Config management & ring

This commit is contained in:
Alex Auvolat 2020-04-07 17:00:48 +02:00
parent 061e676136
commit 82b7fcd280
5 changed files with 95 additions and 19 deletions

View File

@ -15,6 +15,7 @@ pub struct NetworkConfig {
#[derive(Clone, Debug, Serialize, Deserialize)] #[derive(Clone, Debug, Serialize, Deserialize)]
pub struct NetworkConfigEntry { pub struct NetworkConfigEntry {
pub datacenter: String,
pub n_tokens: u32, pub n_tokens: u32,
} }

View File

@ -55,6 +55,9 @@ pub struct ConfigureOpt {
/// Node to configure (prefix of hexadecimal node id) /// Node to configure (prefix of hexadecimal node id)
node_id: String, node_id: String,
/// Location (datacenter) of the node
datacenter: String,
/// Number of tokens /// Number of tokens
n_tokens: u32, n_tokens: u32,
} }
@ -100,7 +103,7 @@ async fn cmd_status(rpc_cli: RpcClient, rpc_host: SocketAddr) -> Result<(), Erro
println!("Healthy nodes:"); println!("Healthy nodes:");
for adv in status.iter() { for adv in status.iter() {
if let Some(cfg) = config.members.get(&adv.id) { if let Some(cfg) = config.members.get(&adv.id) {
println!("{}\t{}\t{}\t{}", hex::encode(adv.id), adv.addr, adv.datacenter, cfg.n_tokens); println!("{}\t{}\t{}\t{}", hex::encode(adv.id), cfg.datacenter, cfg.n_tokens, adv.addr);
} }
} }
@ -109,7 +112,7 @@ async fn cmd_status(rpc_cli: RpcClient, rpc_host: SocketAddr) -> Result<(), Erro
println!("\nFailed nodes:"); println!("\nFailed nodes:");
for (id, cfg) in config.members.iter() { for (id, cfg) in config.members.iter() {
if !status.iter().any(|x| x.id == *id) { if !status.iter().any(|x| x.id == *id) {
println!("{}\t{}", hex::encode(id), cfg.n_tokens); println!("{}\t{}\t{}", hex::encode(id), cfg.datacenter, cfg.n_tokens);
} }
} }
} }
@ -118,7 +121,7 @@ async fn cmd_status(rpc_cli: RpcClient, rpc_host: SocketAddr) -> Result<(), Erro
println!("\nUnconfigured nodes:"); println!("\nUnconfigured nodes:");
for adv in status.iter() { for adv in status.iter() {
if !config.members.contains_key(&adv.id) { if !config.members.contains_key(&adv.id) {
println!("{}\t{}\t{}", hex::encode(adv.id), adv.addr, adv.datacenter); println!("{}\t{}", hex::encode(adv.id), adv.addr);
} }
} }
} }
@ -153,6 +156,7 @@ async fn cmd_configure(rpc_cli: RpcClient, rpc_host: SocketAddr, args: Configure
config.members.insert(candidates[0].clone(), config.members.insert(candidates[0].clone(),
NetworkConfigEntry{ NetworkConfigEntry{
datacenter: args.datacenter,
n_tokens: args.n_tokens, n_tokens: args.n_tokens,
}); });
config.version += 1; config.version += 1;

View File

@ -1,4 +1,6 @@
use std::sync::Arc; use std::sync::Arc;
use std::hash::Hash as StdHash;
use std::hash::Hasher;
use std::path::PathBuf; use std::path::PathBuf;
use std::io::{Read}; use std::io::{Read};
use std::collections::HashMap; use std::collections::HashMap;
@ -34,6 +36,19 @@ pub struct Members {
pub status_hash: Hash, pub status_hash: Hash,
pub config: NetworkConfig, pub config: NetworkConfig,
pub ring: Vec<RingEntry>,
pub n_datacenters: usize,
}
pub struct NodeStatus {
pub addr: SocketAddr,
pub remaining_ping_attempts: usize,
}
pub struct RingEntry {
pub location: Hash,
pub node: UUID,
pub datacenter: u64,
} }
impl Members { impl Members {
@ -43,7 +58,6 @@ impl Members {
NodeStatus{ NodeStatus{
addr: addr.clone(), addr: addr.clone(),
remaining_ping_attempts: MAX_FAILED_PINGS, remaining_ping_attempts: MAX_FAILED_PINGS,
datacenter: info.datacenter.clone(),
}); });
match old_status { match old_status {
None => { None => {
@ -61,18 +75,78 @@ impl Members {
let mut hasher = Sha256::new(); let mut hasher = Sha256::new();
eprintln!("Current set of pingable nodes: --"); eprintln!("Current set of pingable nodes: --");
for (id, status) in nodes { for (id, status) in nodes {
eprintln!("{} {} ({})", hex::encode(id), status.addr, status.datacenter); eprintln!("{} {}", hex::encode(id), status.addr);
hasher.input(format!("{} {}\n", hex::encode(id), status.addr)); hasher.input(format!("{} {}\n", hex::encode(id), status.addr));
} }
eprintln!("END --"); eprintln!("END --");
self.status_hash.copy_from_slice(&hasher.result()[..]); self.status_hash.copy_from_slice(&hasher.result()[..]);
} }
}
pub struct NodeStatus { fn rebuild_ring(&mut self) {
pub addr: SocketAddr, let mut new_ring = vec![];
pub datacenter: String, let mut datacenters = vec![];
pub remaining_ping_attempts: usize,
for (id, config) in self.config.members.iter() {
let mut dc_hasher = std::collections::hash_map::DefaultHasher::new();
config.datacenter.hash(&mut dc_hasher);
let datacenter = dc_hasher.finish();
if !datacenters.contains(&datacenter) {
datacenters.push(datacenter);
}
for i in 0..config.n_tokens {
let mut location_hasher = Sha256::new();
location_hasher.input(format!("{} {}", hex::encode(id), i));
let mut location = [0u8; 32];
location.copy_from_slice(&location_hasher.result()[..]);
new_ring.push(RingEntry{
location,
node: id.clone(),
datacenter,
})
}
}
new_ring.sort_by_key(|x| x.location);
self.ring = new_ring;
self.n_datacenters = datacenters.len();
}
fn walk_ring(&self, from: &Hash, n: usize) -> Vec<UUID> {
if n >= self.config.members.len() {
return self.config.members.keys().cloned().collect::<Vec<_>>();
}
let start = match self.ring.binary_search_by_key(from, |x| x.location) {
Ok(i) => i,
Err(i) => if i == 0 {
self.ring.len() - 1
} else {
i - 1
}
};
let mut ret = vec![];
let mut datacenters = vec![];
for delta in 0..self.ring.len() {
if ret.len() == n {
break;
}
let i = (start + delta) % self.ring.len();
if datacenters.len() == self.n_datacenters && !ret.contains(&self.ring[i].node) {
ret.push(self.ring[i].node.clone());
} else if !datacenters.contains(&self.ring[i].datacenter) {
ret.push(self.ring[i].node.clone());
datacenters.push(self.ring[i].datacenter);
}
}
ret
}
} }
fn read_network_config(metadata_dir: &PathBuf) -> Result<NetworkConfig, Error> { fn read_network_config(metadata_dir: &PathBuf) -> Result<NetworkConfig, Error> {
@ -106,8 +180,11 @@ impl System {
status: HashMap::new(), status: HashMap::new(),
status_hash: [0u8; 32], status_hash: [0u8; 32],
config: net_config, config: net_config,
ring: Vec::new(),
n_datacenters: 0,
}; };
members.recalculate_status_hash(); members.recalculate_status_hash();
members.rebuild_ring();
System{ System{
config, config,
id, id,
@ -135,7 +212,6 @@ impl System {
let members = self.members.read().await; let members = self.members.read().await;
Message::Ping(PingMessage{ Message::Ping(PingMessage{
id: self.id, id: self.id,
datacenter: self.config.datacenter.clone(),
rpc_port: self.config.rpc_port, rpc_port: self.config.rpc_port,
status_hash: members.status_hash.clone(), status_hash: members.status_hash.clone(),
config_version: members.config.version, config_version: members.config.version,
@ -184,7 +260,6 @@ impl System {
to_advertise.push(AdvertisedNode{ to_advertise.push(AdvertisedNode{
id: info.id.clone(), id: info.id.clone(),
addr: addr.clone(), addr: addr.clone(),
datacenter: info.datacenter.clone(),
}); });
} }
if is_new || members.status_hash != info.status_hash { if is_new || members.status_hash != info.status_hash {
@ -247,7 +322,6 @@ impl System {
mem.push(AdvertisedNode{ mem.push(AdvertisedNode{
id: node.clone(), id: node.clone(),
addr: status.addr.clone(), addr: status.addr.clone(),
datacenter: status.datacenter.clone(),
}); });
} }
Ok(Message::AdvertiseNodesUp(mem)) Ok(Message::AdvertiseNodesUp(mem))
@ -274,7 +348,6 @@ impl System {
let old_self = members.status.insert(node.id.clone(), let old_self = members.status.insert(node.id.clone(),
NodeStatus{ NodeStatus{
addr: self_addr, addr: self_addr,
datacenter: self.config.datacenter.clone(),
remaining_ping_attempts: MAX_FAILED_PINGS, remaining_ping_attempts: MAX_FAILED_PINGS,
}); });
has_changed = match old_self { has_changed = match old_self {
@ -303,9 +376,11 @@ impl System {
{ {
let mut members = self.members.write().await; let mut members = self.members.write().await;
if adv.version > members.config.version { if adv.version > members.config.version {
members.config = adv.clone();
tokio::spawn(self.clone().broadcast(Message::AdvertiseConfig(adv.clone()), PING_TIMEOUT)); tokio::spawn(self.clone().broadcast(Message::AdvertiseConfig(adv.clone()), PING_TIMEOUT));
members.config = adv.clone();
self.save_network_config().await; self.save_network_config().await;
members.rebuild_ring();
} }
Ok(Message::Ok) Ok(Message::Ok)

View File

@ -21,7 +21,6 @@ pub enum Message {
#[derive(Debug, Serialize, Deserialize)] #[derive(Debug, Serialize, Deserialize)]
pub struct PingMessage { pub struct PingMessage {
pub id: UUID, pub id: UUID,
pub datacenter: String,
pub rpc_port: u16, pub rpc_port: u16,
pub status_hash: Hash, pub status_hash: Hash,
@ -32,5 +31,4 @@ pub struct PingMessage {
pub struct AdvertisedNode { pub struct AdvertisedNode {
pub id: UUID, pub id: UUID,
pub addr: SocketAddr, pub addr: SocketAddr,
pub datacenter: String,
} }

View File

@ -14,8 +14,6 @@ use crate::rpc_server;
#[derive(Deserialize, Debug)] #[derive(Deserialize, Debug)]
pub struct Config { pub struct Config {
pub datacenter: String,
pub metadata_dir: PathBuf, pub metadata_dir: PathBuf,
pub data_dir: PathBuf, pub data_dir: PathBuf,