pageserver: add pageserver_deltas_per_read_global metric (#10570)

## Problem

We suspect that Postgres checkpoints will limit the number of page
deltas necessary to reconstruct a page, but don't know for certain.

Touches neondatabase/cloud#23283.

## Summary of changes

Add `pageserver_deltas_per_read_global` metric.

This pairs with `pageserver_layers_per_read_global` from #10573.

pageserver/src/metrics.rs

@@ -131,6 +131,16 @@ pub(crate) static LAYERS_PER_READ_GLOBAL: Lazy<Histogram> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
+pub(crate) static DELTAS_PER_READ_GLOBAL: Lazy<Histogram> = Lazy::new(|| {
+    // We expect this to be low because of Postgres checkpoints. Let's see if that holds.
+    register_histogram!(
+        "pageserver_deltas_per_read_global",
+        "Number of delta pages applied to image page per read",
+        vec![0.0, 1.0, 2.0, 4.0, 8.0, 16.0, 32.0, 64.0, 128.0, 256.0],
+    )
+    .expect("failed to define a metric")
+});
+
 pub(crate) static CONCURRENT_INITDBS: Lazy<UIntGauge> = Lazy::new(|| {
     register_uint_gauge!(
         "pageserver_concurrent_initdb",
@@ -3919,6 +3929,7 @@ pub fn preinitialize_metrics(conf: &'static PageServerConf) {
     // histograms
     [
         &LAYERS_PER_READ_GLOBAL,
+        &DELTAS_PER_READ_GLOBAL,
         &WAIT_LSN_TIME,
         &WAL_REDO_TIME,
         &WAL_REDO_RECORDS_HISTOGRAM,

pageserver/src/tenant/storage_layer.rs

@@ -80,6 +80,16 @@ pub(crate) struct ValueReconstructState {
     pub(crate) img: Option<(Lsn, Bytes)>,
 }
 
+impl ValueReconstructState {
+    /// Returns the number of page deltas applied to the page image.
+    pub fn num_deltas(&self) -> usize {
+        match self.img {
+            Some(_) => self.records.len(),
+            None => self.records.len() - 1, // omit will_init record
+        }
+    }
+}
+
 #[derive(Clone, Copy, Debug, Default, Eq, PartialEq)]
 pub(crate) enum ValueReconstructSituation {
     Complete,

pageserver/src/tenant/timeline.rs

@@ -116,7 +116,7 @@ use pageserver_api::config::tenant_conf_defaults::DEFAULT_PITR_INTERVAL;
 
 use crate::config::PageServerConf;
 use crate::keyspace::{KeyPartitioning, KeySpace};
-use crate::metrics::{TimelineMetrics, LAYERS_PER_READ_GLOBAL};
+use crate::metrics::{TimelineMetrics, DELTAS_PER_READ_GLOBAL, LAYERS_PER_READ_GLOBAL};
 use crate::pgdatadir_mapping::CalculateLogicalSizeError;
 use crate::tenant::config::TenantConfOpt;
 use pageserver_api::reltag::RelTag;
@@ -1195,6 +1195,7 @@ impl Timeline {
                             return (key, Err(err));
                         }
                     };
+                    DELTAS_PER_READ_GLOBAL.observe(converted.num_deltas() as f64);
 
                     // The walredo module expects the records to be descending in terms of Lsn.
                     // And we submit the IOs in that order, so, there shuold be no need to sort here.