More like currently stays fairly busy at 80-100M requests per day primarily through the RelatedArticles extension. From the search perspective our main concern is to ensure that Related Articles and this functionality are using the same shape of api requests and thus share the same caches. This will help keep the load down on the search side and ensure better response times to the users by using an already well populated cache.

In terms of total load the search side can handle, it's hard to say for sure.If this has a similar distribution of pages (cachability) to Related Articles, then i wouldn't start to worry about additional load until we are talking about adding 10-20M+ reqs/day. Even then it's not necessarily a worry, but something we will at least keep a closer eye on.

Just met with @EBernhardson and @dcausse to discuss this. Since we are testing based on session id we should see a similar distribution of pages to Related Articles, so for the the A/B test the Search Platform team mainly needs to just be alerted of when it starts and which wikis. Ebernhardson will be out in January but David and others on the team will be around.

David raised concern that MoreLike could be slower on different languages, but we don't have good numbers on it. I will clarify the deployment schedule for the different wikis, and see if we can spread out the deployments.

Lastly, I'm going to followup on getting an estimate for traffic to starting search in general, which should be located in the search satisfaction schema in the data lake.

Regarding traffic estimates i took a quick stab at the data in superset: https://superset.wikimedia.org/sqllab/?savedQueryId=986

This reports the number of page loads that submitted at least one autocomplete event. This seems like a reasonable proxy for the minimum number of autocomplete opens we should expect.

thank you! that's helpful! I also took a look the web team's search dashboard, which has desktop and mobile autocomplete actors raw count

@bwang - curious if we can collect the documentation for this in a single place. Maybe on-wiki? Potentially as a sub-page of https://www.mediawiki.org/wiki/Reading/Web/Content_Discovery_Experiments might work

@ovasileva Published a summary of this task and discussion here: https://www.mediawiki.org/wiki/Reading/Web/Content_Discovery_Experiments/Traffic_impact