perf(pageserver): postpone vectored get fringe keyspace construction (#…

…7904)

Perf shows a significant amount of time is spent on `Keyspace::merge`.
This pull request postpones merging keyspace until retrieving the layer,
which contributes to a 30x improvement in aux keyspace basebackup time.

```
--- old
10000 files found in 0.580569459s
--- new
10000 files found in 0.02995075s
```

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

pageserver/pagebench/src/cmd/aux_files.rs

@@ -5,6 +5,7 @@ use utils::lsn::Lsn;
 
 use std::collections::HashMap;
 use std::sync::Arc;
+use std::time::Instant;
 
 /// Ingest aux files into the pageserver.
 #[derive(clap::Parser)]
@@ -88,11 +89,17 @@ async fn main_impl(args: Args) -> anyhow::Result<()> {
         println!("ingested {file_cnt} files");
     }
 
-    let files = mgmt_api_client
-        .list_aux_files(tenant_shard_id, timeline_id, Lsn(Lsn::MAX.0 - 1))
-        .await?;
-
-    println!("{} files found", files.len());
+    for _ in 0..100 {
+        let start = Instant::now();
+        let files = mgmt_api_client
+            .list_aux_files(tenant_shard_id, timeline_id, Lsn(Lsn::MAX.0 - 1))
+            .await?;
+        println!(
+            "{} files found in {}s",
+            files.len(),
+            start.elapsed().as_secs_f64()
+        );
+    }
 
     anyhow::Ok(())
 }

pageserver/src/tenant/storage_layer.rs

@@ -318,7 +318,7 @@ pub(crate) struct LayerFringe {
 #[derive(Debug)]
 struct LayerKeyspace {
     layer: ReadableLayer,
-    target_keyspace: KeySpace,
+    target_keyspace: Vec<KeySpace>,
 }
 
 impl LayerFringe {
@@ -336,14 +336,23 @@ impl LayerFringe {
         };
 
         let removed = self.layers.remove_entry(&read_desc.layer_id);
+
         match removed {
             Some((
                 _,
                 LayerKeyspace {
                     layer,
                     target_keyspace,
                 },
-            )) => Some((layer, target_keyspace, read_desc.lsn_range)),
+            )) => {
+                let mut keyspace = KeySpaceRandomAccum::new();
+                for ks in target_keyspace {
+                    for part in ks.ranges {
+                        keyspace.add_range(part);
+                    }
+                }
+                Some((layer, keyspace.consume_keyspace(), read_desc.lsn_range))
+            }
             None => unreachable!("fringe internals are always consistent"),
         }
     }
@@ -358,7 +367,7 @@ impl LayerFringe {
         let entry = self.layers.entry(layer_id.clone());
         match entry {
             Entry::Occupied(mut entry) => {
-                entry.get_mut().target_keyspace.merge(&keyspace);
+                entry.get_mut().target_keyspace.push(keyspace);
             }
             Entry::Vacant(entry) => {
                 self.planned_reads_by_lsn.push(ReadDesc {
@@ -367,7 +376,7 @@ impl LayerFringe {
                 });
                 entry.insert(LayerKeyspace {
                     layer,
-                    target_keyspace: keyspace,
+                    target_keyspace: vec![keyspace],
                 });
             }
         }