This is the headline decision of Milestone 0: which piece of WMF storage infrastructure physically holds the pre-rendered articles. Viable candidates span a wide spectrum — a new SQL table on the x1 DB cluster (since it must be readable from every wiki consuming Abstract Content, not just abstract.wikipedia.org), MainStash, ExternalStore, Swift (object storage), a Kask/Cassandra-backed service, or piggybacking on ParserCache — and they differ significantly in durability, size ceiling, multi-DC replication, write amplification, purge semantics, and ops cost. The decision is primarily an MW platform & SRE call, not an AWT one, and we will record it on task once taken.
Acceptance criteria:
- A substrate decision is recorded as an ADR-style note linked from this task, naming the chosen substrate, the reasons it was chosen, and the alternatives that were considered and ruled out.
- The substrate is provisioned in production with SRE sign-off on capacity, cross-DC replication, and ops posture.
- A first write and a first read against the substrate have been exercised end-to-end from a WikiLambda code path, against a representative-shape test entry, on staging or production (not a unit test against a mock).
- The decision inputs listed in the nature/location/access sub-bullet have been collected and recorded alongside the ADR.
- Both "agree with stakeholders" sub-bullets above are marked complete, since the substrate selection depends on their outputs.