MMeent d5b6471fa9 Update prefetch mechanism: (#2687) ...

Prefetch requests and responses are stored in a ringbuffer instead of a
queue, which means we can utilize prefetches of many relations
concurrently -- page reads of un-prefetched relations now don't imply
dropping prefetches.

In a future iteration, this may detect sequential scans based on the
read behavior of sequential scans, and will dynamically prefetch buffers
for such relations as needed. Right now, it still depends on explicit
prefetch requests from PostgreSQL.

The main improvement here is that we now have a buffer for prefetched
pages of 128 entries with random access. Before, we had a similarly sized
cache, but this cache did not allow for random access, which resulted in
dropped entries when multiple systems used the prefetching subsystem
concurrently.

See also: #2544

2022-11-07 18:13:24 +02:00

..

postgres-v14 @ e56b812dd8

Update prefetch mechanism: (#2687)

2022-11-07 18:13:24 +02:00

postgres-v15 @ 39e3d745b3

Update prefetch mechanism: (#2687)

2022-11-07 18:13:24 +02:00