Module stream_batch 

One streamed batch’s CPU work on POST /api/v1/parse/parquet-stream: the cache-writer append and the client blob, in either stream mode.

Split out of parquet_stream.rs (module-size budget) along the seam #5407 added: this is the one place that decides whether a batch is planned against the whole stream (cross-batch) or serialized on its own.

Functions

encode_batch 🔒

Append meshes to the cache fill and encode the client’s blob.

off_the_async_worker 🔒

Run CPU-bound work from inside an async context without starving other connections’ polls: block_in_place on the multi-thread runtime, inline on a current-thread one (where block_in_place panics, as under the #[tokio::test] harness). Shared by the live stream and its cached replay.

Type Aliases

EncodedBatch 🔒

A batch’s client blob, and (vertex_base, index_base) when it was planned against the whole stream.