The invalidation-vs-shared-write split is the real design choice. On a multi-node CDN the same tension shows up as “event-driven purge across PoPs” versus “hope TTLs align.” We have been working on one control plane that operates self-hosted CDN nodes and a plugin marketplace (CDNCA, AGPL-3.0); marketplace features roll out release by release. For India↔UAE, did durable invalidation events beat a shared write path in practice?