Quickstart

Trainy Hub is in private preview. If you are moving from direct cluster submission, follow Submit jobs through Trainy Hub for login, cluster readiness, registry secrets, and logs.

Run “Hello Konduktor”

Run konduktor login first. For the Trainy Hub, login also sets up a Hub Kubernetes context, makes it current, and prints its cluster and namespace. Subsequent konduktor launch, konduktor status, and konduktor logs commands use the Hub by default unless kubernetes.allowed_contexts pins another context in your Konduktor config. In that case, login prints an explicit --context command for Hub submission. To keep your current context, use konduktor login --no-hub-context to sign in without context setup. If you belong to more than one namespace, choose one when prompted, or pass --namespace <name>. Use konduktor logs --context <hub-context> <job-id> for Hub jobs; the CLI reads those logs through the Hub’s OIDC-protected endpoint. On a local context, konduktor logs retains its existing cluster-local read path. Logs from every worker cluster in your Hub namespace are available together. The setup requires gcloud, kubectl, and gke-gcloud-auth-plugin. Hub context setup uses your signed-in session for Google Workforce credentials. If your account has no assigned namespace, ask your organization owner or Trainy to grant access. To submit your first run, copy the below yaml into a file called ‘hello_konduktor.yaml’.

name: hello-world

num_nodes: 1 # scale up your worklaod

resources:
  cpus: 1 # number of cpu cores
  memory: 10 # memory request in GiB (int)
  accelerators: H100:8
  image_id: ubuntu
  labels:
    kueue.x-k8s.io/queue-name: user-queue
    maxRunDurationSeconds: "3200"

run: |
  # print commands ran and exit immediately 
  # with an error if there's any errors
  set -ex
  sleep 2
  echo "Hello Konduktor"
  
  

If you see “Hello World!”, congratulations on setting up Konduktor!

AI-Assisted Workflows: Install the Konduktor Skills plugin to get AI help drafting task YAMLs, writing Python SDK code, and constructing CLI commands directly in your editor.

npx skills add Trainy-ai/konduktor-skills

Requirements

We currently have the following requirements for container images for jobs.

Containers must be Debian/Ubuntu based and use apt as a package manager
have bash in $PATH
permit login/run as root

Get Started

CLI

User Guides

Cluster Operators

Run “Hello Konduktor”

Requirements

​Run “Hello Konduktor”

​Requirements

Run “Hello Konduktor”

Requirements