AttributeError: 'ComputeRecurrenceSchedule' object has no attribute '_to_rest_compute_pattern_object'

[ABrethome]

• Package Name: azure-ai-ml
• Package Version: 1.25.0
• Operating System: Linux abretho-sbx-dsteam-c02 5.15.0-1040-azure 47~20.04.1-Ubuntu SMP Fri Jun 2 21:38:08 UTC 2023 x86_64 x86_64 x86_64 GNU/Linux (generated from uname -a)
• Python Version: 3.11.11

Describe the bug
First, I create a compute instance on behalf of a user using the MLClient begin_create_or_update() with specific compute instance configuration (script, schedule, etc.).
Second, I retrieve the created compute object from the poller of begin_create_or_update(), then I update the compute with new parameters (in my case, enable SSO after creation, as I can't enable SSO during creation).
When I run the update of the compute instance with a second begin_create_or_update(), I get the following error:

AttributeError: 'ComputeRecurrenceSchedule' object has no attribute '_to_rest_compute_pattern_object'

To Reproduce
Steps to reproduce the behavior:

1. Create compute instance object:

    # AssignedUserConfiguration is the class we need to pass
    # when creating compute to tell AML who to assign the compute to
    assigned_user_config = AssignedUserConfiguration(
        user_tenant_id=<MY_USER_TENANT_ID>, user_object_id=<MY_USER_OID>
    )

    # Create shutdown schedule
    # This function is not very well documented, especially what the parameter `start_time` does - from
    # limited trial/error so far it seems to be required, but doesn't actually have an effect on the schedule.
    # The RecurrencePattern is what sets the timing of the schedule.
    stop_schedule = ComputeStartStopSchedule(
        trigger=RecurrenceTrigger(
            start_time=datetime.now().isoformat(
                timespec="seconds"
            ),  # Required with timezone. str with ISO 8601 date format. Naive datetime.
            time_zone=<MY_USER_TIMEZONE>,  # Required with schedule. str or TimeZone
            frequency="day",  # str, e.g. "minute", "hour", "day", "week", "month"
            interval=1,
            schedule=RecurrencePattern(
                hours=20, minutes=0
            ),  # Specifies shutdown at 20:00 each day
        ),
        action=ComputePowerAction.STOP,
        state=ScheduleState.ENABLED,
    )
    compute_schedule = ComputeSchedules(compute_start_stop=[stop_schedule])

    # Configure the setup script
    setup_scripts = SetupScripts(
        creation_script=ScriptReference(
            path=<MY_USER_SCRIPT_PATH>,  # path within the workspace where the compute will be created.
            timeout_minutes=25,  # 25min is the maximum. Any higher time limit will throw an error.
        )
    )

    # Configure the compute
    new_compute = ComputeInstance(
        name=computename,
        description=f"Compute instance created on behalf of {user.name}.",
        size=compute_size.name,  # str
        create_on_behalf_of=assigned_user_config,
        # NetworkSettings doc is bad. Look at code directly (CTRL+F NetworkSettings)
        # https://github.com/Azure/azure-sdk-for-python/blob/main/sdk/ml/azure-ai-ml/azure/ai/ml/entities/_compute/compute.py
        # provide the full URI for subnet and leave vnet_name as None.
        network_settings=NetworkSettings(
            vnet_name=None, subnet=workspace.get_full_uri_subnet()
        ),
        ssh_public_access_enabled=False,
        schedules=compute_schedule,
        idle_time_before_shutdown_minutes=65,
        enable_node_public_ip=False,
        setup_scripts=setup_scripts,
        enable_sso=False,  # By design, AzureML does not allow to enable SSO while creating a compute on behalf of a user
        enable_root_access=True,
    )

ComputeInstance object is successfully init.

2. Create the compute with the MLClient:

    # create the compute instance
    poller = ml_client.begin_create_or_update(new_compute)

    # wait for the operation to complete, returns the created ComputeInstance object
    created_compute: ComputeInstance = poller.result()

    # Check the status of the operation
    if created_compute.provisioning_state != "Succeeded":
        message = f"Compute instance creation of '{computename}' failed. See compute instance in AzureML workspace for more details."
        raise ComputeError(message)

Compute is successfully created.

3. Try enabling SSO (for instance) afterwards:

    # update sso to true (cannot be done at compute creation)
    created_compute.enable_sso = True

    # update the compute instance
    poller = ml_client.begin_create_or_update(created_compute)

    # wait for the operation to complete, returns the created ComputeInstance object
    updated_compute: ComputeInstance = poller.result()

Throws error:
AttributeError: 'ComputeRecurrenceSchedule' object has no attribute '_to_rest_compute_pattern_object'

This confuses me, as the error is related to the schedule, which I didn't touch at that stage.

4. Note: this doesn't work either:

   created_compute = ml_client.compute.get(name=created_compute.name)

    # update sso to true (cannot be done at compute creation)
    created_compute.enable_sso = True

    # update the compute instance
    poller = ml_client.begin_create_or_update(created_compute)

    # wait for the operation to complete, returns the created ComputeInstance object
    updated_compute: ComputeInstance = poller.result()

Throws error:
AttributeError: 'ComputeRecurrenceSchedule' object has no attribute '_to_rest_compute_pattern_object'

This confuses me, as the error is related to the schedule, which I didn't touch at that stage.

Expected behavior
I expect to be able to get a Compute(Instance) object returned from the poller or the compute operations, modify a parameter, then update compute with begin_create_or_update() without having errors.

Screenshots
Not applicable.

Additional context

1. Full error:

  File "/mnt/batch/tasks/shared/LS_root/mounts/clusters/abretho-sbx-dsteam-c02/code/Users/RA_ABRETHO/Git/AML_Workspace_Management/src/inf_aml_mgt/computes.py", line 275, in update_compute_instance_sso
    poller = ml_client.begin_create_or_update(created_compute)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/anaconda/envs/inf-ai-mgt-dev/lib/python3.11/site-packages/azure/ai/ml/_ml_client.py", line 1292, in begin_create_or_update
    return _begin_create_or_update(entity, self._operation_container.all_operations, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/anaconda/envs/inf-ai-mgt-dev/lib/python3.11/functools.py", line 909, in wrapper
    return dispatch(args[0].__class__)(*args, **kw)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/anaconda/envs/inf-ai-mgt-dev/lib/python3.11/site-packages/azure/ai/ml/_ml_client.py", line 1386, in _
    return operations[AzureMLResourceType.COMPUTE].begin_create_or_update(entity, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/anaconda/envs/inf-ai-mgt-dev/lib/python3.11/site-packages/azure/core/tracing/decorator.py", line 116, in wrapper_use_tracer
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/anaconda/envs/inf-ai-mgt-dev/lib/python3.11/site-packages/azure/ai/ml/_telemetry/activity.py", line 288, in wrapper
    return f(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^
  File "/anaconda/envs/inf-ai-mgt-dev/lib/python3.11/site-packages/azure/ai/ml/operations/_compute_operations.py", line 182, in begin_create_or_update
    compute_rest_obj = compute._to_rest_object()
                       ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/anaconda/envs/inf-ai-mgt-dev/lib/python3.11/site-packages/azure/ai/ml/entities/_compute/compute_instance.py", line 322, in _to_rest_object
    compute_instance_prop.schedules = self.schedules._to_rest_object() if self.schedules else None
                                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/anaconda/envs/inf-ai-mgt-dev/lib/python3.11/site-packages/azure/ai/ml/entities/_compute/_schedule.py", line 138, in _to_rest_object
    rest_schedules.append(schedule._to_rest_object())
                          ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/anaconda/envs/inf-ai-mgt-dev/lib/python3.11/site-packages/azure/ai/ml/entities/_compute/_schedule.py", line 82, in _to_rest_object
    rest_object.recurrence = self.trigger._to_rest_compute_recurrence_object()
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/anaconda/envs/inf-ai-mgt-dev/lib/python3.11/site-packages/azure/ai/ml/entities/_schedule/trigger.py", line 276, in _to_rest_compute_recurrence_object
    schedule=self.schedule._to_rest_compute_pattern_object(),
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'ComputeRecurrenceSchedule' object has no attribute '_to_rest_compute_pattern_object'

2. Tactical fix by looking at the doc:
   Taken from: https://learn.microsoft.com/en-us/python/api/azure-ai-ml/azure.ai.ml.operations.computeoperations?view=azure-python#azure-ai-ml-operations-computeoperations-enable-sso

    ml_client.compute.enable_sso(
        name=created_compute.name,
        enable_sso=True
    )

This is experimental though, so not really keen on relying on this long term.

[github-actions]

Thanks for the feedback! We are routing this to the appropriate team for follow-up. cc @Azure/azure-ml-sdk @azureml-github.