Update pageserver to store multixacts per region

libs/postgres_ffi/src/nonrelfile_utils.rs

@@ -85,12 +85,12 @@ pub fn mx_offset_to_member_offset(xid: MultiXactId) -> usize {
             + (xid as u16 % pg_constants::MULTIXACT_MEMBERS_PER_MEMBERGROUP) * 4) as usize
 }
 
-fn mx_offset_to_member_page(xid: u32) -> u32 {
-    xid / pg_constants::MULTIXACT_MEMBERS_PER_PAGE as u32
+fn mx_offset_to_member_page(xid: u32, region: u32) -> u32 {
+    ((xid / pg_constants::MULTIXACT_MEMBERS_PER_PAGE as u32) * pg_constants::MAX_REGIONS) + region
 }
 
-pub fn mx_offset_to_member_segment(xid: u32) -> i32 {
-    (mx_offset_to_member_page(xid) / pg_constants::SLRU_PAGES_PER_SEGMENT) as i32
+pub fn mx_offset_to_member_segment(xid: u32, region: u32) -> i32 {
+    (mx_offset_to_member_page(xid, region) / pg_constants::SLRU_PAGES_PER_SEGMENT) as i32
 }
 
 // See CSNLogPagePrecedes in csn_log.c
@@ -117,28 +117,28 @@ mod tests {
         // corresponding PostgreSQL C macros (MXOffsetTo*). These test values
         // were generated by calling the PostgreSQL macros with a little C
         // program.
-        assert_eq!(mx_offset_to_member_segment(0), 0);
-        assert_eq!(mx_offset_to_member_page(0), 0);
+        assert_eq!(mx_offset_to_member_segment(0, 0), 0);
+        assert_eq!(mx_offset_to_member_page(0, 0), 0);
         assert_eq!(mx_offset_to_flags_offset(0), 0);
         assert_eq!(mx_offset_to_flags_bitshift(0), 0);
         assert_eq!(mx_offset_to_member_offset(0), 4);
-        assert_eq!(mx_offset_to_member_segment(1), 0);
-        assert_eq!(mx_offset_to_member_page(1), 0);
+        assert_eq!(mx_offset_to_member_segment(1, 0), 0);
+        assert_eq!(mx_offset_to_member_page(1, 0), 0);
         assert_eq!(mx_offset_to_flags_offset(1), 0);
         assert_eq!(mx_offset_to_flags_bitshift(1), 8);
         assert_eq!(mx_offset_to_member_offset(1), 8);
-        assert_eq!(mx_offset_to_member_segment(123456789), 2358);
-        assert_eq!(mx_offset_to_member_page(123456789), 75462);
+        assert_eq!(mx_offset_to_member_segment(123456789, 0), 150924);
+        assert_eq!(mx_offset_to_member_page(123456789, 0), 4829568);
         assert_eq!(mx_offset_to_flags_offset(123456789), 4780);
         assert_eq!(mx_offset_to_flags_bitshift(123456789), 8);
         assert_eq!(mx_offset_to_member_offset(123456789), 4788);
-        assert_eq!(mx_offset_to_member_segment(u32::MAX - 1), 82040);
-        assert_eq!(mx_offset_to_member_page(u32::MAX - 1), 2625285);
+        assert_eq!(mx_offset_to_member_segment(u32::MAX - 1, 0), 5250570);
+        assert_eq!(mx_offset_to_member_page(u32::MAX - 1, 0), 168018240);
         assert_eq!(mx_offset_to_flags_offset(u32::MAX - 1), 5160);
         assert_eq!(mx_offset_to_flags_bitshift(u32::MAX - 1), 16);
         assert_eq!(mx_offset_to_member_offset(u32::MAX - 1), 5172);
-        assert_eq!(mx_offset_to_member_segment(u32::MAX), 82040);
-        assert_eq!(mx_offset_to_member_page(u32::MAX), 2625285);
+        assert_eq!(mx_offset_to_member_segment(u32::MAX, 0), 5250570);
+        assert_eq!(mx_offset_to_member_page(u32::MAX, 0), 168018240);
         assert_eq!(mx_offset_to_flags_offset(u32::MAX), 5160);
         assert_eq!(mx_offset_to_flags_bitshift(u32::MAX), 24);
         assert_eq!(mx_offset_to_member_offset(u32::MAX), 5176);

pageserver/src/walingest.rs

@@ -944,7 +944,10 @@ impl<'a> WalIngest<'a> {
         xlrec: &XlMultiXactCreate,
     ) -> Result<()> {
         // Create WAL record for updating the multixact-offsets page
-        let pageno = xlrec.mid / pg_constants::MULTIXACT_OFFSETS_PER_PAGE as u32;
+        let region: u32 = self.timeline.region_id.0 as u32;
+        let pageno = ((xlrec.mid / pg_constants::MULTIXACT_OFFSETS_PER_PAGE as u32)
+            * pg_constants::MAX_REGIONS)
+            + region;
         let segno = pageno / pg_constants::SLRU_PAGES_PER_SEGMENT;
         let rpageno = pageno % pg_constants::SLRU_PAGES_PER_SEGMENT;
 
@@ -955,6 +958,7 @@ impl<'a> WalIngest<'a> {
             NeonWalRecord::MultixactOffsetCreate {
                 mid: xlrec.mid,
                 moff: xlrec.moff,
+                region,
             },
         )?;
 
@@ -989,6 +993,7 @@ impl<'a> WalIngest<'a> {
                 NeonWalRecord::MultixactMembersCreate {
                     moff: offset,
                     members: this_page_members,
+                    region,
                 },
             )?;
 
@@ -1028,9 +1033,11 @@ impl<'a> WalIngest<'a> {
         self.checkpoint_modified = true;
 
         // PerformMembersTruncation
-        let maxsegment: i32 = mx_offset_to_member_segment(pg_constants::MAX_MULTIXACT_OFFSET);
-        let startsegment: i32 = mx_offset_to_member_segment(xlrec.start_trunc_memb);
-        let endsegment: i32 = mx_offset_to_member_segment(xlrec.end_trunc_memb);
+        let region: u32 = self.timeline.region_id.0 as u32;
+        let maxsegment: i32 =
+            mx_offset_to_member_segment(pg_constants::MAX_MULTIXACT_OFFSET, region);
+        let startsegment: i32 = mx_offset_to_member_segment(xlrec.start_trunc_memb, region);
+        let endsegment: i32 = mx_offset_to_member_segment(xlrec.end_trunc_memb, region);
         let mut segment: i32 = startsegment;
 
         // Delete all the segments except the last one. The last segment can still

pageserver/src/walrecord.rs

@@ -37,11 +37,13 @@ pub enum NeonWalRecord {
     MultixactOffsetCreate {
         mid: MultiXactId,
         moff: MultiXactOffset,
+        region: u32,
     },
     /// Extend multixact members SLRU.
     MultixactMembersCreate {
         moff: MultiXactOffset,
         members: Vec<MultiXactMember>,
+        region: u32,
     },
     /// Mark transaction ID as committed with the given LSN on a CsnLog page
     CsnLogSetCommitted {

pageserver/src/walredo.rs

@@ -512,7 +512,7 @@ impl PostgresRedoManager {
                     transaction_id_set_status(xid, pg_constants::TRANSACTION_STATUS_ABORTED, page);
                 }
             }
-            NeonWalRecord::MultixactOffsetCreate { mid, moff } => {
+            NeonWalRecord::MultixactOffsetCreate { mid, moff, region } => {
                 let (slru_kind, segno, blknum) =
                     key_to_slru_block(key).or(Err(WalRedoError::InvalidRecord))?;
                 assert_eq!(
@@ -523,7 +523,9 @@ impl PostgresRedoManager {
                 );
                 // Compute the block and offset to modify.
                 // See RecordNewMultiXact in PostgreSQL sources.
-                let pageno = mid / pg_constants::MULTIXACT_OFFSETS_PER_PAGE as u32;
+                let pageno = ((mid / pg_constants::MULTIXACT_OFFSETS_PER_PAGE as u32)
+                    * pg_constants::MAX_REGIONS)
+                    + *region;
                 let entryno = mid % pg_constants::MULTIXACT_OFFSETS_PER_PAGE as u32;
                 let offset = (entryno * 4) as usize;
 
@@ -545,7 +547,11 @@ impl PostgresRedoManager {
 
                 LittleEndian::write_u32(&mut page[offset..offset + 4], *moff);
             }
-            NeonWalRecord::MultixactMembersCreate { moff, members } => {
+            NeonWalRecord::MultixactMembersCreate {
+                moff,
+                members,
+                region,
+            } => {
                 let (slru_kind, segno, blknum) =
                     key_to_slru_block(key).or(Err(WalRedoError::InvalidRecord))?;
                 assert_eq!(
@@ -559,7 +565,9 @@ impl PostgresRedoManager {
 
                     // Compute the block and offset to modify.
                     // See RecordNewMultiXact in PostgreSQL sources.
-                    let pageno = offset / pg_constants::MULTIXACT_MEMBERS_PER_PAGE as u32;
+                    let pageno = ((offset / pg_constants::MULTIXACT_MEMBERS_PER_PAGE as u32)
+                        * pg_constants::MAX_REGIONS)
+                        + *region;
                     let memberoff = mx_offset_to_member_offset(offset);
                     let flagsoff = mx_offset_to_flags_offset(offset);
                     let bshift = mx_offset_to_flags_bitshift(offset);