xpk 0.7.2__py3-none-any.whl → 0.9.0__py3-none-any.whl

xpk/commands/workload.py

@@ -14,20 +14,22 @@ See the License for the specific language governing permissions and
 limitations under the License.
 """
 
-from ..core.blueprint.blueprint_generator import get_subnetworks_for_a3mega, get_subnetworks_for_a3ultra
 from ..core.cluster import (
+    XPK_SA,
     create_xpk_k8s_service_account,
     get_cluster_credentials,
     setup_k8s_env,
-    XPK_SA,
 )
 from ..core.commands import run_command_with_updates, run_commands
-from ..core.config import VERTEX_TENSORBOARD_FEATURE_FLAG, XPK_CURRENT_VERSION, parse_env_config
+from ..core.config import (
+    VERTEX_TENSORBOARD_FEATURE_FLAG,
+    XPK_CURRENT_VERSION,
+    parse_env_config,
+)
 from ..core.docker_container import (
     get_main_container_docker_image,
     get_user_workload_container,
 )
-
 from ..core.docker_resources import get_volumes
 from ..core.gcloud_context import add_zone_and_project
 from ..core.kueue import LOCAL_QUEUE_NAME
@@ -36,54 +38,58 @@ from ..core.nap import (
     get_autoprovisioning_node_selector_args,
     is_autoprovisioning_enabled,
 )
+from ..core.network import get_cluster_subnetworks
 from ..core.pathways import (
+    append_custom_colocated_python_sidecar,
+    append_custom_pathways_proxy_server,
+    append_custom_pathways_server,
+    append_custom_pathways_worker,
+    check_if_pathways_job_is_installed,
     ensure_pathways_workload_prerequisites,
-    get_pathways_proxy_args,
-    get_pathways_rm_args,
-    get_pathways_sidecar_container,
     get_pathways_unified_query_link,
-    get_pathways_worker_args,
     get_user_workload_for_pathways,
+    try_to_delete_pathwaysjob_first,
 )
 from ..core.resources import CLUSTER_METADATA_CONFIGMAP, get_cluster_configmap
 from ..core.scheduling import (
     check_if_workload_can_schedule,
     create_accelerator_label,
     create_machine_label,
+    create_tpu_machine_type,
+    create_tpu_topology,
     get_cpu_affinity,
     get_gpu_scheduler,
 )
 from ..core.storage import (
-    GCS_FUSE_TYPE,
+    GCE_PD_TYPE,
     GCP_FILESTORE_TYPE,
+    GCS_FUSE_TYPE,
+    PARALLELSTORE_TYPE,
     Storage,
     add_bucket_iam_members,
-    get_storage_volume_mounts_yaml,
-    get_storage_volumes_yaml,
+    get_storage_annotations,
     get_storages_to_mount,
-    get_storage_volume_mounts_yaml_for_gpu,
-    get_storage_volumes_yaml_for_gpu,
-    GCS_FUSE_ANNOTATION,
 )
 from ..core.system_characteristics import (
     AcceleratorType,
-    AcceleratorTypeToAcceleratorCharacteristics,
     get_system_characteristics,
 )
 from ..core.vertex import create_vertex_experiment
 from ..core.workload import (
+    add_gpu_rxdm_container,
     check_if_workload_exists,
-    get_gpu_rxdm_cmd,
-    get_gpu_rxdm_image,
-    get_gpu_tcp_volume,
-    get_gpu_volume,
     get_workload_list,
     wait_for_job_completion,
     zone_to_region,
 )
-from ..core.workload_decorators import rdma_decorator, tcpxo_decorator, storage_decorator
+from ..core.workload_decorators import (
+    rdma_decorator,
+    storage_decorator,
+    tcpxo_decorator,
+)
 from ..utils.console import get_user_input, xpk_exit, xpk_print
 from ..utils.file import write_tmp_file
+from .common import is_TAS_possible
 from . import cluster_gcluster
 
 WORKLOAD_CREATE_YAML = """apiVersion: jobset.x-k8s.io/v1alpha2
@@ -139,7 +145,8 @@ GPU_WORKLOAD_CREATE_YAML = """apiVersion: jobset.x-k8s.io/v1alpha2
 kind: JobSet
 metadata:
   name: {args.workload}
-  annotations: {storage_annotations}
+  annotations:
+    {storage_annotations}
   labels:
     kueue.x-k8s.io/queue-name: multislice-queue  # Name of the LocalQueue
     xpk.google.com/workload: {args.workload}
@@ -176,29 +183,8 @@ spec:
               - operator: "Exists"
                 key: nvidia.com/gpu
               volumes:
-              {gpu_volume}
-              {storage_volumes}
+              {volumes}
               containers:
-              {gpu_rxdm_image}
-                imagePullPolicy: Always
-                command:
-                - "bash"
-                - "-c"
-                - |
-                  {gpu_rxdm_cmd} &
-                  while [ ! -e "/usr/share/workload/workload_terminated" ]; do sleep 10; echo "sleeping"; done
-                securityContext:
-                  privileged: true
-                volumeMounts:
-                {gpu_tcp_volume}
-                {storage_volume_mounts}
-                - name: nvidia-install-dir-host
-                  mountPath: /usr/local/nvidia/lib64
-                - name: workload-terminated-volume
-                  mountPath: /usr/share/workload
-                env:
-                - name: LD_LIBRARY_PATH
-                  value: /usr/local/nvidia/lib64
               {container}
 """
 
@@ -228,7 +214,7 @@ spec:
               labels:
                 xpk.google.com/workload: {args.workload}
               annotations:
-                kueue.x-k8s.io/podset-preferred-topology: "cloud.google.com/gce-topology-host"
+                {kueue_TAS_annotation}
             spec:
               priorityClassName: {args.priority}
               restartPolicy: Never
@@ -241,219 +227,37 @@ spec:
               containers:
               {container}
 """
-
-PW_WORKLOAD_CREATE_YAML = """apiVersion: jobset.x-k8s.io/v1alpha2
-kind: JobSet
-metadata:
-  name: {args.workload}
-  labels:
-    kueue.x-k8s.io/queue-name: {local_queue_name}  # Name of the LocalQueue
-    xpk.google.com/workload: {args.workload}
-spec:
-  ttlSecondsAfterFinished: {args.ttl_seconds_after_finished}
-  failurePolicy:
-    {failure_policy_rules}
-    maxRestarts: {args.max_restarts}
-  successPolicy:
-    operator: "All"
-    targetReplicatedJobs:
-    - {args.targetReplicatedJob}
-  replicatedJobs:
-    - name: worker
-      replicas: {args.num_slices}
-      template:
-        metadata:
-          annotations:
-            alpha.jobset.sigs.k8s.io/exclusive-topology: cloud.google.com/gke-nodepool
-          labels:
-            xpk.google.com/workload: {args.workload}
-        spec:
-          backoffLimit: {backoff_limit}
-          completions: {system.vms_per_slice}
-          parallelism: {system.vms_per_slice}
-          template:
-            metadata:
-              annotations:
-                {storage_annotations}
-            spec:
-              terminationGracePeriodSeconds: {args.termination_grace_period_seconds}
-              serviceAccountName: {service_account}
-              containers:
-              - args:
-                {pathways_worker_args}
-                image: {args.server_image}
-                imagePullPolicy: Always
-                name: pathways-worker
-                ports:
-                - containerPort: 29001
-                - containerPort: 8471
-                - containerPort: 8080
-                resources:
-                  limits:
-                    {resource_type}: {system.chips_per_vm}
-                securityContext:
-                  privileged: true
-                volumeMounts:
-                - mountPath: /tmp
-                  name: shared-tmp
-                {storage_volume_mounts}
-                env:
-                  - name: PROJECT_ID
-                    value: {args.project}
-                  - name: LOCATION
-                    value: {args.zone}
-                  - name: CLUSTER_NAME
-                    value: {args.cluster}
-                  - name: POD_NAME
-                    valueFrom:
-                      fieldRef:
-                        fieldPath: metadata.name
-                  - name: CONTAINER_NAME
-                    value: "pathways-worker"
-                  - name: NAMESPACE
-                    valueFrom:
-                      fieldRef:
-                        fieldPath: metadata.namespace
-                  # Workaround for v6e
-                  - name: MEGASCALE_GRPC_ENABLE_XOR_TRACER
-                    value: "false"
-                  - name: MEGASCALE_NUM_SLICES
-                    valueFrom:
-                      fieldRef:
-                        fieldPath: "metadata.labels['jobset.sigs.k8s.io/replicatedjob-replicas']"
-                  - name: JOBSET_NAME
-                    valueFrom:
-                      fieldRef:
-                        fieldPath: metadata.annotations['jobset.sigs.k8s.io/jobset-name']
-                  - name: REPLICATED_JOB_NAME
-                    valueFrom:
-                      fieldRef:
-                        fieldPath: metadata.annotations['jobset.sigs.k8s.io/replicatedjob-name']
-                  - name: MEGASCALE_SLICE_ID
-                    valueFrom:
-                      fieldRef:
-                        fieldPath: "metadata.labels['jobset.sigs.k8s.io/job-index']"
-                  - name: MEGASCALE_COORDINATOR_ADDRESS
-                    value: "$(JOBSET_NAME)-$(REPLICATED_JOB_NAME)-$(MEGASCALE_SLICE_ID)-0.$(JOBSET_NAME)"
-              {pathways_sidecar_container}
-              nodeSelector:
-                {accelerator_label}
-                {machine_label}
-                {autoprovisioning_args}
-              priorityClassName: {args.priority}
-              hostNetwork: true
-              dnsPolicy: ClusterFirstWithHostNet
-              volumes:
-              - hostPath:
-                  path: /tmp
-                  type: DirectoryOrCreate
-                name: shared-tmp
-              {storage_volumes}
-    - name: rm
-      replicas: 1
-      template:
-        metadata:
-          labels:
-            xpk.google.com/workload: {args.workload}
-        spec:
-          backoffLimit: 0
-          completions: 1
-          parallelism: 1
-          template:
-            spec:
-              containers:
-              - args:
-                {pathways_rm_args}
-                env:
-                - name: PROJECT_ID
-                  value: {args.project}
-                - name: LOCATION
-                  value: {args.zone}
-                - name: CLUSTER_NAME
-                  value: {args.cluster}
-                - name: POD_NAME
-                  valueFrom:
-                    fieldRef:
-                      fieldPath: metadata.name
-                - name: CONTAINER_NAME
-                  value: "pathways-rm"
-                - name: NAMESPACE
-                  valueFrom:
-                    fieldRef:
-                      fieldPath: metadata.namespace
-                - name: REPLICATED_JOB_NAME
-                  valueFrom:
-                    fieldRef:
-                      fieldPath: metadata.annotations['jobset.sigs.k8s.io/replicatedjob-name']
-                - name: JOBSET_NAME
-                  valueFrom:
-                    fieldRef:
-                      fieldPath: metadata.annotations['jobset.sigs.k8s.io/jobset-name']
-                - name: HOST_ADDRESS
-                  value: $(JOBSET_NAME)-$(REPLICATED_JOB_NAME)-0-0.$(JOBSET_NAME)
-                - name: TPU_SKIP_MDS_QUERY
-                  value: "true"
-                image: {args.server_image}
-                imagePullPolicy: Always
-                name: pathways-rm
-                ports:
-                - containerPort: 29001
-                securityContext:
-                  privileged: true
-                volumeMounts:
-                - mountPath: /tmp
-                  name: shared-tmp
-              nodeSelector:
-                cloud.google.com/gke-nodepool: cpu-rm-np
-              hostNetwork: true
-              dnsPolicy: ClusterFirstWithHostNet
-              volumes:
-              - hostPath:
-                  path: /tmp
-                  type: DirectoryOrCreate
-                name: shared-tmp
-    - name: proxy
-      replicas: 1
-      template:
-        metadata:
-          labels:
-            xpk.google.com/workload: {args.workload}
-        spec:
-          backoffLimit: 0
-          completions: 1
-          parallelism: 1
-          template:
-            spec:
-              containers:
-              - args:
-                {pathways_proxy_args}
-                env:
-                - name: PROJECT_ID
-                  value: {args.project}
-                - name: LOCATION
-                  value: {args.zone}
-                - name: CLUSTER_NAME
-                  value: {args.cluster}
-                - name: POD_NAME
-                  valueFrom:
-                    fieldRef:
-                      fieldPath: metadata.name
-                - name: CONTAINER_NAME
-                  value: "pathways-proxy"
-                - name: NAMESPACE
-                  valueFrom:
-                    fieldRef:
-                      fieldPath: metadata.namespace
-                image: {args.proxy_server_image}
-                imagePullPolicy: Always
-                name: pathways-proxy
-                ports:
-                - containerPort: 29000
-              hostNetwork: true
-              dnsPolicy: ClusterFirstWithHostNet
-              nodeSelector:
-                cloud.google.com/gke-nodepool: cpu-proxy-np
-    {user_workload}
+# The indentation of PW_WORKLOAD_CREATE_YAML is intentional to allow reusing the user workload container YAML.
+PW_WORKLOAD_CREATE_YAML = """
+    apiVersion: pathways-job.pathways.domain/v1
+    kind: PathwaysJob
+    metadata:
+      name: {args.workload}
+      labels:
+        kueue.x-k8s.io/queue-name: {local_queue_name}  # Name of the LocalQueue
+        xpk.google.com/workload: {args.workload}
+    spec:
+      maxRestarts: {args.max_restarts}
+      customComponents:
+      {custom_pathways_proxy_server}
+      {custom_pathways_server}
+      {custom_pathways_worker}
+      {colocated_python_sidecar}
+      workers:
+      - type: {machine_type}
+        topology: {topology}
+        numSlices: {args.num_slices}
+        maxSliceRestarts: {args.max_slice_restarts}
+        terminationGracePeriodSeconds: {args.termination_grace_period_seconds}
+        priorityClassName: {args.priority}
+      pathwaysDir: {args.pathways_gcs_location} #This bucket needs to be created in advance.
+      controller:
+        # #Pod template for training, default mode.
+        deploymentMode: default
+        mainContainerName: {args.docker_name}
+        elasticSlices: {args.elastic_slices}
+        template:
+      {user_workload}
 """
 
 
@@ -545,7 +349,6 @@ def workload_create(args) -> None:
 
   parse_env_config(args, tensorboard_config, system)
 
-  # Currently autoprovisioning is not enabled for Pathways workloads.
   autoprovisioning_args = ''
   autoprovisioning_enabled, return_code = is_autoprovisioning_enabled(
       args, system
@@ -560,28 +363,72 @@ def workload_create(args) -> None:
     if return_code != 0:
       xpk_exit(return_code)
 
-  storages: list[Storage] = get_storages_to_mount(k8s_api_client, args.storage)
-  gcs_fuse_storages = list(
-      filter(lambda storage: storage.type == GCS_FUSE_TYPE, storages)
-  )
-  gcpfilestore_storages: list[Storage] = list(
-      filter(lambda storage: storage.type == GCP_FILESTORE_TYPE, storages)
-  )
-  storage_annotations = ''
   service_account = ''
-  if len(gcs_fuse_storages) > 0:
-    storage_annotations = GCS_FUSE_ANNOTATION
-    service_account = XPK_SA
-    xpk_print(f'Detected gcsfuse Storages to add: {gcs_fuse_storages}')
-  else:
-    xpk_print('No gcsfuse Storages to add detected')
-  failure_policy_rules = """rules:
+  all_storages = []
+  # Currently storage customization is not supported for Pathways workloads. b/408468941
+  if not args.use_pathways:
+    storages: list[Storage] = get_storages_to_mount(
+        k8s_api_client, args.storage
+    )
+    gcs_fuse_storages = list(
+        filter(lambda storage: storage.type == GCS_FUSE_TYPE, storages)
+    )
+    gcpfilestore_storages: list[Storage] = list(
+        filter(lambda storage: storage.type == GCP_FILESTORE_TYPE, storages)
+    )
+    parallelstore_storages: list[Storage] = list(
+        filter(lambda storage: storage.type == PARALLELSTORE_TYPE, storages)
+    )
+    pd_storages: list[Storage] = list(
+        filter(lambda storage: storage.type == GCE_PD_TYPE, storages)
+    )
+    if len(gcs_fuse_storages) > 0:
+      service_account = XPK_SA
+      xpk_print(f'Detected gcsfuse Storages to add: {gcs_fuse_storages}')
+    else:
+      xpk_print('No gcsfuse Storages to add detected')
+
+    if len(gcpfilestore_storages) > 0:
+      service_account = XPK_SA
+      xpk_print(
+          f'Detected gcp filestores instances to add: {gcpfilestore_storages}'
+      )
+    else:
+      xpk_print('No gcp filestore instances to add detected.')
+
+    if len(parallelstore_storages) > 0:
+      service_account = XPK_SA
+      xpk_print(
+          'Detected gcp parallelstore instances to add:'
+          f' {parallelstore_storages}'
+      )
+    else:
+      xpk_print('No gcp parallelstore instances to add detected.')
+
+    if len(pd_storages) > 0:
+      service_account = XPK_SA
+      xpk_print(f'Detected gce persistent disk instances to add: {pd_storages}')
+    else:
+      xpk_print('No gce persistent disk instances to add detected.')
+
+    all_storages = (
+        gcs_fuse_storages
+        + gcpfilestore_storages
+        + parallelstore_storages
+        + pd_storages
+    )
+
+  # Currently failure policy rules are supported for Pathways workloads. b/408465881
+  failure_policy_rules = ''
+  pod_failure_policy = ''
+  if not args.use_pathways:
+    failure_policy_rules = """rules:
       - action: FailJobSet
-        onJobFailureReasons: 
+        onJobFailureReasons:
         - PodFailurePolicy"""
-  restart_on_exit_codes = get_restart_exit_codes(args)
-  restart_on_exit_codes = ','.join(map(str, restart_on_exit_codes))
-  pod_failure_policy = f"""
+    restart_on_exit_codes = get_restart_exit_codes(args)
+    restart_on_exit_codes = ','.join(map(str, restart_on_exit_codes))
+    pod_failure_policy = f"""
           podFailurePolicy:
             rules:
             - action: FailJob
@@ -590,14 +437,6 @@ def workload_create(args) -> None:
                 operator: NotIn
                 values: [{restart_on_exit_codes}]"""
 
-  if len(gcpfilestore_storages) > 0:
-    xpk_print(
-        f'Detected gcp filestores instances to add: {gcpfilestore_storages}'
-    )
-    service_account = XPK_SA
-  else:
-    xpk_print('No gcp filestore instances to add detected.')
-  all_storages = gcs_fuse_storages + gcpfilestore_storages
   # Create the workload file based on accelerator type or workload type.
   if system.accelerator_type == AcceleratorType['GPU']:
     container, debugging_dashboard_id = get_user_workload_container(
@@ -609,6 +448,13 @@ def workload_create(args) -> None:
     if return_code != 0:
       xpk_exit(return_code)
 
+    kueue_TAS_annotation = (
+        'kueue.x-k8s.io/podset-preferred-topology:'
+        ' "cloud.google.com/gce-topology-host"'
+    )
+    if not is_TAS_possible(args):
+      kueue_TAS_annotation = ''
+
     if system.device_type in cluster_gcluster.supported_device_types:
       yml_string = A3_GPU_WORKLOAD_CREATE_YAML.format(
           args=args,
@@ -616,38 +462,34 @@ def workload_create(args) -> None:
           service_account=XPK_SA,
           failure_policy_rules=failure_policy_rules,
           pod_failure_policy=pod_failure_policy,
+          kueue_TAS_annotation=kueue_TAS_annotation,
       )
 
+      sub_networks = get_cluster_subnetworks(args)
       if args.device_type == cluster_gcluster.a3mega_device_type:
-        sub_networks = get_subnetworks_for_a3mega(args.cluster)
         yml_string = tcpxo_decorator.decorate_jobset(yml_string, sub_networks)
-
-      if args.device_type == cluster_gcluster.a3ultra_device_type:
-        sub_networks = get_subnetworks_for_a3ultra(args.cluster)
+      elif args.device_type in [
+          cluster_gcluster.a3ultra_device_type,
+          cluster_gcluster.a4_device_type,
+      ]:
         yml_string = rdma_decorator.decorate_jobset(yml_string, sub_networks)
 
-      if len(gcs_fuse_storages) + len(gcpfilestore_storages) > 0:
+      if all_storages:
         yml_string = storage_decorator.decorate_jobset(yml_string, all_storages)
     else:
       yml_string = GPU_WORKLOAD_CREATE_YAML.format(
           args=args,
           container=container,
-          command=args.command,
-          chips_per_vm=system.chips_per_vm,
           gpu_scheduler=gpu_scheduler,
-          gpu_volume=get_gpu_volume(system),
-          gpu_rxdm_image=get_gpu_rxdm_image(system),
-          gpu_rxdm_cmd=get_gpu_rxdm_cmd(system),
-          gpu_tcp_volume=get_gpu_tcp_volume(system),
-          storage_volumes=get_storage_volumes_yaml_for_gpu(all_storages),
-          storage_volume_mounts=get_storage_volume_mounts_yaml_for_gpu(
-              all_storages
+          volumes=get_volumes(args, system),
+          storage_annotations=('\n' + (' ' * 12)).join(
+              get_storage_annotations(all_storages)
           ),
-          storage_annotations=storage_annotations,
           service_account=service_account,
           failure_policy_rules=failure_policy_rules,
           pod_failure_policy=pod_failure_policy,
       )
+      yml_string = add_gpu_rxdm_container(yml_string, system, all_storages)
 
   elif args.use_pathways and ensure_pathways_workload_prerequisites(
       args, system
@@ -655,29 +497,14 @@ def workload_create(args) -> None:
     yml_string = PW_WORKLOAD_CREATE_YAML.format(
         args=args,
         system=system,
-        accelerator_label=create_accelerator_label(
-            system.accelerator_type, system
-        ),
-        machine_label=create_machine_label(system.accelerator_type, system),
-        pathways_worker_args=get_pathways_worker_args(args),
-        pathways_proxy_args=get_pathways_proxy_args(args),
-        pathways_sidecar_container=get_pathways_sidecar_container(args),
-        user_workload=get_user_workload_for_pathways(
-            args, system, pod_failure_policy, storages
-        ),
-        resource_type=AcceleratorTypeToAcceleratorCharacteristics[
-            system.accelerator_type
-        ].resource_type,
+        topology=create_tpu_topology(system.accelerator_type, system),
+        machine_type=create_tpu_machine_type(system.accelerator_type, system),
+        custom_pathways_proxy_server=append_custom_pathways_proxy_server(args),
+        custom_pathways_server=append_custom_pathways_server(args),
+        custom_pathways_worker=append_custom_pathways_worker(args),
+        colocated_python_sidecar=append_custom_colocated_python_sidecar(args),
+        user_workload=get_user_workload_for_pathways(args, system),
         local_queue_name=LOCAL_QUEUE_NAME,
-        autoprovisioning_args=autoprovisioning_args,
-        backoff_limit=system.vms_per_slice * 4,
-        storage_annotations=storage_annotations,
-        storage_volumes=get_storage_volumes_yaml(all_storages),
-        storage_volume_mounts=get_storage_volume_mounts_yaml(all_storages),
-        pathways_rm_args=get_pathways_rm_args(args, system),
-        service_account=service_account,
-        failure_policy_rules=failure_policy_rules,
-        pod_failure_policy=pod_failure_policy,
     )
   else:
     container, debugging_dashboard_id = get_user_workload_container(
@@ -695,7 +522,9 @@ def workload_create(args) -> None:
         local_queue_name=LOCAL_QUEUE_NAME,
         autoprovisioning_args=autoprovisioning_args,
         volumes=get_volumes(args, system),
-        storage_annotations=storage_annotations,
+        storage_annotations=('\n' + (' ' * 16)).join(
+            get_storage_annotations(all_storages)
+        ),
         service_account=service_account,
         failure_policy_rules=failure_policy_rules,
         pod_failure_policy=pod_failure_policy,
@@ -708,7 +537,9 @@ def workload_create(args) -> None:
     xpk_print(f'Create Workload request returned ERROR {return_code}')
     xpk_exit(return_code)
 
-  add_bucket_iam_members(args, storages)
+  if not args.use_pathways:
+    add_bucket_iam_members(args, storages)
+
   # Get GKE outlier dashboard for TPU
   outlier_dashboard_id = None
   if system.accelerator_type == AcceleratorType['TPU']:
@@ -833,6 +664,12 @@ def workload_delete(args) -> None:
   elif not will_delete:
     xpk_print('Skipping delete command.')
   else:
+    # If PathwaysJob exists, delete it.
+    if check_if_pathways_job_is_installed(
+        args
+    ) and try_to_delete_pathwaysjob_first(args, workloads):
+      xpk_exit(0)
+    # PathwaysJob workload does not exist, delete JobSet
     commands = []
     task_names = []
     for workload in workloads:

xpk/core/blueprint/blueprint_definitions.py

@@ -60,3 +60,5 @@ class Blueprint:
   toolkit_modules_url: str
   toolkit_modules_version: str
   vars: dict[str, str | list[str]] | None
+  terraform_providers: Optional[dict[str, Any]] = None
+  validators: Optional[list[Any]] = None