From 2a243f5eb341b09cbd7594dcd76bb1fe137d0692 Mon Sep 17 00:00:00 2001
From: Konstantin Knizhnik <knizhnik@garret.ru>
Date: Tue, 23 May 2023 21:44:15 +0300
Subject: [PATCH] Change rel_size_key prefi to separate data from metadata

---
 pageserver/src/pgdatadir_mapping.rs | 43 ++++++++++++++++-------------
 pageserver/src/tenant/metadata.rs   |  6 +++-
 pageserver/src/tenant/timeline.rs   |  2 ++
 3 files changed, 31 insertions(+), 20 deletions(-)

diff --git a/pageserver/src/pgdatadir_mapping.rs b/pageserver/src/pgdatadir_mapping.rs
index 67f37ee519..879c5ee851 100644
--- a/pageserver/src/pgdatadir_mapping.rs
+++ b/pageserver/src/pgdatadir_mapping.rs
@@ -167,7 +167,7 @@ impl Timeline {
             return Ok(0);
         }
 
-        let key = rel_size_to_key(tag);
+        let key = self.rel_size_to_key(tag);
         let mut buf = self.get(key, lsn, ctx).await?;
         let nblocks = buf.get_u32_le();
 
@@ -516,7 +516,7 @@ impl Timeline {
                 if cancel.is_cancelled() {
                     return Err(CalculateLogicalSizeError::Cancelled);
                 }
-                let relsize_key = rel_size_to_key(rel);
+                let relsize_key = self.rel_size_to_key(rel);
                 let mut buf = self
                     .get(relsize_key, lsn, ctx)
                     .await
@@ -561,7 +561,7 @@ impl Timeline {
                 .collect();
             rels.sort_unstable();
             for rel in rels {
-                let relsize_key = rel_size_to_key(rel);
+                let relsize_key = self.rel_size_to_key(rel);
                 let mut buf = self.get(relsize_key, lsn, ctx).await?;
                 let relsize = buf.get_u32_le();
 
@@ -648,6 +648,22 @@ impl Timeline {
         let mut rel_size_cache = self.rel_size_cache.write().unwrap();
         rel_size_cache.remove(tag);
     }
+
+    fn rel_size_to_key(&self, rel: RelTag) -> Key {
+        let field1 = if self.format_version == 4 {
+            0x00u8
+        } else {
+            0x04u8
+        };
+        Key {
+            field1,
+            field2: rel.spcnode,
+            field3: rel.dbnode,
+            field4: rel.relnode,
+            field5: rel.forknum,
+            field6: 0xffffffff,
+        }
+    }
 }
 
 /// DatadirModification represents an operation to ingest an atomic set of
@@ -890,7 +906,7 @@ impl<'a> DatadirModification<'a> {
         );
 
         // Put size
-        let size_key = rel_size_to_key(rel);
+        let size_key = self.tline.rel_size_to_key(rel);
         let buf = nblocks.to_le_bytes();
         self.put(size_key, Value::Image(Bytes::from(buf.to_vec())));
 
@@ -914,7 +930,7 @@ impl<'a> DatadirModification<'a> {
         anyhow::ensure!(rel.relnode != 0, "invalid relnode");
         let last_lsn = self.tline.get_last_record_lsn();
         if self.tline.get_rel_exists(rel, last_lsn, true, ctx).await? {
-            let size_key = rel_size_to_key(rel);
+            let size_key = self.tline.rel_size_to_key(rel);
             // Fetch the old size first
             let old_size = self.get(size_key, ctx).await?.get_u32_le();
 
@@ -945,7 +961,7 @@ impl<'a> DatadirModification<'a> {
         anyhow::ensure!(rel.relnode != 0, "invalid relnode");
 
         // Put size
-        let size_key = rel_size_to_key(rel);
+        let size_key = self.tline.rel_size_to_key(rel);
         let old_size = self.get(size_key, ctx).await?.get_u32_le();
 
         // only extend relation here. never decrease the size
@@ -977,7 +993,7 @@ impl<'a> DatadirModification<'a> {
         }
 
         // update logical size
-        let size_key = rel_size_to_key(rel);
+        let size_key = self.tline.rel_size_to_key(rel);
         let old_size = self.get(size_key, ctx).await?.get_u32_le();
         self.pending_nblocks -= old_size as i64;
 
@@ -1291,7 +1307,7 @@ static ZERO_PAGE: Bytes = Bytes::from_static(&[0u8; BLCKSZ as usize]);
 // 00 SPCNODE  DBNODE   RELNODE  FORK BLKNUM
 //
 // RelSize:
-// 00 SPCNODE  DBNODE   RELNODE  FORK FFFFFFFF
+// 04 SPCNODE  DBNODE   RELNODE  FORK FFFFFFFF
 //
 // SlruDir:
 // 01 kind     00000000 00000000 00   00000000
@@ -1375,17 +1391,6 @@ fn rel_block_to_key(rel: RelTag, blknum: BlockNumber) -> Key {
     }
 }
 
-fn rel_size_to_key(rel: RelTag) -> Key {
-    Key {
-        field1: 0x00,
-        field2: rel.spcnode,
-        field3: rel.dbnode,
-        field4: rel.relnode,
-        field5: rel.forknum,
-        field6: 0xffffffff,
-    }
-}
-
 fn rel_key_range(rel: RelTag) -> Range<Key> {
     Key {
         field1: 0x00,
diff --git a/pageserver/src/tenant/metadata.rs b/pageserver/src/tenant/metadata.rs
index 1ea61fa26b..66a6d423e2 100644
--- a/pageserver/src/tenant/metadata.rs
+++ b/pageserver/src/tenant/metadata.rs
@@ -23,7 +23,7 @@ use crate::config::PageServerConf;
 use crate::virtual_file::VirtualFile;
 
 /// Use special format number to enable backward compatibility.
-const METADATA_FORMAT_VERSION: u16 = 4;
+const METADATA_FORMAT_VERSION: u16 = 5;
 
 /// Previous supported format versions.
 const METADATA_OLD_FORMAT_VERSION: u16 = 3;
@@ -227,6 +227,10 @@ impl TimelineMetadata {
     pub fn pg_version(&self) -> u32 {
         self.body.pg_version
     }
+
+    pub fn format_version(&self) -> u16 {
+        self.hdr.format_version
+    }
 }
 
 /// Save timeline metadata to file
diff --git a/pageserver/src/tenant/timeline.rs b/pageserver/src/tenant/timeline.rs
index c47f4444f5..5879f7a115 100644
--- a/pageserver/src/tenant/timeline.rs
+++ b/pageserver/src/tenant/timeline.rs
@@ -120,6 +120,7 @@ pub struct Timeline {
     pub timeline_id: TimelineId,
 
     pub pg_version: u32,
+    pub format_version: u16,
 
     pub(super) layers: RwLock<LayerMap<dyn PersistentLayer>>,
 
@@ -1353,6 +1354,7 @@ impl Timeline {
                 timeline_id,
                 tenant_id,
                 pg_version,
+                format_version: metadata.format_version(),
                 layers: RwLock::new(LayerMap::default()),
 
                 walredo_mgr,