When a sketch has a lot of commented events, performance starts to seriously degrade. In my dev environment, a sketch (450k total events) with 1 comment on 3k events would take 14 seconds to load.
The bottleneck was in the way comments are queried from the database via the CommentMixin class, which AFAIU dynamically creates a SQLAlchemy table definition. The Explore API goes through all events in a sketch to populate their comment dictionary. Using subqueryload to eagerly load comments in one go turned out to speed the queries up a lot (went from 14 seconds to 0.5 seconds for the same sketch).
When a sketch has a lot of commented events, performance starts to seriously degrade. In my dev environment, a sketch (450k total events) with 1 comment on 3k events would take 14 seconds to load.
The bottleneck was in the way comments are queried from the database via the
CommentMixinclass, which AFAIU dynamically creates a SQLAlchemy table definition. The Explore API goes through all events in a sketch to populate their comment dictionary. Usingsubqueryloadto eagerly load comments in one go turned out to speed the queries up a lot (went from 14 seconds to 0.5 seconds for the same sketch).