We've demonstrated this new behavior on the beta cluster, and are ready to proceed to more wikis. My initial recommendation is that we just enable everywhere, some time next week. We expect that the impact will be mostly minor and positive: most maps with external data will load faster for readers, and some significant load reduction will happen on the maps servers, which should speed up maps in general. The only potentially negative impact will be on a handful of pages which include complex SPARQL queries: these will take much longer to serve. After smoke testing, for example by saving this page, we see that the final result is as planned, the page contains the same content as before, the external data is left unexpanded. We don't anticipate any other negative impacts. See also T326630.
This will be deployed by enabling the $wgKartographerExternalDataParseTimeFetch feature flag, so it can go out any day and can be immediately reverted if we hear of problems.
What to monitor:
- Volume of geoshape requests: dashboard Geoshapes Service requests
- Expected to drop logarithmically to a permanent, lower value
- Average and P95 time to parse a page: T326637
- Average and P95 time to save a page: T326637
- Number of times parsing has been delayed by > N seconds by a sparql query: T326637
- ParserCache health: https://grafana.wikimedia.org/d/000000106/parser-cache?orgId=1
- Should be no change in hit rate
- Average storage used per parsed page, or total storage used if not bounded: TBD