Introduction
This guide will help you quickly evaluate Gravity by packaging, and installing a sample multi-node Kubernetes application.
We will use Mattermost, an open source chat application for teams. Mattermost represents a fairly typical web application and consists of an HTTP request handling process which connects to a PostgreSQL instance.
Before we start, you may want to go over the Gravity Overview to get familiar with basic concepts of the Gravity solution.
You can also watch the video below which walks through this Quickstart Guide.
Note on Quickstart Video
The video uses version 5.4 of Gravity, so users of newer versions of Gravity may have a slightly different experience.
System Requirements
Gravity is a Kubernetes packaging solution so it only runs on computers capable of running Kubernetes. For this tutorial, you will need:
- A x86_64 Linux machine or a VM for building a Cluster Image that is running one of the supported Linux distributions.
- Docker version 17 or newer. Run
docker info
before continuing to make sure you have Docker up and running. We recommend following instructions on installing Docker CE from Docker.com - You must be a member of the
docker
group. Rungroups
command to make suredocker
group is listed. If not, you can add yourself to the "docker" group viasudo usermod -aG docker $USER
- You must install Helm 2.0 - Helm v2.12 Github Release.
- You must have
git
installed to clone the example application repo. - A target cluster of Linux nodes. It can be just one machine but this example quickstart will use three. The nodes in a target cluster must have at least 2GB of RAM and 40GB of free disk space. They must not have Docker or any other container runtime installed on them.
- You must have
sudo
privileges on all nodes.
Getting the Tools
Start by downloading Gravity and unpacking the archive. You should see the following files:
$ ls -l
-rwxr-xr-x 1 user user 128 Dec 3 13:07 install.sh
-rwxr-xr-x 1 user user 50562960 Dec 3 13:07 tele
-rwxr-xr-x 1 user user 21417992 Dec 3 13:07 tsh
Execute install.sh
to copy tele
and tsh
binaries to
/usr/local/bin/
. Then you can type tele version
to confirm that
everything works:
$ tele version
Edition: open-source
Version: 6.2.0
Git Commit: 59f55954483989fef11bbe47a91c4d09743f65fe
Helm Version: v2.12
Clone the sample Git repository which contains the Kubernetes resources for Mattermost, which we are using in this tutorial as a sample application:
$ git clone https://github.com/gravitational/quickstart.git
$ cd quickstart
Building a Cluster Image
To build a Cluster Image we'll perform the following steps:
- Create Docker containers for application services.
- Create definitions of Kubernetes resources (pods, etc) for application components. This makes an application capable of running on Kubernetes. You can place all Kubernetes resource files (usually in YAML format) in the same directory or you can use Helm.
- Create a Cluster Image Manifest to describe the system requirements for a Kubernetes cluster capable of running your application. A Cluster Image Manifest is a YAML file which allows you to customize the Cluster Image.
- Execute
tele build
CLI command.
Step 1: Containerizing
Run the following to build the Mattermost containers:
$ cd mattermost/worker
$ docker build -t mattermost-worker:2.2.0 .
When Docker finishes building the container, you should be able to see it listed:
$ docker images | grep mattermost
mattermost-worker 2.2.0 ce3ead6dff48 43 seconds ago 405MB
You can now return to the quickstart home directory.
Step 2: Creating the Kubernetes Resources
Making Mattermost run on Kubernetes is easy. The quickstart repository you have cloned above includes the YAML definitions of Kubernetes objects. We'll use a Helm chart for this:
$ tree mattermost/resources/charts/mattermost/
mattermost/resources/charts/mattermost/
├── Chart.yaml
├── templates
│ ├── _helpers.tpl
│ └── mattermost.yaml
└── values.yaml
The most interesting file to take a look at is mattermost.yaml. You are welcome to modify it to your liking.
Tip
In this tutorial, we are packaging a single Helm chart but it is possible to have several of them packaged into a single Cluster Image.
Step 3: Creating the Cluster Image Manifest
In this step, we create an Image Manifest which describes the system requirements for the Cluster.
We have already prepared one for this guide in the cloned repo: mattermost/resources/app.yaml
. You can open it on Github for convenience. We have commented the most important fields
in the example manifest.
Step 4: Building the Cluster Image
Before we build our first Cluster Image, let's make sure Helm is properly initialized.
$ helm init --client-only
Now you can build the Cluster Image, which will consist of a Kubernetes cluster with Mattermost pre-installed inside:
$ tele build -o mattermost.tar mattermost/resources/app.yaml
# the output:
* [1/6] Selecting application runtime
Will use latest runtime version 5.2.4
* [2/6] Downloading dependencies from s3://hub.gravitational.io
Still downloading dependencies from s3://hub.gravitational.io (10 seconds elapsed)
Still downloading dependencies from s3://hub.gravitational.io (20 seconds elapsed)
* [3/6] Embedding application container images
Detected application manifest app.yaml
Detected resource file clusterDeprovision.yaml
Detected resource file clusterProvision.yaml
Detected resource file install.yaml
Detected resource file nodesDeprovision.yaml
Detected resource file nodesProvision.yaml
Detected Helm chart charts/mattermost
Using local image quay.io/gravitational/debian-tall:0.0.1
Using local image quay.io/gravitational/debian-tall:0.0.1
Using local image quay.io/gravitational/provisioner:ci.82
Using local image mattermost-worker:2.2.0
Using local image postgres:9.4.4
Vendored image gravitational/debian-tall:0.0.1
Vendored image gravitational/provisioner:ci.82
Still embedding application container images (10 seconds elapsed)
Vendored image mattermost-worker:2.2.0
Vendored image postgres:9.4.4
* [4/6] Using runtime version 5.2.4
Still using runtime version 5.2.4 (10 seconds elapsed)
* [5/6] Generating the cluster snapshot
* [6/6] Saving the snapshot as mattermost.tar
* [6/6] Build completed in 2 minutes
Let's review what just happened. tele build
did the following:
- Downloaded Kubernetes binaries and Gravity tooling from
s3://hub.gravitational.io
. - Scanned the current directory and the subdirectories for Kubernetes resources and Helm charts.
- Downloaded external container images referenced in the resources discovered in the previous step.
- Packaged (or vendored) Docker images into the Cluster Image.
- Removed the duplicate container image layers, reducing the size of the Cluster Image.
- Saved the Cluster Image as
mattermost.tar
.
Slow Operation Warning
tele build
needs to download hundreds of megabytes of binary dependencies which
can take a considerable amount of time, depending on your Internet connection speed.
The resulting mattermost.tar
file is about 1.6GB and it is entirely
self-sufficient and dependency-free. It contains everything: the Kubernetes
binaries, the Docker engine, the Docker registry and the Mattermost application itself:
everything one needs to get Mattermost up and running on any fleet of Linux
servers (or into an AWS/GCE/Azure account).
Congratulations! You have created your first Kubernetes virtual appliance!
Installing
Installing the mattermost.tar
Cluster Image results in creating a Kubernetes
cluster with the application pre-loaded. This file is the only artifact
one needs to create a Kubernetes cluster with Mattermost running inside.
Copy mattermost.tar
to a clean Linux machine. Let's call it host
. This node
will be used to bootstrap the cluster. Let's untar it and look inside:
$ tar -xf mattermost.tar
$ tree
├── app.yaml
├── gravity
├── gravity.db
├── install
├── packages
│ ├── blobs
│ │ ├── 0e1
│ │ ...
│ │ └── ff1
│ │ └── ff19bcf2dc62f037e0016d5d065150d195f714be3c97a301791365a4ec5a43f0
│ ├── tmp
│ └── unpacked
├── README
├── upgrade
└── upload
Here is a brief description of these files:
File Name | Description |
---|---|
gravity |
Gravity Cluster manager which is a Linux binary (executable). It's responsible for installing, upgrading and managing clusters. |
app.yaml |
The Image Manifest which we've defined earlier and fed to tele build . You'll notice that the build process populated the manifest with additional metadata. |
packages |
The database of Docker image layers for all containers and other binary artifacts, like Kubernetes binaries. |
gravity.db |
The metadata of what's stored in packages . |
upgrade , install , upload |
Helpful bash wrappers around gravity commands. |
README |
Instructions for the end user. |
Gravity supports two modes of installation:
- CLI mode or non-interactive mode is useful for advanced users and for scripting. It allows clusters to be created programmatically or via command line.
- Web mode uses a web browser to guide a user through an installation wizard. This mode is useful for non-technical users, sales demos, etc.
Installing via CLI
To install a Cluster via CLI, you have to execute the ./gravity install
command and
supply two mandatory flags:
Flag | Description |
---|---|
--token |
A secret token of your choosing which will be used to add additional nodes to this Cluster in the future. We'll use word "secret" here. |
--advertise-addr |
The IP address this host will be visible on by other nodes in this Cluster. We'll use 10.5.5.28 . |
The command below will create a single-node Kubernetes cluster with Mattermost running inside:
# We are executing this on the node named 'host' with IP address of 10.5.5.28
$ sudo ./gravity install \
--advertise-addr=10.5.5.28 \
--token=secret
# Output:
Sat Jan 12 05:30:44 UTC Starting installer
Sat Jan 12 05:30:44 UTC Preparing for installation...
Sat Jan 12 05:31:09 UTC Installing application mattermost:2.2.0
Sat Jan 12 05:31:09 UTC Starting non-interactive install
Fri Jan 11 21:31:09 UTC Auto-loaded kernel module: br_netfilter
Fri Jan 11 21:31:09 UTC Auto-loaded kernel module: iptable_nat
Fri Jan 11 21:31:09 UTC Auto-loaded kernel module: iptable_filter
Fri Jan 11 21:31:09 UTC Auto-loaded kernel module: ebtables
Fri Jan 11 21:31:09 UTC Auto-set kernel parameter: net.ipv4.ip_forward=1
Fri Jan 11 21:31:09 UTC Auto-set kernel parameter: net.bridge.bridge-nf-call-iptables=1
Sat Jan 12 05:31:10 UTC All agents have connected!
Sat Jan 12 05:31:10 UTC Starting the installation
Sat Jan 12 05:31:11 UTC Operation has been created
Sat Jan 12 05:31:12 UTC Execute preflight checks
Sat Jan 12 05:31:16 UTC Configure packages for all nodes
Sat Jan 12 05:31:20 UTC Bootstrap master node host
Sat Jan 12 05:31:24 UTC Pull packages on master node host
Sat Jan 12 05:32:20 UTC Install system package teleport:2.4.7 on master node host
Sat Jan 12 05:32:22 UTC Install system package planet:5.2.19-11105 on master node host
Sat Jan 12 05:32:48 UTC Wait for system services to start on all nodes
Sat Jan 12 05:33:24 UTC Label and taint master node host
Sat Jan 12 05:33:25 UTC Bootstrap Kubernetes roles and PSPs
Sat Jan 12 05:33:26 UTC Export applications layers to Docker registries
Sat Jan 12 05:33:27 UTC Populate Docker registry on master node host
Sat Jan 12 05:34:24 UTC Install system application dns-app:0.1.0
Sat Jan 12 05:34:31 UTC Install system application logging-app:5.0.2
Sat Jan 12 05:35:05 UTC Install system application tiller-app:5.2.1
Sat Jan 12 05:35:43 UTC Install system application site:5.2.4
Sat Jan 12 05:36:59 UTC Install system application kubernetes:5.2.4
Sat Jan 12 05:37:00 UTC Install application mattermost:2.2.0
Sat Jan 12 05:37:10 UTC Enable elections
Sat Jan 12 05:37:12 UTC Operation has completed
Sat Jan 12 05:37:13 UTC Installation succeeded in 6m3.257480586s
Congratulations! You have created a fully functional Kubernetes cluster with Mattermost running inside. To check the health and status of the Cluster, execute this command on the target node:
$ gravity status
# the output:
Cluster status: active
Application: mattermost, version 2.2.0
Join token: 3f59d1923ed4e2f1499f3e272c86310b46666c8ddce708f3131b16f256f10004
Last completed operation:
* operation_install (6784ad01-530a-45f0-a303-119ac8cd3417)
started: Tue Jan 22 22:40 UTC (10 minutes ago)
completed: Tue Jan 22 22:40 UTC (10 minutes ago)
Cluster: friendlypoincare4048
Masters:
* host (10.5.5.28, node)
Status: healthy
If a single node cluster is not enough, you can add additional nodes to it:
- Copy
gravity
binary from the bootstrapping node above to another host which is about to be added to the Cluster. Let's assume its IP is10.5.5.29
. - Execute
gravity join
command as shown below. Note that this command will "think" in silence for a few seconds before dumping any output.
# Execute this on the second node with an IP 10.5.5.29
$ sudo ./gravity join 10.5.5.28 --advertise-addr=10.5.5.29 --token=secret
# Output:
Sat Jan 12 06:00:16 UTC Connecting to cluster
Fri Jan 11 22:00:16 UTC Auto-loaded kernel module: br_netfilter
Fri Jan 11 22:00:16 UTC Auto-loaded kernel module: iptable_nat
Fri Jan 11 22:00:16 UTC Auto-loaded kernel module: iptable_filter
Fri Jan 11 22:00:16 UTC Auto-loaded kernel module: ebtables
Fri Jan 11 22:00:16 UTC Auto-set kernel parameter: net.ipv4.ip_forward=1
Fri Jan 11 22:00:16 UTC Auto-set kernel parameter: net.bridge.bridge-nf-call-iptables=1
Sat Jan 12 06:00:16 UTC Connected to existing cluster at 10.5.5.28
Sat Jan 12 06:00:17 UTC Operation has been created
Sat Jan 12 06:00:18 UTC Configure packages for the joining node
Sat Jan 12 06:00:20 UTC Bootstrap the joining node
Sat Jan 12 06:00:21 UTC Pull packages on the joining node
Sat Jan 12 06:01:18 UTC Install system package teleport:2.4.7
Sat Jan 12 06:01:19 UTC Install system package planet:5.2.19-11105
Sat Jan 12 06:01:41 UTC Start RPC agent on the master node 10.5.5.28
Sat Jan 12 06:01:45 UTC Add the joining node to the etcd cluster
Sat Jan 12 06:01:49 UTC Wait for the planet to start
Sat Jan 12 06:02:24 UTC Stop RPC agent on the master node 10.5.5.28
Sat Jan 12 06:02:25 UTC Enable leader election on the joined node
Sat Jan 12 06:02:26 UTC Operation has completed
Sat Jan 12 06:02:26 UTC Joined cluster in 2m10.547146946s
Now you have a two-node Kubernetes cluster with Mattermost running inside. The next step is to create a new Kubernetes user:
# execute this on the K8s master node (running on 10.5.5.28 in our example)
# to create a user "ekontsevoy"
$ gravity users add [email protected] ekontsevoy
# output:
Signup token has been created and is valid for 8h0m0s hours. Share this URL with the user:
https://10.5.5.28:3009/web/newuser/e5b5422da69ff44d41f92e3ce6167659a7fee10e1023acea22062141dfe0238e
Now click on the printed URL and select a password. You are now inside the Cluster
Control Panel. You can bookmark the following URL to access it in the future: https://10.5.5.28:32009/web/
You will also see that this Cluster is running Mattermost inside, accessible as a Kubernetes service
on port 32010
, i.e. it's accessible using IP addresses of both machines in the Cluster:
http://10.5.5.28:32010/
http://10.5.5.29:32010/
Installing via Web Browser
This method of installation launches a graphical installation wizard in a web browser.
To launch a web installer, you will need:
- The Cluster Image
mattermost.tar
which we have prepared earlier. - A Linux computer with a graphical interface and web browser connected to the same network as the target nodes.
First, untar mattermost.tar
and execute the install
script. This command
launches an HTTP server which serves a web UI and acts as a bootstrapping agent
to create a new Cluster. It will print a web URL for you to click
on or paste in your browser.
$ sudo ./install
OPEN THIS IN BROWSER: https://host:61009/web/installer/new/gravitational.io/mattermost/2.2.0?install_token=2a9de4a72ede
If you don't have TLS setup you might see this error message. Click Advanced -> Proceed
The browser-based installer will ask for the following:
- Name of your Cluster. We recommend FQDN-like names like
mattermost.example.com
.
- The network interface to use. This must be the interface which Kubernetes nodes will use to talk to each other.
- The "flavor" of the Cluster, i.e. 1, 2 or 3 nodes. The installer will provide a CLI command to copy to and execute on each node.
- Once all nodes report into the Cluster, the installer will proceed setting up Kubernetes.
The final step is to select the user name and password for the administrator. You will be able to change it later (or configure the SSO). Once you are logged in, you will be placed in Gravity's Control Panel UI where you will find the HTTP end point of Mattermost.
Mattermost Install Complete
You can press Ctrl+C
to stop the install
script.
Conclusion
This is a brief overview of how Kubernetes clusters can be packaged into simple
tar
files and then installed anywhere. Gravity's image-based approach is quite
similar to how virtual machines/instances are treated by using disk images in
virtualized environments.
This dramatically lowers the operational overhead of running multiple Kubernetes clusters within an organization, allows complex SaaS applications to be converted into downloadable Kubernetes appliances and dramatically simplifies implementing compliance in organizations by publishing Kubernetes images that are pre-configured and approved by the security and compliance teams.
If you need additional guidance with packaging your Kubernetes clusters into Gravity appliances, our implementation services team can help ([email protected]).