We simplify the ingestion pipeline by removing a batching layer that batches collections of rows.
Description
Previously, we had a queue of list of list of rows. This represented: queue of batches of sets of rows, where sets of rows are to be committed in one transaction, and the batch holds a series data for transactions to commit. This middle layer of batching was incorporated to to avoid too much contention on the queue that feeds the transaction threads.
However, testing shows this has negligible impact and we can simplify the architecture, reduce memory consumption, and enable later fatures by removing this batching layer.
Goal
We simplify the ingestion pipeline by removing a batching layer that batches collections of rows.
Description
Previously, we had a queue of list of list of rows. This represented: queue of batches of sets of rows, where sets of rows are to be committed in one transaction, and the batch holds a series data for transactions to commit. This middle layer of batching was incorporated to to avoid too much contention on the queue that feeds the transaction threads.
However, testing shows this has negligible impact and we can simplify the architecture, reduce memory consumption, and enable later fatures by removing this batching layer.