Better race condition prevention

This commit is contained in:
Dane Everitt 2020-04-10 18:22:18 -07:00
parent 17d204a631
commit 03045c94be
No known key found for this signature in database
GPG Key ID: EEA66103B3D71F53
9 changed files with 99 additions and 81 deletions

View File

@ -134,7 +134,7 @@ func rootCmdRun(*cobra.Command, []string) {
//
// This will also validate that a server process is running if the last tracked state we have
// is that it was running, but we see that the container process is not currently running.
if r || (!r && (s.State == server.ProcessRunningState || s.State == server.ProcessStartingState)) {
if r || (!r && s.IsRunning()) {
zap.S().Infow("detected server is running, re-attaching to process", zap.String("server", s.Uuid))
if err := s.Environment.Start(); err != nil {
zap.S().Warnw(

View File

@ -227,7 +227,7 @@ func (h *Handler) HandleInbound(m Message) error {
// On every authentication event, send the current server status back
// to the client. :)
h.server.Events().Publish(server.StatusEvent, h.server.State)
h.server.Events().Publish(server.StatusEvent, h.server.GetState())
h.unsafeSendJson(Message{
Event: AuthenticationSuccessEvent,
@ -293,7 +293,7 @@ func (h *Handler) HandleInbound(m Message) error {
return nil
}
if h.server.State == server.ProcessOfflineState {
if h.server.GetState() == server.ProcessOfflineState {
return nil
}

View File

@ -33,7 +33,7 @@ func (s *Server) handleServerCrash() error {
// No point in doing anything here if the server isn't currently offline, there
// is no reason to do a crash detection event. If the server crash detection is
// disabled we want to skip anything after this as well.
if s.State != ProcessOfflineState || !s.CrashDetection.Enabled {
if s.GetState() != ProcessOfflineState || !s.CrashDetection.Enabled {
if !s.CrashDetection.Enabled {
zap.S().Debugw("server triggered crash detection but handler is disabled for server process", zap.String("server", s.Uuid))

View File

@ -271,7 +271,7 @@ func (d *DockerEnvironment) Stop() error {
// does not stop after seconds have passed, an error will be returned, or the instance
// will be terminated forcefully depending on the value of the second argument.
func (d *DockerEnvironment) WaitForStop(seconds int, terminate bool) error {
if d.Server.State == ProcessOfflineState {
if d.Server.GetState() == ProcessOfflineState {
return nil
}
@ -444,7 +444,7 @@ func (d *DockerEnvironment) FollowConsoleOutput() error {
// information, instead just sit there with an async process that lets Docker stream all of this data
// to us automatically.
func (d *DockerEnvironment) EnableResourcePolling() error {
if d.Server.State == ProcessOfflineState {
if d.Server.GetState() == ProcessOfflineState {
return errors.New("cannot enable resource polling on a server that is not running")
}
@ -472,7 +472,7 @@ func (d *DockerEnvironment) EnableResourcePolling() error {
// Disable collection if the server is in an offline state and this process is
// still running.
if s.State == ProcessOfflineState {
if s.GetState() == ProcessOfflineState {
d.DisableResourcePolling()
return
}

View File

@ -39,7 +39,7 @@ func (s *Server) Install() error {
// Reinstalls a server's software by utilizing the install script for the server egg. This
// does not touch any existing files for the server, other than what the script modifies.
func (s *Server) Reinstall() error {
if s.State != ProcessOfflineState {
if s.GetState() != ProcessOfflineState {
zap.S().Debugw("waiting for server instance to enter a stopped state", zap.String("server", s.Uuid))
if err := s.Environment.WaitForStop(10, true); err != nil {
return err

View File

@ -27,7 +27,7 @@ func (s *Server) onConsoleOutput(data string) {
// If the specific line of output is one that would mark the server as started,
// set the server to that state. Only do this if the server is not currently stopped
// or stopping.
if s.State == ProcessStartingState && strings.Contains(data, s.processConfiguration.Startup.Done) {
if s.GetState() == ProcessStartingState && strings.Contains(data, s.processConfiguration.Startup.Done) {
zap.S().Debugw(
"detected server in running state based on line output", zap.String("match", s.processConfiguration.Startup.Done), zap.String("against", data),
)
@ -38,7 +38,7 @@ func (s *Server) onConsoleOutput(data string) {
// If the command sent to the server is one that should stop the server we will need to
// set the server to be in a stopping state, otherwise crash detection will kick in and
// cause the server to unexpectedly restart on the user.
if s.State == ProcessStartingState || s.State == ProcessRunningState {
if s.IsRunning() {
if s.processConfiguration.Stop.Type == api.ProcessStopCommand && data == s.processConfiguration.Stop.Value {
s.SetState(ProcessStoppingState)
}

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)

View File

@ -2,7 +2,9 @@ package server
import (
"encoding/json"
"fmt"
"github.com/pkg/errors"
"go.uber.org/zap"
"io/ioutil"
"os"
"sync"
@ -61,12 +63,12 @@ func getServerStates() (map[string]string, error) {
return states, nil
}
// SaveServerStates .
func SaveServerStates() error {
// saveServerStates .
func saveServerStates() error {
// Get the states of all servers on the daemon.
states := map[string]string{}
for _, s := range GetServers().All() {
states[s.Uuid] = s.State
states[s.Uuid] = s.GetState()
}
// Convert the map to a json object.
@ -85,3 +87,84 @@ func SaveServerStates() error {
return nil
}
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.GetState()
// Obtain a mutex lock and update the current state of the server.
s.Lock()
s.State = state
// Emit the event to any listeners that are currently registered.
zap.S().Debugw("saw server status change event", zap.String("server", s.Uuid), zap.String("status", s.State))
s.Events().Publish(StatusEvent, s.State)
// Release the lock as it is no longer needed for the following actions.
s.Unlock()
// 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 := saveServerStates(); err != nil {
zap.S().Warnw("failed to write server states to disk", zap.Error(err))
}
}()
// 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.GetState() == 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
}
// Returns the current state of the server in a race-safe manner.
func (s *Server) GetState() string {
s.RLock()
defer s.RUnlock()
return s.State
}
// Determines if the server state is running or not. This is different than the
// environment state, it is simply the tracked state from this daemon instance, and
// not the response from Docker.
func (s *Server) IsRunning() bool {
return s.GetState() == ProcessRunningState || s.GetState() == ProcessStartingState
}

View File

@ -102,7 +102,7 @@ func (s *Server) runBackgroundActions() {
// Check if the server is now suspended, and if so and the process is not terminated
// yet, do it immediately.
go func(server *Server) {
if server.Suspended && server.State != ProcessOfflineState {
if server.Suspended && server.GetState() != ProcessOfflineState {
zap.S().Infow("server suspended with running process state, terminating now", zap.String("server", server.Uuid))
/*if err := server.Environment.Terminate(os.Kill); err != nil {