π As of Oct-23 we moved the development of this component to the monorepo. π |
---|
Kubernetes operator to executes Flyte graphs natively on kubernetes
Propeller is a K8s native operator that executes Flyte workflows. Workflow Spec is written in Protobuf for cross-compatibility.
A Mutating Webhook that can be optionally deployed to extend Flyte Propeller's functionality. It currently supports enables injecting secrets into pods launched directly or indirectly through Flyte backend plugins.
A Kubectl-plugin to interact with Flyte Workflow CRDs. It enables retrieving and rendering Flyte Workflows in CLI as well as safely aborting running workflows.
kubectl-flyte is an command line tool that can be used as an extension to kubectl. It is a separate binary that is built from the propeller repo.
This command will install kubectl-flyte and flytepropeller to ~/go/bin
$ make compile
You can also use Krew to install the kubectl-flyte CLI:
$ kubectl krew install flyte
Two ways to execute the command, either standalone kubectl-flyte or as a subcommand of kubectl
$ kubectl-flyte --help
OR
$ kubectl flyte --help
Flyte is a serverless workflow processing platform built for native execution on K8s.
It is extensible and flexible to allow adding new operators and comes with many operators built in
Usage:
kubectl-flyte [flags]
kubectl-flyte [command]
Available Commands:
compile Compile a workflow from core proto-buffer files and output a closure.
config Runs various config commands, look at the help of this command to get a list of available commands..
create Creates a new workflow from proto-buffer files.
delete delete a workflow
get Gets a single workflow or lists all workflows currently in execution
help Help about any command
visualize Get GraphViz dot-formatted output.
To retrieve all workflows in a namespace use the --namespace option, --namespace = "" implies all namespaces.
$ kubectl-flyte get --namespace flytekit-development
workflows
βββ flytekit-development/flytekit-development-f01c74085110840b8827 [ExecId: ... ] (2m34s Succeeded) - Time SinceCreation(30h1m39.683602s)
...
Found 19 workflows
Success: 19, Failed: 0, Running: 0, Waiting: 0
To retrieve a specific workflow, namespace can either be provided in the format namespace/name or using the --namespace argument
$ kubectl-flyte get flytekit-development/flytekit-development-ff806e973581f4508bf1
Workflow
βββ flytekit-development/flytekit-development-ff806e973581f4508bf1 [ExecId: project:"flytekit" domain:"development" name:"ff806e973581f4508bf1" ] (2m32s Succeeded )
βββ start-node start 0s Succeeded
βββ c task 0s Succeeded
βββ b task 0s Succeeded
βββ a task 0s Succeeded
βββ end-node end 0s Succeeded
To delete a specific workflow
$ kubectl-flyte delete --namespace flytekit-development flytekit-development-ff806e973581f4508bf1
To delete all completed workflows - they have to be either success/failed with a special isCompleted label set on them. The Label is set here
$ kubectl-flyte delete --namespace flytekit-development --all-completed
use the config.yaml in root found here. Cd into this folder and then run
$ flytepropeller
Following dependencies need to be met
- Blob store (you can forward minio port to localhost)
- Admin Service endpoint (can be forwarded) OR Disable events to admin and launchplans
- access to kubeconfig and kubeapi
API Server requires the webhook to serve traffic over SSL. To issue self-signed certs to be used for serving traffic, use:
$ flytepropeller webhook init-certs
This will create a ca.crt, tls.crt and key.crt and store them to flyte-pod-webhook secret. If a secret of the same name already exist, it'll not override it.
Starting the webhook can be done by running:
$ flytepropeller webhook
The secret should be mounted and accessible to this command. It'll then create a MutatingWebhookConfiguration object with the details of the webhook and that registers the webhook with ApiServer.
Remember changes to CRD should be carefully done, they should be backwards compatible or else you should use proper operator versioning system. Once you do the changes, you have to follow the following steps.
- Ensure the propeller code is checked out in
$GOPATH/github.com/flyteorg/flytepropeller
- Uncomment https://github.com/flyteorg/flytepropeller/blob/master/hack/tools.go#L5
$ go mod vendor
- Now generate the code
$ make op_code_generate
Why do we have to do this?
Flytepropeller uses old way of writing Custom controllers for K8s. The k8s.io/code-generator only works in the GOPATH
relative code path (sadly). So you have checkout the code in the right place. Also, go mod vendor
is needed to get
code-generator in a discoverable path.
TODO
- We may be able to avoid needing the old style go-path
- Migrate to using controller runtime