Fix crash when deleting an actor

The Cleanup method is called on actors when they exit

actor.go

@@ -127,3 +127,10 @@ type RunShutdownable interface {
 	// should be shut down.
 	Shutdown(ctx context.Context) error
 }
+
+// Cleanupable is any object that must be cleaned up after it has stopped for
+// good. Actors which implement this interface will be cleaned up after they
+// are deleted from the environment.
+type Cleanupable interface {
+	Cleanup(ctx context.Context) error
+}

cron.go

@@ -7,6 +7,7 @@ import "context"
 // tasks on time intervals.
 type cron struct {
 	trimFunc func() bool
+	fastTiming bool
 	timing struct {
 		trimInterval time.Duration
 	}
@@ -28,6 +29,9 @@ func (this *cron) Configure (config Config) error {
 		}
 		this.timing.trimInterval = value
 	}
+	if this.fastTiming {
+		this.timing.trimInterval = time.Second * 10
+	}
 	return nil
 }
 

environment.go

@@ -19,6 +19,7 @@ const defaultRestartInitialInterval  =  8 * time.Second
 const defaultRestartInitialIncrement =  8 * time.Second
 const defaultRestartInitialMaximum   =  1 * time.Hour
 const defaultResetTimeout            =  8 * time.Minute
+const defaultCleanupTimeout          =  1 * time.Minute
 const defaultTrimInterval            =  1 * time.Minute
 const defaultTrimTimeout             =  1 * time.Minute
 const defaultShutdownTimeout         =  8 * time.Minute
@@ -33,6 +34,7 @@ type environment struct {
 	done        context.CancelCauseFunc
 	group       sync.WaitGroup
 	conf        MutableConfig
+	cron        *cron
 
 	// flags stores information from built-in flags.
 	flags struct {
@@ -42,6 +44,8 @@ type environment struct {
 		configFile   string
 		verbose      bool
 		crash        bool
+		crashOnError bool
+		fastTiming   bool
 	}
 
 	// running stores whether the environment is currently running.
@@ -59,6 +63,7 @@ type environment struct {
 		restartIntervalIncrement atomicDuration
 		restartIntervalMaximum   atomicDuration
 		resetTimeout             atomicDuration
+		cleanupTimeout           atomicDuration
 		trimTimeout              atomicDuration
 		shutdownTimeout          atomicDuration
 	}
@@ -79,10 +84,11 @@ func (this *environment) Run(name, description string, actors ...Actor) {
 	this.name = name
 	this.description = description
 	this.actors = usync.NewRWMonitor(&actorSets { })
-	this.addToSets(actors...)
-	this.addToSets(&cron {
+	this.cron = &cron {
 		trimFunc: this.phase70_5Trimming,
-	})
+	}
+	this.addToSets(actors...)
+	this.addToSets(this.cron)
 	
 	if !this.phase10FlagParsing()              { os.Exit(2) }
 	if !this.phase13PidFileCreation()          { os.Exit(1) }
@@ -111,7 +117,12 @@ func (this *environment) Add(ctx context.Context, actors ...Actor) error {
 		}
 	}
 	err := this.initializeActors(ctx, initializable...)
-	if err != nil { return err }
+	if err != nil {
+		if this.flags.crashOnError {
+			panic(fmt.Sprint(err))
+		}
+		return err
+	}
 	for _, actor := range actors {
 		if actor, ok := actor.(Configurable); ok {
 			err := actor.Configure(this.conf)
@@ -140,7 +151,9 @@ func (this *environment) Del(ctx context.Context, actors ...Actor) error {
 		if info.stopped != nil {
 			channels = append(channels, info.stopped)
 		}
-		info.done()
+		if info.done != nil {
+			info.done()
+		}
 	}
 	for _, channel := range channels {
 		if channel == nil { continue }
@@ -229,11 +242,30 @@ func (this *environment) start(actor Actor) {
 // counter will be decremented. note that this function will never increment the
 // wait group counter, so start should usually be used instead.
 func (this *environment) run(actor Actor) {
+	typ := actor.Type()
+	
 	// clean up when done
-	defer this.group.Done()
+	defer func() {
+		this.group.Done()
+		this.delFromSets(actor)
+		if actor, ok := actor.(Cleanupable); ok {
+			ctx, done := context.WithTimeout(
+				context.Background(),
+				defaul(
+					this.timing.cleanupTimeout.Load(),
+					defaultCleanupTimeout))
+			defer done()
+			err := actor.Cleanup(ctx)
+			if err != nil {
+				log.Printf("XXX [%s] failed to cleanup: %v", typ, err)
+				if this.flags.crashOnError {
+					panic(fmt.Sprint(err))
+				}
+			}
+		}
+	}()
 
 	// logging
-	typ := actor.Type()
 	if this.Verb() { log.Printf("(i) [%s] running", typ) }
 	var stopErr error
 	var exited  bool
@@ -246,6 +278,9 @@ func (this *environment) run(actor Actor) {
 			}
 		} else {
 			log.Printf("!!! [%s] stopped with error: %v", typ, stopErr)
+			if this.flags.crashOnError {
+				panic(fmt.Sprint(stopErr))
+			}
 		}
 	}()
 
@@ -311,11 +346,17 @@ func (this *environment) runRunnable(ctx context.Context, actor Runnable) (stopE
 		} else {
 			// failure
 			log.Printf("XXX [%s] failed: %v", typ, err)
+			if this.flags.crashOnError {
+				panic(fmt.Sprint(err))
+			}
 		}
 
 		// restart logic
 		if time.Since(lastStart) < restartThreshold {
 			log.Printf("!!! [%s] failed too soon, restarting in %v", typ, restartInterval)
+			if this.flags.crashOnError {
+				panic("failed too soon")
+			}
 			timer := time.NewTimer(restartInterval)
 			select {
 			case <- timer.C:
@@ -341,6 +382,9 @@ func (this *environment) runRunnable(ctx context.Context, actor Runnable) (stopE
 				err := actor.Reset(ctx)
 				if err != nil {
 					log.Printf("XXX [%s] failed to reset", typ)
+					if this.flags.crashOnError {
+						panic("failed to reset")
+					}
 				}
 			}()
 			if this.Verb() { log.Printf(".// [%s] reset", typ) }
@@ -400,6 +444,7 @@ func (this *environment) applyConfig() error {
 		}
 		return nil
 	}
+	// TODO: trim interval
 	err := parseDuration("init-timeout", &this.timing.initTimeout)
 	if err != nil { return err }
 	err = parseDuration("restart-threshold", &this.timing.restartThreshold)
@@ -412,10 +457,17 @@ func (this *environment) applyConfig() error {
 	if err != nil { return err }
 	err = parseDuration("reset-timeout", &this.timing.resetTimeout)
 	if err != nil { return err }
+	err = parseDuration("cleanup-timeout", &this.timing.cleanupTimeout)
+	if err != nil { return err }
 	err = parseDuration("trim-timeout", &this.timing.trimTimeout)
 	if err != nil { return err }
 	err = parseDuration("shutdown-timeout", &this.timing.shutdownTimeout)
 	if err != nil { return err }
+
+	if this.flags.fastTiming {
+		this.timing.shutdownTimeout.Store(time.Second * 10)
+	}
+	
 	return nil
 }
 

phases.go

@@ -19,13 +19,15 @@ func (this *environment) phase10FlagParsing() bool {
 		name: this.name,
 		description: this.description,
 	}
-	flagHelp          := set.Flag('h', "help",          "Display usage information and exit", nil)
-	flagPidFile       := set.Flag('p', "pid-file",      "Write the PID to the specified file", cli.ValString)
-	flagUser          := set.Flag('u', "user",          "The user:group to run as", cli.ValString)
-	flagLogDirectory  := set.Flag('l', "log-directory", "Write logs to the specified directory", cli.ValString)
-	flagConfigFile    := set.Flag('c', "config-file",   "Use this configuration file", cli.ValString)
-	flagVerbose       := set.Flag('v', "verbose",       "Enable verbose output/logging", nil)
-	flagCrash         := set.Flag(0,   "crash",         "Crash when an actor panics", nil)
+	flagHelp          := set.Flag('h', "help",           "Display usage information and exit", nil)
+	flagPidFile       := set.Flag('p', "pid-file",       "Write the PID to the specified file", cli.ValString)
+	flagUser          := set.Flag('u', "user",           "The user:group to run as", cli.ValString)
+	flagLogDirectory  := set.Flag('l', "log-directory",  "Write logs to the specified directory", cli.ValString)
+	flagConfigFile    := set.Flag('c', "config-file",    "Use this configuration file", cli.ValString)
+	flagVerbose       := set.Flag('v', "verbose",        "(debug) Enable verbose output/logging", nil)
+	flagCrash         := set.Flag(0,   "crash",          "(debug) Crash when an actor panics", nil)
+	flagCrashOnError  := set.Flag(0,   "crash-on-error", "(debug) Crash when an actor experiences any error", nil)
+	flagFastTiming    := set.Flag(0,   "fast-timing",    "(debug) Make timed things happen faster/more often", nil)
 
 	// ask actors to add flags
 	actors, done := this.actors.RBorrow()
@@ -65,6 +67,14 @@ func (this *environment) phase10FlagParsing() bool {
 	if _, ok := flagCrash.First(); ok {
 		this.flags.crash = true
 	}
+	if _, ok := flagCrashOnError.First(); ok {
+		this.flags.crash = true
+		this.flags.crashOnError = true
+	}
+	if _, ok := flagFastTiming.First(); ok {
+		this.flags.fastTiming = true
+		this.cron.fastTiming = true
+	}
 	return true
 }
 
@@ -259,6 +269,9 @@ func (this *environment) phase70_5Trimming() bool {
 	}()
 	if err := this.trimActors(this.ctx, trimmable...); err != nil {
 		log.Println(".// (70.5) failed to trim:", err)
+		if this.flags.crashOnError {
+			panic(err)
+		}
 		return false
 	}
 	if this.Verb() { log.Println(".// (70.5) trimmed") }
@@ -266,6 +279,7 @@ func (this *environment) phase70_5Trimming() bool {
 }
 
 func (this *environment) phase80Shutdown() bool {
+	logActors(All())
 	ctx, done := context.WithTimeout(
 		context.Background(),
 		defaul(this.timing.shutdownTimeout.Load(), defaultShutdownTimeout))
@@ -274,7 +288,7 @@ func (this *environment) phase80Shutdown() bool {
 		<- ctx.Done()
 		if errors.Is(context.Cause(ctx), context.DeadlineExceeded) {
 			log.Println("XXX (80) shutdown timeout expired, performing emergency halt")
-			if Verb() {
+			if Verb() || this.flags.crashOnError {
 				dumpBuffer := make([]byte, 8192)
 				runtime.Stack(dumpBuffer, true)
 				log.Printf("XXX (80) stack trace of all goroutines:\n%s", dumpBuffer)