| | |
|---|
| | | - step005.1 |
|---|
| | | - gpu_install_config |
|---|
| | | tasks: |
|---|
| | | - name: pre-pull relevant images |
|---|
| | | command: docker pull {{ item }} |
|---|
| | | with_items: |
|---|
| | | - "{{ cache_images }}" |
|---|
| | | |
|---|
| | | - name: clone the psap repository onto the bastion |
|---|
| | | git: |
|---|
| | | repo: 'https://github.com/thoraxe/openshift-psap.git' |
|---|
| | | dest: '/root/openshift-psap' |
|---|
| | | version: 'ocp-310-0.6' |
|---|
| | | version: 'ocp-311-0.7' |
|---|
| | | force: yes |
|---|
| | | update: yes |
|---|
| | | |
|---|
| | |
|---|
| | | tasks: |
|---|
| | | # if we do something multi-node this will have to change to look for the total number of gpu nodes |
|---|
| | | - name: Check that the daemonset is ready |
|---|
| | | command: oc get daemonset -n nvidia-device-plugin nvidia-deviceplugin-daemonset -o jsonpath --template='{.status.numberReady}' |
|---|
| | | command: oc get daemonset -n kube-system nvidia-deviceplugin-daemonset -o jsonpath --template='{.status.numberReady}' |
|---|
| | | register: daemonset_ready_out |
|---|
| | | until: 'daemonset_ready_out.stdout | int >= 1' |
|---|
| | | retries: 5 |
|---|
