Add local disk usage to exported prometheus metrics

This commit is contained in:
Alex 2023-01-26 15:30:36 +01:00
parent efb6b6e868
commit c7d0ad0aa0
3 changed files with 75 additions and 12 deletions

View file

@ -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;

View file

@ -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,
@ -281,12 +281,11 @@ impl System {
}
};
let mut local_status = NodeStatus::initial(replication_factor, &cluster_layout);
local_status.update_disk_usage(&config.metadata_dir, &config.data_dir);
#[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));
@ -377,7 +376,6 @@ impl System {
consul_discovery,
#[cfg(feature = "kubernetes-discovery")]
kubernetes_discovery: config.kubernetes_discovery.clone(),
#[cfg(feature = "metrics")]
metrics,
ring,
@ -601,7 +599,7 @@ impl System {
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);
new_si.update_disk_usage(&self.metadata_dir, &self.data_dir, &self.metrics);
self.local_status.swap(Arc::new(new_si));
}
@ -892,7 +890,7 @@ impl NodeStatus {
}
}
fn update_disk_usage(&mut self, meta_dir: &Path, data_dir: &Path) {
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();
@ -906,6 +904,27 @@ impl NodeStatus {
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);
}
}
}

View file

@ -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,
}
}
}