use a WriteLock for write operations on fullcopy tables
All checks were successful
ci/woodpecker/push/debug Pipeline was successful
ci/woodpecker/pr/debug Pipeline was successful

This commit is contained in:
Alex 2025-03-25 13:18:14 +01:00
parent 8ba6454e21
commit 514eb29874
4 changed files with 27 additions and 21 deletions

View file

@ -196,6 +196,19 @@ impl LayoutHelper {
self.current().nodes_of(hash).collect()
}
/// For a given hash, or for all cluster if no hash is given,
/// return for each layout version the set of nodes that writes should be sent to
/// and for which a quorum of OK responses should be awaited.
pub fn write_sets_of(&self, hash: Option<&Hash>) -> Vec<Vec<Uuid>> {
self.versions()
.iter()
.map(|x| match hash {
Some(h) => x.nodes_of(h).collect(),
None => x.all_nodes().to_vec(),
})
.collect()
}
pub fn ack_map_min(&self) -> u64 {
self.ack_map_min
}

View file

@ -143,20 +143,19 @@ impl LayoutManager {
// ---- ACK LOCKING ----
pub fn write_sets_of(self: &Arc<Self>, hash: &Hash) -> WriteLock<Vec<Vec<Uuid>>> {
pub fn write_lock_with<T, F>(self: &Arc<Self>, f: F) -> WriteLock<T>
where
F: FnOnce(&LayoutHelper) -> T,
{
let layout = self.layout();
let version = layout.current().version;
let nodes = layout
.versions()
.iter()
.map(|x| x.nodes_of(hash).collect())
.collect();
let value = f(&layout);
layout
.ack_lock
.get(&version)
.unwrap()
.fetch_add(1, Ordering::Relaxed);
WriteLock::new(version, self, nodes)
WriteLock::new(version, self, value)
}
// ---- INTERNALS ---

View file

@ -25,7 +25,7 @@ pub struct TableFullReplication {
}
impl TableReplication for TableFullReplication {
type WriteSets = Vec<Vec<Uuid>>;
type WriteSets = WriteLock<Vec<Vec<Uuid>>>;
// Do anti-entropy every 10 seconds.
// Compared to sharded tables, anti-entropy is much less costly as there is
@ -52,11 +52,8 @@ impl TableReplication for TableFullReplication {
fn write_sets(&self, _hash: &Hash) -> Self::WriteSets {
self.system
.cluster_layout()
.versions()
.iter()
.map(|ver| ver.all_nodes().to_vec())
.collect()
.layout_manager
.write_lock_with(|l| l.write_sets_of(None))
}
fn write_quorum(&self) -> usize {
let layout = self.system.cluster_layout();
@ -92,7 +89,7 @@ impl TableReplication for TableFullReplication {
partition: 0u16,
first_hash: [0u8; 32].into(),
last_hash: [0xff; 32].into(),
storage_sets: self.write_sets(&[0u8; 32].into()),
storage_sets: layout.write_sets_of(None),
}];
SyncPartitions {

View file

@ -54,7 +54,9 @@ impl TableReplication for TableShardedReplication {
}
fn write_sets(&self, hash: &Hash) -> Self::WriteSets {
self.system.layout_manager.write_sets_of(hash)
self.system
.layout_manager
.write_lock_with(|l| l.write_sets_of(Some(hash)))
}
fn write_quorum(&self) -> usize {
self.write_quorum
@ -72,16 +74,11 @@ impl TableReplication for TableShardedReplication {
.current()
.partitions()
.map(|(partition, first_hash)| {
let storage_sets = layout
.versions()
.iter()
.map(|x| x.nodes_of(&first_hash).collect())
.collect();
SyncPartition {
partition,
first_hash,
last_hash: [0u8; 32].into(), // filled in just after
storage_sets,
storage_sets: layout.write_sets_of(Some(&first_hash)),
}
})
.collect::<Vec<_>>();