aboutsummaryrefslogtreecommitdiff
path: root/src/table/replication
diff options
context:
space:
mode:
authorAlex Auvolat <alex@adnab.me>2023-11-14 14:28:16 +0100
committerAlex Auvolat <alex@adnab.me>2023-11-14 14:28:16 +0100
commit3b361d2959e3d577bdae6f8a5ccb0c9d5526b7ea (patch)
treef5448d44c7d5705c1e31912ca6d101c5998523ef /src/table/replication
parent866196750fca74c1911ade2a90611f3663e60046 (diff)
downloadgarage-3b361d2959e3d577bdae6f8a5ccb0c9d5526b7ea.tar.gz
garage-3b361d2959e3d577bdae6f8a5ccb0c9d5526b7ea.zip
layout: prepare for write sets
Diffstat (limited to 'src/table/replication')
-rw-r--r--src/table/replication/fullcopy.rs9
-rw-r--r--src/table/replication/parameters.rs8
-rw-r--r--src/table/replication/sharded.rs24
3 files changed, 20 insertions, 21 deletions
diff --git a/src/table/replication/fullcopy.rs b/src/table/replication/fullcopy.rs
index beaacc2b..cb5471af 100644
--- a/src/table/replication/fullcopy.rs
+++ b/src/table/replication/fullcopy.rs
@@ -27,6 +27,11 @@ pub struct TableFullReplication {
}
impl TableReplication for TableFullReplication {
+ fn storage_nodes(&self, _hash: &Hash) -> Vec<Uuid> {
+ let layout = self.system.cluster_layout();
+ layout.current().all_nodes().to_vec()
+ }
+
fn read_nodes(&self, _hash: &Hash) -> Vec<Uuid> {
vec![self.system.id]
}
@@ -34,8 +39,8 @@ impl TableReplication for TableFullReplication {
1
}
- fn write_nodes(&self, _hash: &Hash) -> Vec<Uuid> {
- self.system.cluster_layout().current().all_nodes().to_vec()
+ fn write_sets(&self, hash: &Hash) -> Vec<Vec<Uuid>> {
+ vec![self.storage_nodes(hash)]
}
fn write_quorum(&self) -> usize {
let nmembers = self.system.cluster_layout().current().all_nodes().len();
diff --git a/src/table/replication/parameters.rs b/src/table/replication/parameters.rs
index 2a7d3585..2f842409 100644
--- a/src/table/replication/parameters.rs
+++ b/src/table/replication/parameters.rs
@@ -6,21 +6,23 @@ pub trait TableReplication: Send + Sync + 'static {
// See examples in table_sharded.rs and table_fullcopy.rs
// To understand various replication methods
+ /// The entire list of all nodes that store a partition
+ fn storage_nodes(&self, hash: &Hash) -> Vec<Uuid>;
+
/// Which nodes to send read requests to
fn read_nodes(&self, hash: &Hash) -> Vec<Uuid>;
/// Responses needed to consider a read succesfull
fn read_quorum(&self) -> usize;
/// Which nodes to send writes to
- fn write_nodes(&self, hash: &Hash) -> Vec<Uuid>;
- /// Responses needed to consider a write succesfull
+ fn write_sets(&self, hash: &Hash) -> Vec<Vec<Uuid>>;
+ /// Responses needed to consider a write succesfull in each set
fn write_quorum(&self) -> usize;
fn max_write_errors(&self) -> usize;
// Accessing partitions, for Merkle tree & sync
/// Get partition for data with given hash
fn partition_of(&self, hash: &Hash) -> Partition;
-
/// List of partitions and nodes to sync with in current layout
fn sync_partitions(&self) -> SyncPartitions;
}
diff --git a/src/table/replication/sharded.rs b/src/table/replication/sharded.rs
index f02b1d66..1320a189 100644
--- a/src/table/replication/sharded.rs
+++ b/src/table/replication/sharded.rs
@@ -25,21 +25,19 @@ pub struct TableShardedReplication {
}
impl TableReplication for TableShardedReplication {
+ fn storage_nodes(&self, hash: &Hash) -> Vec<Uuid> {
+ self.system.cluster_layout().storage_nodes_of(hash)
+ }
+
fn read_nodes(&self, hash: &Hash) -> Vec<Uuid> {
- self.system
- .cluster_layout()
- .current()
- .nodes_of(hash, self.replication_factor)
+ self.system.cluster_layout().read_nodes_of(hash)
}
fn read_quorum(&self) -> usize {
self.read_quorum
}
- fn write_nodes(&self, hash: &Hash) -> Vec<Uuid> {
- self.system
- .cluster_layout()
- .current()
- .nodes_of(hash, self.replication_factor)
+ fn write_sets(&self, hash: &Hash) -> Vec<Vec<Uuid>> {
+ self.system.cluster_layout().write_sets_of(hash)
}
fn write_quorum(&self) -> usize {
self.write_quorum
@@ -60,13 +58,7 @@ impl TableReplication for TableShardedReplication {
.current()
.partitions()
.map(|(partition, first_hash)| {
- let mut storage_nodes = layout
- .write_sets_of(&first_hash)
- .map(|x| x.into_iter())
- .flatten()
- .collect::<Vec<_>>();
- storage_nodes.sort();
- storage_nodes.dedup();
+ let storage_nodes = layout.storage_nodes_of(&first_hash);
SyncPartition {
partition,
first_hash,