fix: container not registered after restart

bridge/bridge.go

@@ -11,7 +11,6 @@ import (
 	"strconv"
 	"strings"
 	"sync"
-	"time"
 
 	dockerapi "github.com/fsouza/go-dockerclient"
 )
@@ -20,12 +19,11 @@ var serviceIDPattern = regexp.MustCompile(`^(.+?):([a-zA-Z0-9][a-zA-Z0-9_.-]+):[
 
 type Bridge struct {
 	sync.Mutex
-	registry        RegistryAdapter
-	docker          *dockerapi.Client
-	services        map[string][]*Service
-	deadContainers  map[string]*DeadContainer
-	dyingContainers map[string]time.Time
-	config          Config
+	registry       RegistryAdapter
+	docker         *dockerapi.Client
+	services       map[string][]*Service
+	deadContainers map[string]*DeadContainer
+	config         Config
 }
 
 func New(docker *dockerapi.Client, adapterUri string, config Config) (*Bridge, error) {
@@ -40,12 +38,11 @@ func New(docker *dockerapi.Client, adapterUri string, config Config) (*Bridge, e
 
 	log.Println("Using", uri.Scheme, "adapter:", uri)
 	return &Bridge{
-		docker:          docker,
-		config:          config,
-		registry:        factory.New(uri),
-		services:        make(map[string][]*Service),
-		deadContainers:  make(map[string]*DeadContainer),
-		dyingContainers: make(map[string]time.Time),
+		docker:         docker,
+		config:         config,
+		registry:       factory.New(uri),
+		services:       make(map[string][]*Service),
+		deadContainers: make(map[string]*DeadContainer),
 	}, nil
 }
 
@@ -94,18 +91,19 @@ func (b *Bridge) Sync(quiet bool) {
 	b.Lock()
 	defer b.Unlock()
 
-	containers, err := b.docker.ListContainers(dockerapi.ListContainersOptions{})
+	filters := map[string][]string{"status": {"running", "restarting"}}
+	runningContainers, err := b.docker.ListContainers(dockerapi.ListContainersOptions{Filters: filters})
 	if err != nil && quiet {
-		log.Println("error listing containers, skipping sync")
+		log.Println("skipping sync, listing containers failed:", err)
 		return
 	} else if err != nil && !quiet {
 		log.Fatal(err)
 	}
 
-	log.Printf("Syncing services on %d containers", len(containers))
+	log.Printf("Syncing services on %d running containers", len(runningContainers))
 
 	// NOTE: This assumes reregistering will do the right thing, i.e. nothing..
-	for _, listing := range containers {
+	for _, listing := range runningContainers {
 		services := b.services[listing.ID]
 		if services == nil {
 			b.add(listing.ID, quiet)
@@ -124,23 +122,17 @@ func (b *Bridge) Sync(quiet bool) {
 	if b.config.Cleanup {
 		// Remove services if its corresponding container is not running
 		log.Println("Listing non-exited containers")
-		filters := map[string][]string{"status": {"created", "restarting", "running", "paused"}}
-		nonExitedContainers, err := b.docker.ListContainers(dockerapi.ListContainersOptions{Filters: filters})
-		if err != nil {
-			log.Println("error listing nonExitedContainers, skipping sync", err)
-			return
-		}
 		for listingId := range b.services {
 			found := false
-			for _, container := range nonExitedContainers {
+			for _, container := range runningContainers {
 				if listingId == container.ID {
 					found = true
 					break
 				}
 			}
 			// This is a container that does not exist
 			if !found {
-				log.Printf("stale: Removing service %s because it does not exist", listingId)
+				log.Printf("stale: Removing service %s because it's container is not running ", listingId)
 				go b.RemoveOnExit(listingId)
 			}
 		}
@@ -184,11 +176,6 @@ func (b *Bridge) Sync(quiet bool) {
 }
 
 func (b *Bridge) add(containerId string, quiet bool) {
-	if _, ok := b.dyingContainers[containerId]; ok {
-		log.Println("container, ", containerId[:12], ", is dying, ignoring")
-		return
-	}
-
 	if d := b.deadContainers[containerId]; d != nil {
 		b.services[containerId] = d.Services
 		delete(b.deadContainers, containerId)
@@ -389,7 +376,6 @@ func (b *Bridge) remove(containerId string, deregister bool) {
 		b.deadContainers[containerId] = &DeadContainer{b.config.RefreshTtl, b.services[containerId]}
 	}
 	delete(b.services, containerId)
-	b.markContainerAsDying(containerId)
 }
 
 // bit set on ExitCode if it represents an exit via a signal
@@ -423,18 +409,6 @@ func (b *Bridge) shouldRemove(containerId string) bool {
 	return false
 }
 
-func (b *Bridge) markContainerAsDying(containerId string) {
-	// cleanup after CleanupDyingTtl
-	for containerId, t := range b.dyingContainers {
-		if time.Since(t) >= time.Millisecond*time.Duration(b.config.CleanupDyingTtl) {
-			delete(b.dyingContainers, containerId)
-		}
-	}
-
-	// mark container as "dying"
-	b.dyingContainers[containerId] = time.Now()
-}
-
 var Hostname string
 
 func init() {

bridge/types.go

@@ -29,7 +29,6 @@ type Config struct {
 	RefreshInterval int
 	DeregisterCheck string
 	Cleanup         bool
-	CleanupDyingTtl int
 }
 
 type Service struct {

go.mod

@@ -4,6 +4,7 @@ go 1.23.2
 
 require (
 	github.com/coreos/go-etcd v2.0.0+incompatible
+	github.com/docker/docker v27.3.1+incompatible
 	github.com/fsouza/go-dockerclient v1.12.0
 	github.com/gliderlabs/pkg v0.0.0-20161206023812-36f28d47ec7a
 	github.com/hashicorp/consul/api v1.29.5
@@ -19,7 +20,6 @@ require (
 	github.com/armon/go-metrics v0.4.1 // indirect
 	github.com/containerd/log v0.1.0 // indirect
 	github.com/davecgh/go-spew v1.1.2-0.20180830191138-d8f796af33cc // indirect
-	github.com/docker/docker v27.3.1+incompatible // indirect
 	github.com/docker/go-connections v0.5.0 // indirect
 	github.com/docker/go-units v0.5.0 // indirect
 	github.com/fatih/color v1.16.0 // indirect

registrator.go

@@ -10,6 +10,7 @@ import (
 	"strings"
 	"time"
 
+	dockerevents "github.com/docker/docker/api/types/events"
 	dockerapi "github.com/fsouza/go-dockerclient"
 	"github.com/gliderlabs/pkg/usage"
 	"github.com/simplesurance/registrator/bridge"
@@ -33,7 +34,6 @@ var (
 	retryAttempts    = flag.Int("retry-attempts", 0, "Max retry attempts to establish a connection with the backend. Use -1 for infinite retries")
 	retryInterval    = flag.Int("retry-interval", 2000, "Interval (in millisecond) between retry-attempts.")
 	cleanup          = flag.Bool("cleanup", false, "Remove dangling services")
-	cleanupDyingTtl  = flag.Int("ttl-dying-cleanup", 60000, "TTL (in millisecond) for cleaning dying containers cache")
 )
 
 func assert(err error) {
@@ -109,7 +109,6 @@ func main() {
 		RefreshInterval: *refreshInterval,
 		DeregisterCheck: *deregister,
 		Cleanup:         *cleanup,
-		CleanupDyingTtl: *cleanupDyingTtl,
 	})
 
 	assert(err)
@@ -132,7 +131,7 @@ func main() {
 	}
 
 	// Start event listener before listing containers to avoid missing anything
-	events := make(chan *dockerapi.APIEvents)
+	events := make(chan *dockerapi.APIEvents, 128)
 	assert(docker.AddEventListener(events))
 	log.Println("Listening for Docker events ...")
 
@@ -174,12 +173,12 @@ func main() {
 
 	// Process Docker events
 	for msg := range events {
-		switch msg.Status {
-		case "start":
+		switch dockerevents.Action(msg.Action) {
+		case dockerevents.ActionStart, dockerevents.ActionUnPause:
 			go b.Add(msg.ID)
-		case "die":
+		case dockerevents.ActionDie:
 			go b.RemoveOnExit(msg.ID)
-		case "kill":
+		case dockerevents.ActionKill:
 			if *deregisterOnStop {
 				go b.RemoveOnExit(msg.ID)
 			}