bhearsum and tarek met to compare and contrast Balrog and absearch.

Notes

• absearch and Balrog public API is very similar, both use an url with client information substituted in.
  • absearch format: https://search.services.mozilla.com/1/%PRODUCT%/%VERSION%/%CHANNEL%/%LOCALE%/%REGION%/%DISTRIBUTION%/%DISTRIBUTION_VERSION%/<cohort>
• Key difference in public API is that cohort is independent of system and product information, and is controllable by the server
• Cohorts are defined on the server, and can be grouped by locations, locale, and other info. They can also have size limits, and sample rates (ie: put 10% of users in cohort X until it reaches size Y).
•  %REGION% is the location (not locale), and is calculated by the client using MLS or Google's location service. This takes a lot of burden off of the server.
• Balrog has more advanced permissions, auditing, and rules.
• absearch's was of forming cohorts is very clever, and respects privacy (no per user tokens).
• It may make sense to use Balrog for the absearch backend at some point - Tarek isn't tied to the absearch server in particular
• Alternatively, we may want to look at implementing a/b testing in Balrog by somehow backending it on the existing absearch server. No idea if it's possible, but probably worth looking into.

Action Items

Nothing immediately actionable coming out of this meeting, but when we're ready to implement a/b testing in Balrog, we should look at more at implementing absearch's method, or backending a/b testing on it.