vdk-control-service: fix race condition in test

[murphp15]

Why?

There is an intermittent failure in the integration tests for the control plane.
The stacktrace can be found here #1129.
It happens because the job watcher sees a job in kubernetes before a row exists in the database.
So it creates a row from nothing instead of updating the existing row.
When the watcher creates a row from scratch it doesn't have the startedBy value.
Because deploying the job and updating the database can't be an atomic operation we need to make sure that the job isn't submitted unless the row exists in the database.

What

We now add the row into the database before deploying to k8s to prevent the race condition.
If there is a failure during the deployment we roll back the commit to the database.

How has this been tested?

I added a test to make sure that we rollback correctly.
I also added back in the assertion to check that the started by is set correctly.

What type of change are you making?

bug fix.

Signed-off-by: murphp15 murphp15@tcd.ie

[antoniivanov]

I will try to provide a review by end of the day or Monday morning.

But please fix the title of the PR to follow the recommended format: https://github.com/vmware/versatile-data-kit/blob/main/support/git-commit-template.txt (add component and move the link to the ticket somewhere else).

That format is used later when making release documentation so it's kinda important.

[murphp15]

But please fix the title of the PR to follow the recommended format: https://github.com/vmware/versatile-data-kit/blob/main/support/git-commit-template.txt (add component and move the link to the ticket somewhere else).

Done.

[mivanov1988]

Good catch.

[antoniivanov]

Looks ok to me. If you can explain the unit test.