Support sticky sessions under SWARM Mode.

SWARM Mode has it's own built in Load balancer, so if we want to leverage sticky sessions, or if we would just prefer to bypass it and go directly to the containers (aka tasks), via --label traefik.backend.disable.swarm.loadbalancer=true then we need to let Traefik know about the underlying tasks and register them as services within it's backend.
traefik · Jan 13, 2017 · 59cec1c · 59cec1c
1 parent dd85cbc
commit 59cec1c
Show file tree

Hide file tree

Showing 3 changed files with 156 additions and 5 deletions.
diff --git a/docs/toml.md b/docs/toml.md
@@ -795,6 +795,7 @@ Labels can be used on containers to override default behaviour:
 - `traefik.backend.maxconn.extractorfunc=client.ip`: set the function to be used against the request to determine what to limit maximum connections to the backend by. Must be used in conjunction with the above label to take effect.
 - `traefik.backend.loadbalancer.method=drr`: override the default `wrr` load balancer algorithm
 - `traefik.backend.loadbalancer.sticky=true`: enable backend sticky sessions
+- `traefik.backend.loadbalancer.swarm=true `: use Swarm's inbuilt load balncer (only relevant under Swarm Mode).
 - `traefik.backend.circuitbreaker.expression=NetworkErrorRatio() > 0.5`: create a [circuit breaker](/basics/#backends) to be used against the backend
 - `traefik.port=80`: register this port. Useful when the container exposes multiples ports.
 - `traefik.protocol=https`: override the default `http` protocol

diff --git a/docs/user-guide/swarm-mode.md b/docs/user-guide/swarm-mode.md
@@ -47,6 +47,7 @@ docker-machine ssh worker1 "docker swarm join \
 	--listen-addr $(docker-machine ip worker1) \
 	--advertise-addr $(docker-machine ip worker1) \
 	$(docker-machine ip manager)"
+
 docker-machine ssh worker2 "docker swarm join \
 	--token=${worker_token} \
 	--listen-addr $(docker-machine ip worker2) \
@@ -119,14 +120,22 @@ docker-machine ssh manager "docker service create \
 	--name whoami0 \
 	--label traefik.port=80 \
 	--network traefik-net \
+	--label traefik.frontend.rule=Host:whoami0.traefik \
+	--label traefik.backend=whoami0 \
 	emilevauge/whoami"
+
 docker-machine ssh manager "docker service create \
 	--name whoami1 \
 	--label traefik.port=80 \
 	--network traefik-net \
+	--label traefik.frontend.rule=Host:whoam1.traefik \
+	--label traefik.backend=whoami1 \
+	--label traefik.backend.loadbalancer.sticky=true \
 	emilevauge/whoami"
 ```
 
+Note that we set whoami1 to use sticky sessions (`--label traefik.backend.loadbalancer.sticky=true`).  We'll demonstrate that later.
+
 Check that everything is scheduled and started:
 
 ```sh
@@ -218,6 +227,84 @@ X-Forwarded-Proto: http
 X-Forwarded-Server: 8fbc39271b4c
 ```
 
+## Scale both services
+
+```sh
+docker-machine ssh manager "docker service scale whoami0=5"
+
+docker-machine ssh manager "docker service scale whoami1=5"
+```
+
+
+Check that we now have 5 replicas of each `whoami` service:
+
+```sh
+docker-machine ssh manager "docker service ls"
+ID            NAME     REPLICAS  IMAGE              COMMAND
+ab046gpaqtln  whoami0  5/5       emilevauge/whoami
+cgfg5ifzrpgm  whoami1  5/5       emilevauge/whoami
+dtpl249tfghc  traefik  1/1       traefik            --docker --docker.swarmmode --docker.domain=traefik --docker.watch --web
+```
+## Access to your whoami0 through Træfɪk multiple times.
+
+Repeat the following command multiple times and note that the Hostname changes each time as Traefik load balances each request against the 5 tasks.
+```sh
+curl -H Host:whoami0.traefik http://$(docker-machine ip manager)
+Hostname: 8147a7746e7a
+IP: 127.0.0.1
+IP: ::1
+IP: 10.0.9.3
+IP: fe80::42:aff:fe00:903
+IP: 172.18.0.3
+IP: fe80::42:acff:fe12:3
+GET / HTTP/1.1
+Host: 10.0.9.3:80
+User-Agent: curl/7.35.0
+Accept: */*
+Accept-Encoding: gzip
+X-Forwarded-For: 192.168.99.1
+X-Forwarded-Host: 10.0.9.3:80
+X-Forwarded-Proto: http
+X-Forwarded-Server: 8fbc39271b4c
+```
+
+Do the same against whoami1.  
+```sh
+curl -H Host:whoami1.traefik http://$(docker-machine ip manager)
+Hostname: ba2c21488299
+IP: 127.0.0.1
+IP: ::1
+IP: 10.0.9.4
+IP: fe80::42:aff:fe00:904
+IP: 172.18.0.2
+IP: fe80::42:acff:fe12:2
+GET / HTTP/1.1
+Host: 10.0.9.4:80
+User-Agent: curl/7.35.0
+Accept: */*
+Accept-Encoding: gzip
+X-Forwarded-For: 192.168.99.1
+X-Forwarded-Host: 10.0.9.4:80
+X-Forwarded-Proto: http
+X-Forwarded-Server: 8fbc39271b4c
+```
+Wait, I thought we added the sticky flag to whoami1?  Traefik relies on a cookie to maintain stickyness so you'll need to test this with a browser.
+
+First you need to add whoami1.traefik to your hosts file:
+```ssh
+if [ -n "$(grep whoami1.traefik /etc/hosts)" ];  
+then 
+echo "whoami1.traefik already exists (make sure the ip is current)"; 
+else 
+sudo -- sh -c -e "echo '$(docker-machine ip manager)\twhoami1.traefik' 
+>> /etc/hosts"; 
+fi
+```
+
+Now open your browser and go to http://whoami1.traefik/
+
+You will now see that stickyness is maintained.
+
 ![](http://i.giphy.com/ujUdrdpX7Ok5W.gif)
 
 
diff --git a/provider/docker.go b/provider/docker.go
@@ -131,7 +131,7 @@ func (provider *Docker) Provide(configurationChan chan<- types.ConfigMessage, po
 			log.Debugf("Docker connection established with docker %s (API %s)", version.Version, version.APIVersion)
 			var dockerDataList []dockerData
 			if provider.SwarmMode {
-				dockerDataList, err = listServices(ctx, dockerClient)
+				dockerDataList, err = provider.listServices(ctx, dockerClient)
 				if err != nil {
 					log.Errorf("Failed to list services for docker swarm mode, error %s", err)
 					return err
@@ -158,7 +158,7 @@ func (provider *Docker) Provide(configurationChan chan<- types.ConfigMessage, po
 						for {
 							select {
 							case <-ticker.C:
-								services, err := listServices(ctx, dockerClient)
+								services, err := provider.listServices(ctx, dockerClient)
 								if err != nil {
 									log.Errorf("Failed to list services for docker, error %s", err)
 									return
@@ -258,8 +258,8 @@ func (provider *Docker) loadDockerConfig(containersInspected []dockerData) *type
 		"getMaxConnAmount":            provider.getMaxConnAmount,
 		"getMaxConnExtractorFunc":     provider.getMaxConnExtractorFunc,
 		"getSticky":                   provider.getSticky,
+		"getIsBackendLBSwarm":         provider.getIsBackendLBSwarm,
 	}
-
 	// filter containers
 	filteredContainers := fun.Filter(func(container dockerData) bool {
 		return provider.containerFilter(container)
@@ -463,6 +463,13 @@ func (provider *Docker) getSticky(container dockerData) string {
 	return "false"
 }
 
+func (provider *Docker) getIsBackendLBSwarm(container dockerData) string {
+	if _, err := getLabel(container, "traefik.backend.loadbalancer.swarm"); err == nil {
+		return "true"
+	}
+	return "false"
+}
+
 func (provider *Docker) getDomain(container dockerData) string {
 	if label, err := getLabel(container, "traefik.domain"); err == nil {
 		return label
@@ -593,7 +600,7 @@ func (provider *Docker) getSubDomain(name string) string {
 	return strings.Replace(strings.TrimPrefix(name, "/"), "/", "-", -1)
 }
 
-func listServices(ctx context.Context, dockerClient client.APIClient) ([]dockerData, error) {
+func (provider *Docker) listServices(ctx context.Context, dockerClient client.APIClient) ([]dockerData, error) {
 	serviceList, err := dockerClient.ServiceList(ctx, dockertypes.ServiceListOptions{})
 	if err != nil {
 		return []dockerData{}, err
@@ -614,11 +621,21 @@ func listServices(ctx context.Context, dockerClient client.APIClient) ([]dockerD
 	}
 
 	var dockerDataList []dockerData
+	var dockerDataListTasks []dockerData
 
 	for _, service := range serviceList {
 		dockerData := parseService(service, networkMap)
+		useSwarmLB, _ := strconv.ParseBool(provider.getIsBackendLBSwarm(dockerData))
+
+		if useSwarmLB {
+			dockerDataList = append(dockerDataList, dockerData)
+		} else {
+			dockerDataListTasks, err = listTasks(ctx, dockerClient, service.ID, dockerData, networkMap)
 
-		dockerDataList = append(dockerDataList, dockerData)
+			for _, dockerDataTask := range dockerDataListTasks {
+				dockerDataList = append(dockerDataList, dockerDataTask)
+			}
+		}
 	}
 	return dockerDataList, err
 
@@ -656,3 +673,49 @@ func parseService(service swarmtypes.Service, networkMap map[string]*dockertypes
 	}
 	return dockerData
 }
+
+func listTasks(ctx context.Context, dockerClient client.APIClient, serviceID string,
+	serviceDockerData dockerData, networkMap map[string]*dockertypes.NetworkResource) ([]dockerData, error) {
+	taskList, err := dockerClient.TaskList(ctx, dockertypes.TaskListOptions{})
+	if err != nil {
+		return []dockerData{}, err
+	}
+	var dockerDataList []dockerData
+
+	for _, task := range taskList {
+		if task.ServiceID == serviceID {
+			dockerData := parseTasks(task, serviceDockerData, networkMap)
+			dockerDataList = append(dockerDataList, dockerData)
+		}
+	}
+	return dockerDataList, err
+}
+
+func parseTasks(task swarmtypes.Task, serviceDockerData dockerData, networkMap map[string]*dockertypes.NetworkResource) dockerData {
+	dockerData := dockerData{
+		Name:            serviceDockerData.Name + "." + strconv.Itoa(task.Slot),
+		Labels:          serviceDockerData.Labels,
+		NetworkSettings: networkSettings{},
+	}
+
+	if task.NetworksAttachments != nil {
+		dockerData.NetworkSettings.Networks = make(map[string]*networkData)
+		for _, virtualIP := range task.NetworksAttachments {
+			networkService := networkMap[virtualIP.Network.ID]
+			if networkService != nil {
+				// Not sure about this next loop - when would a task have multiple IP's for the same network?
+				for _, addr := range virtualIP.Addresses {
+					ip, _, _ := net.ParseCIDR(addr)
+					network := &networkData{
+						ID:   virtualIP.Network.ID,
+						Name: networkService.Name,
+						Addr: ip.String(),
+					}
+					dockerData.NetworkSettings.Networks[network.Name] = network
+				}
+			}
+		}
+	}
+
+	return dockerData
+}