Merge pull request 'system metrics improvements' (#726) from peer-metrics into main

Reviewed-on: https://git.deuxfleurs.fr/Deuxfleurs/garage/pulls/726

4 files changed, 412 insertions, 104 deletions

diff --git a/doc/book/reference-manual/monitoring.md b/doc/book/reference-manual/monitoring.md
index 97c533d3..f392c133 100644
--- a/doc/book/reference-manual/monitoring.md
+++ b/doc/book/reference-manual/monitoring.md
@@ -27,6 +27,112 @@ Exposes the Garage replication factor configured on the node
 garage_replication_factor 3
 ```
 
+#### `garage_local_disk_avail` and `garage_local_disk_total` (gauge)
+
+Reports the available and total disk space on each node, for data and metadata separately.
+
+```
+garage_local_disk_avail{volume="data"} 540341960704
+garage_local_disk_avail{volume="metadata"} 540341960704
+garage_local_disk_total{volume="data"} 763063566336
+garage_local_disk_total{volume="metadata"} 763063566336
+```
+
+### Cluster health status metrics
+
+#### `cluster_healthy` (gauge)
+
+Whether all storage nodes are connected (0 or 1)
+
+```
+cluster_healthy 0
+```
+
+#### `cluster_available` (gauge)
+
+Whether all requests can be served, even if some storage nodes are disconnected
+
+```
+cluster_available 1
+```
+
+#### `cluster_connected_nodes` (gauge)
+
+Number of nodes currently connected
+
+```
+cluster_connected_nodes 3
+```
+
+#### `cluster_known_nodes` (gauge)
+
+Number of nodes already seen once in the cluster
+
+```
+cluster_known_nodes 3
+```
+
+#### `cluster_layout_node_connected` (gauge)
+
+Connection status for individual nodes of the cluster layout
+
+```
+cluster_layout_node_connected{id="62b218d848e86a64",role_capacity="1000000000",role_gateway="0",role_zone="dc1"} 1
+cluster_layout_node_connected{id="a11c7cf18af29737",role_capacity="1000000000",role_gateway="0",role_zone="dc1"} 0
+cluster_layout_node_connected{id="a235ac7695e0c54d",role_capacity="1000000000",role_gateway="0",role_zone="dc1"} 1
+cluster_layout_node_connected{id="b10c110e4e854e5a",role_capacity="1000000000",role_gateway="0",role_zone="dc1"} 1
+```
+
+#### `cluster_layout_node_disconnected_time` (gauge)
+
+Time (in seconds) since last connection to individual nodes of the cluster layout
+
+```
+cluster_layout_node_disconnected_time{id="62b218d848e86a64",role_capacity="1000000000",role_gateway="0",role_zone="dc1"} 0
+cluster_layout_node_disconnected_time{id="a235ac7695e0c54d",role_capacity="1000000000",role_gateway="0",role_zone="dc1"} 0
+cluster_layout_node_disconnected_time{id="b10c110e4e854e5a",role_capacity="1000000000",role_gateway="0",role_zone="dc1"} 0
+```
+
+#### `cluster_storage_nodes` (gauge)
+
+Number of storage nodes declared in the current layout
+
+```
+cluster_storage_nodes 4
+```
+
+#### `cluster_storage_nodes_ok` (gauge)
+
+Number of storage nodes currently connected
+
+```
+cluster_storage_nodes_ok 3
+```
+
+#### `cluster_partitions` (gauge)
+
+Number of partitions in the layout (this is always 256)
+
+```
+cluster_partitions 256
+```
+
+#### `cluster_partitions_all_ok` (gauge)
+
+Number of partitions for which all storage nodes are connected
+
+```
+cluster_partitions_all_ok 64
+```
+
+#### `cluster_partitions_quorum` (gauge)
+
+Number of partitions for which we have a quorum of connected nodes and all requests can be served
+
+```
+cluster_partitions_quorum 256
+```
+
 ### Metrics of the API endpoints
 
 #### `api_admin_request_counter` (counter)
diff --git a/src/garage/server.rs b/src/garage/server.rs
index 51b06b8e..6323f957 100644
--- a/src/garage/server.rs
+++ b/src/garage/server.rs
@@ -162,6 +162,7 @@ pub async fn run_server(config_file: PathBuf, secrets: Secrets) -> Result<(), Er
 	info!("Netapp exited");
 
 	// Drop all references so that stuff can terminate properly
+	garage.system.cleanup();
 	drop(garage);
 
 	// Await for all background tasks to end
diff --git a/src/rpc/system.rs b/src/rpc/system.rs
index 998517de..9e475717 100644
--- a/src/rpc/system.rs
+++ b/src/rpc/system.rs
@@ -3,11 +3,10 @@ use std::collections::HashMap;
 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};
 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};
@@ -88,7 +87,7 @@ pub struct System {
 	persist_cluster_layout: Persister<ClusterLayout>,
 	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>,
@@ -106,10 +105,10 @@ pub struct System {
 	#[cfg(feature = "kubernetes-discovery")]
 	kubernetes_discovery: Option<KubernetesDiscoveryConfig>,
 
-	metrics: SystemMetrics,
+	metrics: ArcSwapOption<SystemMetrics>,
 
 	replication_mode: ReplicationMode,
-	replication_factor: usize,
+	pub(crate) replication_factor: usize,
 
 	/// The ring
 	pub ring: watch::Receiver<Arc<Ring>>,
@@ -170,7 +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,
@@ -280,10 +279,8 @@ impl System {
 			}
 		};
 
-		let metrics = SystemMetrics::new(replication_factor);
-
 		let mut local_status = NodeStatus::initial(replication_factor, &cluster_layout);
-		local_status.update_disk_usage(&config.metadata_dir, &config.data_dir, &metrics);
+		local_status.update_disk_usage(&config.metadata_dir, &config.data_dir);
 
 		let ring = Ring::new(cluster_layout, replication_factor);
 		let (update_ring, ring) = watch::channel(Arc::new(ring));
@@ -357,7 +354,7 @@ impl System {
 			id: netapp.id.into(),
 			persist_cluster_layout,
 			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(),
@@ -377,14 +374,19 @@ impl System {
 			consul_discovery,
 			#[cfg(feature = "kubernetes-discovery")]
 			kubernetes_discovery: config.kubernetes_discovery.clone(),
-			metrics,
+			metrics: ArcSwapOption::new(None),
 
 			ring,
 			update_ring: Mutex::new(update_ring),
 			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)
 	}
 
@@ -402,6 +404,11 @@ impl System {
 		);
 	}
 
+	pub fn cleanup(&self) {
+		// Break reference cycle
+		self.metrics.store(None);
+	}
+
 	// ---- Administrative operations (directly available and
 	//      also available through RPC) ----
 
@@ -546,12 +553,9 @@ 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,
-				rpc_public_addr,
-			)
+			.publish_consul_service(self.netapp.id, &hostname, rpc_public_addr)
 			.await
 		{
 			error!("Error while publishing Consul service: {}", e);
@@ -573,13 +577,8 @@ impl System {
 			}
 		};
 
-		if let Err(e) = publish_kubernetes_node(
-			k,
-			self.netapp.id,
-			&self.local_status.load_full().hostname,
-			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);
 		}
@@ -596,15 +595,13 @@ impl System {
 	}
 
 	fn update_local_status(&self) {
-		let mut new_si: NodeStatus = self.local_status.load().as_ref().clone();
+		let mut local_status = self.local_status.write().unwrap();
 
 		let ring = self.ring.borrow();
-		new_si.cluster_layout_version = ring.layout.version;
-		new_si.cluster_layout_staging_hash = ring.layout.staging_hash;
-
-		new_si.update_disk_usage(&self.metadata_dir, &self.data_dir, &self.metrics);
+		local_status.cluster_layout_version = ring.layout.version;
+		local_status.cluster_layout_staging_hash = ring.layout.staging_hash;
 
-		self.local_status.swap(Arc::new(new_si));
+		local_status.update_disk_usage(&self.metadata_dir, &self.data_dir);
 	}
 
 	// --- RPC HANDLERS ---
@@ -629,7 +626,7 @@ 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.",
@@ -644,6 +641,8 @@ impl System {
 			tokio::spawn(self.clone().pull_cluster_layout(from));
 		}
 
+		drop(local_info);
+
 		self.node_status
 			.write()
 			.unwrap()
@@ -707,8 +706,10 @@ 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
 				.broadcast(
@@ -904,12 +905,7 @@ 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) => {
@@ -945,27 +941,6 @@ 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..ffbef6df 100644
--- a/src/rpc/system_metrics.rs
+++ b/src/rpc/system_metrics.rs
@@ -1,32 +1,58 @@
-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::ring::Ring;
+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 +65,239 @@ 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 ring: Arc<Ring> = system.ring.borrow().clone();
+						let nodes = system.get_known_nodes();
+						for (id, _, config) in ring.layout.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 ring: Arc<Ring> = system.ring.borrow().clone();
+						let nodes = system.get_known_nodes();
+						for (id, _, config) in ring.layout.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()
+			},
 		}
 	}
 }