Taclib stands for Task as Containers Library, and it was created to ease running luigi tasks in Kubernetes.
Each luigi task will be running in a pod as a containerized application. The luigi controller
pod needs to have a permission to create new pods in the cluster, in order to do that we have
to create a Cluster Role and bind it to an active service account.
The resource file cluster_role.yaml contain the necessary roles which allow
the controller pod to create task pods.
To grant the controller this permission we need to apply this role to our cluster through
kubectl apply -f cluster_role.yaml
In the directory example there are two python scripts that will serve as an example on how to use taclib.
._ example_task.py
import luigi
from taclib.task import KubernetesTask
class ExampleTask(KubernetesTask):
name = luigi.Parameter(default="taclib")
@property
def image(self):
return "my-custom-image:0.1"
@property
def command(self):
return ["python", "-m", "script.py", "--name", self.name]
def output(self):
return luigi.LocalTarget("_SUCCESS")
If you are familiar with Luigi, you will recognize the structure above.
We define a custom task that will inherit KubernetesTask. task will spawn a pod
on Kubernetes using the image you define either through the function image
or through TACLIB_IMAGE environment variable, if none of them are defined taclib
will set the default image as python:3-7-alpine.
Once the pod is running, the task will fire the command defined in
def command(self):. Note that in the example the given command executes
script.py, i.e, this file should be in the image you create.
Taclib tasks can be either started from a luigi command on you local environment,
or right in the Kubernetes cluster using kubectl.
luigi --module example.example_task ExampleTask --local-scheduler
kubectl run my-task \
--restart Never \
--image-pull-policy Always \
--image my-image \
--serviceaccount taclib-controller-sa
-- luigi --module example.example_task ExampleTask \
--scheduler-host my-luigi-scheduler \