An environment for RESTBase now exists to proof large, or disruptive changes that require performance characteristics comparable to that of production. One of the first use-cases for this environment is the testing of proposed alternative data models. Seeing an appropriate subset of update requests, with the same data sizes and overwrite distribution would help us determine whether compaction will behave as we have predicted, and assist in establishing a baseline configuration.
The most straightforward way to accomplish this would be to run a separate instance of change-propagation, updated to propagate changes to the dev environment for an apropos subset of titles.
When replaying update requests, a corresponding request load will be generated on the Parsoid and API clusters. We will need to establish whether the additional request load will be a problem for these systems. If the additional request load is a problem, one possibility would be the addition of a RESTBase endpoint to emulate Parsoid using cached responses.
See also: T129682: Look into solutions for replaying traffic to testing environment(s)