pageserver: add vectored get implementation (#6576)

This PR introduces a new vectored implementation of the read path.

The search is basically a DFS if you squint at it long enough.
LayerFringe tracks the next layers to visit and acts as our stack.
Vertices are tuples of (layer, keyspace, lsn range). Continuously
pop the top of the stack (most recent layer) and do all the reads
for one layer at once.

The search maintains a fringe (`LayerFringe`) which tracks all the
layers that intersect the current keyspace being searched. Continuously
pop the top of the fringe (layer with highest LSN) and get all the data
required from the layer in one go.

Said search is done on one timeline at a time. If data is still required for
some keys, then search the ancestor timeline.

Apart from the high level layer traversal, vectored variants have been
introduced for grabbing data from each layer type. They still suffer from
read amplification issues and that will be addressed in a different PR.

You might notice that in some places we duplicate the code for the
existing read path. All of that code will be removed when we switch
the non-vectored read path to proxy into the vectored read path.
In the meantime, we'll have to contend with the extra cruft for the sake
of testing and gentle releasing.
This commit is contained in:
Vlad Lazar
2024-02-21 09:49:46 +00:00
committed by GitHub
parent 3882f57001
commit 5d6083bfc6
17 changed files with 1284 additions and 127 deletions
+39 -13
View File
@@ -2,6 +2,7 @@ use postgres_ffi::BLCKSZ;
use std::ops::Range;
use crate::key::Key;
use itertools::Itertools;
///
/// Represents a set of Keys, in a compact form.
@@ -63,9 +64,36 @@ impl KeySpace {
KeyPartitioning { parts }
}
/// Update the keyspace such that it doesn't contain any range
/// that is overlapping with `other`. This can involve splitting or
/// removing of existing ranges.
/// Merge another keyspace into the current one.
/// Note: the keyspaces must not ovelap (enforced via assertions)
pub fn merge(&mut self, other: &KeySpace) {
let all_ranges = self
.ranges
.iter()
.merge_by(other.ranges.iter(), |lhs, rhs| lhs.start < rhs.start);
let mut accum = KeySpaceAccum::new();
let mut prev: Option<&Range<Key>> = None;
for range in all_ranges {
if let Some(prev) = prev {
let overlap =
std::cmp::max(range.start, prev.start) < std::cmp::min(range.end, prev.end);
assert!(
!overlap,
"Attempt to merge ovelapping keyspaces: {:?} overlaps {:?}",
prev, range
);
}
accum.add_range(range.clone());
prev = Some(range);
}
self.ranges = accum.to_keyspace().ranges;
}
/// Remove all keys in `other` from `self`.
/// This can involve splitting or removing of existing ranges.
pub fn remove_overlapping_with(&mut self, other: &KeySpace) {
let (self_start, self_end) = match (self.start(), self.end()) {
(Some(start), Some(end)) => (start, end),
@@ -220,16 +248,7 @@ impl KeySpaceAccum {
}
pub fn consume_keyspace(&mut self) -> KeySpace {
if let Some(accum) = self.accum.take() {
self.ranges.push(accum);
}
let mut prev_accum = KeySpaceAccum::new();
std::mem::swap(self, &mut prev_accum);
KeySpace {
ranges: prev_accum.ranges,
}
std::mem::take(self).to_keyspace()
}
pub fn size(&self) -> u64 {
@@ -279,6 +298,13 @@ impl KeySpaceRandomAccum {
}
KeySpace { ranges }
}
pub fn consume_keyspace(&mut self) -> KeySpace {
let mut prev_accum = KeySpaceRandomAccum::new();
std::mem::swap(self, &mut prev_accum);
prev_accum.to_keyspace()
}
}
pub fn key_range_size(key_range: &Range<Key>) -> u32 {