gsadmin/rustfs
1
0
Fork 0
mirror of https://github.com/rustfs/rustfs.git synced 2026-09-01 01:38:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat(obs): complete metric dimension coverage (#6314)

Browse Source 
Co-authored-by: heihutu <heihutu@gmail.com>
This commit is contained in:
houseme
2026-08-20 22:43:40 +08:00
committed by GitHub
parent d65fba9142
commit 830e553a3c
15 changed files with 821 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/obs/src/metrics/schema/cluster_notification.rs
+40
View File
@@ -15,6 +15,10 @@
use crate::{MetricDescriptor, MetricName, new_counter_md, new_gauge_md, subsystems};
use std::sync::LazyLock;
pub const SERVER: &str = "server";
const SERVER_LABELS: [&str; 1] = [SERVER];
pub static NOTIFICATION_CURRENT_SEND_IN_PROGRESS_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_gauge_md(
MetricName::NotificationCurrentSendInProgress,
@@ -24,6 +28,15 @@ pub static NOTIFICATION_CURRENT_SEND_IN_PROGRESS_MD: LazyLock<MetricDescriptor>
)
});
pub static NOTIFICATION_CURRENT_SEND_IN_PROGRESS_BY_SERVER_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_gauge_md(
MetricName::Custom("current_send_in_progress_by_server".to_string()),
"Number of concurrent async Send calls active to all targets by server",
&SERVER_LABELS,
subsystems::NOTIFICATION,
)
});
pub static NOTIFICATION_EVENTS_ERRORS_TOTAL_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::NotificationEventsErrorsTotal,
@@ -33,6 +46,15 @@ pub static NOTIFICATION_EVENTS_ERRORS_TOTAL_MD: LazyLock<MetricDescriptor> = Laz
)
});
pub static NOTIFICATION_EVENTS_ERRORS_TOTAL_BY_SERVER_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::Custom("events_errors_total_by_server".to_string()),
"Events that failed to be sent to the targets by server",
&SERVER_LABELS,
subsystems::NOTIFICATION,
)
});
pub static NOTIFICATION_EVENTS_SENT_TOTAL_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::NotificationEventsSentTotal,
@@ -42,6 +64,15 @@ pub static NOTIFICATION_EVENTS_SENT_TOTAL_MD: LazyLock<MetricDescriptor> = LazyL
)
});
pub static NOTIFICATION_EVENTS_SENT_TOTAL_BY_SERVER_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::Custom("events_sent_total_by_server".to_string()),
"Total number of events sent to the targets by server",
&SERVER_LABELS,
subsystems::NOTIFICATION,
)
});
pub static NOTIFICATION_EVENTS_SKIPPED_TOTAL_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::NotificationEventsSkippedTotal,
@@ -50,3 +81,12 @@ pub static NOTIFICATION_EVENTS_SKIPPED_TOTAL_MD: LazyLock<MetricDescriptor> = La
subsystems::NOTIFICATION,
)
});
pub static NOTIFICATION_EVENTS_SKIPPED_TOTAL_BY_SERVER_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::Custom("events_skipped_total_by_server".to_string()),
"Notification dispatch attempts skipped before delivery by server",
&SERVER_LABELS,
subsystems::NOTIFICATION,
)
});
crates/obs/src/metrics/schema/entry/metric_name.rs
+4
View File
@@ -371,6 +371,8 @@ pub enum MetricName {
DriveWaitingIO,
DriveAPILatencyMicros,
DriveHealth,
DriveWritesTotal,
DriveDeletesTotal,
DriveOfflineCount,
DriveOnlineCount,
@@ -780,6 +782,8 @@ impl MetricName {
Self::DriveWaitingIO => "waiting_io".to_string(),
Self::DriveAPILatencyMicros => "api_latency_micros".to_string(),
Self::DriveHealth => "health".to_string(),
Self::DriveWritesTotal => "writes_total".to_string(),
Self::DriveDeletesTotal => "deletes_total".to_string(),
Self::DriveOfflineCount => "offline_count".to_string(),
Self::DriveOnlineCount => "online_count".to_string(),
crates/obs/src/metrics/schema/ilm.rs
+40
View File
@@ -18,6 +18,10 @@ use std::sync::LazyLock;
pub const SERVER_LABEL: &str = "server";
pub const ACTION_LABEL: &str = "action";
pub const STATE_LABEL: &str = "state";
pub const QUEUE_STATE_LABEL: &str = "queue_state";
pub const RESULT_LABEL: &str = "result";
pub const REASON_LABEL: &str = "reason";
pub const SOURCE_LABEL: &str = "source";
pub static ILM_ACTION_TASKS_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_gauge_md(
@@ -28,6 +32,33 @@ pub static ILM_ACTION_TASKS_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
)
});
pub static ILM_TASKS_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_gauge_md(
MetricName::Custom("tasks".to_string()),
"Current ILM task counts by server, action, and queue state",
&[SERVER_LABEL, ACTION_LABEL, QUEUE_STATE_LABEL],
subsystems::ILM,
)
});
pub static ILM_TASK_EVENTS_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::Custom("task_events_total".to_string()),
"ILM task events by server, action, and result",
&[SERVER_LABEL, ACTION_LABEL, RESULT_LABEL],
subsystems::ILM,
)
});
pub static ILM_QUEUE_BACKPRESSURE_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::Custom("queue_backpressure_total".to_string()),
"ILM queue backpressure events by server, action, and reason",
&[SERVER_LABEL, ACTION_LABEL, REASON_LABEL],
subsystems::ILM,
)
});
pub static ILM_EXPIRY_PENDING_TASKS_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_gauge_md(
MetricName::IlmExpiryPendingTasks,
@@ -108,3 +139,12 @@ pub static ILM_VERSIONS_SCANNED_MD: LazyLock<MetricDescriptor> = LazyLock::new(|
subsystems::ILM,
)
});
pub static ILM_VERSIONS_SCANNED_BY_SERVER_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::Custom("versions_scanned_by_server".to_string()),
"ILM lifecycle-checked object versions by server and source",
&[SERVER_LABEL, SOURCE_LABEL],
subsystems::ILM,
)
});
crates/obs/src/metrics/schema/scanner.rs
+18
View File
@@ -59,6 +59,24 @@ pub static SCANNER_CYCLE_BUCKET_DRIVE_RESULT_MD: LazyLock<MetricDescriptor> = La
)
});
pub static SCANNER_ACTIVE_BUCKET_DRIVE_SCANS_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_gauge_md(
MetricName::Custom("active_bucket_drive_scans".to_string()),
"Current active scanner bucket-drive scans by server, source, bucket, and drive",
&[SERVER_LABEL, SOURCE_LABEL, BUCKET_LABEL, DRIVE_LABEL],
subsystems::SCANNER,
)
});
pub static SCANNER_ACTIVE_BUCKET_DRIVE_SCAN_AGE_SECONDS_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_gauge_md(
MetricName::Custom("active_bucket_drive_scan_age_seconds".to_string()),
"Age of the oldest active scanner bucket-drive scan by server, source, bucket, and drive",
&[SERVER_LABEL, SOURCE_LABEL, BUCKET_LABEL, DRIVE_LABEL],
subsystems::SCANNER,
)
});
pub static SCANNER_BUCKET_SCANS_FINISHED_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::ScannerBucketScansFinished,
crates/obs/src/metrics/schema/system_drive.rs
+18
View File
@@ -259,6 +259,24 @@ pub static DRIVE_HEALTH_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
)
});
pub static DRIVE_WRITES_TOTAL_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::DriveWritesTotal,
"Total successful write operations on a drive",
&ALL_DRIVE_LABELS[..],
subsystems::SYSTEM_DRIVE,
)
});
pub static DRIVE_DELETES_TOTAL_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::DriveDeletesTotal,
"Total successful delete operations on a drive",
&ALL_DRIVE_LABELS[..],
subsystems::SYSTEM_DRIVE,
)
});
pub static DRIVE_OFFLINE_COUNT_MD: LazyLock<MetricDescriptor> =
LazyLock::new(|| new_gauge_md(MetricName::DriveOfflineCount, "Count of offline drives", &[], subsystems::SYSTEM_DRIVE));