Hi, Parallel B-tree, GIN, and BRIN index builds currently use effective_io_concurrency for heap scans because the scans do not pass READ_STREAM_MAINTENANCE to the read stream.
The attached patch passes the maintenance context through the existing scan flags, making these scans use maintenance_io_concurrency. Serial index builds appear to have the same issue. Passing the flag there seems to require extending the table scan interface, which could be follow-up work. --- Regards, Jingtang Zhang
v1-parallel-index-maintenance-io.patch
Description: Binary data
