Phase3
Some checks failed
CI / Node — lint, typecheck, test, build (push) Failing after 9s
CI / Go — node-agent tests (push) Failing after 9s
CI / Go — edge-gateway build (push) Successful in 13s

This commit is contained in:
smueller
2026-06-26 12:17:26 +02:00
parent c4077d4673
commit 9b061c3ee7
92 changed files with 4128 additions and 146 deletions

View File

@@ -2,11 +2,13 @@ package runtime
import (
"context"
"fmt"
"encoding/json"
"errors"
"log/slog"
"sync"
"github.com/hexahost/gamecloud/node-agent/internal/docker"
"github.com/hexahost/gamecloud/node-agent/internal/files"
"github.com/hexahost/gamecloud/node-agent/internal/protocol"
)
@@ -27,6 +29,7 @@ const (
// Responder sends protocol responses back to the control plane.
type Responder interface {
SendState(ctx context.Context, correlationID string, payload protocol.ServerStatePayload) error
SendLog(ctx context.Context, correlationID string, payload protocol.ServerLogPayload) error
SendOperationComplete(ctx context.Context, correlationID string, payload protocol.ServerOperationResultPayload) error
SendOperationFailed(ctx context.Context, correlationID string, payload protocol.ServerOperationResultPayload) error
}
@@ -42,20 +45,22 @@ type serverRecord struct {
// 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
mu sync.RWMutex
docker *docker.Client
servers map[string]*serverRecord
logStreams map[string]context.CancelFunc
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,
docker: dockerClient,
servers: make(map[string]*serverRecord),
logStreams: make(map[string]context.CancelFunc),
resp: resp,
log: log,
}
}
@@ -184,6 +189,8 @@ func (m *Manager) HandleDelete(ctx context.Context, correlationID string, meta p
m.emitState(ctx, correlationID, meta, ServerStateDeleting, "", "")
m.stopLogStream(meta.ServerID)
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())
@@ -204,6 +211,147 @@ func (m *Manager) HandleDelete(ctx context.Context, correlationID string, meta p
m.complete(ctx, correlationID, meta, ServerStateDeleted)
}
// HandleLogsSubscribe starts streaming container logs for a server.
func (m *Manager) HandleLogsSubscribe(ctx context.Context, correlationID string, payload protocol.ServerLogsSubscribePayload) {
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
}
streamCtx, cancel := context.WithCancel(ctx)
m.mu.Lock()
if existing, ok := m.logStreams[meta.ServerID]; ok {
existing()
}
m.logStreams[meta.ServerID] = cancel
m.mu.Unlock()
go func() {
defer func() {
m.mu.Lock()
if current, ok := m.logStreams[meta.ServerID]; ok && current == cancel {
delete(m.logStreams, meta.ServerID)
}
m.mu.Unlock()
}()
err := m.docker.StreamLogs(streamCtx, rec.ContainerID, payload.TailLines, payload.Follow, func(stream, line string) {
logPayload := protocol.ServerLogPayload{
OperationMeta: meta,
Line: line,
Stream: stream,
}
if err := m.resp.SendLog(ctx, correlationID, logPayload); err != nil {
m.log.Warn("send server.log failed",
"server_id", meta.ServerID,
"error", err,
)
}
})
if err != nil && streamCtx.Err() == nil {
m.log.Warn("log stream ended with error",
"server_id", meta.ServerID,
"error", err,
)
}
}()
}
// HandleCommand executes a console command via RCON.
func (m *Manager) HandleCommand(ctx context.Context, correlationID string, payload protocol.ServerCommandPayload) {
meta := payload.OperationMeta
rec, ok := m.getServer(meta.ServerID)
if !ok {
m.failOperation(ctx, correlationID, meta, "command", "SERVER_NOT_FOUND", "server is not provisioned on this node")
return
}
stdout, stderr, err := m.docker.ExecRcon(ctx, rec.ContainerID, payload.Command)
if err != nil {
m.failOperation(ctx, correlationID, meta, "command", "COMMAND_FAILED", err.Error())
return
}
m.completeOperation(ctx, correlationID, meta, "command", map[string]string{
"stdout": stdout,
"stderr": stderr,
})
}
// HandleFilesList lists files in the server data directory.
func (m *Manager) HandleFilesList(ctx context.Context, correlationID string, payload protocol.ServerFilesListPayload) {
meta := payload.OperationMeta
rec, ok := m.getServer(meta.ServerID)
if !ok {
m.failOperation(ctx, correlationID, meta, "files.list", "SERVER_NOT_FOUND", "server is not provisioned on this node")
return
}
entries, err := files.ListDir(rec.DataPath, payload.Path)
if err != nil {
m.failOperation(ctx, correlationID, meta, "files.list", filesErrorCode(err), err.Error())
return
}
result := protocol.ServerFilesListResultPayload{
Entries: make([]protocol.FileEntry, 0, len(entries)),
}
for _, entry := range entries {
result.Entries = append(result.Entries, protocol.FileEntry{
Name: entry.Name,
Path: entry.Path,
IsDir: entry.IsDir,
Size: entry.Size,
ModifiedAt: entry.ModifiedAt.Format("2006-01-02T15:04:05Z"),
})
}
m.completeOperation(ctx, correlationID, meta, "files.list", result)
}
// HandleFilesRead reads a file from the server data directory.
func (m *Manager) HandleFilesRead(ctx context.Context, correlationID string, payload protocol.ServerFilesReadPayload) {
meta := payload.OperationMeta
rec, ok := m.getServer(meta.ServerID)
if !ok {
m.failOperation(ctx, correlationID, meta, "files.read", "SERVER_NOT_FOUND", "server is not provisioned on this node")
return
}
content, truncated, err := files.ReadFile(rec.DataPath, payload.Path, payload.MaxBytes)
if err != nil {
m.failOperation(ctx, correlationID, meta, "files.read", filesErrorCode(err), err.Error())
return
}
m.completeOperation(ctx, correlationID, meta, "files.read", protocol.ServerFilesReadResultPayload{
Path: payload.Path,
Content: content,
Truncated: truncated,
})
}
// HandleFilesWrite writes a file in the server data directory.
func (m *Manager) HandleFilesWrite(ctx context.Context, correlationID string, payload protocol.ServerFilesWritePayload) {
meta := payload.OperationMeta
rec, ok := m.getServer(meta.ServerID)
if !ok {
m.failOperation(ctx, correlationID, meta, "files.write", "SERVER_NOT_FOUND", "server is not provisioned on this node")
return
}
if err := files.WriteFile(rec.DataPath, payload.Path, payload.Content, payload.Create); err != nil {
m.failOperation(ctx, correlationID, meta, "files.write", filesErrorCode(err), err.Error())
return
}
m.completeOperation(ctx, correlationID, meta, "files.write", map[string]string{
"path": payload.Path,
})
}
func (m *Manager) getServer(serverID string) (*serverRecord, bool) {
m.mu.RLock()
defer m.mu.RUnlock()
@@ -223,6 +371,15 @@ func (m *Manager) setState(serverID, state string) {
}
}
func (m *Manager) stopLogStream(serverID string) {
m.mu.Lock()
defer m.mu.Unlock()
if cancel, ok := m.logStreams[serverID]; ok {
cancel()
delete(m.logStreams, serverID)
}
}
func (m *Manager) emitState(ctx context.Context, correlationID string, meta protocol.OperationMeta, state, errorCode, errorMessage string) {
payload := protocol.ServerStatePayload{
OperationMeta: meta,
@@ -267,3 +424,59 @@ func (m *Manager) fail(ctx context.Context, correlationID string, meta protocol.
)
}
}
func (m *Manager) completeOperation(ctx context.Context, correlationID string, meta protocol.OperationMeta, operation string, result any) {
raw, err := json.Marshal(result)
if err != nil {
m.failOperation(ctx, correlationID, meta, operation, "RESULT_MARSHAL_FAILED", err.Error())
return
}
payload := protocol.ServerOperationResultPayload{
OperationMeta: meta,
ResultCode: ResultCodeOK,
Operation: operation,
Result: raw,
}
if err := m.resp.SendOperationComplete(ctx, correlationID, payload); err != nil {
m.log.Warn("send operation complete failed",
"server_id", meta.ServerID,
"operation", operation,
"error", err,
)
}
}
func (m *Manager) failOperation(ctx context.Context, correlationID string, meta protocol.OperationMeta, operation, errorCode, errorMessage string) {
payload := protocol.ServerOperationResultPayload{
OperationMeta: meta,
ResultCode: ResultCodeError,
Operation: operation,
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,
"operation", operation,
"error", err,
)
}
}
func filesErrorCode(err error) string {
switch {
case errors.Is(err, files.ErrPathTraversal):
return "PATH_TRAVERSAL"
case errors.Is(err, files.ErrSymlink):
return "SYMLINK_NOT_ALLOWED"
case errors.Is(err, files.ErrNotFound):
return "FILE_NOT_FOUND"
case errors.Is(err, files.ErrNotDirectory):
return "NOT_A_DIRECTORY"
case errors.Is(err, files.ErrIsDirectory):
return "IS_DIRECTORY"
default:
return "FILES_ERROR"
}
}