PathwaysJob API is an OSS Kubernetes-native API, to deploy ML training and batch inference workloads, using Pathways on GKE. //ToDo(roshanin) - add intro for Pathways.

The PathwaysJob is an API that provides an easy way to run JAX workloads using Pathways. It support two modes of deployment.

The 'colocate' mode bundles the Pathways resource manager(RM), the Pathways proxy and the user workload containers into a single pod called "leader" and deploys them besides a "worker" pod on one of the TPU workers. This is preferred for Pathways batch inference workloads, where latency is crucial.

The default mode is preferred for Pathways training workloads where the worker utilizes the TPUs completely. The Pathways RM and Pathways proxy are scheduled as pods on a CPU nodepool and the "workers" are scheduled on TPUs.

The user workload is also scheduled as a pod on the CPU nodepool.

The user workload is typically on a Vertex AI notebook, so users can connect to the PathwaysJob via port-forwarding.

• go version v1.22.0+
• docker version 17.03+.
• kubectl version v1.11.3+.
• Access to a Kubernetes v1.11.3+ cluster.

Build and push your image to the location specified by IMG:

make docker-build docker-push IMG=<some-registry>/pathways-job:tag

NOTE: This image ought to be published in the personal registry you specified. And it is required to have access to pull the image from the working environment. Make sure you have the proper permission to the registry if the above commands don’t work.

Install the CRDs into the cluster:

make install

Deploy the Manager to the cluster with the image specified by IMG:

make deploy IMG=<some-registry>/pathways-job:tag

NOTE: If you encounter RBAC errors, you may need to grant yourself cluster-admin privileges or be logged in as admin.

Create instances of your solution You can apply the samples (examples) from the config/sample:

kubectl apply -k config/samples/

NOTE: Ensure that the samples has default values to test it out.

Delete the instances (CRs) from the cluster:

kubectl delete -k config/samples/

Delete the APIs(CRDs) from the cluster:

make uninstall

UnDeploy the controller from the cluster:

make undeploy

Following are the steps to build the installer and distribute this project to users.

1. Build the installer for the image built and published in the registry:

make build-installer IMG=<some-registry>/pathways-job:tag

NOTE: The makefile target mentioned above generates an 'install.yaml' file in the dist directory. This file contains all the resources built with Kustomize, which are necessary to install this project without its dependencies.

2. Using the installer

Users can just run kubectl apply -f to install the project, i.e.:

kubectl apply -f https://raw.githubusercontent.com/<org>/pathways-job/<tag or branch>/dist/install.yaml

// TODO(user): Add detailed information on how you would like others to contribute to this project

NOTE: Run make help for more information on all potential make targets

More information can be found via the Kubebuilder Documentation

Copyright 2025.

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.