Deuxfleurs/garage
48
56
Fork 73
Code Issues 100 Pull requests 13 Releases 30 Wiki Activity

system metrics improvements #726

Merged
lx merged 5 commits from peer-metrics into main 2024-02-20 15:35:13 +00:00
Conversation 0 Commits 5 Files changed 4 +83 -72
3 changed files with 83 additions and 72 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files
Showing only changes of commit 182a23cc12 - Show all commits

1
src/garage/server.rs
View file

@ -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

28
src/rpc/system.rs
View file

@ -6,6 +6,7 @@ use std::path::{Path, PathBuf};
use std::sync::{Arc, RwLock};
use std::time::{Duration, Instant};
use arc_swap::ArcSwapOption;
use async_trait::async_trait;
use futures::join;
use serde::{Deserialize, Serialize};
@ -86,7 +87,7 @@ pub struct System {
persist_cluster_layout: Persister<ClusterLayout>,
persist_peer_list: Persister<PeerList>,
local_status: Arc<RwLock<NodeStatus>>,
pub(crate) local_status: RwLock<NodeStatus>,
node_status: RwLock<HashMap<Uuid, (u64, NodeStatus)>>,
pub netapp: Arc<NetApp>,
@ -104,10 +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>>,
@ -280,9 +281,6 @@ impl System {
let mut local_status = NodeStatus::initial(replication_factor, &cluster_layout);
local_status.update_disk_usage(&config.metadata_dir, &config.data_dir);
let local_status = Arc::new(RwLock::new(local_status));
let metrics = SystemMetrics::new(replication_factor, local_status.clone());
let ring = Ring::new(cluster_layout, replication_factor);
let (update_ring, ring) = watch::channel(Arc::new(ring));
@ -356,7 +354,7 @@ impl System {
id: netapp.id.into(),
persist_cluster_layout,
persist_peer_list,
local_status,
local_status: RwLock::new(local_status),
node_status: RwLock::new(HashMap::new()),
netapp: netapp.clone(),
peering: peering.clone(),
@ -376,14 +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)
}
@ -401,6 +404,11 @@ impl System {
);
}
pub fn cleanup(&self) {
// Break reference cycle
self.metrics.store(None);
}
// ---- Administrative operations (directly available and
// also available through RPC) ----
@ -699,11 +707,7 @@ impl System {
let restart_at = Instant::now() + STATUS_EXCHANGE_INTERVAL;
// Update local node status that is exchanged.
// Status variables are exported into Prometheus in SystemMetrics,
// so we take the opportunity to also update here the health status
// that is reported in those metrics.
self.update_local_status();
*self.metrics.health.write().unwrap() = Some(self.health());
let local_status: NodeStatus = self.local_status.read().unwrap().clone();
let _ = self

88
src/rpc/system_metrics.rs
View file

@ -1,13 +1,12 @@
use std::sync::{Arc, RwLock};
use std::time::{Duration, Instant};
use opentelemetry::{global, metrics::*, KeyValue};
use crate::system::{ClusterHealth, ClusterHealthStatus, NodeStatus};
use crate::system::{ClusterHealthStatus, System};
/// TableMetrics reference all counter used for metrics
pub struct SystemMetrics {
pub(crate) health: Arc<RwLock<Option<ClusterHealth>>>,
// Static values
pub(crate) _garage_build_info: ValueObserver<u64>,
pub(crate) _replication_factor: ValueObserver<u64>,
@ -29,12 +28,25 @@ pub struct SystemMetrics {
}
impl SystemMetrics {
pub fn new(replication_factor: usize, local_status: Arc<RwLock<NodeStatus>>) -> Self {
pub fn new(system: Arc<System>) -> Self {
let meter = global::meter("garage_system");
let health = Arc::new(RwLock::new(None));
Self {
health: health.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| {
@ -48,19 +60,22 @@ impl SystemMetrics {
})
.with_description("Garage build info")
.init(),
_replication_factor: meter
_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(),
.init()
},
// Disk space values from System::local_status
_disk_avail: {
let status = local_status.clone();
let system = system.clone();
meter
.u64_value_observer("garage_local_disk_avail", move |observer| {
let st = status.read().unwrap();
let st = system.local_status.read().unwrap();
if let Some((avail, _total)) = st.data_disk_avail {
observer.observe(avail, &[KeyValue::new("volume", "data")]);
}
@ -72,10 +87,10 @@ impl SystemMetrics {
.init()
},
_disk_total: {
let status = local_status.clone();
let system = system.clone();
meter
.u64_value_observer("garage_local_disk_total", move |observer| {
let st = status.read().unwrap();
let st = system.local_status.read().unwrap();
if let Some((_avail, total)) = st.data_disk_avail {
observer.observe(total, &[KeyValue::new("volume", "data")]);
}
@ -87,98 +102,90 @@ impl SystemMetrics {
.init()
},
// Health report from System::health()
// Health report from System::()
_cluster_healthy: {
let health = health.clone();
let get_health = get_health.clone();
meter
.u64_value_observer("cluster_healthy", move |observer| {
if let Some(h) = health.read().unwrap().as_ref() {
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 health = health.clone();
let get_health = get_health.clone();
meter.u64_value_observer("cluster_available", move |observer| {
if let Some(h) = health.read().unwrap().as_ref() {
let h = get_health();
if h.status != ClusterHealthStatus::Unavailable {
observer.observe(1, &[]);
} else {
observer.observe(0, &[]);
}
}
})
.with_description("Whether all requests can be served, even if some storage nodes are disconnected")
.init()
},
_known_nodes: {
let health = health.clone();
let get_health = get_health.clone();
meter
.u64_value_observer("cluster_known_nodes", move |observer| {
if let Some(h) = health.read().unwrap().as_ref() {
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 health = health.clone();
let get_health = get_health.clone();
meter
.u64_value_observer("cluster_connected_nodes", move |observer| {
if let Some(h) = health.read().unwrap().as_ref() {
let h = get_health();
observer.observe(h.connected_nodes as u64, &[]);
}
})
.with_description("Number of nodes currently connected")
.init()
},
_storage_nodes: {
let health = health.clone();
let get_health = get_health.clone();
meter
.u64_value_observer("cluster_storage_nodes", move |observer| {
if let Some(h) = health.read().unwrap().as_ref() {
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 health = health.clone();
let get_health = get_health.clone();
meter
.u64_value_observer("cluster_storage_nodes_ok", move |observer| {
if let Some(h) = health.read().unwrap().as_ref() {
let h = get_health();
observer.observe(h.storage_nodes_ok as u64, &[]);
}
})
.with_description("Number of storage nodes currently connected")
.init()
},
_partitions: {
let health = health.clone();
let get_health = get_health.clone();
meter
.u64_value_observer("cluster_partitions", move |observer| {
if let Some(h) = health.read().unwrap().as_ref() {
let h = get_health();
observer.observe(h.partitions as u64, &[]);
}
})
.with_description("Number of partitions in the layout")
.init()
},
_partitions_quorum: {
let health = health.clone();
let get_health = get_health.clone();
meter
.u64_value_observer("cluster_partitions_quorum", move |observer| {
if let Some(h) = health.read().unwrap().as_ref() {
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",
@ -186,12 +193,11 @@ impl SystemMetrics {
.init()
},
_partitions_all_ok: {
let health = health.clone();
let get_health = get_health.clone();
meter
.u64_value_observer("cluster_partitions_all_ok", move |observer| {
if let Some(h) = health.read().unwrap().as_ref() {
let h = get_health();
observer.observe(h.partitions_all_ok as u64, &[]);
}
})
.with_description(
"Number of partitions for which all storage nodes are connected",