HODA/internal/discovery/discovery.go

// Package discovery keeps the store in step with this host's labeled containers.
package discovery

import (
	"context"
	"log/slog"
	"strings"
	"time"

	"git.mariosem.es/msemes/hoda/internal/apps"
	"git.mariosem.es/msemes/hoda/internal/docker"
)

const (
	settle     = 300 * time.Millisecond // lets a burst of events become one re-list
	minBackoff = time.Second
	maxBackoff = 30 * time.Second
)

// IconQueue is told about the current apps so their icons get resolved.
type IconQueue interface {
	Ensure([]apps.App)
}

// Discovery lists this host's labeled containers into the store.
type Discovery struct {
	docker *docker.Client
	store  *apps.Store
	icons  IconQueue
	prefix string
	host   string
	resync time.Duration
	log    *slog.Logger

	reachable bool
	count     int
	logged    map[string]bool // container + problem already logged
}

// New returns a Discovery for the Docker host behind dc, named host on the dashboard.
func New(dc *docker.Client, store *apps.Store, icons IconQueue, prefix, host string, resync time.Duration, log *slog.Logger) *Discovery {
	return &Discovery{
		docker:    dc,
		store:     store,
		icons:     icons,
		prefix:    prefix,
		host:      host,
		resync:    resync,
		log:       log,
		reachable: true,
		count:     -1,
		logged:    map[string]bool{},
	}
}

// Run lists containers now, again after every relevant Docker event and every
// resync interval, until ctx ends.
func (d *Discovery) Run(ctx context.Context) {
	changed := make(chan struct{}, 1)
	go d.watch(ctx, func() {
		select {
		case changed <- struct{}{}:
		default:
		}
	})
	ticker := time.NewTicker(d.resync)
	defer ticker.Stop()
	d.sync(ctx)
	for {
		select {
		case <-ctx.Done():
			return
		case <-ticker.C:
		case <-changed:
			select {
			case <-ctx.Done():
				return
			case <-time.After(settle):
			}
			select {
			case <-changed:
			default:
			}
		}
		d.sync(ctx)
	}
}

func (d *Discovery) sync(ctx context.Context) {
	listCtx, cancel := context.WithTimeout(ctx, 10*time.Second)
	defer cancel()
	// Every container, not only opted-in ones: labels without enable are a problem to report.
	containers, err := d.docker.ListContainers(listCtx, "")
	if err != nil {
		if ctx.Err() != nil {
			return
		}
		d.store.SetError(apps.LocalSource, d.host)
		if d.reachable {
			d.log.Warn("can't reach Docker, keeping the last known apps", "err", err)
		}
		d.reachable = false
		return
	}
	if !d.reachable {
		d.log.Info("Docker is reachable again")
	}
	d.reachable = true

	list := []apps.App{}
	issues := []apps.Issue{}
	current := map[string]bool{}
	for _, c := range containers {
		app, show, problems := apps.Parse(c, d.prefix, d.host)
		for _, p := range problems {
			issues = append(issues, apps.Issue{Host: d.host, Container: c.Name(), Message: p})
			key := c.Name() + "\x00" + p
			current[key] = true
			if !d.logged[key] {
				d.log.Warn("label problem", "container", c.Name(), "problem", p)
				d.logged[key] = true
			}
		}
		if show {
			list = append(list, app)
		}
	}
	for key := range d.logged {
		if !current[key] {
			delete(d.logged, key)
		}
	}
	if len(list) != d.count {
		d.log.Info("apps on this host", "count", len(list))
		d.count = len(list)
	}
	d.store.SetApps(apps.LocalSource, d.host, list, issues)
	d.icons.Ensure(list)
}

// watch follows Docker events and reconnects with backoff when the stream breaks.
func (d *Discovery) watch(ctx context.Context, changed func()) {
	backoff := minBackoff
	for {
		started := time.Now()
		err := d.docker.Events(ctx, d.prefix+".enable", func(ev docker.Event) {
			if relevant(ev.Action) {
				changed()
			}
		})
		if ctx.Err() != nil {
			return
		}
		if time.Since(started) > time.Minute {
			backoff = minBackoff
		}
		d.log.Debug("Docker event stream ended, reconnecting", "err", err, "in", backoff)
		select {
		case <-ctx.Done():
			return
		case <-time.After(backoff):
		}
		backoff = min(backoff*2, maxBackoff)
		changed() // events may have been missed while disconnected
	}
}

// relevant drops the event chatter of healthchecks and `docker exec`, which
// never changes what the dashboard shows.
func relevant(action string) bool {
	switch {
	case strings.HasPrefix(action, "exec_"), strings.HasPrefix(action, "archive-path"):
		return false
	case action == "attach", action == "detach", action == "resize", action == "top",
		action == "commit", action == "copy", action == "export", action == "extract-to-dir":
		return false
	}
	return true
}