From 9f097af0fcc57413190ed4468c0777a278ec0e55 Mon Sep 17 00:00:00 2001
From: Vlad Lazar <vlad@neon.tech>
Date: Wed, 4 Dec 2024 12:33:53 +0100
Subject: [PATCH] pageserver: add disk consistent and remote lsn metrics

---
 pageserver/src/metrics.rs                     | 24 +++++++++++++++++++
 .../src/tenant/remote_timeline_client.rs      |  8 +++++++
 pageserver/src/tenant/timeline.rs             |  4 ++++
 3 files changed, 36 insertions(+)

diff --git a/pageserver/src/metrics.rs b/pageserver/src/metrics.rs
index e5707356110f..f2f00dc67395 100644
--- a/pageserver/src/metrics.rs
+++ b/pageserver/src/metrics.rs
@@ -464,6 +464,24 @@ static LAST_RECORD_LSN: Lazy<IntGaugeVec> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
+static DISK_CONSISTENT_LSN: Lazy<IntGaugeVec> = Lazy::new(|| {
+    register_int_gauge_vec!(
+        "pageserver_disk_consistent_lsn",
+        "Disk consistent LSN grouped by timeline",
+        &["tenant_id", "shard_id", "timeline_id"]
+    )
+    .expect("failed to define a metric")
+});
+
+pub(crate) static PROJECTED_REMOTE_CONSISTENT_LSN: Lazy<IntGaugeVec> = Lazy::new(|| {
+    register_int_gauge_vec!(
+        "pageserver_projected_remote_consistent_lsn",
+        "Projected remote consistent LSN grouped by timeline",
+        &["tenant_id", "shard_id", "timeline_id"]
+    )
+    .expect("failed to define a metric")
+});
+
 static PITR_HISTORY_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
     register_uint_gauge_vec!(
         "pageserver_pitr_history_size",
@@ -2395,6 +2413,7 @@ pub(crate) struct TimelineMetrics {
     pub garbage_collect_histo: StorageTimeMetrics,
     pub find_gc_cutoffs_histo: StorageTimeMetrics,
     pub last_record_lsn_gauge: IntGauge,
+    pub disk_consistent_lsn_gauge: IntGauge,
     pub pitr_history_size: UIntGauge,
     pub archival_size: UIntGauge,
     pub(crate) layer_size_image: UIntGauge,
@@ -2479,6 +2498,10 @@ impl TimelineMetrics {
             .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
             .unwrap();
 
+        let disk_consistent_lsn_gauge = DISK_CONSISTENT_LSN
+            .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
+            .unwrap();
+
         let pitr_history_size = PITR_HISTORY_SIZE
             .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
             .unwrap();
@@ -2579,6 +2602,7 @@ impl TimelineMetrics {
             find_gc_cutoffs_histo,
             load_layer_map_histo,
             last_record_lsn_gauge,
+            disk_consistent_lsn_gauge,
             pitr_history_size,
             archival_size,
             layer_size_image,
diff --git a/pageserver/src/tenant/remote_timeline_client.rs b/pageserver/src/tenant/remote_timeline_client.rs
index 007bd3eef083..94f69256d74a 100644
--- a/pageserver/src/tenant/remote_timeline_client.rs
+++ b/pageserver/src/tenant/remote_timeline_client.rs
@@ -2190,6 +2190,14 @@ impl RemoteTimelineClient {
                     upload_queue.clean.1 = Some(task.task_id);
 
                     let lsn = upload_queue.clean.0.metadata.disk_consistent_lsn();
+                    crate::metrics::PROJECTED_REMOTE_CONSISTENT_LSN
+                        .get_metric_with_label_values(&[
+                            &self.tenant_shard_id.tenant_id.to_string(),
+                            &format!("{}", self.tenant_shard_id.shard_slug()),
+                            &self.timeline_id.to_string(),
+                        ])
+                        .unwrap()
+                        .set(lsn.0 as i64);
 
                     if self.generation.is_none() {
                         // Legacy mode: skip validating generation
diff --git a/pageserver/src/tenant/timeline.rs b/pageserver/src/tenant/timeline.rs
index 02a839a83bce..444c0b762384 100644
--- a/pageserver/src/tenant/timeline.rs
+++ b/pageserver/src/tenant/timeline.rs
@@ -3850,6 +3850,10 @@ impl Timeline {
     fn set_disk_consistent_lsn(&self, new_value: Lsn) -> bool {
         let old_value = self.disk_consistent_lsn.fetch_max(new_value);
         assert!(new_value >= old_value, "disk_consistent_lsn must be growing monotonously at runtime; current {old_value}, offered {new_value}");
+
+        self.metrics
+            .disk_consistent_lsn_gauge
+            .set(new_value.0 as i64);
         new_value != old_value
     }