logical size calculation metrics: differentiate regular vs imitated (#4197)

I want this distinction so I can prove my assumption that the disk IO
peaks which we see every 24h on prod are due to eviction's imitate
synthetic size calculations.

refs https://github.com/neondatabase/neon/issues/4154

pageserver/src/metrics.rs

@@ -30,6 +30,7 @@ const STORAGE_TIME_OPERATIONS: &[&str] = &[
     "create images",
     "init logical size",
     "logical size",
+    "imitate logical size",
     "load layer map",
     "gc",
 ];
@@ -688,6 +689,7 @@ pub struct TimelineMetrics {
     pub compact_time_histo: StorageTimeMetrics,
     pub create_images_time_histo: StorageTimeMetrics,
     pub logical_size_histo: StorageTimeMetrics,
+    pub imitate_logical_size_histo: StorageTimeMetrics,
     pub load_layer_map_histo: StorageTimeMetrics,
     pub garbage_collect_histo: StorageTimeMetrics,
     pub last_record_gauge: IntGauge,
@@ -720,6 +722,8 @@ impl TimelineMetrics {
         let create_images_time_histo =
             StorageTimeMetrics::new("create images", &tenant_id, &timeline_id);
         let logical_size_histo = StorageTimeMetrics::new("logical size", &tenant_id, &timeline_id);
+        let imitate_logical_size_histo =
+            StorageTimeMetrics::new("imitate logical size", &tenant_id, &timeline_id);
         let load_layer_map_histo =
             StorageTimeMetrics::new("load layer map", &tenant_id, &timeline_id);
         let garbage_collect_histo = StorageTimeMetrics::new("gc", &tenant_id, &timeline_id);
@@ -756,6 +760,7 @@ impl TimelineMetrics {
             compact_time_histo,
             create_images_time_histo,
             logical_size_histo,
+            imitate_logical_size_histo,
             garbage_collect_histo,
             load_layer_map_histo,
             last_record_gauge,

pageserver/src/tenant/timeline.rs

@@ -49,7 +49,7 @@ use crate::tenant::{
 
 use crate::config::PageServerConf;
 use crate::keyspace::{KeyPartitioning, KeySpace};
-use crate::metrics::{TimelineMetrics, UNEXPECTED_ONDEMAND_DOWNLOADS};
+use crate::metrics::{StorageTimeMetrics, TimelineMetrics, UNEXPECTED_ONDEMAND_DOWNLOADS};
 use crate::pgdatadir_mapping::LsnForTimestamp;
 use crate::pgdatadir_mapping::{is_rel_fsm_block_key, is_rel_vm_block_key};
 use crate::pgdatadir_mapping::{BlockNumber, CalculateLogicalSizeError};
@@ -1938,7 +1938,12 @@ impl Timeline {
             let cancel = cancel.child_token();
             let ctx = ctx.attached_child();
             self_calculation
-                .calculate_logical_size(lsn, cancel, &ctx)
+                .calculate_logical_size(
+                    lsn,
+                    &self_calculation.metrics.logical_size_histo,
+                    cancel,
+                    &ctx,
+                )
                 .await
         });
         let timeline_state_cancellation = async {
@@ -1993,6 +1998,7 @@ impl Timeline {
     pub async fn calculate_logical_size(
         &self,
         up_to_lsn: Lsn,
+        storage_time_metrics: &StorageTimeMetrics,
         cancel: CancellationToken,
         ctx: &RequestContext,
     ) -> Result<u64, CalculateLogicalSizeError> {
@@ -2026,7 +2032,7 @@ impl Timeline {
         if let Some(size) = self.current_logical_size.initialized_size(up_to_lsn) {
             return Ok(size);
         }
-        let timer = self.metrics.logical_size_histo.start_timer();
+        let timer = storage_time_metrics.start_timer();
         let logical_size = self
             .get_current_logical_size_non_incremental(up_to_lsn, cancel, ctx)
             .await?;

pageserver/src/tenant/timeline/eviction_task.rs

@@ -335,7 +335,12 @@ impl Timeline {
 
         // imitiate on-restart initial logical size
         let size = self
-            .calculate_logical_size(lsn, cancel.clone(), ctx)
+            .calculate_logical_size(
+                lsn,
+                &self.metrics.imitate_logical_size_histo,
+                cancel.clone(),
+                ctx,
+            )
             .instrument(info_span!("calculate_logical_size"))
             .await;