unip-controller/controller/templates/experiment-pipeline/simple-pipeline-trial.yaml

# ============================================================
# Система: Единая библиотека, Центр ИИ НИУ ВШЭ
# Модуль: ExperimentPipeline
# Авторы: Полежаев В.А., Хританков А.С.
# Дата создания: 2024 г.
# ============================================================
piVersion: batch/v1
kind: Job
metadata:
  name: {{ job_name }}
  namespace: {{ namespace }}
  {% if annotations %}
  annotations:
  {% for key, value in annotations.items() %}
    {{ key }}: "{{ value }}"
  {% endfor %}
  {% endif %}
  {% if labels %}
  labels:
  {% for key, value in labels.items() %}
    {{ key }}: "{{ value }}"
  {% endfor %}
  {% endif %}
spec:
  completions: 1
  backoffLimit: 3
  template:
    {% if labels %}
    metadata:
      labels:
      {% for key, value in labels.items() %}
        {{ key }}: "{{ value }}"
      {% endfor %}
    {% endif %}
    spec:
      restartPolicy: Never
      initContainers:
      {% for stage in stages %}
        - name: {{ stage.name }}
          image: {{ stage.image }}
          {% if stage.cmd %}
          command:
            {% for line in stage.cmd %}
            - {{ line }}
            {% endfor %}
          {% endif %}
          {% if stage.volume_mounts %}
          volumeMounts:
            {% for volume_mount in stage.volume_mounts %}
            - name: {{ volume_mount['volume_name'] }}
              mountPath: {{ volume_mount['mount_path'] }}
              {% if volume_mount['read_only'] %}
              readOnly: True
              {% endif %}
              {% if volume_mount['sub_path'] %}
              subPath: {{ volume_mount['sub_path'] }}
              {% endif %}
            {% endfor %}
          {% endif %}
          {% if stage.resource_limits or stage.resource_requests %}
          resources:
            {% if stage.resource_limits %}
            limits:
              {% if stage.resource_limits.memory %}
              memory: {{ stage.resource_limits.memory }}
              {% endif %}
              {% if stage.resource_limits.cpu %}
              cpu: {{ stage.resource_limits.cpu }}
              {% endif %}
              {% if stage.resource_limits.gpu %}
              nvidia.com/gpu: {{ stage.resource_limits.gpu }}
              {% endif %}
            {% endif %}
            {% if stage.resource_requests %}
            requests:
              {% if stage.resource_requests.memory %}
                memory: {{ stage.resource_requests.memory }}
              {% endif %}
              {% if stage.resource_requests.cpu %}
                cpu: {{ stage.resource_requests.cpu }}
              {% endif %}
              {% if stage.resource_requests.gpu %}
                nvidia.com/gpu: {{ stage.resource_requests.gpu }}
              {% endif %}
            {% endif %}
          {% endif %}
          {% if stage.env %}
          env:
            {% for env_var in stage.env %}
            - name: {{ env_var['name'] }}
              value: "{{ env_var['value'] }}"
            {% endfor %}
          {% endif %}
          {% if stage.env_from %}
          envFrom:
            {% for env_from_elem in stage.env_from %}
            - {{ env_from_elem['key'] }}:
                {{ env_from_elem['body'] }}
            {% endfor %}
          {% endif %}
      {% endfor %}
      containers:
        - name: validate-trial-results
          image: {{ validation.image_name }}
          {% if validation %}
          command: ["python3", "-m"]
          args:
            - exp_pipeline.results.main
          {% if validation.volume_mounts %}
          volumeMounts:
            {% for volume_mount in validation.volume_mounts %}
            - name: {{ volume_mount['volume_name'] }}
              mountPath: {{ volume_mount['mount_path'] }}
              {% if volume_mount['read_only'] %}
              readOnly: True
              {% endif %}
              {% if volume_mount['sub_path'] %}
              subPath: {{ volume_mount['sub_path'] }}
              {% endif %}
            {% endfor %}
          {% endif %}
          {% if validation.resource_limits %}
          resources:
            limits:
              {% if validation.resource_limits.memory %}
              memory: {{ validation.resource_limits.memory }}
              {% endif %}
              {% if validation.resource_limits.cpu %}
              cpu: {{ validation.resource_limits.cpu }}
              {% endif %}
              {% if validation.resource_limits.gpu %}
              nvidia.com/gpu: {{ validation.resource_limits.gpu }}
              {% endif %}
          {% endif %}
          {% if validation.env %}
          env:
            {% for env_var in validation.env %}
            - name: {{ env_var[ 'name' ] }}
              value: "{{ env_var['value'] }}"
            {% endfor %}
          {% endif %}
          {% if validation.env_from %}
          envFrom:
            {% for env_from_elem in validation.env_from %}
            - {{ env_from_elem['key'] }}:
                {{ env_from_elem['body'] }}
            {% endfor %}
          {% endif %}
          {% else %}
          command: [ "python3", "-c" ]
          args:
            - print('Results validation disabled')
          {% endif %}
      {% if volumes %}
      volumes:
        {% for volume in volumes %}
        - name: {{ volume['volume_name'] }}
          {% if volume['pvc_name'] %}
          persistentVolumeClaim:
            claimName: {{ volume['pvc_name'] }}
          {% endif %}
          {% if volume['secret_name'] %}
          secret:
            secretName: {{ volume['secret_name'] }}
            {% if volume['volume_items'] %}
            items:
            {% for vi in volume['volume_items'] %}
              - key: "{{ vi.key }}"
                path: "{{ vi.path }}"
            {% endfor %}
            {% endif %}
          {% endif %}
          {% if volume['configmap_name'] %}
          configMap:
            name: {{ volume['configmap_name'] }}
            {% if volume['volume_items'] %}
            items:
            {% for vi in volume['volume_items'] %}
              - key: "{{ vi.key }}"
                path: "{{ vi.path }}"
            {% endfor %}
            {% endif %}
          {% endif %}
        {% endfor %}
      {% endif %}
      {% if image_reg_creds %}
      imagePullPolicy: Always
      imagePullSecrets:
        {% for secret_name in image_reg_creds %}
        - name: {{ secret_name }}
        {% endfor %}
      {% endif %}
      {% if tolerations %}
      tolerations:
        {% for tol in tolerations %}
        - key: "{{ tol.key }}"
          operator: "{{ tol.operator }}"
          value: "{{ tol.value }}"
          effect: "{{ tol.effect }}"
        {% endfor %}
      {% endif %}
      {% if node_selector %}
      nodeSelector:
      {% for key, value in node_selector.items() %}
        {{ key }}: "{{ value }}"
      {% endfor %}
      {% endif %}
      {% if node_affinity %}
      affinity:
        nodeAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:
            nodeSelectorTerms:
              - matchExpressions:
                  {% for e in node_affinity.expressions %}
                  - key: {{ e.key }}
                    operator: {{ e.operator }}
                    values:
                      {% for v in e.vals %}
                      - "{{ v }}"
                      {% endfor %}
                  {% endfor %}
      {% endif %}