aboutsummaryrefslogtreecommitdiff
path: root/src/rpc
diff options
context:
space:
mode:
Diffstat (limited to 'src/rpc')
-rw-r--r--src/rpc/Cargo.toml3
-rw-r--r--src/rpc/lib.rs6
-rw-r--r--src/rpc/system.rs102
-rw-r--r--src/rpc/system_metrics.rs44
4 files changed, 132 insertions, 23 deletions
diff --git a/src/rpc/Cargo.toml b/src/rpc/Cargo.toml
index e9a0929a..87ae15ac 100644
--- a/src/rpc/Cargo.toml
+++ b/src/rpc/Cargo.toml
@@ -20,9 +20,10 @@ arc-swap = "1.0"
bytes = "1.0"
gethostname = "0.2"
hex = "0.4"
-tracing = "0.1.30"
+tracing = "0.1"
rand = "0.8"
sodiumoxide = { version = "0.2.5-0", package = "kuska-sodiumoxide" }
+systemstat = "0.2.3"
async-trait = "0.1.7"
serde = { version = "1.0", default-features = false, features = ["derive", "rc"] }
diff --git a/src/rpc/lib.rs b/src/rpc/lib.rs
index a8cc0030..5aec92c0 100644
--- a/src/rpc/lib.rs
+++ b/src/rpc/lib.rs
@@ -3,6 +3,9 @@
#[macro_use]
extern crate tracing;
+mod metrics;
+mod system_metrics;
+
#[cfg(feature = "consul-discovery")]
mod consul;
#[cfg(feature = "kubernetes-discovery")]
@@ -13,9 +16,6 @@ pub mod replication_mode;
pub mod ring;
pub mod system;
-mod metrics;
pub mod rpc_helper;
pub use rpc_helper::*;
-
-pub mod system_metrics;
diff --git a/src/rpc/system.rs b/src/rpc/system.rs
index 90f6a4c2..e0ced8cc 100644
--- a/src/rpc/system.rs
+++ b/src/rpc/system.rs
@@ -3,6 +3,7 @@ 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};
@@ -38,7 +39,6 @@ use crate::replication_mode::*;
use crate::ring::*;
use crate::rpc_helper::*;
-#[cfg(feature = "metrics")]
use crate::system_metrics::*;
const DISCOVERY_INTERVAL: Duration = Duration::from_secs(60);
@@ -106,7 +106,7 @@ pub struct System {
consul_discovery: Option<ConsulDiscovery>,
#[cfg(feature = "kubernetes-discovery")]
kubernetes_discovery: Option<KubernetesDiscoveryConfig>,
- #[cfg(feature = "metrics")]
+
metrics: SystemMetrics,
replication_mode: ReplicationMode,
@@ -118,18 +118,28 @@ pub struct System {
/// Path to metadata directory
pub metadata_dir: PathBuf,
+ /// Path to data directory
+ pub data_dir: PathBuf,
}
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct NodeStatus {
/// Hostname of the node
pub hostname: String,
+
/// Replication factor configured on the node
pub replication_factor: usize,
/// Cluster layout version
pub cluster_layout_version: u64,
/// Hash of cluster layout staging data
pub cluster_layout_staging_hash: Hash,
+
+ /// Disk usage on partition containing metadata directory (tuple: `(avail, total)`)
+ #[serde(default)]
+ pub meta_disk_avail: Option<(u64, u64)>,
+ /// Disk usage on partition containing data directory (tuple: `(avail, total)`)
+ #[serde(default)]
+ pub data_disk_avail: Option<(u64, u64)>,
}
#[derive(Debug, Clone, Serialize, Deserialize)]
@@ -271,18 +281,11 @@ impl System {
}
};
- let local_status = NodeStatus {
- hostname: gethostname::gethostname()
- .into_string()
- .unwrap_or_else(|_| "<invalid utf-8>".to_string()),
- replication_factor,
- cluster_layout_version: cluster_layout.version,
- cluster_layout_staging_hash: cluster_layout.staging_hash,
- };
-
- #[cfg(feature = "metrics")]
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);
+
let ring = Ring::new(cluster_layout, replication_factor);
let (update_ring, ring) = watch::channel(Arc::new(ring));
@@ -373,12 +376,12 @@ impl System {
consul_discovery,
#[cfg(feature = "kubernetes-discovery")]
kubernetes_discovery: config.kubernetes_discovery.clone(),
- #[cfg(feature = "metrics")]
metrics,
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());
Ok(sys)
@@ -416,12 +419,7 @@ impl System {
.get(&n.id.into())
.cloned()
.map(|(_, st)| st)
- .unwrap_or(NodeStatus {
- hostname: "?".to_string(),
- replication_factor: 0,
- cluster_layout_version: 0,
- cluster_layout_staging_hash: Hash::from([0u8; 32]),
- }),
+ .unwrap_or(NodeStatus::unknown()),
})
.collect::<Vec<_>>();
known_nodes
@@ -600,6 +598,9 @@ impl System {
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);
+
self.local_status.swap(Arc::new(new_si));
}
@@ -864,6 +865,69 @@ impl EndpointHandler<SystemRpc> for System {
}
}
+impl NodeStatus {
+ fn initial(replication_factor: usize, layout: &ClusterLayout) -> Self {
+ NodeStatus {
+ hostname: gethostname::gethostname()
+ .into_string()
+ .unwrap_or_else(|_| "<invalid utf-8>".to_string()),
+ replication_factor,
+ cluster_layout_version: layout.version,
+ cluster_layout_staging_hash: layout.staging_hash,
+ meta_disk_avail: None,
+ data_disk_avail: None,
+ }
+ }
+
+ fn unknown() -> Self {
+ NodeStatus {
+ hostname: "?".to_string(),
+ replication_factor: 0,
+ cluster_layout_version: 0,
+ cluster_layout_staging_hash: Hash::from([0u8; 32]),
+ meta_disk_avail: None,
+ data_disk_avail: None,
+ }
+ }
+
+ fn update_disk_usage(&mut self, meta_dir: &Path, data_dir: &Path, metrics: &SystemMetrics) {
+ use systemstat::{Platform, System};
+ let mounts = System::new().mounts().unwrap_or_default();
+
+ let mount_avail = |path: &Path| {
+ mounts
+ .iter()
+ .filter(|x| path.starts_with(&x.fs_mounted_on))
+ .max_by_key(|x| x.fs_mounted_on.len())
+ .map(|x| (x.avail.as_u64(), x.total.as_u64()))
+ };
+
+ self.meta_disk_avail = mount_avail(meta_dir);
+ self.data_disk_avail = mount_avail(data_dir);
+
+ 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);
+ }
+ }
+}
+
fn get_default_ip() -> Option<IpAddr> {
pnet_datalink::interfaces()
.iter()
diff --git a/src/rpc/system_metrics.rs b/src/rpc/system_metrics.rs
index d96b67e4..83f5fa97 100644
--- a/src/rpc/system_metrics.rs
+++ b/src/rpc/system_metrics.rs
@@ -1,14 +1,31 @@
+use std::sync::atomic::{AtomicU64, Ordering};
+use std::sync::Arc;
+
use opentelemetry::{global, metrics::*, KeyValue};
/// TableMetrics reference all counter used for metrics
pub struct SystemMetrics {
pub(crate) _garage_build_info: ValueObserver<u64>,
pub(crate) _replication_factor: ValueObserver<u64>,
+ 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,
}
impl SystemMetrics {
pub fn new(replication_factor: usize) -> Self {
let meter = global::meter("garage_system");
+ let values = Arc::new(SystemMetricsValues::default());
+ let values1 = values.clone();
+ let values2 = values.clone();
Self {
_garage_build_info: meter
.u64_value_observer("garage_build_info", move |observer| {
@@ -28,6 +45,33 @@ impl SystemMetrics {
})
.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")]),
+ };
+ })
+ .with_description("Garage total disk space on each node")
+ .init(),
+ values,
}
}
}
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-02 16:38:32 +0000