Christian Schwarz 5443c73d20 layer impls: run get_value_reconstruct_data in spawn_blocking ...

Effectively, this means we use the tokio runtime's spawn_blocking-thread-pool
to execute the layer reads, instead of doing the reads on the
tokio runtime's main executor threads.

The use of the thread pool adds some overhead, but, not blocking
the main executor threads is more important, because they can now
execute other async tasks while we do the IO.

With a sufficiently large spawn_blocking-thread-pool, we also get more
IO parallelism between timelines than with blocking the main executor
threads. So, we might push the pageserver's NVMe closer to its limits.
But right now, there's lots of headroom.

2023-05-11 17:10:15 +02:00

..

benches

revert: "Add check for duplicates of generated image layers" (#4104)

2023-04-28 17:20:18 +03:00

fixtures

walredo: simple tests and bench updates (#3045)

2023-01-16 18:24:45 +02:00

src

layer impls: run get_value_reconstruct_data in spawn_blocking

2023-05-11 17:10:15 +02:00

Cargo.toml

make parking_lot lock guards Send to be able to hold them across .await

2023-05-10 20:08:49 +02:00