Alex Chi clarified that RisingWave automatically handles the distribution of data. The ingest_batch command is called by streaming executors, which put the write batch into the shared buffer. The compute worker node then checkpoints it onto shared storage, like S3. The distribution of data is not done at the storage layer but rather at the compute layer through the stream plan. For example, when creating a materialized view with an exchange by clause, RisingWave ensures that each executor reads only its own portion of data. The shared buffer is local to each compute node and allows executors on the same node to see each other's writes, but it does not control data distribution across nodes. The actual distribution is defined by the SQL plan, which specifies how data is partitioned (e.g., hash(c) % n == x).