aboutsummaryrefslogtreecommitdiff
path: root/src/rpc
diff options
context:
space:
mode:
Diffstat (limited to 'src/rpc')
-rw-r--r--src/rpc/Cargo.toml5
-rw-r--r--src/rpc/layout.rs4
-rw-r--r--src/rpc/rpc_helper.rs18
-rw-r--r--src/rpc/system.rs123
4 files changed, 75 insertions, 75 deletions
diff --git a/src/rpc/Cargo.toml b/src/rpc/Cargo.toml
index 1b411c6a..3d4d3ff5 100644
--- a/src/rpc/Cargo.toml
+++ b/src/rpc/Cargo.toml
@@ -1,6 +1,6 @@
[package]
name = "garage_rpc"
-version = "0.8.0"
+version = "0.8.1"
authors = ["Alex Auvolat <alex@adnab.me>"]
edition = "2018"
license = "AGPL-3.0"
@@ -14,7 +14,7 @@ path = "lib.rs"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[dependencies]
-garage_util = { version = "0.8.0", path = "../util" }
+garage_util = { version = "0.8.1", path = "../util" }
arc-swap = "1.0"
bytes = "1.0"
@@ -27,7 +27,6 @@ itertools="0.10"
sodiumoxide = { version = "0.2.5-0", package = "kuska-sodiumoxide" }
async-trait = "0.1.7"
-rmp-serde = "0.15"
serde = { version = "1.0", default-features = false, features = ["derive", "rc"] }
serde_bytes = "0.11"
serde_json = "1.0"
diff --git a/src/rpc/layout.rs b/src/rpc/layout.rs
index 1cef44d1..d756f0aa 100644
--- a/src/rpc/layout.rs
+++ b/src/rpc/layout.rs
@@ -9,6 +9,7 @@ use serde::{Deserialize, Serialize};
use garage_util::crdt::{AutoCrdt, Crdt, Lww, LwwMap};
use garage_util::data::*;
+use garage_util::encode::nonversioned_encode;
use garage_util::error::*;
use crate::graph_algo::*;
@@ -58,6 +59,7 @@ pub struct ClusterLayout {
pub staging_roles: LwwMap<Uuid, NodeRoleV>,
pub staging_hash: Hash,
}
+impl garage_util::migrate::InitialFormat for ClusterLayout {}
/// This struct is used to set the parameters to be used in the assignation computation
/// algorithm. It is stored as a Crdt.
@@ -134,7 +136,7 @@ impl ClusterLayout {
fn calculate_staging_hash(&self) -> Hash {
let hashed_tuple = (&self.staging_roles, &self.staging_parameters);
- blake2sum(&rmp_to_vec_all_named(&hashed_tuple).unwrap()[..])
+ blake2sum(&nonversioned_encode(&hashed_tuple).unwrap()[..])
}
pub fn merge(&mut self, other: &ClusterLayout) -> bool {
diff --git a/src/rpc/rpc_helper.rs b/src/rpc/rpc_helper.rs
index 949aced6..1ec250c3 100644
--- a/src/rpc/rpc_helper.rs
+++ b/src/rpc/rpc_helper.rs
@@ -5,7 +5,6 @@ use std::time::Duration;
use futures::future::join_all;
use futures::stream::futures_unordered::FuturesUnordered;
use futures::stream::StreamExt;
-use futures_util::future::FutureExt;
use tokio::select;
use tokio::sync::watch;
@@ -24,7 +23,6 @@ pub use netapp::message::{
use netapp::peering::fullmesh::FullMeshPeeringStrategy;
pub use netapp::{self, NetApp, NodeID};
-use garage_util::background::BackgroundRunner;
use garage_util::data::*;
use garage_util::error::Error;
use garage_util::metrics::RecordDuration;
@@ -94,7 +92,6 @@ pub struct RpcHelper(Arc<RpcHelperInner>);
struct RpcHelperInner {
our_node_id: Uuid,
fullmesh: Arc<FullMeshPeeringStrategy>,
- background: Arc<BackgroundRunner>,
ring: watch::Receiver<Arc<Ring>>,
metrics: RpcMetrics,
rpc_timeout: Duration,
@@ -104,7 +101,6 @@ impl RpcHelper {
pub(crate) fn new(
our_node_id: Uuid,
fullmesh: Arc<FullMeshPeeringStrategy>,
- background: Arc<BackgroundRunner>,
ring: watch::Receiver<Arc<Ring>>,
rpc_timeout: Option<Duration>,
) -> Self {
@@ -113,7 +109,6 @@ impl RpcHelper {
Self(Arc::new(RpcHelperInner {
our_node_id,
fullmesh,
- background,
ring,
metrics,
rpc_timeout: rpc_timeout.unwrap_or(DEFAULT_TIMEOUT),
@@ -377,16 +372,13 @@ impl RpcHelper {
if !resp_stream.is_empty() {
// Continue remaining requests in background.
- // Continue the remaining requests immediately using tokio::spawn
- // but enqueue a task in the background runner
- // to ensure that the process won't exit until the requests are done
- // (if we had just enqueued the resp_stream.collect directly in the background runner,
- // the requests might have been put on hold in the background runner's queue,
- // in which case they might timeout or otherwise fail)
- let wait_finished_fut = tokio::spawn(async move {
+ // Note: these requests can get interrupted on process shutdown,
+ // we must not count on them being executed for certain.
+ // For all background things that have to happen with certainty,
+ // they have to be put in a proper queue that is persisted to disk.
+ tokio::spawn(async move {
resp_stream.collect::<Vec<Result<_, _>>>().await;
});
- self.0.background.spawn(wait_finished_fut.map(|_| Ok(())));
}
}
diff --git a/src/rpc/system.rs b/src/rpc/system.rs
index ea92b570..1f4d86e7 100644
--- a/src/rpc/system.rs
+++ b/src/rpc/system.rs
@@ -21,7 +21,6 @@ use netapp::peering::fullmesh::FullMeshPeeringStrategy;
use netapp::util::parse_and_resolve_peer_addr_async;
use netapp::{NetApp, NetworkKey, NodeID, NodeKey};
-use garage_util::background::BackgroundRunner;
use garage_util::config::Config;
#[cfg(feature = "kubernetes-discovery")]
use garage_util::config::KubernetesDiscoveryConfig;
@@ -50,8 +49,6 @@ pub const GARAGE_VERSION_TAG: u64 = 0x6761726167650008; // garage 0x0008
/// RPC endpoint used for calls related to membership
pub const SYSTEM_RPC_PATH: &str = "garage_rpc/membership.rs/SystemRpc";
-pub const CONNECT_ERROR_MESSAGE: &str = "Error establishing RPC connection to remote node. This can happen if the remote node is not reachable on the network, but also if the two nodes are not configured with the same rpc_secret";
-
/// RPC messages related to membership
#[derive(Debug, Serialize, Deserialize, Clone)]
pub enum SystemRpc {
@@ -76,13 +73,17 @@ impl Rpc for SystemRpc {
type Response = Result<SystemRpc, Error>;
}
+#[derive(Serialize, Deserialize)]
+pub struct PeerList(Vec<(Uuid, SocketAddr)>);
+impl garage_util::migrate::InitialFormat for PeerList {}
+
/// This node's membership manager
pub struct System {
/// The id of this node
pub id: Uuid,
persist_cluster_layout: Persister<ClusterLayout>,
- persist_peer_list: Persister<Vec<(Uuid, SocketAddr)>>,
+ persist_peer_list: Persister<PeerList>,
local_status: ArcSwap<NodeStatus>,
node_status: RwLock<HashMap<Uuid, (u64, NodeStatus)>>,
@@ -110,9 +111,6 @@ pub struct System {
pub ring: watch::Receiver<Arc<Ring>>,
update_ring: Mutex<watch::Sender<Arc<Ring>>>,
- /// The job runner of this node
- pub background: Arc<BackgroundRunner>,
-
/// Path to metadata directory
pub metadata_dir: PathBuf,
}
@@ -232,7 +230,6 @@ impl System {
/// Create this node's membership manager
pub fn new(
network_key: NetworkKey,
- background: Arc<BackgroundRunner>,
replication_mode: ReplicationMode,
config: &Config,
) -> Result<Arc<Self>, Error> {
@@ -354,7 +351,6 @@ impl System {
rpc: RpcHelper::new(
netapp.id.into(),
fullmesh,
- background.clone(),
ring.clone(),
config.rpc_timeout_msec.map(Duration::from_millis),
),
@@ -372,7 +368,6 @@ impl System {
ring,
update_ring: Mutex::new(update_ring),
- background,
metadata_dir: config.metadata_dir.clone(),
});
sys.system_endpoint.set_handler(sys.clone());
@@ -444,17 +439,14 @@ impl System {
))
})?;
let mut errors = vec![];
- for ip in addrs.iter() {
- match self
- .netapp
- .clone()
- .try_connect(*ip, pubkey)
- .await
- .err_context(CONNECT_ERROR_MESSAGE)
- {
+ for addr in addrs.iter() {
+ match self.netapp.clone().try_connect(*addr, pubkey).await {
Ok(()) => return Ok(()),
Err(e) => {
- errors.push((*ip, e));
+ errors.push((
+ *addr,
+ Error::Message(connect_error_message(*addr, pubkey, e)),
+ ));
}
}
}
@@ -529,56 +521,61 @@ impl System {
// ---- INTERNALS ----
#[cfg(feature = "consul-discovery")]
- async fn advertise_to_consul(self: Arc<Self>) -> Result<(), Error> {
+ async fn advertise_to_consul(self: Arc<Self>) {
let c = match &self.consul_discovery {
Some(c) => c,
- _ => return Ok(()),
+ _ => return,
};
let rpc_public_addr = match self.rpc_public_addr {
Some(addr) => addr,
None => {
warn!("Not advertising to Consul because rpc_public_addr is not defined in config file and could not be autodetected.");
- return Ok(());
+ return;
}
};
- c.publish_consul_service(
- self.netapp.id,
- &self.local_status.load_full().hostname,
- rpc_public_addr,
- )
- .await
- .err_context("Error while publishing Consul service")
+ if let Err(e) = c
+ .publish_consul_service(
+ self.netapp.id,
+ &self.local_status.load_full().hostname,
+ rpc_public_addr,
+ )
+ .await
+ {
+ error!("Error while publishing Consul service: {}", e);
+ }
}
#[cfg(feature = "kubernetes-discovery")]
- async fn advertise_to_kubernetes(self: Arc<Self>) -> Result<(), Error> {
+ async fn advertise_to_kubernetes(self: Arc<Self>) {
let k = match &self.kubernetes_discovery {
Some(k) => k,
- _ => return Ok(()),
+ _ => return,
};
let rpc_public_addr = match self.rpc_public_addr {
Some(addr) => addr,
None => {
warn!("Not advertising to Kubernetes because rpc_public_addr is not defined in config file and could not be autodetected.");
- return Ok(());
+ return;
}
};
- publish_kubernetes_node(
+ if let Err(e) = publish_kubernetes_node(
k,
self.netapp.id,
&self.local_status.load_full().hostname,
rpc_public_addr,
)
.await
- .err_context("Error while publishing node to kubernetes")
+ {
+ error!("Error while publishing node to Kubernetes: {}", e);
+ }
}
/// Save network configuration to disc
- async fn save_cluster_layout(self: Arc<Self>) -> Result<(), Error> {
+ async fn save_cluster_layout(&self) -> Result<(), Error> {
let ring: Arc<Ring> = self.ring.borrow().clone();
self.persist_cluster_layout
.save_async(&ring.layout)
@@ -630,11 +627,7 @@ impl System {
if info.cluster_layout_version > local_info.cluster_layout_version
|| info.cluster_layout_staging_hash != local_info.cluster_layout_staging_hash
{
- let self2 = self.clone();
- self.background.spawn_cancellable(async move {
- self2.pull_cluster_layout(from).await;
- Ok(())
- });
+ tokio::spawn(self.clone().pull_cluster_layout(from));
}
self.node_status
@@ -676,18 +669,21 @@ impl System {
drop(update_ring);
let self2 = self.clone();
- self.background.spawn_cancellable(async move {
- self2
+ tokio::spawn(async move {
+ if let Err(e) = self2
.rpc
.broadcast(
&self2.system_endpoint,
SystemRpc::AdvertiseClusterLayout(layout),
RequestStrategy::with_priority(PRIO_HIGH),
)
- .await?;
- Ok(())
+ .await
+ {
+ warn!("Error while broadcasting new cluster layout: {}", e);
+ }
});
- self.background.spawn(self.clone().save_cluster_layout());
+
+ self.save_cluster_layout().await?;
}
Ok(SystemRpc::Ok)
@@ -734,7 +730,7 @@ impl System {
// Add peer list from list stored on disk
if let Ok(peers) = self.persist_peer_list.load_async().await {
- ping_list.extend(peers.iter().map(|(id, addr)| ((*id).into(), *addr)))
+ ping_list.extend(peers.0.iter().map(|(id, addr)| ((*id).into(), *addr)))
}
// Fetch peer list from Consul
@@ -773,12 +769,12 @@ impl System {
}
for (node_id, node_addr) in ping_list {
- tokio::spawn(
- self.netapp
- .clone()
- .try_connect(node_addr, node_id)
- .map(|r| r.err_context(CONNECT_ERROR_MESSAGE)),
- );
+ let self2 = self.clone();
+ tokio::spawn(async move {
+ if let Err(e) = self2.netapp.clone().try_connect(node_addr, node_id).await {
+ error!("{}", connect_error_message(node_addr, node_id, e));
+ }
+ });
}
}
@@ -787,11 +783,10 @@ impl System {
}
#[cfg(feature = "consul-discovery")]
- self.background.spawn(self.clone().advertise_to_consul());
+ tokio::spawn(self.clone().advertise_to_consul());
#[cfg(feature = "kubernetes-discovery")]
- self.background
- .spawn(self.clone().advertise_to_kubernetes());
+ tokio::spawn(self.clone().advertise_to_kubernetes());
let restart_at = tokio::time::sleep(DISCOVERY_INTERVAL);
select! {
@@ -815,12 +810,16 @@ impl System {
// and append it to the list we are about to save,
// so that no peer ID gets lost in the process.
if let Ok(mut prev_peer_list) = self.persist_peer_list.load_async().await {
- prev_peer_list.retain(|(id, _ip)| peer_list.iter().all(|(id2, _ip2)| id2 != id));
- peer_list.extend(prev_peer_list);
+ prev_peer_list
+ .0
+ .retain(|(id, _ip)| peer_list.iter().all(|(id2, _ip2)| id2 != id));
+ peer_list.extend(prev_peer_list.0);
}
// Save new peer list to file
- self.persist_peer_list.save_async(&peer_list).await
+ self.persist_peer_list
+ .save_async(&PeerList(peer_list))
+ .await
}
async fn pull_cluster_layout(self: Arc<Self>, peer: Uuid) {
@@ -881,3 +880,11 @@ async fn resolve_peers(peers: &[String]) -> Vec<(NodeID, SocketAddr)> {
ret
}
+
+fn connect_error_message(
+ addr: SocketAddr,
+ pubkey: ed25519::PublicKey,
+ e: netapp::error::Error,
+) -> String {
+ format!("Error establishing RPC connection to remote node: {}@{}.\nThis can happen if the remote node is not reachable on the network, but also if the two nodes are not configured with the same rpc_secret.\n{}", hex::encode(pubkey), addr, e)
+}
generated by cgit v1.2.3 (git 2.25.1) at 2024-09-20 08:39:11 +0000