Optune servo driver for Rancher 1.6.
This driver supports adjusting Services deployed via Rancher in a Stack.
A Rancher Stack may contain multiple Services (components in Servo parlance). Each may be individually modified by the adjust driver. Each service in the Stack is considered a separately adjustable component.
To describe a Stack:
./adjust --describe <stack-name-or-id>
To adjust a Service
./adjust <service-name-or-id> < adjust.json
See: USAGE-adjust.md for full documentation.
Adjust requires several configuration variables to operate. These can be provided via config.yaml
or environment settings. You can either place these into the environment directly, or supply
a file called .env in the root directory with adjust.
.env
OPTUNE_API_URL='http://rancher.api/'
OPTUNE_API_KEY=myrancherapikey
OPTUNE_API_SECRET=myranchersecretkey
OPTUNE_PROJECT=Default
OPTUNE_STACK=my-stack
OPTUNE_CONFIG=an *optional* path to a config.yaml file
There is also a sample_config.yaml file which you can tweak as needed.
If provided, the yaml file takes precidence over the environment variables. If the file is named
config.yaml and coxists with adjust, it will automatically be loaded.
For each service of the stack, the following settings are automatically available when
adjust --describe is run:
cpu: number of reserved coresmem: memory reservation for the servicereplicas: number of service instances
In addition, an optional configuration file may be used to define which environment variables can be modified. By default, NO environment variables may be changed.
The environment settings is a whitelist hash of the environment variables adjust is allowed to
modify. It is a strict replace of the current value, no substring matching currently occurs.
However, if the configuration provides a units key/value as part of the variable description,
integer values passed during an adjust operation will automatically be converted from Gb into the
provided units.
For example, a config.yaml configured with:
MEMORY:
units: M
Will convert an adjust JSON:
MEMORY: 2
Into:
MEMORY: 2048M
All of these settings will be returned when calling adjust --describe. Each maps to a
launchConfig
setting of the matching container as follows:
cpu(int) ->vcpumem(int) ->memoryMbreplicas(int) ->countenvironment(map) ->environment
A service may be excluded from operation by having an exclude key provided within its key/value
set. If a service is configured as such, it will be ignored upon an adjust if it is requested.
In addition, if a launchConfig label has been configed with com.opsani.exclude it will be
excluded, even if it does not have a matching exclude tag in the config.yaml.
See sample_config.yaml for an example configuration.
To perform an adjust, a JSON file must be supplied on STDIN. The format should be similar to other adjust.json files.
An adjust may list multiple services within it. But it must define the service to upgrade otherwise no upgrade will take place.
Adjust is called by providing the service name and passing JSON on stdin.
adjust service < adjust.json
The service hash should include key/values from those defined above
(cpu, mem, replicas, environment). The first three keys should have integer values.
If environment is passed, any keys within that hash must be included in a supplied config.yaml or
they will be ignored.
The following sample configures the front service to have a single cpu and sets the MEMORY
environment variable to 1Gb.
{
"front": {
"cpu": 1
"environment": {
"MEMORY": 1
}
}
}
See http-test.json for an example configuration.
The simplest Rancher 1.6 environment is a single node non-HA configuration launched as a Docker container. Install docker on a current Linux system (we recommend Ubuntu 18.04 LTS), possibly using the bash script from Docker:
bash <(curl -sL https://get.docker.com)
Note that if you continue to use a "default" user other than root, you will also want to provide that user with docker access (e.g. for the "ubuntu" default user):
sudo usermod -aG docker ubuntu
Or use any other method that is appropriate to your operating system.
Then simply launch the Rancher service:
sudo docker run -d --restart=unless-stopped -p 8080:8080 rancher/server
At this point, you should have Rancher running on your server on port 8080 (make sure you open port 8080 on any server associated security-group or firewall settings). But this is not enough, you will also need to add a host to the Rancher environment in order to launch a service. And the simplest approach is to launch the rancher agent on the same host that his running the rancher server, and for testing purposes this should be adequate. There is one glitch that needs to be resolved. In order for the rancher agent to start properly, the /etc/resolv.conf file needs to be overwritten if it contains a pointer to the 127.0.0.0/8 network, as is currently being done in many of the latest linux operating systems. In order to remedy this, do the following:
sudo rm /etc/resolv.conf
echo nameserver 8.8.8.8 | sudo tee - /etc/resolv.conf
Then you can go to the hosts page in the Rancher UI and copy the docker run command to add the host to the system. Note that if you are running behind a firewall/NAT gateway (e.g. in Amazon) you will need to use the public address for the host as the target for your connection request. Depending on your gateway configuration, you may be able to discover this address with a curl request like:
curl -4 icanhazip.com
In addition to the host configuration, the system does not have user authentication enabled initially. For the test enviornment, we recommend using the Local configuration option (Admin->Authentication) and creating an admin user with a password.
And finally, we need to enable the secrets service (this is a plugin for Rancher available in the application catalog), and create and set the API key and the key and secret secrets. To load the Secrets plugin, select the Catalog->Library from the UI. Then search for Secrets and then "select" and "launch" the Secrets service.
Now we can create and distribute secrets to our services. Before we do that, we need to create a new API Key and Secret pair, which is available via the UI at API->Keys. Create a secret pair called servo-secret, and then note the Key and Secret.
NOTE: the Secret is only available in this display window and is not recorded in an accessible location after being displayed.
Finally we can create the secrets that our servo will read. Select Infrastructure->Secrets from the UI. Then select Add Secret, and create a secret called api_key, and add the API key value to this secret. Do the same for a secret called api_secret, but install the API secret in this key.
The Opsani test app co-http can be used to provide a test target for the Opsani Optune engine within the rancher environment. And a docker_compose and rancher_compose file are provided in the example directory to support launching the application directly (the app can also be defined and deployed manually).
To launch the app, select the Stacks->User from the UI, and then select Add Stack. Provide a name (e.g. http-test) for the stack, and then either copy and paste the contents of the docker_compose.yml and rancher_compose.yml into the respective sections of the UI, or browse to the examples directory and upload the file contents. Select Create to build out the test app.
Note that the default service created is available on port 8090, and it will only be available on the local rancher host unless a port mapping, security group rule, or filrewall ACL is applied to make this port available in most cloud based VM environments (e.g. AWS).
A quick check can be made to ensure that the application is running by running a curl command on the local Rancher host:
curl http://localhost:8090/?call=http%3A%2F%2Fback%3A8080%2F%3Fbusy%3D40\&busy=10
Simple instructions if you want to set up a local development environment:
python3 -m venv .venv
source .venv/bin/activate
pip install -r ./requirements.txt