The Android team is currently rolling out RESTBase content loading to the beta channel. To accurately gauge performance, we should pre-generate mobile content accessed by the app, so that don't need to wait for relatively slow on-demand content massaging.
The main concerns with doing so are
a) load created on the content service, and
b) storage needs in RESTBase while the conversion to the multi-instance setup is not finished yet (T95253).
With the performance gains & reduced storage usage from the upgrade to Cassandra 2.1.12 (T120803) I'm quite confident that we can manage b). Based on past dump sizes, I would expect at most 150G of extra storage usage per node once storage is completely filled. In the worst case we could quickly delete all stored mobile apps content to free up space quickly.
For a), my information is that we currently have two instances of the content service deployed. This will likely be a bit tight to process about 100 updates per second, so we should carefully ramp it up to see how it does in practice. I would propose to start with pre-generating 10% of requests, and then gradually ramping it up.
@mobrovac, @pchelolo, @eevans, @fgiunchedi: Does this sound like a good plan to you?