The goal of this milestone is to stand up a durable storage layer for whole pre-rendered Abstract article sections, keyed by topic, language and section id, sized to survive the volumes we expect once multiple wikis are consuming from it. This is a new tier sitting alongside — not replacing — the existing fragment-level memcached caching in AbstractWikiRequest / CacheAbstractContentFragmentJob, which will continue to serve its original purpose of absorbing repeat fragment calls close to the orchestrator.

The new tier is at section granularity, populated by a pre-generation job that composes fragments into article sections, so that readers never pay composition cost on the read path.

A subsequent (non-blocking) consumer of this storage layer will be read-mode on abstract.wikipedia.org itself: T422635: Migrate abstract.wikipedia.org's read path to server-side generated pages pulling from the durable storage layer where available