av/cmd/vidforward/notifier.go

/*
DESCRIPTION
  notifier.go provides a tool for notifying a systemd watchdog under healthy
  operation of the vidforward service.

AUTHORS
  Saxon A. Nelson-Milton <saxon@ausocean.org>

LICENSE
  Copyright (C) 2022 the Australian Ocean Lab (AusOcean)

  It is free software: you can redistribute it and/or modify them
  under the terms of the GNU General Public License as published by the
  Free Software Foundation, either version 3 of the License, or (at your
  option) any later version.

  It is distributed in the hope that it will be useful, but WITHOUT
  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
  for more details.

  You should have received a copy of the GNU General Public License
  along with revid in gpl.txt. If not, see http://www.gnu.org/licenses.
*/

package main

import (
	"log"
	"os"
	"os/signal"
	"sync"
	"syscall"
	"time"

	"bitbucket.org/ausocean/utils/logging"
	"github.com/coreos/go-systemd/daemon"
)

// By default we assume we should be notifying a systemd watchdog. This can be
// toggled off by using the nowatchdog build tag (see nowatchdog.go file).
var notifyWatchdog = true

// watchdogNotifier keeps track of the watchdog interval from the external
// sysd service settings, the currently active request handlers and a curId
// field that is is incremented to generate new handler ids for storage.
type watchdogNotifier struct {
	watchdogInterval time.Duration
	activeHandlers   map[int]handlerInfo
	curId            int
	termCallback     func()
	log              logging.Logger
	mu               sync.Mutex
	haveRun          bool
}

// handlerInfo keeps track of a handlers name (for any logging purposes) and
// time at which the handler was invoked, which is later used to calculate time
// active and therefore heatlh.
type handlerInfo struct {
	name string
	time time.Time
}

// newWatchdogNotifier creates a new watchdogNotifier with the provided logger
// and termination callback that is called if a SIGINT or SIGTERM signal is
// received. Recommended use of this is an attempted state save.
func newWatchdogNotifier(l logging.Logger, termCallback func()) (*watchdogNotifier, error) {
	interval := 1 * time.Minute

	return &watchdogNotifier{
		activeHandlers:   make(map[int]handlerInfo),
		watchdogInterval: interval,
		log:              l,
		termCallback:     termCallback,
	}, nil
}

// notify is to be called as a routine. This is responsible for checking if the
// handlers are healthy and then notifying the watchdog if so, otherwise we
// wait, continue and check again until they are. If the handlers take too long to
// become healthy, we risk exceeding the watchdog interval causing a process restart.
// notify also starts a routine to monitor for any SIGINT or SIGTERM, upon which
// a callback that's provided at initialisation is called.
func (n *watchdogNotifier) notify() {
	notifyTicker := time.NewTicker(n.watchdogInterval / 2.0)

	go func() {
		sigs := make(chan os.Signal, 1)
		signal.Notify(sigs, syscall.SIGINT, syscall.SIGTERM)
		sig := <-sigs
		n.log.Warning("received termination signal, calling termination callback", "signal", sig.String())
		n.termCallback()
	}()

	for {
		if n.handlersUnhealthy() {
			const unhealthyHandlerWait = 1 * time.Second
			time.Sleep(unhealthyHandlerWait)
			continue
		}

		<-notifyTicker.C

		if !notifyWatchdog {
			continue
		}

		if !n.haveRun {
			n.haveRun = true

			const clearEnvVars = false
			ok, err := daemon.SdNotify(clearEnvVars, daemon.SdNotifyReady)
			if err != nil {
				n.log.Fatal("unexpected watchog notify read error", "error", err)
			}

			if !ok {
				n.log.Fatal("watchdog notification not supported")
			}

			n.watchdogInterval, err = daemon.SdWatchdogEnabled(clearEnvVars)
			if err != nil {
				n.log.Fatal("unexpected watchdog error", "error", err)
			}

			if n.watchdogInterval == 0 {
				n.log.Fatal("Watchdog not enabled or this is the wrong PID")
			}

		}

		// If this fails for any reason it indicates a systemd service configuration
		// issue, and therefore programmer error, so do fatal log to cause crash.
		supported, err := daemon.SdNotify(false, daemon.SdNotifyWatchdog)
		if err != nil {
			n.log.Fatal("error from systemd watchdog notify", "error", err)
		}

		if !supported {
			n.log.Fatal("watchdog notification not supported")
		}
	}
}

// handlersUnhealthy returns true if it is detected that any handlers are unhealthy,
// that is, if they have been handling for longer than the unhealthyHandleDuration.
func (n *watchdogNotifier) handlersUnhealthy() bool {
	n.mu.Lock()
	defer n.mu.Unlock()
	for _, info := range n.activeHandlers {
		const unhealthyHandleDuration = 30 * time.Second
		if time.Now().Sub(info.time) > unhealthyHandleDuration {
			n.log.Warning("handler unhealthy", "name", info.name)
			return true
		}
	}
	return false
}

// handlerInvoked is to be called at the start of a request handler to indicate
// that handling has begun. The name and start time of the handler is recorded
// in the active handlers map with a unique ID as the key. A function is returned
// that must be called at exit of the handler to indicate that handling has
// finished. It is recommended this be done using a defer statement immediately
// after receiveing it.
func (n *watchdogNotifier) handlerInvoked(name string) func() {
	n.mu.Lock()
	defer n.mu.Unlock()

	id := n.curId
	n.curId++
	n.activeHandlers[id] = handlerInfo{time: time.Now(), name: name}

	return func() {
		n.mu.Lock()
		defer n.mu.Unlock()

		if _, ok := n.activeHandlers[id]; !ok {
			log.Fatal("handler id not in map", "name", name)
		}

		delete(n.activeHandlers, id)
	}
}