gsadmin/rustfs
1
0
Fork 0
mirror of https://github.com/rustfs/rustfs.git synced 2026-08-08 22:33:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix(replication): harden backlog observability (#5564)

Browse Source 
Add RAII guards for replication runtime backlog tickets so active worker and queue counters unwind on every terminal path.

Expose node-local MRF pending, dropped, missed, and flush-failure metrics through the bucket replication Prometheus collector while keeping the existing current backlog and durable MRF gauges additive.

Update durable MRF summary maintenance to aggregate incrementally during the persister loop, avoiding repeated full-entry scans on each successful flush.

Co-authored-by: heihutu <heihutu@gmail.com>
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
This commit is contained in:
houseme
2026-08-01 19:12:58 +08:00
committed by GitHub
parent b7805caa58
commit da389c0e21
7 changed files with 1024 additions and 164 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/obs/src/metrics/collectors/bucket_replication.rs
+98 -2
View File
@@ -20,6 +20,8 @@ use crate::metrics::schema::bucket_replication::{
BUCKET_REPL_CURRENT_BACKLOG_COUNT_MD, BUCKET_REPL_DURABLE_MRF_AVAILABLE_MD, BUCKET_REPL_DURABLE_MRF_BACKLOG_BYTES_MD,
BUCKET_REPL_DURABLE_MRF_BACKLOG_COUNT_MD, BUCKET_REPL_LAST_HR_FAILED_BYTES_MD, BUCKET_REPL_LAST_HR_FAILED_COUNT_MD,
BUCKET_REPL_LAST_MIN_FAILED_BYTES_MD, BUCKET_REPL_LAST_MIN_FAILED_COUNT_MD, BUCKET_REPL_LATENCY_MS_MD,
BUCKET_REPL_MRF_DROPPED_COUNT_MD, BUCKET_REPL_MRF_FLUSH_FAILURES_MD, BUCKET_REPL_MRF_LAST_FLUSH_DURATION_MILLIS_MD,
BUCKET_REPL_MRF_MISSED_COUNT_MD, BUCKET_REPL_MRF_PENDING_BYTES_MD, BUCKET_REPL_MRF_PENDING_COUNT_MD,
BUCKET_REPL_PROXIED_DELETE_TAGGING_REQUESTS_FAILURES_MD, BUCKET_REPL_PROXIED_DELETE_TAGGING_REQUESTS_TOTAL_MD,
BUCKET_REPL_PROXIED_GET_REQUESTS_FAILURES_MD, BUCKET_REPL_PROXIED_GET_REQUESTS_TOTAL_MD,
BUCKET_REPL_PROXIED_GET_TAGGING_REQUESTS_FAILURES_MD, BUCKET_REPL_PROXIED_GET_TAGGING_REQUESTS_TOTAL_MD,
@@ -34,7 +36,7 @@ use crate::metrics::schema::bucket_replication::{
use std::borrow::Cow;
const BASE_BUCKET_REPLICATION_METRICS_PER_BUCKET: usize = 25;
const BASE_BUCKET_REPLICATION_BACKLOG_METRICS_PER_BUCKET: usize = 5;
const BASE_BUCKET_REPLICATION_BACKLOG_METRICS_PER_BUCKET: usize = 11;
#[derive(Debug, Clone, Default)]
pub struct BucketReplicationTargetStats {
@@ -91,6 +93,12 @@ pub(crate) struct BucketReplicationBacklogStats {
pub(crate) durable_mrf_available: bool,
pub(crate) durable_mrf_backlog_count: u64,
pub(crate) durable_mrf_backlog_bytes: u64,
pub(crate) mrf_pending_count: u64,
pub(crate) mrf_pending_bytes: u64,
pub(crate) mrf_dropped_count: u64,
pub(crate) mrf_missed_count: u64,
pub(crate) mrf_flush_failures: u64,
pub(crate) mrf_last_flush_duration_millis: u64,
}
pub fn collect_bucket_replication_bandwidth_metrics(stats: &[BucketReplicationBandwidthStats]) -> Vec<PrometheusMetric> {
@@ -309,6 +317,34 @@ pub(crate) fn collect_bucket_replication_backlog_metrics(stats: &[BucketReplicat
PrometheusMetric::from_descriptor(&BUCKET_REPL_DURABLE_MRF_BACKLOG_BYTES_MD, stat.durable_mrf_backlog_bytes as f64)
.with_label(BUCKET_L, bucket_label),
);
let bucket_label: Cow<'static, str> = Cow::Owned(stat.bucket.clone());
metrics.push(
PrometheusMetric::from_descriptor(&BUCKET_REPL_MRF_PENDING_COUNT_MD, stat.mrf_pending_count as f64)
.with_label(BUCKET_L, bucket_label.clone()),
);
metrics.push(
PrometheusMetric::from_descriptor(&BUCKET_REPL_MRF_PENDING_BYTES_MD, stat.mrf_pending_bytes as f64)
.with_label(BUCKET_L, bucket_label.clone()),
);
metrics.push(
PrometheusMetric::from_descriptor(&BUCKET_REPL_MRF_DROPPED_COUNT_MD, stat.mrf_dropped_count as f64)
.with_label(BUCKET_L, bucket_label.clone()),
);
metrics.push(
PrometheusMetric::from_descriptor(&BUCKET_REPL_MRF_MISSED_COUNT_MD, stat.mrf_missed_count as f64)
.with_label(BUCKET_L, bucket_label.clone()),
);
metrics.push(
PrometheusMetric::from_descriptor(&BUCKET_REPL_MRF_FLUSH_FAILURES_MD, stat.mrf_flush_failures as f64)
.with_label(BUCKET_L, bucket_label.clone()),
);
metrics.push(
PrometheusMetric::from_descriptor(
&BUCKET_REPL_MRF_LAST_FLUSH_DURATION_MILLIS_MD,
stat.mrf_last_flush_duration_millis as f64,
)
.with_label(BUCKET_L, bucket_label),
);
}
metrics
@@ -427,10 +463,16 @@ mod tests {
durable_mrf_available: true,
durable_mrf_backlog_count: 2,
durable_mrf_backlog_bytes: 2048,
mrf_pending_count: 1,
mrf_pending_bytes: 512,
mrf_dropped_count: 3,
mrf_missed_count: 4,
mrf_flush_failures: 5,
mrf_last_flush_duration_millis: 6,
}];
let metrics = collect_bucket_replication_backlog_metrics(&stats);
assert_eq!(metrics.len(), 5);
assert_eq!(metrics.len(), 11);
let backlog_count_name = BUCKET_REPL_CURRENT_BACKLOG_COUNT_MD.get_full_metric_name();
assert!(metrics.iter().any(|metric| {
@@ -466,6 +508,48 @@ mod tests {
&& metric.value == 2048.0
&& metric.labels.iter().any(|(key, value)| *key == BUCKET_L && value == "b1")
}));
let pending_count_name = BUCKET_REPL_MRF_PENDING_COUNT_MD.get_full_metric_name();
assert!(metrics.iter().any(|metric| {
metric.name == pending_count_name
&& metric.value == 1.0
&& metric.labels.iter().any(|(key, value)| *key == BUCKET_L && value == "b1")
}));
let pending_bytes_name = BUCKET_REPL_MRF_PENDING_BYTES_MD.get_full_metric_name();
assert!(metrics.iter().any(|metric| {
metric.name == pending_bytes_name
&& metric.value == 512.0
&& metric.labels.iter().any(|(key, value)| *key == BUCKET_L && value == "b1")
}));
let dropped_count_name = BUCKET_REPL_MRF_DROPPED_COUNT_MD.get_full_metric_name();
assert!(metrics.iter().any(|metric| {
metric.name == dropped_count_name
&& metric.value == 3.0
&& metric.labels.iter().any(|(key, value)| *key == BUCKET_L && value == "b1")
}));
let missed_count_name = BUCKET_REPL_MRF_MISSED_COUNT_MD.get_full_metric_name();
assert!(metrics.iter().any(|metric| {
metric.name == missed_count_name
&& metric.value == 4.0
&& metric.labels.iter().any(|(key, value)| *key == BUCKET_L && value == "b1")
}));
let flush_failures_name = BUCKET_REPL_MRF_FLUSH_FAILURES_MD.get_full_metric_name();
assert!(metrics.iter().any(|metric| {
metric.name == flush_failures_name
&& metric.value == 5.0
&& metric.labels.iter().any(|(key, value)| *key == BUCKET_L && value == "b1")
}));
let flush_duration_name = BUCKET_REPL_MRF_LAST_FLUSH_DURATION_MILLIS_MD.get_full_metric_name();
assert!(metrics.iter().any(|metric| {
metric.name == flush_duration_name
&& metric.value == 6.0
&& metric.labels.iter().any(|(key, value)| *key == BUCKET_L && value == "b1")
}));
}
#[test]
@@ -484,6 +568,12 @@ mod tests {
durable_mrf_available: true,
durable_mrf_backlog_count: 2,
durable_mrf_backlog_bytes: 4096,
mrf_pending_count: 1,
mrf_pending_bytes: 2,
mrf_dropped_count: 3,
mrf_missed_count: 4,
mrf_flush_failures: 5,
mrf_last_flush_duration_millis: 6,
}];
let metrics = collect_bucket_replication_backlog_metrics(&stats);
@@ -493,6 +583,12 @@ mod tests {
BUCKET_REPL_DURABLE_MRF_AVAILABLE_MD.get_full_metric_name(),
BUCKET_REPL_DURABLE_MRF_BACKLOG_COUNT_MD.get_full_metric_name(),
BUCKET_REPL_DURABLE_MRF_BACKLOG_BYTES_MD.get_full_metric_name(),
BUCKET_REPL_MRF_PENDING_COUNT_MD.get_full_metric_name(),
BUCKET_REPL_MRF_PENDING_BYTES_MD.get_full_metric_name(),
BUCKET_REPL_MRF_DROPPED_COUNT_MD.get_full_metric_name(),
BUCKET_REPL_MRF_MISSED_COUNT_MD.get_full_metric_name(),
BUCKET_REPL_MRF_FLUSH_FAILURES_MD.get_full_metric_name(),
BUCKET_REPL_MRF_LAST_FLUSH_DURATION_MILLIS_MD.get_full_metric_name(),
];
for name in backlog_names {
crates/obs/src/metrics/schema/bucket_replication.rs
+60
View File
@@ -38,6 +38,12 @@ const CURRENT_BACKLOG_BYTES: &str = "current_backlog_bytes";
const DURABLE_MRF_AVAILABLE: &str = "durable_mrf_available";
const DURABLE_MRF_BACKLOG_COUNT: &str = "durable_mrf_backlog_count";
const DURABLE_MRF_BACKLOG_BYTES: &str = "durable_mrf_backlog_bytes";
const MRF_PENDING_COUNT: &str = "mrf_pending_count";
const MRF_PENDING_BYTES: &str = "mrf_pending_bytes";
const MRF_DROPPED_COUNT: &str = "mrf_dropped_count";
const MRF_MISSED_COUNT: &str = "mrf_missed_count";
const MRF_FLUSH_FAILURES: &str = "mrf_flush_failures";
const MRF_LAST_FLUSH_DURATION_MILLIS: &str = "mrf_last_flush_duration_millis";
pub static BUCKET_REPL_LAST_HR_FAILED_BYTES_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_gauge_md(
@@ -129,6 +135,60 @@ pub static BUCKET_REPL_DURABLE_MRF_BACKLOG_BYTES_MD: LazyLock<MetricDescriptor>
)
});
pub static BUCKET_REPL_MRF_PENDING_COUNT_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_gauge_md(
MetricName::from(MRF_PENDING_COUNT),
"Current number of MRF entries waiting to be flushed to the durable recovery file for a bucket on this node",
&[BUCKET_L],
subsystems::BUCKET_REPLICATION,
)
});
pub static BUCKET_REPL_MRF_PENDING_BYTES_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_gauge_md(
MetricName::from(MRF_PENDING_BYTES),
"Current bytes represented by MRF entries waiting to be flushed to the durable recovery file for a bucket on this node",
&[BUCKET_L],
subsystems::BUCKET_REPLICATION,
)
});
pub static BUCKET_REPL_MRF_DROPPED_COUNT_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::from(MRF_DROPPED_COUNT),
"Total number of MRF entries dropped after the bounded pending backlog cap was reached for a bucket on this node",
&[BUCKET_L],
subsystems::BUCKET_REPLICATION,
)
});
pub static BUCKET_REPL_MRF_MISSED_COUNT_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::from(MRF_MISSED_COUNT),
"Total number of MRF entries that could not be admitted to the save channel for a bucket on this node",
&[BUCKET_L],
subsystems::BUCKET_REPLICATION,
)
});
pub static BUCKET_REPL_MRF_FLUSH_FAILURES_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::from(MRF_FLUSH_FAILURES),
"Total number of durable MRF flush failures observed while a bucket had pending MRF entries on this node",
&[BUCKET_L],
subsystems::BUCKET_REPLICATION,
)
});
pub static BUCKET_REPL_MRF_LAST_FLUSH_DURATION_MILLIS_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_gauge_md(
MetricName::from(MRF_LAST_FLUSH_DURATION_MILLIS),
"Duration in milliseconds of the last durable MRF flush that touched pending entries for a bucket on this node",
&[BUCKET_L],
subsystems::BUCKET_REPLICATION,
)
});
pub static BUCKET_REPL_PROXIED_DELETE_TAGGING_REQUESTS_TOTAL_MD: LazyLock<MetricDescriptor> = LazyLock::new(|| {
new_counter_md(
MetricName::ProxiedDeleteTaggingRequestsTotal,
crates/obs/src/metrics/stats_collector.rs
+6
View File
@@ -206,6 +206,12 @@ async fn obs_bucket_replication_stats_bundle() -> (Vec<BucketReplicationStats>,
durable_mrf_available: stats.durable_mrf_available,
durable_mrf_backlog_count: stats.durable_mrf_backlog_count,
durable_mrf_backlog_bytes: stats.durable_mrf_backlog_bytes,
mrf_pending_count: stats.mrf_pending_count,
mrf_pending_bytes: stats.mrf_pending_bytes,
mrf_dropped_count: stats.mrf_dropped_count,
mrf_missed_count: stats.mrf_missed_count,
mrf_flush_failures: stats.mrf_flush_failures,
mrf_last_flush_duration_millis: stats.mrf_last_flush_duration_millis,
});
detail_stats.push(bucket_replication_detail_from_snapshot(stats));
}
crates/obs/src/metrics/storage_api.rs
+91 -3
View File
@@ -18,7 +18,8 @@ use std::time::Duration;
pub(crate) use rustfs_ecstore::api::bucket::bandwidth::monitor::Monitor as ObsBucketBandwidthMonitor;
pub(crate) use rustfs_ecstore::api::bucket::metadata_sys::get_quota_config as obs_get_quota_config;
use rustfs_ecstore::api::bucket::replication::{
DurableMrfBucketBacklog, durable_mrf_backlog_summary_snapshot, get_global_replication_stats,
DurableMrfBucketBacklog, MrfBucketBacklogObservability, durable_mrf_backlog_summary_snapshot, get_global_replication_stats,
mrf_backlog_observability_snapshot,
};
pub(crate) use rustfs_ecstore::api::capacity::{
get_total_usable_capacity as obs_get_total_usable_capacity,
@@ -76,6 +77,12 @@ pub(crate) struct ObsBucketReplicationStatsSnapshot {
pub(crate) durable_mrf_available: bool,
pub(crate) durable_mrf_backlog_count: u64,
pub(crate) durable_mrf_backlog_bytes: u64,
pub(crate) mrf_pending_count: u64,
pub(crate) mrf_pending_bytes: u64,
pub(crate) mrf_dropped_count: u64,
pub(crate) mrf_missed_count: u64,
pub(crate) mrf_flush_failures: u64,
pub(crate) mrf_last_flush_duration_millis: u64,
pub(crate) targets: Vec<ObsBucketReplicationTargetStatsSnapshot>,
}
@@ -152,6 +159,7 @@ fn bucket_replication_stats_snapshot_from_parts(
proxy: ObsBucketReplicationProxySnapshot,
durable_mrf_available: bool,
durable_bucket: DurableMrfBucketBacklog,
mrf_observability: MrfBucketBacklogObservability,
) -> ObsBucketReplicationStatsSnapshot {
ObsBucketReplicationStatsSnapshot {
bucket,
@@ -185,6 +193,12 @@ fn bucket_replication_stats_snapshot_from_parts(
durable_mrf_available,
durable_mrf_backlog_count: durable_bucket.count,
durable_mrf_backlog_bytes: durable_bucket.bytes,
mrf_pending_count: mrf_observability.pending_count,
mrf_pending_bytes: mrf_observability.pending_bytes,
mrf_dropped_count: mrf_observability.dropped_count,
mrf_missed_count: mrf_observability.missed_count,
mrf_flush_failures: mrf_observability.flush_failure_count,
mrf_last_flush_duration_millis: mrf_observability.last_flush_duration_millis,
targets: runtime.targets,
}
}
@@ -196,7 +210,8 @@ pub(crate) async fn obs_bucket_replication_stats_snapshot() -> Vec<ObsBucketRepl
} else {
HashMap::new()
};
let durable_mrf_summary = if obs_resolve_object_store_handle().is_some() {
let replication_storage_available = obs_resolve_object_store_handle().is_some();
let durable_mrf_summary = if replication_storage_available {
durable_mrf_backlog_summary_snapshot()
} else {
Default::default()
@@ -207,7 +222,22 @@ pub(crate) async fn obs_bucket_replication_stats_snapshot() -> Vec<ObsBucketRepl
.into_iter()
.map(|bucket| (bucket.bucket.clone(), bucket))
.collect::<HashMap<String, DurableMrfBucketBacklog>>();
let mut bucket_names = Vec::with_capacity(all_bucket_stats.len().saturating_add(durable_buckets.len()));
let mrf_observability = if replication_storage_available {
mrf_backlog_observability_snapshot()
} else {
Default::default()
};
let mrf_observability_buckets = mrf_observability
.buckets
.into_iter()
.map(|bucket| (bucket.bucket.clone(), bucket))
.collect::<HashMap<String, MrfBucketBacklogObservability>>();
let mut bucket_names = Vec::with_capacity(
all_bucket_stats
.len()
.saturating_add(durable_buckets.len())
.saturating_add(mrf_observability_buckets.len()),
);
bucket_names.extend(all_bucket_stats.keys().cloned());
bucket_names.extend(
durable_buckets
@@ -215,6 +245,12 @@ pub(crate) async fn obs_bucket_replication_stats_snapshot() -> Vec<ObsBucketRepl
.filter(|bucket| !all_bucket_stats.contains_key(*bucket))
.cloned(),
);
bucket_names.extend(
mrf_observability_buckets
.keys()
.filter(|bucket| !all_bucket_stats.contains_key(*bucket) && !durable_buckets.contains_key(*bucket))
.cloned(),
);
let mut buckets = Vec::with_capacity(bucket_names.len());
for bucket in bucket_names {
@@ -291,12 +327,14 @@ pub(crate) async fn obs_bucket_replication_stats_snapshot() -> Vec<ObsBucketRepl
runtime.current_backlog_bytes = i64_to_u64_floor_zero(bucket_stats.q_stat.curr.bytes);
}
let durable_bucket = durable_buckets.get(&bucket).cloned().unwrap_or_default();
let mrf_observability = mrf_observability_buckets.get(&bucket).cloned().unwrap_or_default();
buckets.push(bucket_replication_stats_snapshot_from_parts(
bucket,
runtime,
proxy,
durable_mrf_available,
durable_bucket,
mrf_observability,
));
}
@@ -394,6 +432,15 @@ mod tests {
count: 5,
bytes: 8192,
},
MrfBucketBacklogObservability {
bucket: "runtime-bucket".to_string(),
pending_count: 1,
pending_bytes: 512,
dropped_count: 2,
missed_count: 3,
flush_failure_count: 4,
last_flush_duration_millis: 5,
},
);
assert_eq!(snapshot.bucket, "runtime-bucket");
@@ -402,6 +449,12 @@ mod tests {
assert!(snapshot.durable_mrf_available);
assert_eq!(snapshot.durable_mrf_backlog_count, 5);
assert_eq!(snapshot.durable_mrf_backlog_bytes, 8192);
assert_eq!(snapshot.mrf_pending_count, 1);
assert_eq!(snapshot.mrf_pending_bytes, 512);
assert_eq!(snapshot.mrf_dropped_count, 2);
assert_eq!(snapshot.mrf_missed_count, 3);
assert_eq!(snapshot.mrf_flush_failures, 4);
assert_eq!(snapshot.mrf_last_flush_duration_millis, 5);
assert_eq!(snapshot.resync_failed_count, 2);
assert_eq!(snapshot.proxied_get_requests_total, 7);
assert_eq!(snapshot.proxied_get_requests_failures, 1);
@@ -419,6 +472,7 @@ mod tests {
count: 11,
bytes: 2048,
},
MrfBucketBacklogObservability::default(),
);
assert_eq!(snapshot.bucket, "durable-only");
@@ -428,6 +482,39 @@ mod tests {
assert_eq!(snapshot.durable_mrf_backlog_bytes, 2048);
}
#[test]
fn bucket_replication_snapshot_reports_mrf_observability_only_bucket() {
let snapshot = bucket_replication_stats_snapshot_from_parts(
"mrf-observability-only".to_string(),
ObsBucketReplicationRuntimeSnapshot::default(),
ObsBucketReplicationProxySnapshot::default(),
true,
DurableMrfBucketBacklog::default(),
MrfBucketBacklogObservability {
bucket: "mrf-observability-only".to_string(),
pending_count: 13,
pending_bytes: 4096,
dropped_count: 1,
missed_count: 2,
flush_failure_count: 3,
last_flush_duration_millis: 4,
},
);
assert_eq!(snapshot.bucket, "mrf-observability-only");
assert_eq!(snapshot.current_backlog_count, 0);
assert_eq!(snapshot.current_backlog_bytes, 0);
assert!(snapshot.durable_mrf_available);
assert_eq!(snapshot.durable_mrf_backlog_count, 0);
assert_eq!(snapshot.durable_mrf_backlog_bytes, 0);
assert_eq!(snapshot.mrf_pending_count, 13);
assert_eq!(snapshot.mrf_pending_bytes, 4096);
assert_eq!(snapshot.mrf_dropped_count, 1);
assert_eq!(snapshot.mrf_missed_count, 2);
assert_eq!(snapshot.mrf_flush_failures, 3);
assert_eq!(snapshot.mrf_last_flush_duration_millis, 4);
}
#[test]
fn bucket_replication_snapshot_preserves_durable_mrf_unavailable_state() {
let snapshot = bucket_replication_stats_snapshot_from_parts(
@@ -440,6 +527,7 @@ mod tests {
ObsBucketReplicationProxySnapshot::default(),
false,
DurableMrfBucketBacklog::default(),
MrfBucketBacklogObservability::default(),
);
assert_eq!(snapshot.current_backlog_count, 1);