Turbo-hipster

Turbo-hipster works with the existing OpenStack code review system to implement testing-related plugins. Historically, whenever code has been written for Nova it has been tested against trivial datasets rather than real data. This can mean that when users run the updated code on their databases they can run into issues that were not found during testing. A variety of real-world databases have been collected, anonymized, and added to the database migration plugin used by turbo-hipster. Turbo-hipster is integrated into the existing code review system, and automatically runs tests against these larger test datasets. Turbo-hipster is specifically designed to flag issues where changes to the database schema may not work due to outliers in real datasets, and to identify situations where a migration may take an unreasonable amount of time against a large database.

Note

Database anonymity is important, and can be very time consuming. The databases used by turbo-hipster to test against are real-world databases that have been anonymized with a database anonymization tool called Fuzzy Happiness. Fuzzy Happiness takes markup in the sqlalchemy models file and uses that to decide what values to anonymize, and how to do so. This feature is still in development, and until it is complete turbo-hipster will not report back to Zuul automatically.

Additionally, turbo-hipster has been designed to be extensible, so it is possible to write other plugins to expand its capabilities.

Turbo-hipster and Zuul

Turbo-hipster is a Gearman worker. Zuul provides arguments that turbo- hipster uses to check out the patch, perform the database testing, and then report back with success or failure. Zuul allows you to specify which jobs should be run against which projects. You can create a rule in Zuul for it to select jobs that require testing against a database. Turbo-hipster will then register as being able to complete that type of job. Gearman handles the connection between Zuul and turbo-hipster, recognizing when a job matches the rule, and passing it to turbo-hipster for testing. When turbo- hipster receives the patchset for the job, it creates a virtual environment to test it. The result of the test is sent back to Gearman as a json string, which contains links to compiled logfiles.

The simplified workflow for turbo-hipster:

  1. Registers as a worker against Zuul’s Gearman server
  2. Receives jobs from Zuul as they arrive
  3. Checks out the patchset
  4. Sets up a new virtual environment for testing
  5. Loads a representative subset of the available datasets
  6. Runs the migration against each dataset, and checks the result
  7. Reports the results to Zuul, using the Gearman protocol

Typical workflow diagram