Merge branch 'main' into next-0.10

2 files changed, 328 insertions, 118 deletions

diff --git a/src/rpc/system.rs b/src/rpc/system.rs
index 21156d15..e8844f29 100644
--- a/src/rpc/system.rs
+++ b/src/rpc/system.rs
@@ -3,11 +3,10 @@ use std::collections::{HashMap, HashSet};
 use std::io::{Read, Write};
 use std::net::{IpAddr, SocketAddr};
 use std::path::{Path, PathBuf};
-use std::sync::atomic::Ordering;
 use std::sync::{Arc, RwLock, RwLockReadGuard};
 use std::time::{Duration, Instant};
 
-use arc_swap::ArcSwap;
+use arc_swap::ArcSwapOption;
 use async_trait::async_trait;
 use futures::join;
 use serde::{Deserialize, Serialize};
@@ -92,7 +91,7 @@ pub struct System {
 
 	persist_peer_list: Persister<PeerList>,
 
-	local_status: ArcSwap<NodeStatus>,
+	pub(crate) local_status: RwLock<NodeStatus>,
 	node_status: RwLock<HashMap<Uuid, (u64, NodeStatus)>>,
 
 	pub netapp: Arc<NetApp>,
@@ -101,7 +100,6 @@ pub struct System {
 	pub(crate) system_endpoint: Arc<Endpoint<SystemRpc, System>>,
 
 	rpc_listen_addr: SocketAddr,
-	#[cfg(any(feature = "consul-discovery", feature = "kubernetes-discovery"))]
 	rpc_public_addr: Option<SocketAddr>,
 	bootstrap_peers: Vec<String>,
 
@@ -112,10 +110,10 @@ pub struct System {
 
 	pub layout_manager: Arc<LayoutManager>,
 
-	metrics: SystemMetrics,
+	metrics: ArcSwapOption<SystemMetrics>,
 
 	replication_mode: ReplicationMode,
-	replication_factor: usize,
+	pub(crate) replication_factor: usize,
 
 	/// Path to metadata directory
 	pub metadata_dir: PathBuf,
@@ -171,7 +169,7 @@ pub struct ClusterHealth {
 	pub partitions_all_ok: usize,
 }
 
-#[derive(Debug, Clone, Copy)]
+#[derive(Debug, Clone, Copy, Eq, PartialEq)]
 pub enum ClusterHealthStatus {
 	/// All nodes are available
 	Healthy,
@@ -256,7 +254,10 @@ impl System {
 			hex::encode(&node_key.public_key()[..8])
 		);
 
-		let netapp = NetApp::new(GARAGE_VERSION_TAG, network_key, node_key);
+		let bind_outgoing_to = Some(config)
+			.filter(|x| x.rpc_bind_outgoing)
+			.map(|x| x.rpc_bind_addr.ip());
+		let netapp = NetApp::new(GARAGE_VERSION_TAG, network_key, node_key, bind_outgoing_to);
 		let system_endpoint = netapp.endpoint(SYSTEM_RPC_PATH.into());
 
 		// ---- setup netapp public listener and full mesh peering strategy ----
@@ -283,11 +284,8 @@ impl System {
 			replication_mode,
 		)?;
 
-		// ---- set up metrics and status exchange ----
-		let metrics = SystemMetrics::new(replication_factor);
-
 		let mut local_status = NodeStatus::initial(replication_factor, &layout_manager);
-		local_status.update_disk_usage(&config.metadata_dir, &config.data_dir, &metrics);
+		local_status.update_disk_usage(&config.metadata_dir, &config.data_dir);
 
 		// ---- if enabled, set up additionnal peer discovery methods ----
 		#[cfg(feature = "consul-discovery")]
@@ -308,11 +306,11 @@ impl System {
 			warn!("Kubernetes discovery is not enabled in this build.");
 		}
 
-		// ---- done ----
+		// ---- almost done ----
 		let sys = Arc::new(System {
 			id: netapp.id.into(),
 			persist_peer_list,
-			local_status: ArcSwap::new(Arc::new(local_status)),
+			local_status: RwLock::new(local_status),
 			node_status: RwLock::new(HashMap::new()),
 			netapp: netapp.clone(),
 			peering: peering.clone(),
@@ -320,7 +318,6 @@ impl System {
 			replication_mode,
 			replication_factor,
 			rpc_listen_addr: config.rpc_bind_addr,
-			#[cfg(any(feature = "consul-discovery", feature = "kubernetes-discovery"))]
 			rpc_public_addr,
 			bootstrap_peers: config.bootstrap_peers.clone(),
 			#[cfg(feature = "consul-discovery")]
@@ -328,27 +325,39 @@ impl System {
 			#[cfg(feature = "kubernetes-discovery")]
 			kubernetes_discovery: config.kubernetes_discovery.clone(),
 			layout_manager,
-			metrics,
+			metrics: ArcSwapOption::new(None),
 
 			metadata_dir: config.metadata_dir.clone(),
 			data_dir: config.data_dir.clone(),
 		});
+
 		sys.system_endpoint.set_handler(sys.clone());
+
+		let metrics = SystemMetrics::new(sys.clone());
+		sys.metrics.store(Some(Arc::new(metrics)));
+
 		Ok(sys)
 	}
 
 	/// Perform bootstraping, starting the ping loop
 	pub async fn run(self: Arc<Self>, must_exit: watch::Receiver<bool>) {
 		join!(
-			self.netapp
-				.clone()
-				.listen(self.rpc_listen_addr, None, must_exit.clone()),
+			self.netapp.clone().listen(
+				self.rpc_listen_addr,
+				self.rpc_public_addr,
+				must_exit.clone()
+			),
 			self.peering.clone().run(must_exit.clone()),
 			self.discovery_loop(must_exit.clone()),
 			self.status_exchange_loop(must_exit.clone()),
 		);
 	}
 
+	pub fn cleanup(&self) {
+		// Break reference cycle
+		self.metrics.store(None);
+	}
+
 	// ---- Public utilities / accessors ----
 
 	pub fn cluster_layout(&self) -> RwLockReadGuard<'_, LayoutHelper> {
@@ -511,12 +520,9 @@ impl System {
 			}
 		};
 
+		let hostname = self.local_status.read().unwrap().hostname.clone();
 		if let Err(e) = c
-			.publish_consul_service(
-				self.netapp.id,
-				&self.local_status.load_full().hostname.as_deref().unwrap(),
-				rpc_public_addr,
-			)
+			.publish_consul_service(self.netapp.id, &hostname, rpc_public_addr)
 			.await
 		{
 			error!("Error while publishing Consul service: {}", e);
@@ -538,26 +544,17 @@ impl System {
 			}
 		};
 
-		if let Err(e) = publish_kubernetes_node(
-			k,
-			self.netapp.id,
-			&self.local_status.load_full().hostname.as_deref().unwrap(),
-			rpc_public_addr,
-		)
-		.await
+		let hostname = self.local_status.read().unwrap().hostname.clone();
+		if let Err(e) = publish_kubernetes_node(k, self.netapp.id, &hostname, rpc_public_addr).await
 		{
 			error!("Error while publishing node to Kubernetes: {}", e);
 		}
 	}
 
 	fn update_local_status(&self) {
-		let mut new_si: NodeStatus = self.local_status.load().as_ref().clone();
-
-		new_si.layout_digest = self.layout_manager.layout().digest();
-
-		new_si.update_disk_usage(&self.metadata_dir, &self.data_dir, &self.metrics);
-
-		self.local_status.swap(Arc::new(new_si));
+		let mut local_status = self.local_status.write().unwrap();
+		local_status.layout_digest = self.layout_manager.layout().digest();
+		local_status.update_disk_usage(&self.metadata_dir, &self.data_dir);
 	}
 
 	// --- RPC HANDLERS ---
@@ -577,7 +574,7 @@ impl System {
 		from: Uuid,
 		info: &NodeStatus,
 	) -> Result<SystemRpc, Error> {
-		let local_info = self.local_status.load();
+		let local_info = self.local_status.read().unwrap();
 
 		if local_info.replication_factor < info.replication_factor {
 			error!("Some node have a higher replication factor ({}) than this one ({}). This is not supported and will lead to data corruption. Shutting down for safety.",
@@ -589,6 +586,8 @@ impl System {
 		self.layout_manager
 			.handle_advertise_status(from, &info.layout_digest);
 
+		drop(local_info);
+
 		self.node_status
 			.write()
 			.unwrap()
@@ -601,8 +600,10 @@ impl System {
 		while !*stop_signal.borrow() {
 			let restart_at = Instant::now() + STATUS_EXCHANGE_INTERVAL;
 
+			// Update local node status that is exchanged.
 			self.update_local_status();
-			let local_status: NodeStatus = self.local_status.load().as_ref().clone();
+
+			let local_status: NodeStatus = self.local_status.read().unwrap().clone();
 			let _ = self
 				.rpc_helper()
 				.broadcast(
@@ -622,15 +623,17 @@ impl System {
 
 	async fn discovery_loop(self: &Arc<Self>, mut stop_signal: watch::Receiver<bool>) {
 		while !*stop_signal.borrow() {
-			let not_configured = self.cluster_layout().check().is_err();
-			let no_peers = self.peering.get_peer_list().len() < self.replication_factor;
-			let expected_n_nodes = self.cluster_layout().all_nodes().len();
-			let bad_peers = self
+			let n_connected = self
 				.peering
 				.get_peer_list()
 				.iter()
 				.filter(|p| p.is_up())
-				.count() != expected_n_nodes;
+				.count();
+
+			let not_configured = self.cluster_layout().check().is_err();
+			let no_peers = n_connected < self.replication_factor;
+			let expected_n_nodes = self.cluster_layout().all_nodes().len();
+			let bad_peers = n_connected != expected_n_nodes;
 
 			if not_configured || no_peers || bad_peers {
 				info!("Doing a bootstrap/discovery step (not_configured: {}, no_peers: {}, bad_peers: {})", not_configured, no_peers, bad_peers);
@@ -677,6 +680,14 @@ impl System {
 					}
 				}
 
+				if !not_configured && !no_peers {
+					// If the layout is configured, and we already have some connections
+					// to other nodes in the cluster, we can skip trying to connect to
+					// nodes that are not in the cluster layout.
+					let layout = self.cluster_layout();
+					ping_list.retain(|(id, _)| layout.all_nodes().contains(&(*id).into()));
+				}
+
 				for (node_id, node_addr) in ping_list {
 					let self2 = self.clone();
 					tokio::spawn(async move {
@@ -787,12 +798,7 @@ impl NodeStatus {
 		}
 	}
 
-	fn update_disk_usage(
-		&mut self,
-		meta_dir: &Path,
-		data_dir: &DataDirEnum,
-		metrics: &SystemMetrics,
-	) {
+	fn update_disk_usage(&mut self, meta_dir: &Path, data_dir: &DataDirEnum) {
 		use nix::sys::statvfs::statvfs;
 		let mount_avail = |path: &Path| match statvfs(path) {
 			Ok(x) => {
@@ -828,27 +834,6 @@ impl NodeStatus {
 				)
 			})(),
 		};
-
-		if let Some((avail, total)) = self.meta_disk_avail {
-			metrics
-				.values
-				.meta_disk_avail
-				.store(avail, Ordering::Relaxed);
-			metrics
-				.values
-				.meta_disk_total
-				.store(total, Ordering::Relaxed);
-		}
-		if let Some((avail, total)) = self.data_disk_avail {
-			metrics
-				.values
-				.data_disk_avail
-				.store(avail, Ordering::Relaxed);
-			metrics
-				.values
-				.data_disk_total
-				.store(total, Ordering::Relaxed);
-		}
 	}
 }
 
diff --git a/src/rpc/system_metrics.rs b/src/rpc/system_metrics.rs
index af81b71f..fb3c983c 100644
--- a/src/rpc/system_metrics.rs
+++ b/src/rpc/system_metrics.rs
@@ -1,32 +1,57 @@
-use std::sync::atomic::{AtomicU64, Ordering};
-use std::sync::Arc;
+use std::sync::{Arc, RwLock};
+use std::time::{Duration, Instant};
 
 use opentelemetry::{global, metrics::*, KeyValue};
 
+use crate::system::{ClusterHealthStatus, System};
+
 /// TableMetrics reference all counter used for metrics
 pub struct SystemMetrics {
+	// Static values
 	pub(crate) _garage_build_info: ValueObserver<u64>,
 	pub(crate) _replication_factor: ValueObserver<u64>,
+
+	// Disk space values from System::local_status
 	pub(crate) _disk_avail: ValueObserver<u64>,
 	pub(crate) _disk_total: ValueObserver<u64>,
-	pub(crate) values: Arc<SystemMetricsValues>,
-}
 
-#[derive(Default)]
-pub struct SystemMetricsValues {
-	pub(crate) data_disk_total: AtomicU64,
-	pub(crate) data_disk_avail: AtomicU64,
-	pub(crate) meta_disk_total: AtomicU64,
-	pub(crate) meta_disk_avail: AtomicU64,
+	// Health report from System::health()
+	pub(crate) _cluster_healthy: ValueObserver<u64>,
+	pub(crate) _cluster_available: ValueObserver<u64>,
+	pub(crate) _known_nodes: ValueObserver<u64>,
+	pub(crate) _connected_nodes: ValueObserver<u64>,
+	pub(crate) _storage_nodes: ValueObserver<u64>,
+	pub(crate) _storage_nodes_ok: ValueObserver<u64>,
+	pub(crate) _partitions: ValueObserver<u64>,
+	pub(crate) _partitions_quorum: ValueObserver<u64>,
+	pub(crate) _partitions_all_ok: ValueObserver<u64>,
+
+	// Status report for individual cluster nodes
+	pub(crate) _layout_node_connected: ValueObserver<u64>,
+	pub(crate) _layout_node_disconnected_time: ValueObserver<u64>,
 }
 
 impl SystemMetrics {
-	pub fn new(replication_factor: usize) -> Self {
+	pub fn new(system: Arc<System>) -> Self {
 		let meter = global::meter("garage_system");
-		let values = Arc::new(SystemMetricsValues::default());
-		let values1 = values.clone();
-		let values2 = values.clone();
+
+		let health_cache = RwLock::new((Instant::now(), system.health()));
+		let system2 = system.clone();
+		let get_health = Arc::new(move || {
+			{
+				let cache = health_cache.read().unwrap();
+				if cache.0 > Instant::now() - Duration::from_secs(1) {
+					return cache.1;
+				}
+			}
+
+			let health = system2.health();
+			*health_cache.write().unwrap() = (Instant::now(), health);
+			health
+		});
+
 		Self {
+			// Static values
 			_garage_build_info: meter
 				.u64_value_observer("garage_build_info", move |observer| {
 					observer.observe(
@@ -39,39 +64,239 @@ impl SystemMetrics {
 				})
 				.with_description("Garage build info")
 				.init(),
-			_replication_factor: meter
-				.u64_value_observer("garage_replication_factor", move |observer| {
-					observer.observe(replication_factor as u64, &[])
-				})
-				.with_description("Garage replication factor setting")
-				.init(),
-			_disk_avail: meter
-				.u64_value_observer("garage_local_disk_avail", move |observer| {
-					match values1.data_disk_avail.load(Ordering::Relaxed) {
-						0 => (),
-						x => observer.observe(x, &[KeyValue::new("volume", "data")]),
-					};
-					match values1.meta_disk_avail.load(Ordering::Relaxed) {
-						0 => (),
-						x => observer.observe(x, &[KeyValue::new("volume", "metadata")]),
-					};
-				})
-				.with_description("Garage available disk space on each node")
-				.init(),
-			_disk_total: meter
-				.u64_value_observer("garage_local_disk_total", move |observer| {
-					match values2.data_disk_total.load(Ordering::Relaxed) {
-						0 => (),
-						x => observer.observe(x, &[KeyValue::new("volume", "data")]),
-					};
-					match values2.meta_disk_total.load(Ordering::Relaxed) {
-						0 => (),
-						x => observer.observe(x, &[KeyValue::new("volume", "metadata")]),
-					};
+			_replication_factor: {
+				let replication_factor = system.replication_factor;
+				meter
+					.u64_value_observer("garage_replication_factor", move |observer| {
+						observer.observe(replication_factor as u64, &[])
+					})
+					.with_description("Garage replication factor setting")
+					.init()
+			},
+
+			// Disk space values from System::local_status
+			_disk_avail: {
+				let system = system.clone();
+				meter
+					.u64_value_observer("garage_local_disk_avail", move |observer| {
+						let st = system.local_status.read().unwrap();
+						if let Some((avail, _total)) = st.data_disk_avail {
+							observer.observe(avail, &[KeyValue::new("volume", "data")]);
+						}
+						if let Some((avail, _total)) = st.meta_disk_avail {
+							observer.observe(avail, &[KeyValue::new("volume", "metadata")]);
+						}
+					})
+					.with_description("Garage available disk space on each node")
+					.init()
+			},
+			_disk_total: {
+				let system = system.clone();
+				meter
+					.u64_value_observer("garage_local_disk_total", move |observer| {
+						let st = system.local_status.read().unwrap();
+						if let Some((_avail, total)) = st.data_disk_avail {
+							observer.observe(total, &[KeyValue::new("volume", "data")]);
+						}
+						if let Some((_avail, total)) = st.meta_disk_avail {
+							observer.observe(total, &[KeyValue::new("volume", "metadata")]);
+						}
+					})
+					.with_description("Garage total disk space on each node")
+					.init()
+			},
+
+			// Health report from System::()
+			_cluster_healthy: {
+				let get_health = get_health.clone();
+				meter
+					.u64_value_observer("cluster_healthy", move |observer| {
+						let h = get_health();
+						if h.status == ClusterHealthStatus::Healthy {
+							observer.observe(1, &[]);
+						} else {
+							observer.observe(0, &[]);
+						}
+					})
+					.with_description("Whether all storage nodes are connected")
+					.init()
+			},
+			_cluster_available: {
+				let get_health = get_health.clone();
+				meter.u64_value_observer("cluster_available", move |observer| {
+					let h = get_health();
+					if h.status != ClusterHealthStatus::Unavailable {
+						observer.observe(1, &[]);
+					} else {
+						observer.observe(0, &[]);
+					}
 				})
-				.with_description("Garage total disk space on each node")
-				.init(),
-			values,
+				.with_description("Whether all requests can be served, even if some storage nodes are disconnected")
+				.init()
+			},
+			_known_nodes: {
+				let get_health = get_health.clone();
+				meter
+					.u64_value_observer("cluster_known_nodes", move |observer| {
+						let h = get_health();
+						observer.observe(h.known_nodes as u64, &[]);
+					})
+					.with_description("Number of nodes already seen once in the cluster")
+					.init()
+			},
+			_connected_nodes: {
+				let get_health = get_health.clone();
+				meter
+					.u64_value_observer("cluster_connected_nodes", move |observer| {
+						let h = get_health();
+						observer.observe(h.connected_nodes as u64, &[]);
+					})
+					.with_description("Number of nodes currently connected")
+					.init()
+			},
+			_storage_nodes: {
+				let get_health = get_health.clone();
+				meter
+					.u64_value_observer("cluster_storage_nodes", move |observer| {
+						let h = get_health();
+						observer.observe(h.storage_nodes as u64, &[]);
+					})
+					.with_description("Number of storage nodes declared in the current layout")
+					.init()
+			},
+			_storage_nodes_ok: {
+				let get_health = get_health.clone();
+				meter
+					.u64_value_observer("cluster_storage_nodes_ok", move |observer| {
+						let h = get_health();
+						observer.observe(h.storage_nodes_ok as u64, &[]);
+					})
+					.with_description("Number of storage nodes currently connected")
+					.init()
+			},
+			_partitions: {
+				let get_health = get_health.clone();
+				meter
+					.u64_value_observer("cluster_partitions", move |observer| {
+						let h = get_health();
+						observer.observe(h.partitions as u64, &[]);
+					})
+					.with_description("Number of partitions in the layout")
+					.init()
+			},
+			_partitions_quorum: {
+				let get_health = get_health.clone();
+				meter
+					.u64_value_observer("cluster_partitions_quorum", move |observer| {
+						let h = get_health();
+						observer.observe(h.partitions_quorum as u64, &[]);
+					})
+					.with_description(
+						"Number of partitions for which we have a quorum of connected nodes",
+					)
+					.init()
+			},
+			_partitions_all_ok: {
+				let get_health = get_health.clone();
+				meter
+					.u64_value_observer("cluster_partitions_all_ok", move |observer| {
+						let h = get_health();
+						observer.observe(h.partitions_all_ok as u64, &[]);
+					})
+					.with_description(
+						"Number of partitions for which all storage nodes are connected",
+					)
+					.init()
+			},
+
+			// Status report for individual cluster nodes
+			_layout_node_connected: {
+				let system = system.clone();
+				meter
+					.u64_value_observer("cluster_layout_node_connected", move |observer| {
+						let layout = system.cluster_layout();
+						let nodes = system.get_known_nodes();
+						for (id, _, config) in layout.current().roles.items().iter() {
+							if let Some(role) = &config.0 {
+								let mut kv = vec![
+									KeyValue::new("id", format!("{:?}", id)),
+									KeyValue::new("role_zone", role.zone.clone()),
+								];
+								match role.capacity {
+									Some(cap) => {
+										kv.push(KeyValue::new("role_capacity", cap as i64));
+										kv.push(KeyValue::new("role_gateway", 0));
+									}
+									None => {
+										kv.push(KeyValue::new("role_gateway", 1));
+									}
+								}
+
+								let value;
+								if let Some(node) = nodes.iter().find(|n| n.id == *id) {
+									value = if node.is_up { 1 } else { 0 };
+								// TODO: if we add address and hostname, and those change, we
+								// get duplicate metrics, due to bad otel aggregation :(
+								// Can probably be fixed when we upgrade opentelemetry
+								// kv.push(KeyValue::new("address", node.addr.to_string()));
+								// kv.push(KeyValue::new(
+								// 	"hostname",
+								// 	node.status.hostname.clone(),
+								// ));
+								} else {
+									value = 0;
+								}
+
+								observer.observe(value, &kv);
+							}
+						}
+					})
+					.with_description("Connection status for nodes in the cluster layout")
+					.init()
+			},
+			_layout_node_disconnected_time: {
+				let system = system.clone();
+				meter
+					.u64_value_observer("cluster_layout_node_disconnected_time", move |observer| {
+						let layout = system.cluster_layout();
+						let nodes = system.get_known_nodes();
+						for (id, _, config) in layout.current().roles.items().iter() {
+							if let Some(role) = &config.0 {
+								let mut kv = vec![
+									KeyValue::new("id", format!("{:?}", id)),
+									KeyValue::new("role_zone", role.zone.clone()),
+								];
+								match role.capacity {
+									Some(cap) => {
+										kv.push(KeyValue::new("role_capacity", cap as i64));
+										kv.push(KeyValue::new("role_gateway", 0));
+									}
+									None => {
+										kv.push(KeyValue::new("role_gateway", 1));
+									}
+								}
+
+								if let Some(node) = nodes.iter().find(|n| n.id == *id) {
+									// TODO: see comment above
+									// kv.push(KeyValue::new("address", node.addr.to_string()));
+									// kv.push(KeyValue::new(
+									// 	"hostname",
+									// 	node.status.hostname.clone(),
+									// ));
+									if node.is_up {
+										observer.observe(0, &kv);
+									} else if let Some(secs) = node.last_seen_secs_ago {
+										observer.observe(secs, &kv);
+									}
+								}
+							}
+						}
+					})
+					.with_description(
+						"Time (in seconds) since last connection to nodes in the cluster layout",
+					)
+					.init()
+			},
 		}
 	}
 }