandleStop(ctx context.Context, correlationID string, payload protocol.ServerStopPayload) {
meta := payload.OperationMeta
rec, ok := m.getServer(meta.ServerID)
if !ok {
m.fail(ctx, correlationID, meta, "SERVER_NOT_FOUND", "server is not provisioned on this node")
return
}
m.emitState(ctx, correlationID, meta, ServerStateStopping, "", "")
if err := m.docker.StopContainer(ctx, rec.ContainerID); err != nil {
m.setState(meta.ServerID, ServerStateError)
m.emitState(ctx, correlationID, meta, ServerStateError, "STOP_FAILED", err.Error())
m.fail(ctx, correlationID, meta, "STOP_FAILED", err.Error())
return
}
m.setState(meta.ServerID, ServerStateStopped)
m.emitState(ctx, correlationID, meta, ServerStateStopped, "", "")
m.complete(ctx, correlationID, meta, ServerStateStopped)
}
// HandleDelete removes a server container and drops local state.
func (m *Manager) HandleDelete(ctx context.Context, correlationID string, meta protocol.OperationMeta) {
rec, ok := m.getServer(meta.ServerID)
if !ok {
m.complete(ctx, correlationID, meta, ServerStateDeleted)
return
}
m.emitState(ctx, correlationID, meta, ServerStateDeleting, "", "")
if state, err := m.docker.GetContainerState(ctx, rec.ContainerID); err == nil && state.Running {
if err := m.docker.StopContainer(ctx, rec.ContainerID); err != nil {
m.fail(ctx, correlationID, meta, "DELETE_STOP_FAILED", err.Error())
return
}
}
if err := m.docker.RemoveContainer(ctx, rec.ContainerID); err != nil {
m.fail(ctx, correlationID, meta, "DELETE_FAILED", err.Error())
return
}
m.mu.Lock()
delete(m.servers, meta.ServerID)
m.mu.Unlock()
m.emitState(ctx, correlationID, meta, ServerStateDeleted, "", "")
m.complete(ctx, correlationID, meta, ServerStateDeleted)
}
func (m *Manager) getServer(serverID string) (*serverRecord, bool) {
m.mu.RLock()
defer m.mu.RUnlock()
rec, ok := m.servers[serverID]
if !ok {
return nil, false
}
copy := *rec
return &copy, true
}
func (m *Manager) setState(serverID, state string) {
m.mu.Lock()
defer m.mu.Unlock()
if rec, ok := m.servers[serverID]; ok {
rec.State = state
}
}
func (m *Manager) emitState(ctx context.Context, correlationID string, meta protocol.OperationMeta, state, errorCode, errorMessage string) {
payload := protocol.ServerStatePayload{
OperationMeta: meta,
State: state,
ErrorCode: errorCode,
ErrorMessage: errorMessage,
}
if err := m.resp.SendState(ctx, correlationID, payload); err != nil {
m.log.Warn("send server.state failed",
"server_id", meta.ServerID,
"state", state,
"error", err,
)
}
}
func (m *Manager) complete(ctx context.Context, correlationID string, meta protocol.OperationMeta, _ string) {
payload := protocol.ServerOperationResultPayload{
OperationMeta: meta,
ResultCode: ResultCodeOK,
}
if err := m.resp.SendOperationComplete(ctx, correlationID, payload); err != nil {
m.log.Warn("send operation complete failed",
"server_id", meta.ServerID,
"error", err,
)
}
}
func (m *Manager) fail(ctx context.Context, correlationID string, meta protocol.OperationMeta, errorCode, errorMessage string) {
m.emitState(ctx, correlationID, meta, ServerStateError, errorCode, errorMessage)
payload := protocol.ServerOperationResultPayload{
OperationMeta: meta,
ResultCode: ResultCodeError,
ErrorCode: errorCode,
ErrorMessage: errorMessage,
}
if err := m.resp.SendOperationFailed(ctx, correlationID, payload); err != nil {
m.log.Warn("send operation failed failed",
"server_id", meta.ServerID,
"error", err,
)
}
}