hnakra/routines/routines.go

// Package routines provides utilities for managing long-running goroutines.
package routines

import "io"
import "fmt"
import "log"
import "time"
import "sync"

// Routine is a long-running function that does not return until it is finished.
// An error is returned if the routine exited due to an error.
type Routine func () error

// Manager is a system capable of managing multiple routines, and restarting
// them if they fail.
type Manager struct {
	// Routines specifies a list of routines to manage. These are started
	// when Run() is called.
	Routines []Routine

	// RestartDeadline specifies the amount of time a routine has to be
	// running before failing to be restarted. This is to prevent routines
	// that immediately fail from just being restarted over and over again.
	RestartDeadline time.Duration

	// Logger, if non-nil, is where log messages will be written to. If it
	// is nil, messages will be written to the standard logger. To disable
	// logging altogether, this can be set to io.Discard.
	Logger io.Writer
}

// Run spawns all routines in the Routines slice. If a routine exits with an
// error and it was running for longer than RestartDeadline, it is restarted.
// Run returns  only when all routines have exited.
func (manager *Manager) Run () error {
	var waitGroup sync.WaitGroup
	var errExit   error
	
	for _, routine := range manager.Routines {
		if routine != nil {
			waitGroup.Add(1)
			go manager.runRoutine(routine, &waitGroup, &errExit)
		}
	}
	
	waitGroup.Wait()
	return errExit
}

// Append adds one or more routines to the Routines slice. This has no effect if
// the manager is already running.
func (manager *Manager) Append (routines ...Routine) {
	manager.Routines = append(manager.Routines, routines...)
}

func (manager *Manager) log (message ...any) {
	if manager.Logger == nil {
		log.Println(message...)
	} else {
		fmt.Fprintln(manager.Logger, message...)
	}
}

func (manager *Manager) runRoutine (routine Routine, group *sync.WaitGroup, errExit *error) {
	defer group.Done()

	var err error
	for {
		// TODO: recover from panics
		lastStart := time.Now()
		err = routine()

		if err == nil {
			manager.log("(i) routine exited")
			break
		} else {
			manager.log("XXX routine failed:", err)
		}
		
		if time.Since(lastStart) < manager.RestartDeadline {
			manager.log("!!! not restarting routine, failed too soon")
			break
		} else {
			manager.log("(i) routine is being restarted")
		}
	}
	
	if err != nil {
		*errExit = err
	}
}