2020-04-19 13:14:23 +00:00
|
|
|
use std::sync::Arc;
|
|
|
|
|
2023-11-08 14:41:24 +00:00
|
|
|
use garage_rpc::layout::*;
|
2021-10-14 09:50:12 +00:00
|
|
|
use garage_rpc::system::System;
|
2020-04-24 10:10:01 +00:00
|
|
|
use garage_util::data::*;
|
|
|
|
|
2021-03-11 15:54:15 +00:00
|
|
|
use crate::replication::*;
|
2020-04-19 13:14:23 +00:00
|
|
|
|
2023-11-11 11:08:32 +00:00
|
|
|
// TODO: find a way to track layout changes for this as well
|
|
|
|
// The hard thing is that this data is stored also on gateway nodes,
|
|
|
|
// whereas sharded data is stored only on non-Gateway nodes (storage nodes)
|
|
|
|
// Also we want to be more tolerant to failures of gateways so we don't
|
|
|
|
// want to do too much holding back of data when progress of gateway
|
|
|
|
// nodes is not reported in the layout history's ack/sync/sync_ack maps.
|
|
|
|
|
2021-03-26 18:41:46 +00:00
|
|
|
/// Full replication schema: all nodes store everything
|
|
|
|
/// Advantage: do all reads locally, extremely fast
|
|
|
|
/// Inconvenient: only suitable to reasonably small tables
|
2023-11-11 11:08:32 +00:00
|
|
|
/// Inconvenient: if some writes fail, nodes will read outdated data
|
2020-04-19 13:14:23 +00:00
|
|
|
#[derive(Clone)]
|
|
|
|
pub struct TableFullReplication {
|
2021-03-26 18:41:46 +00:00
|
|
|
/// The membership manager of this node
|
2021-03-16 10:14:27 +00:00
|
|
|
pub system: Arc<System>,
|
2020-04-19 13:14:23 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
impl TableReplication for TableFullReplication {
|
2023-11-15 14:40:44 +00:00
|
|
|
type WriteSets = Vec<Vec<Uuid>>;
|
|
|
|
|
2023-11-14 13:28:16 +00:00
|
|
|
fn storage_nodes(&self, _hash: &Hash) -> Vec<Uuid> {
|
|
|
|
let layout = self.system.cluster_layout();
|
|
|
|
layout.current().all_nodes().to_vec()
|
|
|
|
}
|
|
|
|
|
2021-10-15 09:05:09 +00:00
|
|
|
fn read_nodes(&self, _hash: &Hash) -> Vec<Uuid> {
|
2021-03-16 10:14:27 +00:00
|
|
|
vec![self.system.id]
|
2020-04-19 13:14:23 +00:00
|
|
|
}
|
|
|
|
fn read_quorum(&self) -> usize {
|
|
|
|
1
|
|
|
|
}
|
|
|
|
|
2023-11-15 14:40:44 +00:00
|
|
|
fn write_sets(&self, hash: &Hash) -> Self::WriteSets {
|
2023-11-14 13:28:16 +00:00
|
|
|
vec![self.storage_nodes(hash)]
|
2020-04-19 13:14:23 +00:00
|
|
|
}
|
2021-03-16 10:14:27 +00:00
|
|
|
fn write_quorum(&self) -> usize {
|
2023-11-14 12:06:16 +00:00
|
|
|
let nmembers = self.system.cluster_layout().current().all_nodes().len();
|
2024-03-04 17:37:00 +00:00
|
|
|
|
|
|
|
let max_faults = if nmembers > 1 { 1 } else { 0 };
|
|
|
|
|
|
|
|
if nmembers > max_faults {
|
|
|
|
nmembers - max_faults
|
2021-03-11 18:06:27 +00:00
|
|
|
} else {
|
|
|
|
1
|
|
|
|
}
|
2020-04-19 13:14:23 +00:00
|
|
|
}
|
|
|
|
|
2021-03-16 11:18:03 +00:00
|
|
|
fn partition_of(&self, _hash: &Hash) -> Partition {
|
|
|
|
0u16
|
|
|
|
}
|
2023-11-11 11:08:32 +00:00
|
|
|
|
|
|
|
fn sync_partitions(&self) -> SyncPartitions {
|
|
|
|
let layout = self.system.cluster_layout();
|
|
|
|
let layout_version = layout.current().version;
|
|
|
|
SyncPartitions {
|
|
|
|
layout_version,
|
|
|
|
partitions: vec![SyncPartition {
|
|
|
|
partition: 0u16,
|
|
|
|
first_hash: [0u8; 32].into(),
|
|
|
|
last_hash: [0xff; 32].into(),
|
2023-12-07 10:16:10 +00:00
|
|
|
storage_sets: vec![layout.current().all_nodes().to_vec()],
|
2023-11-11 11:08:32 +00:00
|
|
|
}],
|
|
|
|
}
|
2020-04-19 13:14:23 +00:00
|
|
|
}
|
|
|
|
}
|