We saw many blank static thumbnails while parse-time geoshape expansion was enabled. Our working hypothesis is that expansion changes the group ID, and this makes old, cached page HTML and image URL incompatible with new ParserOutput for that same revision. To get around this, we need to make the code robust to any of the possible combinations of new and old content.

Alternatives:

• Do nothing and let thumbnails break while caches repopulate.
  • Most disruptive to users.
  • Need to monitor this breakage so that we can predict when it will finish.
  • Maybe some statistics can be gleaned from the Jan 30th test enablement.
• Serve both the expanded and unexpanded mapdata from the API.
  • Easy to implement, low overhead.
  • Messy and possibly confusing to diagnose whether any code is using the deprecated copy.
  • Kartographer mapdata API would be deployed ahead of client changes or expansion feature.
• T329170: Kill the hashed group ID concept entirely
  • It's caused enough pain already.
  • Doesn't seem to be delivering any value at the moment, it could also be an autoincrementing ID given to each mapframe on a page.