Deployment using OCI Data Science Job¶

ML Monitoring Application can be quickly set up using OCI Data Science Job and ML Monitoring Application managed container image.

Configure required policy and dynamic rules
Provision OCI Data Science Job
Trigger Data science job run passing the required runtime configuration

Add Dynamic Groups Rules¶

User needs to create Dynamic Groups and add following rules for setting up OCI resources

ALL {resource.type='datasciencejobrun', resource.compartment.id = '<COMPARTMENT_ID>'}

Here COMPARTMENT_ID is the OCID of Compartment in which user need to setup OCI Data Science Job resource and push logs for ML Monitoring Application run .

Add Policy Rules¶

Policy rules required to be added for authorizing and giving access permission to OCI resources within user’s compartment via dynamic-group . Here user is giving required permission using following policy to all the resources defined in the dynamic group created above.

Allow dynamic-group <ml-monitoring-application-dynamic-group> to manage log-groups in compartment <COMPARTMENT NAME>
Allow dynamic-group <ml-monitoring-application-dynamic-group> to use log-content in compartment <COMPARTMENT NAME>
Allow dynamic-group <ml-monitoring-application-dynamic-group> to manage objects in compartment <COMPARTMENT NAME>
Allow dynamic-group <ml-monitoring-application-dynamic-group> to read buckets in compartment <COMPARTMENT NAME>

Configure OCI Data Science Job¶

Steps to setup OCI Data Science Job :

Create OCI Data Science Job using the link.
Note
- Do not use BYOC while creating the job for ML Monitoring Application.
- Do not use Fast Job. Only Custom Configuration is supported.
- Do not use ARM shapes. Only Intel / AMD shapes are supported.
Set the below environment variable while configuring the job:
Key

Description

Value Example

CONTAINER_CUSTOM_IMAGE

Managed Container moniker for ML Monitoring Application. Available versions

"dsmc://odsc-ml-monitoring-application:<version>"

Key	Description	Value Example
CONTAINER_CUSTOM_IMAGE	Managed Container moniker for ML Monitoring Application. Available versions	"dsmc://odsc-ml-monitoring-application:<version>"

Example

Run OCI Data Science Job¶

After the OCI Data Science Job is created using ML Monitoring Application managed container, we can use OCI Data Science Job Run to repeatedly run the job.

Other ways of creating Job and Job Runs¶

CLI

Use the Data Science CLI to create a job as in this example:

Create a job with:

oci data-science job create \
--display-name <job_name>\
--compartment-id <compartment_ocid>\
--project-id <project_ocid> \
--configuration-details file://<jobs_configuration_json_file> \
--infrastructure-configuration-details file://<jobs_infrastructure_configuration_json_file> \
--log-configuration-details file://<optional_jobs_infrastructure_configuration_json_file>

Use this jobs configuration JSON file:

{
  "jobType": "DEFAULT",
  "environmentVariables": {
        "CONTAINER_CUSTOM_IMAGE": "dsmc://odsc-ml-monitoring-application:<version>"
  }
}

Use this jobs infrastructure configuration JSON file:

{
  "jobInfrastructureType":"ME_STANDALONE",
  "shapeName": "<shape_name>",
  "blockStorageSizeInGBs": "block_storage_size_in_GBs"
}

(Optional) Use this jobs logging configuration JSON file:

{
  "enableLogging": true,
  "enableAutoLogCreation": true,
  "logGroupId": "<log_group_ocid>"
}

Create a job run with:

oci data-science job-run create \
--display-name <job_name>\
--compartment-id <compartment_ocid>\
--project-id <project_ocid> \
--job-id <job_ocid> \
--configuration-override-details file://<jobs_override_configuration_json_file>

Use this job run override configuration JSON file:

{
  "jobType": "DEFAULT",
  "environmentVariables": {
        "CONFIG_LOCATION": "<HTTP_LOCATION_OF_OCI_STORAGE_APPLICATION_CONFIG_FILE_PATH>"
        "RUNTIME_PARAMETER": "{"ACTION_TYPE":"RUN_CONFIG_VALIDATION"}"
  }
}

API

The ADS SDK is also a publicly available Python library that you can install with this command:

pip install oracle-ads

It provides the wrapper that makes the creation and running jobs from notebooks or on your client machine easy.

Use the ADS SDK to create job and run jobs

job_and_job_run_example.py

import ads
import os
from ads.jobs import Job, DataScienceJob, PythonRuntime

# Setting OCI_ODSC_SERVICE_ENDPOINT
os.environ['OCI_ODSC_SERVICE_ENDPOINT'] = 'https://datascience-int.us-ashburn-1.oci.oc-test.com'

def create_job_and_job_run():
        rp_version = os.environ.get("OCI_RESOURCE_PRINCIPAL_VERSION", "UNDEFINED")

        # Setting authentication

        # Setting local authentication
        if not rp_version or rp_version == "UNDEFINED":
                ads.set_auth("security_token", oci_config_location="~/.oci/config", profile="DEFAULT")
        else:
                # Setting Resource Principal authentication
                ads.set_auth("resource_principal")

        job = (
                Job(name="Job_creation_using_Ads")
                        .with_infrastructure(
                        DataScienceJob()
                                .with_compartment_id(COMPARTMENT_OCID)
                                .with_project_id(PROJECT_OCID)
                                .with_job_type('DEFAULT')
                                .with_shape_name("SHAPE_NAME")
                                .with_block_storage_size(BLOCK_STORAGE_SIZE)
                                .with_log_group_id(LOG_GROUP_OCID)
                                .with_job_infrastructure_type('ME_STANDALONE')
                )
                        .with_runtime(
                        PythonRuntime()
                                .with_environment_variable(CONTAINER_CUSTOM_IMAGE="dsmc://odsc-ml-monitoring-application:<version>")
                )
        )

        # Create Job
        job.create()

        # Create Job Run
        job_run = job.run(
                name="ML_Monitoring_Validation_Run_ads_1",
                env_var={
                        "CONFIG_LOCATION": "<HTTP_LOCATION_OF_OCI_STORAGE_APPLICATION_CONFIG_FILE_PATH>",
                        "RUNTIME_PARAMETER": "{\"ACTION_TYPE\":\"RUN_CONFIG_VALIDATION\"}"
                }
        )

Terraform

Use the Data Science Create Job Terraform to create a job.

Use the Data Science Create Job Run Terraform to create a job run.