[BUG]: ucx install Fails on Azure with AWS-linked validation error

mokane-db commented 3 months ago

Is there an existing issue for this?

[X] I have searched the existing issues

Current Behavior

Attempting to run install ucx on an Azure environment with defaults results in a failure after the Creating dashboards... step, indicating an input validation error on a Cluster. This validation error indicates the aws_attributes.availability attribute is required, which seems erroneous given this is Azure.

Expected Behavior

Expectation is for ucx install to complete successfully

Steps To Reproduce

❯ databricks --version Databricks CLI v0.220.0 ❯ databricks labs installed Name Description Version ucx Unity Catalog Migration Toolkit (UCX) v0.23.1

❯ databricks labs install ucx Using workspace profile: 17:09:51 INFO [d.l.ucx.install] Installing UCX v0.23.1 17:09:53 INFO [d.l.ucx.install] UCX v0.23.1 is already installed on this workspace Do you want to update the existing installation? (default: no): yes 17:09:58 INFO [d.l.ucx.install] Installing UCX v0.23.1 17:09:58 INFO [d.l.ucx.install][installing_components_0] Creating ucx schemas... 17:09:58 INFO [d.l.ucx.install][installing_components_1] Creating dashboards... Traceback (most recent call last): ... databricks.sdk.errors.platform.InvalidParameterValue: Cluster validation error: Validation failed for aws_attributes.availability, the value must be present

Cloud

Azure

Operating System

macOS

Version

latest via Databricks CLI

Relevant log output

❯ databricks labs install ucx
Using workspace profile: <redacted>>
17:09:51  INFO [d.l.ucx.install] Installing UCX v0.23.1
17:09:53  INFO [d.l.ucx.install] UCX v0.23.1 is already installed on this workspace
Do you want to update the existing installation? (default: no): yes
17:09:58  INFO [d.l.ucx.install] Installing UCX v0.23.1
17:09:58  INFO [d.l.ucx.install][installing_components_0] Creating ucx schemas...
17:09:58  INFO [d.l.ucx.install][installing_components_1] Creating dashboards...
Traceback (most recent call last):
  File "/redacted/path/.databricks/labs/ucx/lib/src/databricks/labs/ucx/install.py", line 667, in <module>
    workspace_installer.run()
  File "/redacted/path/.databricks/labs/ucx/lib/src/databricks/labs/ucx/install.py", line 191, in run
    workspace_installation.run()
  File "/redacted/path/.databricks/labs/ucx/lib/src/databricks/labs/ucx/install.py", line 475, in run
    readme_url = self._workflows_installer.create_jobs()
  File "/redacted/path/.databricks/labs/ucx/lib/src/databricks/labs/ucx/installer/workflows.py", line 420, in create_jobs
    self._deploy_workflow(workflow_name, settings)
  File "/redacted/path/.databricks/labs/ucx/lib/src/databricks/labs/ucx/installer/workflows.py", line 522, in _deploy_workflow
    new_job = self._ws.jobs.create(**settings)
  File "/redacted/path/.databricks/labs/ucx/state/venv/lib/python3.10/site-packages/databricks/sdk/service/jobs.py", line 5257, in create
    res = self._api.do('POST', '/api/2.1/jobs/create', body=body, headers=headers)
  File "/redacted/path/.databricks/labs/ucx/state/venv/lib/python3.10/site-packages/databricks/sdk/core.py", line 130, in do
    response = retryable(self._perform)(method,
  File "/redacted/path/.databricks/labs/ucx/state/venv/lib/python3.10/site-packages/databricks/sdk/retries.py", line 54, in wrapper
    raise err
  File "/redacted/path/.databricks/labs/ucx/state/venv/lib/python3.10/site-packages/databricks/sdk/retries.py", line 33, in wrapper
    return func(*args, **kwargs)
  File "/redacted/path/.databricks/labs/ucx/state/venv/lib/python3.10/site-packages/databricks/sdk/core.py", line 241, in _perform
    raise self._make_nicer_error(response=response, **payload) from None
databricks.sdk.errors.platform.InvalidParameterValue: Cluster validation error: Validation failed for aws_attributes.availability, the value must be present
Error: installer: exit status 1

mokane-db commented 3 months ago

Capturing some further debugging context: This appears to be down to improper handling of the suffix-matching approach for determining which environment the client is connecting through within the SDK, rather than UCX itself. When no match is found the SDK defaults to AWS. In the two scenarios I ran into:

Using native token auth in DBR15 results in the DATABRICKS_HOST environment variable being suffixed with :443 causing a failure
Using my local repro a trailing slash on the host cli config appeared to cause a similar failure.

In both cases the issue may be worked around by correcting the value of the host config or removing the erroneous aws_attributes.availability property from the generated UCX cluster policy.

HariGS-DB commented 3 weeks ago

SDK version 0.30 fixes the bug described in this issue. Closing

databrickslabs / ucx