feat(pageserver): generate image layers for sparse keyspace

Signed-off-by: Alex Chi Z <chi@neon.tech>

collect read bytes

Signed-off-by: Alex Chi Z <chi@neon.tech>

pageserver/src/context.rs

@@ -86,6 +86,8 @@
 //! [`RequestContext`] argument. Functions in the middle of the call chain
 //! only need to pass it on.
 
+use std::sync::atomic::AtomicUsize;
+
 use crate::task_mgr::TaskKind;
 
 pub(crate) mod optional_counter;
@@ -98,6 +100,8 @@ pub struct RequestContext {
     access_stats_behavior: AccessStatsBehavior,
     page_content_kind: PageContentKind,
     pub micros_spent_throttled: optional_counter::MicroSecondsCounterU32,
+    /// Total number of kilobytes of layer files processed in this request.
+    pub vectored_access_delta_file_size_kb: AtomicUsize,
 }
 
 /// The kind of access to the page cache.
@@ -154,6 +158,7 @@ impl RequestContextBuilder {
                 access_stats_behavior: AccessStatsBehavior::Update,
                 page_content_kind: PageContentKind::Unknown,
                 micros_spent_throttled: Default::default(),
+                vectored_access_delta_file_size_kb: AtomicUsize::new(0),
             },
         }
     }
@@ -168,6 +173,7 @@ impl RequestContextBuilder {
                 access_stats_behavior: original.access_stats_behavior,
                 page_content_kind: original.page_content_kind,
                 micros_spent_throttled: Default::default(),
+                vectored_access_delta_file_size_kb: AtomicUsize::new(0),
             },
         }
     }

pageserver/src/tenant.rs

@@ -5534,7 +5534,7 @@ mod tests {
             .await?;
 
         const NUM_KEYS: usize = 1000;
-        const STEP: usize = 100; // random update + scan base_key + idx * STEP
+        const STEP: usize = 10000; // random update + scan base_key + idx * STEP
 
         let cancel = CancellationToken::new();
 
@@ -5567,7 +5567,7 @@ mod tests {
 
         let keyspace = KeySpace::single(base_key..base_key.add((NUM_KEYS * STEP) as u32));
 
-        for _ in 0..10 {
+        for iter in 0..=10 {
             // Read all the blocks
             for (blknum, last_lsn) in updated.iter().enumerate() {
                 test_key.field6 = (blknum * STEP) as u32;
@@ -5618,12 +5618,27 @@ mod tests {
                 updated[blknum] = lsn;
             }
 
-            // Perform a cycle of flush, compact, and GC
-            tline.freeze_and_flush().await?;
-            tline.compact(&cancel, EnumSet::empty(), &ctx).await?;
-            tenant
-                .gc_iteration(Some(tline.timeline_id), 0, Duration::ZERO, &cancel, &ctx)
-                .await?;
+            // Perform two cycles of flush, compact, and GC
+            for round in 0..2 {
+                tline.freeze_and_flush().await?;
+                tline
+                    .compact(
+                        &cancel,
+                        if iter % 5 == 0 && round == 0 {
+                            let mut flags = EnumSet::new();
+                            flags.insert(CompactFlags::ForceImageLayerCreation);
+                            flags.insert(CompactFlags::ForceRepartition);
+                            flags
+                        } else {
+                            EnumSet::empty()
+                        },
+                        &ctx,
+                    )
+                    .await?;
+                tenant
+                    .gc_iteration(Some(tline.timeline_id), 0, Duration::ZERO, &cancel, &ctx)
+                    .await?;
+            }
         }
 
         Ok(())

pageserver/src/tenant/storage_layer.rs

@@ -119,6 +119,7 @@ pub(crate) struct ValuesReconstructState {
 
     keys_done: KeySpaceRandomAccum,
     layers_visited: u32,
+    delta_file_kb_visited: usize,
 }
 
 impl ValuesReconstructState {
@@ -127,6 +128,7 @@ impl ValuesReconstructState {
             keys: HashMap::new(),
             keys_done: KeySpaceRandomAccum::new(),
             layers_visited: 0,
+            delta_file_kb_visited: 0,
         }
     }
 
@@ -140,8 +142,17 @@ impl ValuesReconstructState {
         }
     }
 
-    pub(crate) fn on_layer_visited(&mut self) {
+    pub(crate) fn on_layer_visited(&mut self, layer: &ReadableLayer) {
         self.layers_visited += 1;
+        if let ReadableLayer::PersistentLayer(layer) = layer {
+            if layer.layer_desc().is_delta() {
+                self.delta_file_kb_visited += layer.layer_desc().file_size as usize / 1024;
+            }
+        }
+    }
+
+    pub(crate) fn get_vectored_access_delta_file_size_kb(&self) -> usize {
+        self.delta_file_kb_visited
     }
 
     pub(crate) fn get_layers_visited(&self) -> u32 {