Better race condition prevention

This commit is contained in:
Dane Everitt
2020-04-10 18:22:18 -07:00
parent 17d204a631
commit 03045c94be
9 changed files with 99 additions and 81 deletions

View File

@@ -181,8 +181,8 @@ func LoadDirectory() error {
}
if state, exists := states[s.Uuid]; exists {
s.State = state
zap.S().Debugw("loaded server state from cache", zap.String("server", s.Uuid), zap.String("state", s.State))
s.SetState(state)
zap.S().Debugw("loaded server state from cache", zap.String("server", s.Uuid), zap.String("state", s.GetState()))
}
servers.Add(s)
@@ -317,71 +317,6 @@ func (s *Server) CreateEnvironment() error {
return s.Environment.Create()
}
const (
ProcessOfflineState = "offline"
ProcessStartingState = "starting"
ProcessRunningState = "running"
ProcessStoppingState = "stopping"
)
// Sets the state of the server internally. This function handles crash detection as
// well as reporting to event listeners for the server.
func (s *Server) SetState(state string) error {
if state != ProcessOfflineState && state != ProcessStartingState && state != ProcessRunningState && state != ProcessStoppingState {
return errors.New(fmt.Sprintf("invalid server state received: %s", state))
}
prevState := s.State
s.State = state
// Persist this change to the disk immediately so that should the Daemon be stopped or
// crash we can immediately restore the server state.
//
// This really only makes a difference if all of the Docker containers are also stopped,
// but this was a highly requested feature and isn't hard to work with, so lets do it.
//
// We also get the benefit of server status changes always propagating corrected configurations
// to the disk should we forget to do it elsewhere.
go func() {
/*if _, err := server.WriteConfigurationToDisk(); err != nil {
zap.S().Warnw("failed to write server state change to disk", zap.String("server", server.Uuid), zap.Error(err))
}*/
if err := SaveServerStates(); err != nil {
zap.S().Warnw("failed to write server states to disk", zap.Error(err))
}
}()
zap.S().Debugw("saw server status change event", zap.String("server", s.Uuid), zap.String("status", s.State))
// Emit the event to any listeners that are currently registered.
s.Events().Publish(StatusEvent, s.State)
// If server was in an online state, and is now in an offline state we should handle
// that as a crash event. In that scenario, check the last crash time, and the crash
// counter.
//
// In the event that we have passed the thresholds, don't do anything, otherwise
// automatically attempt to start the process back up for the user. This is done in a
// separate thread as to not block any actions currently taking place in the flow
// that called this function.
if (prevState == ProcessStartingState || prevState == ProcessRunningState) && s.State == ProcessOfflineState {
zap.S().Infow("detected server as entering a potentially crashed state; running handler", zap.String("server", s.Uuid))
go func(server *Server) {
if err := server.handleServerCrash(); err != nil {
if IsTooFrequentCrashError(err) {
zap.S().Infow("did not restart server after crash; occurred too soon after last", zap.String("server", server.Uuid))
} else {
zap.S().Errorw("failed to handle server crash state", zap.String("server", server.Uuid), zap.Error(err))
}
}
}(s)
}
return nil
}
// Gets the process configuration data for the server.
func (s *Server) GetProcessConfiguration() (*api.ServerConfigurationResponse, *api.RequestError, error) {
return api.NewRequester().GetServerConfiguration(s.Uuid)