This example shows how to build a simple, multi-tier web application using Kubernetes and Docker.
The example consists of:
- A web frontend
- A redis master (for storage and a replicated set of redis slaves)
The web front end interacts with the redis master via javascript redis API calls.
This example assumes that you have a basic understanding of kubernetes services and that you have forked the repository and turned up a Kubernetes cluster: This example requires a kubernetes cluster.
See the companion Setup Kubernetes for some quick notes on how to get started.
If you are running from source, replace commands such as "kubectl" below with calls to cluster/kubectl.sh.
Note: This redis-master is not highly available. Making it highly available would be a very interesting, but intricate exersize - redis doesn't actually support multi-master deployments at the time of this writing, so high availability would be a somewhat tricky thing implement, and might involve periodic serialization to disk, and so on.
Use (or just create) the file examples/guestbook/redis-master-controller.json which describes a single pod running a redis key-value server in a container:
Note that, although the redis server runs just with a single replica, we use replication controller to enforce that exactly one pod keeps running (e.g. in a event of node going down, the replication controller will ensure that the redis master gets restarted on a healthy node).
{
"id": "redis-master-controller",
"kind": "ReplicationController",
"apiVersion": "v1beta1",
"desiredState": {
"replicas": 1,
"replicaSelector": {"name": "redis-master"},
"podTemplate": {
"desiredState": {
"manifest": {
"version": "v1beta1",
"id": "redis-master",
"containers": [{
"name": "redis-master",
"image": "dockerfile/redis",
"cpu": 100,
"ports": [{
"containerPort": 6379, # containerPort: Where traffic to redis ultimately is routed to.
}]
}]
}
},
"labels": {
"name": "redis-master",
"app": "redis"
}
}
},
"labels": {
"name": "redis-master" # This label needed for when we start our redis-master service.
}
}Now, create the redis pod in your Kubernetes cluster by running:
$ kubectl create -f examples/guestbook/redis-master-controller.json
$ cluster/kubectl.sh get rc
CONTROLLER CONTAINER(S) IMAGE(S) SELECTOR REPLICAS
redis-master-controller redis-master dockerfile/redis name=redis-master 1Once that's up you can list the pods in the cluster, to verify that the master is running:
$ kubectl get podsYou'll see all kubernetes components, most importantly the redis master pod. It will also display the machine that the pod is running on once it gets placed (may take up to thirty seconds):
POD IP CONTAINER(S) IMAGE(S) HOST LABELS STATUS
redis-master-controller-gb50a 10.244.3.7 redis-master dockerfile/redis kubernetes-minion-7agi.c.hazel-mote-834.internal/104.154.54.203 app=redis,name=redis-master RunningIf you ssh to that machine, you can run docker ps to see the actual pod:
me@workstation$ gcloud compute ssh kubernetes-minion-7agi
me@kubernetes-minion-7agi:~$ sudo docker ps
CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
0ffef9649265 dockerfile/redis:latest "redis-server /etc/r About a minute ago Up About a minute k8s_redis-master.767aef46_redis-master-controller-gb50a.default.api_4530d7b3-ae5d-11e4-bf77-42010af0d719_579ee964 (Note that initial docker pull may take a few minutes, depending on network conditions. You can monitor the status of this by running journalctl -f -u docker to check when the image is being downloaded. Of course, you can also run journalctl -f -u kubelet to see what state the kubelet is in as well during this time.
A Kubernetes 'service' is a named load balancer that proxies traffic to one or more containers. This is done using the labels metadata which we defined in the redis-master pod above. As mentioned, in redis there is only one master, but we nevertheless still want to create a service for it. Why? Because it gives us a deterministic way to route to the single master using an elastic IP.
The services in a Kubernetes cluster are discoverable inside other containers via environment variables.
Services find the containers to load balance based on pod labels.
The pod that you created in Step One has the label name=redis-master. The selector field of the service determines which pods will receive the traffic sent to the service, and the port and containerPort information defines what port the service proxy will run at.
Use the file examples/guestbook/redis-master-service.json:
{
"id": "redis-master",
"kind": "Service",
"apiVersion": "v1beta1",
"port": 6379,
"containerPort": 6379,
"selector": {
"name": "redis-master"
},
"labels": {
"name": "redis-master"
}
}to create the service by running:
$ kubectl create -f examples/guestbook/redis-master-service.json
redis-master
$ kubectl get services
NAME LABELS SELECTOR IP PORT
redis-master name=redis-master name=redis-master 10.0.246.242 6379This will cause all pods to see the redis master apparently running on :6379. The traffic flow from slaves to masters can be described in two steps, like so.
- A redis slave will connect to "port" on the redis master service
- Traffic will be forwarded from the service "port" (on the service node) to the containerPort on the pod which (a node the service listens to).
Thus, once created, the service proxy on each minion is configured to set up a proxy on the specified port (in this case port 6379).
Although the redis master is a single pod, the redis read slaves are a 'replicated' pod. In Kubernetes, a replication controller is responsible for managing multiple instances of a replicated pod. The replicationController will automatically launch new Pods if the number of replicas falls (this is quite easy - and fun - to test, just kill the docker processes for your pods at will and watch them come back online on a new node shortly thereafter).
Use the file examples/guestbook/redis-slave-controller.json, which looks like this:
{
"id": "redis-slave-controller",
"kind": "ReplicationController",
"apiVersion": "v1beta1",
"desiredState": {
"replicas": 2,
"replicaSelector": {"name": "redis-slave"},
"podTemplate": {
"desiredState": {
"manifest": {
"version": "v1beta1",
"id": "redis-slave",
"containers": [{
"name": "redis-slave",
"image": "brendanburns/redis-slave",
"cpu": 200,
"ports": [{"containerPort": 6379}]
}]
}
},
"labels": {
"name": "redis-slave",
"uses": "redis-master",
"app": "redis"
}
}
},
"labels": {"name": "redis-slave"}
}to create the replication controller by running:
$ kubectl create -f examples/guestbook/redis-slave-controller.json
redis-slave-controller
$ kubectl get rc
CONTROLLER CONTAINER(S) IMAGE(S) SELECTOR REPLICAS
redis-master-controller redis-master dockerfile/redis name=redis-master 1
redis-slave-controller redis-slave brendanburns/redis-slave name=redis-slave 2The redis slave configures itself by looking for the Kubernetes service environment variables in the container environment. In particular, the redis slave is started with the following command:
redis-server --slaveof ${REDIS_MASTER_SERVICE_HOST:-$SERVICE_HOST} $REDIS_MASTER_SERVICE_PORTYou might be curious about where the REDIS_MASTER_SERVICE_HOST is coming from. It is provided to this container when it is launched via the kubernetes services, which create environment variables (there is a well defined syntax for how service names get transformed to environment variable names in the documentation linked above).
Once that's up you can list the pods in the cluster, to verify that the master and slaves are running:
$ kubectl get pods
POD IP CONTAINER(S) IMAGE(S) HOST LABELS STATUS
redis-master-controller-gb50a 10.244.3.7 redis-master dockerfile/redis kubernetes-minion-7agi.c.hazel-mote-834.internal/104.154.54.203 app=redis,name=redis-master Running
redis-slave-controller-182tv 10.244.3.6 redis-slave brendanburns/redis-slave kubernetes-minion-7agi.c.hazel-mote-834.internal/104.154.54.203 app=redis,name=redis-slave,uses=redis-master Running
redis-slave-controller-zwk1b 10.244.2.8 redis-slave brendanburns/redis-slave kubernetes-minion-3vxa.c.hazel-mote-834.internal/104.154.54.6 app=redis,name=redis-slave,uses=redis-master RunningYou will see a single redis master pod and two redis slave pods.
Just like the master, we want to have a service to proxy connections to the read slaves. In this case, in addition to discovery, the slave service provides transparent load balancing to web app clients.
The service specification for the slaves is in examples/guestbook/redis-slave-service.json:
{
"id": "redisslave",
"kind": "Service",
"apiVersion": "v1beta1",
"port": 6379,
"containerPort": 6379,
"labels": {
"name": "redis-slave"
},
"selector": {
"name": "redis-slave"
}
}This time the selector for the service is name=redis-slave, because that identifies the pods running redis slaves. It may also be helpful to set labels on your service itself as we've done here to make it easy to locate them with the cluster/kubectl.sh get services -l "label=value" command.
Now that you have created the service specification, create it in your cluster by running:
$ kubectl create -f examples/guestbook/redis-slave-service.json
redisslave
$ kubectl get services
NAME LABELS SELECTOR IP PORT
redis-master name=redis-master name=redis-master 10.0.246.242 6379
redisslave name=redis-slave name=redis-slave 10.0.72.62 6379This is a simple PHP server that is configured to talk to either the slave or master services depending on whether the request is a read or a write. It exposes a simple AJAX interface, and serves an angular-based UX. Like the redis read slaves it is a replicated service instantiated by a replication controller.
It can now leverage writes to the load balancing redis-slaves, which can be highly replicated.
The pod is described in the file examples/guestbook/frontend-controller.json:
{
"id": "frontend-controller",
"kind": "ReplicationController",
"apiVersion": "v1beta1",
"desiredState": {
"replicas": 3,
"replicaSelector": {"name": "frontend"},
"podTemplate": {
"desiredState": {
"manifest": {
"version": "v1beta1",
"id": "frontend",
"containers": [{
"name": "php-redis",
"image": "kubernetes/example-guestbook-php-redis",
"cpu": 100,
"memory": 50000000,
"ports": [{"name": "http-server", "containerPort": 80}]
}]
}
},
"labels": {
"name": "frontend",
"uses": "redis-slave,redis-master",
"app": "frontend"
}
}},
"labels": {"name": "frontend"}
}Using this file, you can turn up your frontend with:
$ kubectl create -f examples/guestbook/frontend-controller.json
frontend-controller
$ kubectl get rc
CONTROLLER CONTAINER(S) IMAGE(S) SELECTOR REPLICAS
frontend-controller php-redis kubernetes/example-guestbook-php-redis name=frontend 3
redis-master-controller redis-master dockerfile/redis name=redis-master 1
redis-slave-controller redis-slave brendanburns/redis-slave name=redis-slave 2Once that's up (it may take ten to thirty seconds to create the pods) you can list the pods in the cluster, to verify that the master, slaves and frontends are running:
$ kubectl get pods
POD IP CONTAINER(S) IMAGE(S) HOST LABELS STATUS
frontend-controller-5m1zc 10.244.1.131 php-redis kubernetes/example-guestbook-php-redis kubernetes-minion-3vxa.c.hazel-mote-834.internal/146.148.71.71 app=frontend,name=frontend,uses=redis-slave,redis-master Running
frontend-controller-ckn42 10.244.2.134 php-redis kubernetes/example-guestbook-php-redis kubernetes-minion-by92.c.hazel-mote-834.internal/104.154.54.6 app=frontend,name=frontend,uses=redis-slave,redis-master Running
frontend-controller-v5drx 10.244.0.128 php-redis kubernetes/example-guestbook-php-redis kubernetes-minion-wilb.c.hazel-mote-834.internal/23.236.61.63 app=frontend,name=frontend,uses=redis-slave,redis-master Running
redis-master-controller-gb50a 10.244.3.7 redis-master dockerfile/redis kubernetes-minion-7agi.c.hazel-mote-834.internal/104.154.54.203 app=redis,name=redis-master Running
redis-slave-controller-182tv 10.244.3.6 redis-slave brendanburns/redis-slave kubernetes-minion-7agi.c.hazel-mote-834.internal/104.154.54.203 app=redis,name=redis-slave,uses=redis-master Running
redis-slave-controller-zwk1b 10.244.2.8 redis-slave brendanburns/redis-slave kubernetes-minion-3vxa.c.hazel-mote-834.internal/104.154.54.6 app=redis,name=redis-slave,uses=redis-master RunningYou will see a single redis master pod, two redis slaves, and three frontend pods.
The code for the PHP service looks like this:
<?
set_include_path('.:/usr/share/php:/usr/share/pear:/vendor/predis');
error_reporting(E_ALL);
ini_set('display_errors', 1);
require 'predis/autoload.php';
if (isset($_GET['cmd']) === true) {
header('Content-Type: application/json');
if ($_GET['cmd'] == 'set') {
$client = new Predis\Client([
'scheme' => 'tcp',
'host' => getenv('REDIS_MASTER_SERVICE_HOST') ?: getenv('SERVICE_HOST'),
'port' => getenv('REDIS_MASTER_SERVICE_PORT'),
]);
$client->set($_GET['key'], $_GET['value']);
print('{"message": "Updated"}');
} else {
$read_port = getenv('REDIS_MASTER_SERVICE_PORT');
if (isset($_ENV['REDISSLAVE_SERVICE_PORT'])) {
$read_port = getenv('REDISSLAVE_SERVICE_PORT');
}
$client = new Predis\Client([
'scheme' => 'tcp',
'host' => getenv('REDIS_MASTER_SERVICE_HOST') ?: getenv('SERVICE_HOST'),
'port' => $read_port,
]);
$value = $client->get($_GET['key']);
print('{"data": "' . $value . '"}');
}
} else {
phpinfo();
} ?>Just like the others, you want a service to group your frontend pods.
The service is described in the file examples/guestbook/frontend-service.json:
{
"id": "frontend",
"kind": "Service",
"apiVersion": "v1beta1",
"port": 8000,
"containerPort": "http-server",
"selector": {
"name": "frontend"
},
"labels": {
"name": "frontend"
},
"createExternalLoadBalancer": true
}$ kubectl create -f examples/guestbook/frontend-service.json
frontend
$ kubectl get services
NAME LABELS SELECTOR IP PORT
frontend name=frontend name=frontend 10.0.93.211 8000
redis-master name=redis-master name=redis-master 10.0.246.242 6379
redisslave name=redis-slave name=redis-slave 10.0.72.62 6379In GCE, cluster/kubectl.sh automatically creates forwarding rule for services with createExternalLoadBalancer.
$ gcloud compute forwarding-rules list
NAME REGION IP_ADDRESS IP_PROTOCOL TARGET
frontend us-central1 130.211.188.51 TCP us-central1/targetPools/frontendYou can grab the external IP of the load balancer associated with that rule and visit http://130.211.188.51:8000.
In GCE, you also may need to open the firewall for port 8000 using the console or the gcloud tool. The following command will allow traffic from any source to instances tagged kubernetes-minion:
$ gcloud compute firewall-rules create --allow=tcp:8000 --target-tags=kubernetes-minion kubernetes-minion-8000For GCE details about limiting traffic to specific sources, see the GCE firewall documentation.
In other environments, you can get the service IP from looking at the output of kubectl get pods,services, and modify your firewall using standard tools and services (firewalld, iptables, selinux) which you are already familar with.
And of course, finally, if you are running Kubernetes locally, you can just visit http://localhost:8000.
To turn down a Kubernetes cluster, if you ran this from source, you can use
$ cluster/kube-down.shIf you are in a live kubernetes cluster, you can just kill the pods, using a script such as this (obviously, read through it and make sure you understand it before running it blindly, as it will kill several pods automatically for you).
### First, kill services and controllers.
kubectl stop rc examples/guestbook/redis-slave-controller.json
kubectl stop rc examples/guestbook/frontend-controller.json
kubectl delete -f examples/guestbook/redis-master-service.json
kubectl delete -f examples/guestbook/redis-slave-service.json
kubectl delete pod redis-master # This is the only pod that requires manual removal.the Guestbook example can fail for a variety of reasons, which makes it an effective test. Lets test the web app simply using curl, so we can see whats going on.
Before we proceed, what are some setup idiosyncracies that might cause the app to fail (or, appear to fail, when merely you have a cold start issue.
- running kubernetes from HEAD, in which case, there may be subtle bugs in the kubernetes core component interactions.
- running kubernetes with security turned on, in such a way that containers are restricted from doing their job.
- starting the kubernetes and not allowing enough time for all services and pods to come online, before doing testing.
To post a message (Note that this call overwrites the messages field), so it will be reset to just one entry.
curl "localhost:8000/index.php?cmd=set&key=messages&value=jay_sais_hi"
And, to get messages afterwards...
curl "localhost:8000/index.php?cmd=get&key=messages"
- When the Web page hasn't come up yet:
When you go to localhost:8000, you might not see the page at all. Testing it with curl...
==> default: curl: (56) Recv failure: Connection reset by peerThis means the web frontend isn't up yet. Wait a while, possibly about 2 minutes or more, depending on your set up. Also, run a watch on docker ps, to see if containers are cycling on and off or not starting.
$> watch -n 1 docker ps
If you run this on a node to which the frontend is assigned, you will eventually see the frontend container turns on. At that point, this basic error will likely go away.
- When Redis can't connect to master:
<b>Fatal error</b>: Uncaught exception 'Predis\Connection\ConnectionException' with message 'php_network_getaddresses: getaddrinfo failed: Name or service not known [tcp://:0]' in /vendor/predis/predis/lib/Predis/Connection/AbstractConnection.php:141The fix: Make sure that environmental variables are being set correctly. In particular, the PHP containers need to be started with the environment variables for the redis master (i.e. REDIS_MASTER_SERVICE_HOST and REDIS_MASTER_PORT).
- Temporarily, while waiting for the app to come up , you might see a few of these:
==> default: <br />
==> default: <b>Fatal error</b>: Uncaught exception 'Predis\Connection\ConnectionException' with message 'Error while reading line from the server [tcp://10.254.168.69:6379]' in /vendor/predis/predis/lib/Predis/Connection/AbstractConnection.php:141The fix, just go get some coffee. When you come back, there is a good chance the service endpoint will eventually be up. If not, make sure its running and that the redis master / slave docker logs show something like this.
$> docker logs 26af6bd5ac12
...
[9] 20 Feb 23:47:51.015 # WARNING: The TCP backlog setting of 511 cannot be enforced because /proc/sys/net/core/somaxconn is set to the lower value of 128.
[9] 20 Feb 23:47:51.015 * The server is now ready to accept connections on port 6379
[9] 20 Feb 23:47:52.005 * Connecting to MASTER 10.254.168.69:6379
[9] 20 Feb 23:47:52.005 * MASTER <-> SLAVE sync started- When security issues cause redis writes to fail you may have to run docker logs on the redis containers:
==> default: <b>Fatal error</b>: Uncaught exception 'Predis\ServerException' with message 'MISCONF Redis is configured to save RDB snapshots, but is currently not able to persist on disk. Commands that may modify the data set are disabled. Please check Redis logs for details about the error.' in /vendor/predis/predis/lib/Predis/Client.php:282" The fix is to setup SE Linux properly (don't just turn it off). Remember that you can also rebuild this entire app from scratch, using the dockerfiles, and modify while redeploying. Reach out on the mailing list if you need help doing so!