rust/neon
1
0
Fork 0
mirror of https://github.com/neondatabase/neon.git synced 2026-05-14 03:30:36 +00:00
Code Issues Packages Projects Releases Wiki Activity

clippy-allow await while get_value_reconstruct_data calls + explainer

Browse Source
This commit is contained in:
Christian Schwarz
2023-05-11 17:25:25 +02:00
parent 5443c73d20
commit d629e136fa
1 changed files with 18 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
pageserver/src/tenant/timeline.rs
View File

@@ -2143,6 +2143,24 @@ impl Timeline {
///
/// This function takes the current timeline's locked LayerMap as an argument,
/// so callers can avoid potential race conditions.
///
// TODO: find a way to not hold the Timeline::layers lock during get_value_reconstruct_data calls.
//
// Since these calls do local disk IO, they'll be reasonably fast, until come disk IOPS bound.
// We have lots of headroom on current pageservers, so, it's going to be fine for now.
//
// We can't use tokio::sync::RwLock that easily because its guard is not Send, but,
// many tasks that access Timeline::layers run inside task_mgr tasks, which are required
// to be Send. It has been tried in origin/problame/asyncify-get-reconstruct-data--tokio-sync.
//
// The solution will probably be to have an immutable + multi-versioned layer map, allowing
// us to grab a snapshot of the layer map once and execute this function on the snapshot.
//
// Or, we could invest time to figure out whether we can drop the layer map lock after
// we grabbed the layer, do the IO, re-aquire, and continue the traversal.
//
// (Why is this allow() not inside the function? Because clippy doesn't respect it then).
#[allow(clippy::await_holding_lock)]
async fn get_reconstruct_data(
&self,
key: Key,