Documentation›Azure ML
Integrations

Azure ML

Add a CertOne gate between training and packaging. A failing verdict fails the step, so the model is never packaged.

Before you start

  • An Azure ML workspace and a compute cluster with a system-assigned managed identity.
  • A member API key stored in the workspace Key Vault as the secret certone-api-key.
  • The cluster's identity granted Key Vault Secrets User on that vault.
Azure CLI
# The cluster's managed identity reads the key from the workspace Key Vault.
az keyvault secret set --vault-name <workspace-vault> --name certone-api-key --value "$CERTONE_API_KEY"
# Grant the compute cluster's system-assigned identity "Key Vault Secrets User" on that vault.

In the CertOne console, open Developer Hub, then API keys. Name the key, choose the member role and select New key.

  • Copy the secret right away: it is shown only once.
  • Keys start with ck_ and expire after at most 365 days.
  • Creating keys requires the admin role in your organization.
  • A viewer key can read results but cannot submit scans.

Add the gate step

Declare the vault URL as a pipeline input and add a command job that runs certone-submit. The packaging job takes the gate's output as its input, so it only runs after a pass.

pipeline.yml
inputs:
  certone_key_vault_url:
    type: string
jobs:
  # train: ...   evaluate: ...
  certone_gate:
    type: command
    inputs:
      certone_key_vault_url: ${{parent.inputs.certone_key_vault_url}}
      model: ${{parent.jobs.train.outputs.model}}
    outputs:
      report: { type: uri_folder }
    code: ..
    command: >-
      pip install -q "certone-sdk[azure]>=0.2.1" &&
      certone-submit --project . --asset "Card Fraud Scorer" --environment ci
      --country-code CA --tee kata_cc --fail-on high
      --api-key-vault "${{inputs.certone_key_vault_url}}/secrets/certone-api-key"
      --report-out ${{outputs.report}}/report.json
      --attestation-out ${{outputs.report}}/attestation.json
  package:
    inputs:
      gate: ${{parent.jobs.certone_gate.outputs.report}}   # runs only if the gate passed

Submit the pipeline with the vault URL:

Azure CLI
az ml job create -f pipeline.yml --set inputs.certone_key_vault_url=https://<workspace-vault>.vault.azure.net

Options

The gate uses certone-submit; every option is described in the SDKs and CLI guide. The ones used here:

OptionMeaning
--api-key-vaultRead the key from Key Vault with the compute's managed identity.
--teeamd_sev_snp (default) or kata_cc.
--country-codeWhere the system operates.
--fail-onLowest severity that fails, high by default.
--report-out, --attestation-outWrite the report and, on a pass, the attestation to the step's output.

What you see

  • In Studio, the pipeline graph shows the gate step completed or failed, and the packaging step runs only after a pass.
  • The gate step's user_logs/std_log.txt shows the verdict, where the scan ran, and one line per finding with its severity.

How it gates

  • Exit 0 on a passing verdict: the pipeline continues.
  • Exit 1 on a failing verdict, or 2 on an error such as an unreadable key: the step fails and nothing downstream runs.