SRE would like to know what kind of load we will be seeing on the mwaddlink-query tool. Some considerations:
- We want to link recommendations for 19,500 articles (39 ORES topics * 500 articles per topic) per wiki
- We will launch with 4 wikis, but will scale to ~30 in Q3, and in theory to as many as 100 wikis by end of Q4
- Cached recommendations are invalidated when an article is edited (excluding null edits) or deleted
- At the moment I think we can ignore edits to templates that are present on an article page, but if we use Parsoid HTML for mwaddlink-query then we probably need to take those into account.
- We will have a maintenance script running regularly (hourly?) to ensure we have the desired 500 articles per ORES topic for each wiki
So the goal of this task to is to figure out, roughly:
- how often would we be issuing requests mwaddlink-query to generate new recommendations?