pageserver: rate limited warning on too many layers visited

* Give it a better name
* Track all layers
* Larger buckets

.github/workflows/build_and_test.yml

@@ -477,7 +477,6 @@ jobs:
           BUILD_TAG: ${{ needs.tag.outputs.build-tag }}
           PAGESERVER_VIRTUAL_FILE_IO_ENGINE: tokio-epoll-uring
           PAGESERVER_GET_VECTORED_IMPL: vectored
-          PAGESERVER_GET_IMPL: vectored
 
       # Temporary disable this step until we figure out why it's so flaky
       # Ref https://github.com/neondatabase/neon/issues/4540

Cargo.lock

@@ -3658,7 +3658,6 @@ dependencies = [
  "tokio-util",
  "toml_edit",
  "tracing",
- "twox-hash",
  "url",
  "utils",
  "walkdir",

control_plane/src/local_env.rs

@@ -129,7 +129,6 @@ pub struct PageServerConf {
 
     pub(crate) virtual_file_io_engine: Option<String>,
     pub(crate) get_vectored_impl: Option<String>,
-    pub(crate) get_impl: Option<String>,
 }
 
 impl Default for PageServerConf {
@@ -142,7 +141,6 @@ impl Default for PageServerConf {
             http_auth_type: AuthType::Trust,
             virtual_file_io_engine: None,
             get_vectored_impl: None,
-            get_impl: None,
         }
     }
 }

control_plane/src/pageserver.rs

@@ -92,7 +92,6 @@ impl PageServerNode {
             http_auth_type,
             virtual_file_io_engine,
             get_vectored_impl,
-            get_impl,
         } = &self.conf;
 
         let id = format!("id={}", id);
@@ -112,11 +111,6 @@ impl PageServerNode {
         } else {
             String::new()
         };
-        let get_impl = if let Some(get_impl) = get_impl {
-            format!("get_impl='{get_impl}'")
-        } else {
-            String::new()
-        };
 
         let broker_endpoint_param = format!("broker_endpoint='{}'", self.env.broker.client_url());
 
@@ -130,7 +124,6 @@ impl PageServerNode {
             broker_endpoint_param,
             virtual_file_io_engine,
             get_vectored_impl,
-            get_impl,
         ];
 
         if let Some(control_plane_api) = &self.env.control_plane_api {
@@ -441,11 +434,6 @@ impl PageServerNode {
                 .map(serde_json::from_str)
                 .transpose()
                 .context("parse `timeline_get_throttle` from json")?,
-            switch_to_aux_file_v2: settings
-                .remove("switch_to_aux_file_v2")
-                .map(|x| x.parse::<bool>())
-                .transpose()
-                .context("Failed to parse 'switch_to_aux_file_v2' as bool")?,
         };
         if !settings.is_empty() {
             bail!("Unrecognized tenant settings: {settings:?}")
@@ -564,11 +552,6 @@ impl PageServerNode {
                     .map(serde_json::from_str)
                     .transpose()
                     .context("parse `timeline_get_throttle` from json")?,
-                switch_to_aux_file_v2: settings
-                    .remove("switch_to_aux_file_v2")
-                    .map(|x| x.parse::<bool>())
-                    .transpose()
-                    .context("Failed to parse 'switch_to_aux_file_v2' as bool")?,
             }
         };
 

libs/metrics/src/lib.rs

@@ -256,16 +256,7 @@ fn update_rusage_metrics() {
     DISK_IO_BYTES
         .with_label_values(&["write"])
         .set(rusage_stats.ru_oublock * BYTES_IN_BLOCK);
-
-    // On macOS, the unit of maxrss is bytes; on Linux, it's kilobytes. https://stackoverflow.com/a/59915669
-    #[cfg(target_os = "macos")]
-    {
-        MAXRSS_KB.set(rusage_stats.ru_maxrss / 1024);
-    }
-    #[cfg(not(target_os = "macos"))]
-    {
-        MAXRSS_KB.set(rusage_stats.ru_maxrss);
-    }
+    MAXRSS_KB.set(rusage_stats.ru_maxrss);
 }
 
 fn get_rusage_stats() -> libc::rusage {

libs/pageserver_api/src/key.rs

@@ -1,6 +1,5 @@
 use anyhow::{bail, Result};
 use byteorder::{ByteOrder, BE};
-use bytes::BufMut;
 use postgres_ffi::relfile_utils::{FSM_FORKNUM, VISIBILITYMAP_FORKNUM};
 use postgres_ffi::{Oid, TransactionId};
 use serde::{Deserialize, Serialize};
@@ -22,107 +21,15 @@ pub struct Key {
     pub field6: u32,
 }
 
-/// The storage key size.
 pub const KEY_SIZE: usize = 18;
 
-/// The metadata key size. 2B fewer than the storage key size because field2 is not fully utilized.
-/// See [`Key::to_i128`] for more information on the encoding.
-pub const METADATA_KEY_SIZE: usize = 16;
-
-/// The key prefix start range for the metadata keys. All keys with the first byte >= 0x40 is a metadata key.
-pub const METADATA_KEY_BEGIN_PREFIX: u8 = 0x60;
-pub const METADATA_KEY_END_PREFIX: u8 = 0x7F;
-
-/// The (reserved) key prefix of relation sizes.
-pub const RELATION_SIZE_PREFIX: u8 = 0x61;
-
-/// The key prefix of AUX file keys.
-pub const AUX_KEY_PREFIX: u8 = 0x62;
-
-/// Check if the key falls in the range of metadata keys.
-pub const fn is_metadata_key_slice(key: &[u8]) -> bool {
-    key[0] >= METADATA_KEY_BEGIN_PREFIX && key[0] < METADATA_KEY_END_PREFIX
-}
-
 impl Key {
-    /// Check if the key falls in the range of metadata keys.
-    pub const fn is_metadata_key(&self) -> bool {
-        self.field1 >= METADATA_KEY_BEGIN_PREFIX && self.field1 < METADATA_KEY_END_PREFIX
-    }
-
-    /// Encode a metadata key to a storage key.
-    pub fn from_metadata_key_fixed_size(key: &[u8; METADATA_KEY_SIZE]) -> Self {
-        assert!(is_metadata_key_slice(key), "key not in metadata key range");
-        Key {
-            field1: key[0],
-            field2: u16::from_be_bytes(key[1..3].try_into().unwrap()) as u32,
-            field3: u32::from_be_bytes(key[3..7].try_into().unwrap()),
-            field4: u32::from_be_bytes(key[7..11].try_into().unwrap()),
-            field5: key[11],
-            field6: u32::from_be_bytes(key[12..16].try_into().unwrap()),
-        }
-    }
-
-    /// Encode a metadata key to a storage key.
-    pub fn from_metadata_key(key: &[u8]) -> Self {
-        Self::from_metadata_key_fixed_size(key.try_into().expect("expect 16 byte metadata key"))
-    }
-
-    /// Extract a metadata key to a writer. The result should always be 16 bytes.
-    pub fn extract_metadata_key_to_writer(&self, mut writer: impl BufMut) {
-        writer.put_u8(self.field1);
-        assert!(self.field2 <= 0xFFFF);
-        writer.put_u16(self.field2 as u16);
-        writer.put_u32(self.field3);
-        writer.put_u32(self.field4);
-        writer.put_u8(self.field5);
-        writer.put_u32(self.field6);
-    }
-
-    /// Get the range of metadata keys.
-    pub fn metadata_key_range() -> Range<Self> {
-        Key {
-            field1: METADATA_KEY_BEGIN_PREFIX,
-            field2: 0,
-            field3: 0,
-            field4: 0,
-            field5: 0,
-            field6: 0,
-        }..Key {
-            field1: METADATA_KEY_END_PREFIX,
-            field2: 0,
-            field3: 0,
-            field4: 0,
-            field5: 0,
-            field6: 0,
-        }
-    }
-
-    /// Get the range of aux keys.
-    pub fn metadata_aux_key_range() -> Range<Self> {
-        Key {
-            field1: AUX_KEY_PREFIX,
-            field2: 0,
-            field3: 0,
-            field4: 0,
-            field5: 0,
-            field6: 0,
-        }..Key {
-            field1: AUX_KEY_PREFIX + 1,
-            field2: 0,
-            field3: 0,
-            field4: 0,
-            field5: 0,
-            field6: 0,
-        }
-    }
-
     /// 'field2' is used to store tablespaceid for relations and small enum numbers for other relish.
     /// As long as Neon does not support tablespace (because of lack of access to local file system),
     /// we can assume that only some predefined namespace OIDs are used which can fit in u16
     pub fn to_i128(&self) -> i128 {
         assert!(self.field2 < 0xFFFF || self.field2 == 0xFFFFFFFF || self.field2 == 0x22222222);
-        (((self.field1 & 0x7F) as i128) << 120)
+        (((self.field1 & 0xf) as i128) << 120)
             | (((self.field2 & 0xFFFF) as i128) << 104)
             | ((self.field3 as i128) << 72)
             | ((self.field4 as i128) << 40)
@@ -132,7 +39,7 @@ impl Key {
 
     pub const fn from_i128(x: i128) -> Self {
         Key {
-            field1: ((x >> 120) & 0x7F) as u8,
+            field1: ((x >> 120) & 0xf) as u8,
             field2: ((x >> 104) & 0xFFFF) as u32,
             field3: (x >> 72) as u32,
             field4: (x >> 40) as u32,
@@ -141,11 +48,11 @@ impl Key {
         }
     }
 
-    pub const fn next(&self) -> Key {
+    pub fn next(&self) -> Key {
         self.add(1)
     }
 
-    pub const fn add(&self, x: u32) -> Key {
+    pub fn add(&self, x: u32) -> Key {
         let mut key = *self;
 
         let r = key.field6.overflowing_add(x);
@@ -174,8 +81,6 @@ impl Key {
         key
     }
 
-    /// Convert a 18B slice to a key. This function should not be used for metadata keys because field2 is handled differently.
-    /// Use [`Key::from_metadata_key`] instead.
     pub fn from_slice(b: &[u8]) -> Self {
         Key {
             field1: b[0],
@@ -187,8 +92,6 @@ impl Key {
         }
     }
 
-    /// Convert a key to a 18B slice. This function should not be used for metadata keys because field2 is handled differently.
-    /// Use [`Key::extract_metadata_key_to_writer`] instead.
     pub fn write_to_byte_slice(&self, buf: &mut [u8]) {
         buf[0] = self.field1;
         BE::write_u32(&mut buf[1..5], self.field2);
@@ -572,14 +475,12 @@ pub const AUX_FILES_KEY: Key = Key {
 // Reverse mappings for a few Keys.
 // These are needed by WAL redo manager.
 
-pub const NON_INHERITED_RANGE: Range<Key> = AUX_FILES_KEY..AUX_FILES_KEY.next();
-
 // AUX_FILES currently stores only data for logical replication (slots etc), and
 // we don't preserve these on a branch because safekeepers can't follow timeline
 // switch (and generally it likely should be optional), so ignore these.
 #[inline(always)]
 pub fn is_inherited_key(key: Key) -> bool {
-    !NON_INHERITED_RANGE.contains(&key)
+    key != AUX_FILES_KEY
 }
 
 #[inline(always)]
@@ -655,14 +556,11 @@ impl std::str::FromStr for Key {
 mod tests {
     use std::str::FromStr;
 
-    use crate::key::is_metadata_key_slice;
     use crate::key::Key;
 
     use rand::Rng;
     use rand::SeedableRng;
 
-    use super::AUX_KEY_PREFIX;
-
     #[test]
     fn display_fromstr_bijection() {
         let mut rng = rand::rngs::StdRng::seed_from_u64(42);
@@ -678,16 +576,4 @@ mod tests {
 
         assert_eq!(key, Key::from_str(&format!("{key}")).unwrap());
     }
-
-    #[test]
-    fn test_metadata_keys() {
-        let mut metadata_key = vec![AUX_KEY_PREFIX];
-        metadata_key.extend_from_slice(&[0xFF; 15]);
-        let encoded_key = Key::from_metadata_key(&metadata_key);
-        let mut output_key = Vec::new();
-        encoded_key.extract_metadata_key_to_writer(&mut output_key);
-        assert_eq!(metadata_key, output_key);
-        assert!(encoded_key.is_metadata_key());
-        assert!(is_metadata_key_slice(&metadata_key));
-    }
 }

libs/pageserver_api/src/keyspace.rs

@@ -15,13 +15,7 @@ pub struct KeySpace {
 }
 
 impl KeySpace {
-    /// Create a key space with a single range.
-    pub fn single(key_range: Range<Key>) -> Self {
-        Self {
-            ranges: vec![key_range],
-        }
-    }
-
+    ///
     /// Partition a key space into roughly chunks of roughly 'target_size' bytes
     /// in each partition.
     ///
@@ -70,10 +64,6 @@ impl KeySpace {
         KeyPartitioning { parts }
     }
 
-    pub fn is_empty(&self) -> bool {
-        self.total_size() == 0
-    }
-
     /// Merge another keyspace into the current one.
     /// Note: the keyspaces must not ovelap (enforced via assertions)
     pub fn merge(&mut self, other: &KeySpace) {
@@ -104,13 +94,12 @@ impl KeySpace {
 
     /// Remove all keys in `other` from `self`.
     /// This can involve splitting or removing of existing ranges.
-    /// Returns the removed keyspace
-    pub fn remove_overlapping_with(&mut self, other: &KeySpace) -> KeySpace {
+    pub fn remove_overlapping_with(&mut self, other: &KeySpace) {
         let (self_start, self_end) = match (self.start(), self.end()) {
             (Some(start), Some(end)) => (start, end),
             _ => {
                 // self is empty
-                return KeySpace::default();
+                return;
             }
         };
 
@@ -123,37 +112,30 @@ impl KeySpace {
             .skip_while(|range| self_start >= range.end)
             .take_while(|range| self_end > range.start);
 
-        let mut removed_accum = KeySpaceRandomAccum::new();
         for range in other_ranges {
             while let Some(overlap_at) = self.overlaps_at(range) {
                 let overlapped = self.ranges[overlap_at].clone();
 
                 if overlapped.start < range.start && overlapped.end <= range.end {
                     // Higher part of the range is completely overlapped.
-                    removed_accum.add_range(range.start..self.ranges[overlap_at].end);
                     self.ranges[overlap_at].end = range.start;
                 }
                 if overlapped.start >= range.start && overlapped.end > range.end {
                     // Lower part of the range is completely overlapped.
-                    removed_accum.add_range(self.ranges[overlap_at].start..range.end);
                     self.ranges[overlap_at].start = range.end;
                 }
                 if overlapped.start < range.start && overlapped.end > range.end {
                     // Middle part of the range is overlapped.
-                    removed_accum.add_range(range.clone());
                     self.ranges[overlap_at].end = range.start;
                     self.ranges
                         .insert(overlap_at + 1, range.end..overlapped.end);
                 }
                 if overlapped.start >= range.start && overlapped.end <= range.end {
                     // Whole range is overlapped
-                    removed_accum.add_range(self.ranges[overlap_at].clone());
                     self.ranges.remove(overlap_at);
                 }
             }
         }
-
-        removed_accum.to_keyspace()
     }
 
     pub fn start(&self) -> Option<Key> {
@@ -188,11 +170,6 @@ impl KeySpace {
     pub fn overlaps(&self, range: &Range<Key>) -> bool {
         self.overlaps_at(range).is_some()
     }
-
-    /// Check if the keyspace contains a key
-    pub fn contains(&self, key: &Key) -> bool {
-        self.overlaps(&(*key..key.next()))
-    }
 }
 
 ///
@@ -576,16 +553,7 @@ mod tests {
                 Key::from_i128(11)..Key::from_i128(13),
             ],
         };
-        let removed = key_space1.remove_overlapping_with(&key_space2);
-        let removed_expected = KeySpace {
-            ranges: vec![
-                Key::from_i128(2)..Key::from_i128(3),
-                Key::from_i128(6)..Key::from_i128(7),
-                Key::from_i128(11)..Key::from_i128(12),
-            ],
-        };
-        assert_eq!(removed, removed_expected);
-
+        key_space1.remove_overlapping_with(&key_space2);
         assert_eq!(
             key_space1.ranges,
             vec![
@@ -615,17 +583,7 @@ mod tests {
                 Key::from_i128(14)..Key::from_i128(17),
             ],
         };
-
-        let removed = key_space1.remove_overlapping_with(&key_space2);
-        let removed_expected = KeySpace {
-            ranges: vec![
-                Key::from_i128(3)..Key::from_i128(5),
-                Key::from_i128(8)..Key::from_i128(10),
-                Key::from_i128(14)..Key::from_i128(15),
-            ],
-        };
-        assert_eq!(removed, removed_expected);
-
+        key_space1.remove_overlapping_with(&key_space2);
         assert_eq!(
             key_space1.ranges,
             vec![
@@ -652,11 +610,7 @@ mod tests {
                 Key::from_i128(15)..Key::from_i128(17),
             ],
         };
-
-        let removed = key_space1.remove_overlapping_with(&key_space2);
-        let removed_expected = KeySpace::default();
-        assert_eq!(removed, removed_expected);
-
+        key_space1.remove_overlapping_with(&key_space2);
         assert_eq!(
             key_space1.ranges,
             vec![
@@ -683,17 +637,7 @@ mod tests {
         let key_space2 = KeySpace {
             ranges: vec![Key::from_i128(9)..Key::from_i128(19)],
         };
-
-        let removed = key_space1.remove_overlapping_with(&key_space2);
-        let removed_expected = KeySpace {
-            ranges: vec![
-                Key::from_i128(9)..Key::from_i128(10),
-                Key::from_i128(12)..Key::from_i128(15),
-                Key::from_i128(17)..Key::from_i128(19),
-            ],
-        };
-        assert_eq!(removed, removed_expected);
-
+        key_space1.remove_overlapping_with(&key_space2);
         assert_eq!(
             key_space1.ranges,
             vec![

libs/pageserver_api/src/models.rs

@@ -303,7 +303,6 @@ pub struct TenantConfig {
     pub lazy_slru_download: Option<bool>,
     pub timeline_get_throttle: Option<ThrottleConfig>,
     pub image_layer_creation_check_threshold: Option<u8>,
-    pub switch_to_aux_file_v2: Option<bool>,
 }
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
@@ -430,7 +429,6 @@ pub struct StatusResponse {
 #[derive(Serialize, Deserialize, Debug)]
 #[serde(deny_unknown_fields)]
 pub struct TenantLocationConfigRequest {
-    #[serde(skip_serializing_if = "Option::is_none")]
     pub tenant_id: Option<TenantShardId>,
     #[serde(flatten)]
     pub config: LocationConfig, // as we have a flattened field, we should reject all unknown fields in it
@@ -849,72 +847,39 @@ impl TryFrom<u8> for PagestreamBeMessageTag {
     }
 }
 
-// In the V2 protocol version, a GetPage request contains two LSN values:
-//
-// request_lsn: Get the page version at this point in time.  Lsn::Max is a special value that means
-// "get the latest version present". It's used by the primary server, which knows that no one else
-// is writing WAL. 'not_modified_since' must be set to a proper value even if request_lsn is
-// Lsn::Max. Standby servers use the current replay LSN as the request LSN.
-//
-// not_modified_since: Hint to the pageserver that the client knows that the page has not been
-// modified between 'not_modified_since' and the request LSN. It's always correct to set
-// 'not_modified_since equal' to 'request_lsn' (unless Lsn::Max is used as the 'request_lsn'), but
-// passing an earlier LSN can speed up the request, by allowing the pageserver to process the
-// request without waiting for 'request_lsn' to arrive.
-//
-// The legacy V1 interface contained only one LSN, and a boolean 'latest' flag. The V1 interface was
-// sufficient for the primary; the 'lsn' was equivalent to the 'not_modified_since' value, and
-// 'latest' was set to true. The V2 interface was added because there was no correct way for a
-// standby to request a page at a particular non-latest LSN, and also include the
-// 'not_modified_since' hint. That led to an awkward choice of either using an old LSN in the
-// request, if the standby knows that the page hasn't been modified since, and risk getting an error
-// if that LSN has fallen behind the GC horizon, or requesting the current replay LSN, which could
-// require the pageserver unnecessarily to wait for the WAL to arrive up to that point. The new V2
-// interface allows sending both LSNs, and let the pageserver do the right thing. There is no
-// difference in the responses between V1 and V2.
-//
-// The Request structs below reflect the V2 interface. If V1 is used, the parse function
-// maps the old format requests to the new format.
-//
-#[derive(Clone, Copy)]
-pub enum PagestreamProtocolVersion {
-    V1,
-    V2,
-}
-
 #[derive(Debug, PartialEq, Eq)]
 pub struct PagestreamExistsRequest {
-    pub request_lsn: Lsn,
-    pub not_modified_since: Lsn,
+    pub latest: bool,
+    pub lsn: Lsn,
     pub rel: RelTag,
 }
 
 #[derive(Debug, PartialEq, Eq)]
 pub struct PagestreamNblocksRequest {
-    pub request_lsn: Lsn,
-    pub not_modified_since: Lsn,
+    pub latest: bool,
+    pub lsn: Lsn,
     pub rel: RelTag,
 }
 
 #[derive(Debug, PartialEq, Eq)]
 pub struct PagestreamGetPageRequest {
-    pub request_lsn: Lsn,
-    pub not_modified_since: Lsn,
+    pub latest: bool,
+    pub lsn: Lsn,
     pub rel: RelTag,
     pub blkno: u32,
 }
 
 #[derive(Debug, PartialEq, Eq)]
 pub struct PagestreamDbSizeRequest {
-    pub request_lsn: Lsn,
-    pub not_modified_since: Lsn,
+    pub latest: bool,
+    pub lsn: Lsn,
     pub dbnode: u32,
 }
 
 #[derive(Debug, PartialEq, Eq)]
 pub struct PagestreamGetSlruSegmentRequest {
-    pub request_lsn: Lsn,
-    pub not_modified_since: Lsn,
+    pub latest: bool,
+    pub lsn: Lsn,
     pub kind: u8,
     pub segno: u32,
 }
@@ -961,16 +926,14 @@ pub struct TenantHistorySize {
 }
 
 impl PagestreamFeMessage {
-    /// Serialize a compute -> pageserver message. This is currently only used in testing
-    /// tools. Always uses protocol version 2.
     pub fn serialize(&self) -> Bytes {
         let mut bytes = BytesMut::new();
 
         match self {
             Self::Exists(req) => {
                 bytes.put_u8(0);
-                bytes.put_u64(req.request_lsn.0);
-                bytes.put_u64(req.not_modified_since.0);
+                bytes.put_u8(u8::from(req.latest));
+                bytes.put_u64(req.lsn.0);
                 bytes.put_u32(req.rel.spcnode);
                 bytes.put_u32(req.rel.dbnode);
                 bytes.put_u32(req.rel.relnode);
@@ -979,8 +942,8 @@ impl PagestreamFeMessage {
 
             Self::Nblocks(req) => {
                 bytes.put_u8(1);
-                bytes.put_u64(req.request_lsn.0);
-                bytes.put_u64(req.not_modified_since.0);
+                bytes.put_u8(u8::from(req.latest));
+                bytes.put_u64(req.lsn.0);
                 bytes.put_u32(req.rel.spcnode);
                 bytes.put_u32(req.rel.dbnode);
                 bytes.put_u32(req.rel.relnode);
@@ -989,8 +952,8 @@ impl PagestreamFeMessage {
 
             Self::GetPage(req) => {
                 bytes.put_u8(2);
-                bytes.put_u64(req.request_lsn.0);
-                bytes.put_u64(req.not_modified_since.0);
+                bytes.put_u8(u8::from(req.latest));
+                bytes.put_u64(req.lsn.0);
                 bytes.put_u32(req.rel.spcnode);
                 bytes.put_u32(req.rel.dbnode);
                 bytes.put_u32(req.rel.relnode);
@@ -1000,15 +963,15 @@ impl PagestreamFeMessage {
 
             Self::DbSize(req) => {
                 bytes.put_u8(3);
-                bytes.put_u64(req.request_lsn.0);
-                bytes.put_u64(req.not_modified_since.0);
+                bytes.put_u8(u8::from(req.latest));
+                bytes.put_u64(req.lsn.0);
                 bytes.put_u32(req.dbnode);
             }
 
             Self::GetSlruSegment(req) => {
                 bytes.put_u8(4);
-                bytes.put_u64(req.request_lsn.0);
-                bytes.put_u64(req.not_modified_since.0);
+                bytes.put_u8(u8::from(req.latest));
+                bytes.put_u64(req.lsn.0);
                 bytes.put_u8(req.kind);
                 bytes.put_u32(req.segno);
             }
@@ -1017,40 +980,18 @@ impl PagestreamFeMessage {
         bytes.into()
     }
 
-    pub fn parse<R: std::io::Read>(
-        body: &mut R,
-        protocol_version: PagestreamProtocolVersion,
-    ) -> anyhow::Result<PagestreamFeMessage> {
+    pub fn parse<R: std::io::Read>(body: &mut R) -> anyhow::Result<PagestreamFeMessage> {
+        // TODO these gets can fail
+
         // these correspond to the NeonMessageTag enum in pagestore_client.h
         //
         // TODO: consider using protobuf or serde bincode for less error prone
         // serialization.
         let msg_tag = body.read_u8()?;
-
-        let (request_lsn, not_modified_since) = match protocol_version {
-            PagestreamProtocolVersion::V2 => (
-                Lsn::from(body.read_u64::<BigEndian>()?),
-                Lsn::from(body.read_u64::<BigEndian>()?),
-            ),
-            PagestreamProtocolVersion::V1 => {
-                // In the old protocol, each message starts with a boolean 'latest' flag,
-                // followed by 'lsn'. Convert that to the two LSNs, 'request_lsn' and
-                // 'not_modified_since', used in the new protocol version.
-                let latest = body.read_u8()? != 0;
-                let request_lsn = Lsn::from(body.read_u64::<BigEndian>()?);
-                if latest {
-                    (Lsn::MAX, request_lsn) // get latest version
-                } else {
-                    (request_lsn, request_lsn) // get version at specified LSN
-                }
-            }
-        };
-
-        // The rest of the messages are the same between V1 and V2
         match msg_tag {
             0 => Ok(PagestreamFeMessage::Exists(PagestreamExistsRequest {
-                request_lsn,
-                not_modified_since,
+                latest: body.read_u8()? != 0,
+                lsn: Lsn::from(body.read_u64::<BigEndian>()?),
                 rel: RelTag {
                     spcnode: body.read_u32::<BigEndian>()?,
                     dbnode: body.read_u32::<BigEndian>()?,
@@ -1059,8 +1000,8 @@ impl PagestreamFeMessage {
                 },
             })),
             1 => Ok(PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
-                request_lsn,
-                not_modified_since,
+                latest: body.read_u8()? != 0,
+                lsn: Lsn::from(body.read_u64::<BigEndian>()?),
                 rel: RelTag {
                     spcnode: body.read_u32::<BigEndian>()?,
                     dbnode: body.read_u32::<BigEndian>()?,
@@ -1069,8 +1010,8 @@ impl PagestreamFeMessage {
                 },
             })),
             2 => Ok(PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
-                request_lsn,
-                not_modified_since,
+                latest: body.read_u8()? != 0,
+                lsn: Lsn::from(body.read_u64::<BigEndian>()?),
                 rel: RelTag {
                     spcnode: body.read_u32::<BigEndian>()?,
                     dbnode: body.read_u32::<BigEndian>()?,
@@ -1080,14 +1021,14 @@ impl PagestreamFeMessage {
                 blkno: body.read_u32::<BigEndian>()?,
             })),
             3 => Ok(PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
-                request_lsn,
-                not_modified_since,
+                latest: body.read_u8()? != 0,
+                lsn: Lsn::from(body.read_u64::<BigEndian>()?),
                 dbnode: body.read_u32::<BigEndian>()?,
             })),
             4 => Ok(PagestreamFeMessage::GetSlruSegment(
                 PagestreamGetSlruSegmentRequest {
-                    request_lsn,
-                    not_modified_since,
+                    latest: body.read_u8()? != 0,
+                    lsn: Lsn::from(body.read_u64::<BigEndian>()?),
                     kind: body.read_u8()?,
                     segno: body.read_u32::<BigEndian>()?,
                 },
@@ -1215,8 +1156,8 @@ mod tests {
         // Test serialization/deserialization of PagestreamFeMessage
         let messages = vec![
             PagestreamFeMessage::Exists(PagestreamExistsRequest {
-                request_lsn: Lsn(4),
-                not_modified_since: Lsn(3),
+                latest: true,
+                lsn: Lsn(4),
                 rel: RelTag {
                     forknum: 1,
                     spcnode: 2,
@@ -1225,8 +1166,8 @@ mod tests {
                 },
             }),
             PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
-                request_lsn: Lsn(4),
-                not_modified_since: Lsn(4),
+                latest: false,
+                lsn: Lsn(4),
                 rel: RelTag {
                     forknum: 1,
                     spcnode: 2,
@@ -1235,8 +1176,8 @@ mod tests {
                 },
             }),
             PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
-                request_lsn: Lsn(4),
-                not_modified_since: Lsn(3),
+                latest: true,
+                lsn: Lsn(4),
                 rel: RelTag {
                     forknum: 1,
                     spcnode: 2,
@@ -1246,16 +1187,14 @@ mod tests {
                 blkno: 7,
             }),
             PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
-                request_lsn: Lsn(4),
-                not_modified_since: Lsn(3),
+                latest: true,
+                lsn: Lsn(4),
                 dbnode: 7,
             }),
         ];
         for msg in messages {
             let bytes = msg.serialize();
-            let reconstructed =
-                PagestreamFeMessage::parse(&mut bytes.reader(), PagestreamProtocolVersion::V2)
-                    .unwrap();
+            let reconstructed = PagestreamFeMessage::parse(&mut bytes.reader()).unwrap();
             assert!(msg == reconstructed);
         }
     }

libs/remote_storage/src/lib.rs

@@ -21,13 +21,11 @@ use std::{
     fmt::Debug,
     num::{NonZeroU32, NonZeroUsize},
     pin::Pin,
-    str::FromStr,
     sync::Arc,
     time::{Duration, SystemTime},
 };
 
 use anyhow::{bail, Context};
-use aws_sdk_s3::types::StorageClass;
 use camino::{Utf8Path, Utf8PathBuf};
 
 use bytes::Bytes;
@@ -136,11 +134,6 @@ impl RemotePath {
     pub fn strip_prefix(&self, p: &RemotePath) -> Result<&Utf8Path, std::path::StripPrefixError> {
         self.0.strip_prefix(&p.0)
     }
-
-    pub fn add_trailing_slash(&self) -> Self {
-        // Unwrap safety inputs are guararnteed to be valid UTF-8
-        Self(format!("{}/", self.0).try_into().unwrap())
-    }
 }
 
 /// We don't need callers to be able to pass arbitrary delimiters: just control
@@ -164,21 +157,47 @@ pub struct Listing {
 /// providing basic CRUD operations for storage files.
 #[allow(async_fn_in_trait)]
 pub trait RemoteStorage: Send + Sync + 'static {
-    /// List objects in remote storage, with semantics matching AWS S3's ListObjectsV2.
-    /// (see `<https://docs.aws.amazon.com/AmazonS3/latest/API/API_ListObjectsV2.html>`)
-    ///
-    /// Note that the prefix is relative to any `prefix_in_bucket` configured for the client, not
-    /// from the absolute root of the bucket.
-    ///
-    /// `mode` configures whether to use a delimiter.  Without a delimiter all keys
-    /// within the prefix are listed in the `keys` of the result.  With a delimiter, any "directories" at the top level of
-    /// the prefix are returned in the `prefixes` of the result, and keys in the top level of the prefix are
-    /// returned in `keys` ().
-    ///
-    /// `max_keys` controls the maximum number of keys that will be returned.  If this is None, this function
-    /// will iteratively call listobjects until it runs out of keys.  Note that this is not safe to use on
-    /// unlimted size buckets, as the full list of objects is allocated into a monolithic data structure.
+    /// Lists all top level subdirectories for a given prefix
+    /// Note: here we assume that if the prefix is passed it was obtained via remote_object_id
+    /// which already takes into account any kind of global prefix (prefix_in_bucket for S3 or storage_root for LocalFS)
+    /// so this method doesnt need to.
+    async fn list_prefixes(
+        &self,
+        prefix: Option<&RemotePath>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        let result = self
+            .list(prefix, ListingMode::WithDelimiter, None, cancel)
+            .await?
+            .prefixes;
+        Ok(result)
+    }
+    /// Lists all files in directory "recursively"
+    /// (not really recursively, because AWS has a flat namespace)
+    /// Note: This is subtely different than list_prefixes,
+    /// because it is for listing files instead of listing
+    /// names sharing common prefixes.
+    /// For example,
+    /// list_files("foo/bar") = ["foo/bar/cat123.txt",
+    /// "foo/bar/cat567.txt", "foo/bar/dog123.txt", "foo/bar/dog456.txt"]
+    /// whereas,
+    /// list_prefixes("foo/bar/") = ["cat", "dog"]
+    /// See `test_real_s3.rs` for more details.
     ///
+    /// max_keys limits max number of keys returned; None means unlimited.
+    async fn list_files(
+        &self,
+        prefix: Option<&RemotePath>,
+        max_keys: Option<NonZeroU32>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        let result = self
+            .list(prefix, ListingMode::NoDelimiter, max_keys, cancel)
+            .await?
+            .keys;
+        Ok(result)
+    }
+
     async fn list(
         &self,
         prefix: Option<&RemotePath>,
@@ -317,6 +336,41 @@ impl<Other: RemoteStorage> GenericRemoteStorage<Arc<Other>> {
         }
     }
 
+    // A function for listing all the files in a "directory"
+    // Example:
+    // list_files("foo/bar") = ["foo/bar/a.txt", "foo/bar/b.txt"]
+    //
+    // max_keys limits max number of keys returned; None means unlimited.
+    pub async fn list_files(
+        &self,
+        folder: Option<&RemotePath>,
+        max_keys: Option<NonZeroU32>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        match self {
+            Self::LocalFs(s) => s.list_files(folder, max_keys, cancel).await,
+            Self::AwsS3(s) => s.list_files(folder, max_keys, cancel).await,
+            Self::AzureBlob(s) => s.list_files(folder, max_keys, cancel).await,
+            Self::Unreliable(s) => s.list_files(folder, max_keys, cancel).await,
+        }
+    }
+
+    // lists common *prefixes*, if any of files
+    // Example:
+    // list_prefixes("foo123","foo567","bar123","bar432") = ["foo", "bar"]
+    pub async fn list_prefixes(
+        &self,
+        prefix: Option<&RemotePath>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        match self {
+            Self::LocalFs(s) => s.list_prefixes(prefix, cancel).await,
+            Self::AwsS3(s) => s.list_prefixes(prefix, cancel).await,
+            Self::AzureBlob(s) => s.list_prefixes(prefix, cancel).await,
+            Self::Unreliable(s) => s.list_prefixes(prefix, cancel).await,
+        }
+    }
+
     /// See [`RemoteStorage::upload`]
     pub async fn upload(
         &self,
@@ -565,7 +619,6 @@ pub struct S3Config {
     /// See [`DEFAULT_REMOTE_STORAGE_S3_CONCURRENCY_LIMIT`] for more details.
     pub concurrency_limit: NonZeroUsize,
     pub max_keys_per_list_response: Option<i32>,
-    pub upload_storage_class: Option<StorageClass>,
 }
 
 impl Debug for S3Config {
@@ -694,18 +747,6 @@ impl RemoteStorageConfig {
                     endpoint,
                     concurrency_limit,
                     max_keys_per_list_response,
-                    upload_storage_class: toml
-                        .get("upload_storage_class")
-                        .map(|prefix_in_bucket| -> anyhow::Result<_> {
-                            let s = parse_toml_string("upload_storage_class", prefix_in_bucket)?;
-                            let storage_class = StorageClass::from_str(&s).expect("infallible");
-                            #[allow(deprecated)]
-                            if matches!(storage_class, StorageClass::Unknown(_)) {
-                                bail!("Specified storage class unknown to SDK: '{s}'. Allowed values: {:?}", StorageClass::values());
-                            }
-                            Ok(storage_class)
-                        })
-                        .transpose()?,
                 })
             }
             (_, _, _, Some(_), None) => {

libs/remote_storage/src/local_fs.rs

@@ -5,9 +5,11 @@
 //! volume is mounted to the local FS.
 
 use std::{
-    collections::HashSet,
+    borrow::Cow,
+    future::Future,
     io::ErrorKind,
     num::NonZeroU32,
+    pin::Pin,
     time::{Duration, SystemTime, UNIX_EPOCH},
 };
 
@@ -20,11 +22,11 @@ use tokio::{
     io::{self, AsyncReadExt, AsyncSeekExt, AsyncWriteExt},
 };
 use tokio_util::{io::ReaderStream, sync::CancellationToken};
-use utils::crashsafe::path_with_suffix_extension;
+use tracing::*;
+use utils::{crashsafe::path_with_suffix_extension, fs_ext::is_directory_empty};
 
 use crate::{
     Download, DownloadError, Listing, ListingMode, RemotePath, TimeTravelError, TimeoutOrCancel,
-    REMOTE_STORAGE_PREFIX_SEPARATOR,
 };
 
 use super::{RemoteStorage, StorageMetadata};
@@ -91,47 +93,7 @@ impl LocalFs {
 
     #[cfg(test)]
     async fn list_all(&self) -> anyhow::Result<Vec<RemotePath>> {
-        use std::{future::Future, pin::Pin};
-        fn get_all_files<'a, P>(
-            directory_path: P,
-        ) -> Pin<Box<dyn Future<Output = anyhow::Result<Vec<Utf8PathBuf>>> + Send + Sync + 'a>>
-        where
-            P: AsRef<Utf8Path> + Send + Sync + 'a,
-        {
-            Box::pin(async move {
-                let directory_path = directory_path.as_ref();
-                if directory_path.exists() {
-                    if directory_path.is_dir() {
-                        let mut paths = Vec::new();
-                        let mut dir_contents = fs::read_dir(directory_path).await?;
-                        while let Some(dir_entry) = dir_contents.next_entry().await? {
-                            let file_type = dir_entry.file_type().await?;
-                            let entry_path =
-                                Utf8PathBuf::from_path_buf(dir_entry.path()).map_err(|pb| {
-                                    anyhow::Error::msg(format!(
-                                        "non-Unicode path: {}",
-                                        pb.to_string_lossy()
-                                    ))
-                                })?;
-                            if file_type.is_symlink() {
-                                tracing::debug!("{entry_path:?} is a symlink, skipping")
-                            } else if file_type.is_dir() {
-                                paths.extend(get_all_files(&entry_path).await?.into_iter())
-                            } else {
-                                paths.push(entry_path);
-                            }
-                        }
-                        Ok(paths)
-                    } else {
-                        bail!("Path {directory_path:?} is not a directory")
-                    }
-                } else {
-                    Ok(Vec::new())
-                }
-            })
-        }
-
-        Ok(get_all_files(&self.storage_root)
+        Ok(get_all_files(&self.storage_root, true)
             .await?
             .into_iter()
             .map(|path| {
@@ -158,14 +120,6 @@ impl LocalFs {
         // S3 object list prefixes can be arbitrary strings, but when reading
         // the local filesystem we need a directory to start calling read_dir on.
         let mut initial_dir = full_path.clone();
-
-        // If there's no trailing slash, we have to start looking from one above: even if
-        // `initial_dir` is a directory, we should still list any prefixes in the parent
-        // that start with the same string.
-        if !full_path.to_string().ends_with('/') {
-            initial_dir.pop();
-        }
-
         loop {
             // Did we make it to the root?
             if initial_dir.parent().is_none() {
@@ -341,66 +295,61 @@ impl RemoteStorage for LocalFs {
         let op = async {
             let mut result = Listing::default();
 
-            // Filter out directories: in S3 directories don't exist, only the keys within them do.
-            let keys = self
-                .list_recursive(prefix)
+            if let ListingMode::NoDelimiter = mode {
+                let keys = self
+                    .list_recursive(prefix)
+                    .await
+                    .map_err(DownloadError::Other)?;
+
+                result.keys = keys
+                    .into_iter()
+                    .filter(|k| {
+                        let path = k.with_base(&self.storage_root);
+                        !path.is_dir()
+                    })
+                    .collect();
+
+                if let Some(max_keys) = max_keys {
+                    result.keys.truncate(max_keys.get() as usize);
+                }
+
+                return Ok(result);
+            }
+
+            let path = match prefix {
+                Some(prefix) => Cow::Owned(prefix.with_base(&self.storage_root)),
+                None => Cow::Borrowed(&self.storage_root),
+            };
+
+            let prefixes_to_filter = get_all_files(path.as_ref(), false)
                 .await
                 .map_err(DownloadError::Other)?;
-            let keys = keys
-                .into_iter()
-                .filter(|k| {
-                    let path = k.with_base(&self.storage_root);
-                    !path.is_dir()
-                })
-                .collect();
 
-            if let ListingMode::NoDelimiter = mode {
-                result.keys = keys;
-            } else {
-                let mut prefixes = HashSet::new();
-                for key in keys {
-                    // If the part after the prefix includes a "/", take only the first part and put it in `prefixes`.
-                    let relative_key = if let Some(prefix) = prefix {
-                        let mut prefix = prefix.clone();
-                        // We only strip the dirname of the prefix, so that when we strip it from the start of keys we
-                        // end up with full file/dir names.
-                        let prefix_full_local_path = prefix.with_base(&self.storage_root);
-                        let has_slash = prefix.0.to_string().ends_with('/');
-                        let strip_prefix = if prefix_full_local_path.is_dir() && has_slash {
-                            prefix
-                        } else {
-                            prefix.0.pop();
-                            prefix
-                        };
-
-                        RemotePath::new(key.strip_prefix(&strip_prefix).unwrap()).unwrap()
-                    } else {
-                        key
-                    };
-
-                    let relative_key = format!("{}", relative_key);
-                    if relative_key.contains(REMOTE_STORAGE_PREFIX_SEPARATOR) {
-                        let first_part = relative_key
-                            .split(REMOTE_STORAGE_PREFIX_SEPARATOR)
-                            .next()
-                            .unwrap()
-                            .to_owned();
-                        prefixes.insert(first_part);
-                    } else {
-                        result
-                            .keys
-                            .push(RemotePath::from_string(&relative_key).unwrap());
-                    }
+            // filter out empty directories to mirror s3 behavior.
+            for prefix in prefixes_to_filter {
+                if prefix.is_dir()
+                    && is_directory_empty(&prefix)
+                        .await
+                        .map_err(DownloadError::Other)?
+                {
+                    continue;
+                }
+
+                let stripped = prefix
+                    .strip_prefix(&self.storage_root)
+                    .context("Failed to strip prefix")
+                    .and_then(RemotePath::new)
+                    .expect(
+                        "We list files for storage root, hence should be able to remote the prefix",
+                    );
+
+                if prefix.is_dir() {
+                    result.prefixes.push(stripped);
+                } else {
+                    result.keys.push(stripped);
                 }
-                result.prefixes = prefixes
-                    .into_iter()
-                    .map(|s| RemotePath::from_string(&s).unwrap())
-                    .collect();
             }
 
-            if let Some(max_keys) = max_keys {
-                result.keys.truncate(max_keys.get() as usize);
-            }
             Ok(result)
         };
 
@@ -611,6 +560,50 @@ fn storage_metadata_path(original_path: &Utf8Path) -> Utf8PathBuf {
     path_with_suffix_extension(original_path, "metadata")
 }
 
+fn get_all_files<'a, P>(
+    directory_path: P,
+    recursive: bool,
+) -> Pin<Box<dyn Future<Output = anyhow::Result<Vec<Utf8PathBuf>>> + Send + Sync + 'a>>
+where
+    P: AsRef<Utf8Path> + Send + Sync + 'a,
+{
+    Box::pin(async move {
+        let directory_path = directory_path.as_ref();
+        if directory_path.exists() {
+            if directory_path.is_dir() {
+                let mut paths = Vec::new();
+                let mut dir_contents = fs::read_dir(directory_path).await?;
+                while let Some(dir_entry) = dir_contents.next_entry().await? {
+                    let file_type = dir_entry.file_type().await?;
+                    let entry_path =
+                        Utf8PathBuf::from_path_buf(dir_entry.path()).map_err(|pb| {
+                            anyhow::Error::msg(format!(
+                                "non-Unicode path: {}",
+                                pb.to_string_lossy()
+                            ))
+                        })?;
+                    if file_type.is_symlink() {
+                        debug!("{entry_path:?} is a symlink, skipping")
+                    } else if file_type.is_dir() {
+                        if recursive {
+                            paths.extend(get_all_files(&entry_path, true).await?.into_iter())
+                        } else {
+                            paths.push(entry_path)
+                        }
+                    } else {
+                        paths.push(entry_path);
+                    }
+                }
+                Ok(paths)
+            } else {
+                bail!("Path {directory_path:?} is not a directory")
+            }
+        } else {
+            Ok(Vec::new())
+        }
+    })
+}
+
 async fn create_target_directory(target_file_path: &Utf8Path) -> anyhow::Result<()> {
     let target_dir = match target_file_path.parent() {
         Some(parent_dir) => parent_dir,
@@ -930,18 +923,13 @@ mod fs_tests {
         // No delimiter: should recursively list everything
         let (storage, cancel) = create_storage()?;
         let child = upload_dummy_file(&storage, "grandparent/parent/child", None, &cancel).await?;
-        let child_sibling =
-            upload_dummy_file(&storage, "grandparent/parent/child_sibling", None, &cancel).await?;
         let uncle = upload_dummy_file(&storage, "grandparent/uncle", None, &cancel).await?;
 
         let listing = storage
             .list(None, ListingMode::NoDelimiter, None, &cancel)
             .await?;
         assert!(listing.prefixes.is_empty());
-        assert_eq!(
-            listing.keys.into_iter().collect::<HashSet<_>>(),
-            HashSet::from([uncle.clone(), child.clone(), child_sibling.clone()])
-        );
+        assert_eq!(listing.keys, [uncle.clone(), child.clone()].to_vec());
 
         // Delimiter: should only go one deep
         let listing = storage
@@ -954,25 +942,7 @@ mod fs_tests {
         );
         assert!(listing.keys.is_empty());
 
-        // Delimiter & prefix with a trailing slash
-        let listing = storage
-            .list(
-                Some(&RemotePath::from_string("timelines/some_timeline/grandparent/").unwrap()),
-                ListingMode::WithDelimiter,
-                None,
-                &cancel,
-            )
-            .await?;
-        assert_eq!(
-            listing.keys,
-            [RemotePath::from_string("uncle").unwrap()].to_vec()
-        );
-        assert_eq!(
-            listing.prefixes,
-            [RemotePath::from_string("parent").unwrap()].to_vec()
-        );
-
-        // Delimiter and prefix without a trailing slash
+        // Delimiter & prefix
         let listing = storage
             .list(
                 Some(&RemotePath::from_string("timelines/some_timeline/grandparent").unwrap()),
@@ -981,66 +951,12 @@ mod fs_tests {
                 &cancel,
             )
             .await?;
-        assert_eq!(listing.keys, [].to_vec());
         assert_eq!(
             listing.prefixes,
-            [RemotePath::from_string("grandparent").unwrap()].to_vec()
-        );
-
-        // Delimiter and prefix that's partway through a path component
-        let listing = storage
-            .list(
-                Some(&RemotePath::from_string("timelines/some_timeline/grandp").unwrap()),
-                ListingMode::WithDelimiter,
-                None,
-                &cancel,
-            )
-            .await?;
-        assert_eq!(listing.keys, [].to_vec());
-        assert_eq!(
-            listing.prefixes,
-            [RemotePath::from_string("grandparent").unwrap()].to_vec()
-        );
-
-        Ok(())
-    }
-
-    #[tokio::test]
-    async fn list_part_component() -> anyhow::Result<()> {
-        // No delimiter: should recursively list everything
-        let (storage, cancel) = create_storage()?;
-
-        // Imitates what happens in a tenant path when we have an unsharded path and a sharded path, and do a listing
-        // of the unsharded path: although there is a "directory" at the unsharded path, it should be handled as
-        // a freeform prefix.
-        let _child_a =
-            upload_dummy_file(&storage, "grandparent/tenant-01/child", None, &cancel).await?;
-        let _child_b =
-            upload_dummy_file(&storage, "grandparent/tenant/child", None, &cancel).await?;
-
-        // Delimiter and prefix that's partway through a path component
-        let listing = storage
-            .list(
-                Some(
-                    &RemotePath::from_string("timelines/some_timeline/grandparent/tenant").unwrap(),
-                ),
-                ListingMode::WithDelimiter,
-                None,
-                &cancel,
-            )
-            .await?;
-        assert_eq!(listing.keys, [].to_vec());
-
-        let mut found_prefixes = listing.prefixes.clone();
-        found_prefixes.sort();
-        assert_eq!(
-            found_prefixes,
-            [
-                RemotePath::from_string("tenant").unwrap(),
-                RemotePath::from_string("tenant-01").unwrap(),
-            ]
-            .to_vec()
+            [RemotePath::from_string("timelines/some_timeline/grandparent/parent").unwrap()]
+                .to_vec()
         );
+        assert_eq!(listing.keys, [uncle.clone()].to_vec());
 
         Ok(())
     }

libs/remote_storage/src/s3_bucket.rs

@@ -30,7 +30,7 @@ use aws_sdk_s3::{
     config::{AsyncSleep, Builder, IdentityCache, Region, SharedAsyncSleep},
     error::SdkError,
     operation::get_object::GetObjectError,
-    types::{Delete, DeleteMarkerEntry, ObjectIdentifier, ObjectVersion, StorageClass},
+    types::{Delete, DeleteMarkerEntry, ObjectIdentifier, ObjectVersion},
     Client,
 };
 use aws_smithy_async::rt::sleep::TokioSleep;
@@ -62,7 +62,6 @@ pub struct S3Bucket {
     bucket_name: String,
     prefix_in_bucket: Option<String>,
     max_keys_per_list_response: Option<i32>,
-    upload_storage_class: Option<StorageClass>,
     concurrency_limiter: ConcurrencyLimiter,
     // Per-request timeout. Accessible for tests.
     pub timeout: Duration,
@@ -155,7 +154,6 @@ impl S3Bucket {
             max_keys_per_list_response: aws_config.max_keys_per_list_response,
             prefix_in_bucket,
             concurrency_limiter: ConcurrencyLimiter::new(aws_config.concurrency_limit.get()),
-            upload_storage_class: aws_config.upload_storage_class.clone(),
             timeout,
         })
     }
@@ -180,7 +178,10 @@ impl S3Bucket {
 
     pub fn relative_path_to_s3_object(&self, path: &RemotePath) -> String {
         assert_eq!(std::path::MAIN_SEPARATOR, REMOTE_STORAGE_PREFIX_SEPARATOR);
-        let path_string = path.get_path().as_str();
+        let path_string = path
+            .get_path()
+            .as_str()
+            .trim_end_matches(REMOTE_STORAGE_PREFIX_SEPARATOR);
         match &self.prefix_in_bucket {
             Some(prefix) => prefix.clone() + "/" + path_string,
             None => path_string.to_string(),
@@ -470,11 +471,16 @@ impl RemoteStorage for S3Bucket {
         // get the passed prefix or if it is not set use prefix_in_bucket value
         let list_prefix = prefix
             .map(|p| self.relative_path_to_s3_object(p))
-            .or_else(|| {
-                self.prefix_in_bucket.clone().map(|mut s| {
-                    s.push(REMOTE_STORAGE_PREFIX_SEPARATOR);
-                    s
-                })
+            .or_else(|| self.prefix_in_bucket.clone())
+            .map(|mut p| {
+                // required to end with a separator
+                // otherwise request will return only the entry of a prefix
+                if matches!(mode, ListingMode::WithDelimiter)
+                    && !p.ends_with(REMOTE_STORAGE_PREFIX_SEPARATOR)
+                {
+                    p.push(REMOTE_STORAGE_PREFIX_SEPARATOR);
+                }
+                p
             });
 
         let _permit = self.permit(kind, cancel).await?;
@@ -543,15 +549,11 @@ impl RemoteStorage for S3Bucket {
                 }
             }
 
-            // S3 gives us prefixes like "foo/", we return them like "foo"
-            result.prefixes.extend(prefixes.iter().filter_map(|o| {
-                Some(
-                    self.s3_object_to_relative_path(
-                        o.prefix()?
-                            .trim_end_matches(REMOTE_STORAGE_PREFIX_SEPARATOR),
-                    ),
-                )
-            }));
+            result.prefixes.extend(
+                prefixes
+                    .iter()
+                    .filter_map(|o| Some(self.s3_object_to_relative_path(o.prefix()?))),
+            );
 
             continuation_token = match response.next_continuation_token {
                 Some(new_token) => Some(new_token),
@@ -584,7 +586,6 @@ impl RemoteStorage for S3Bucket {
             .bucket(self.bucket_name.clone())
             .key(self.relative_path_to_s3_object(to))
             .set_metadata(metadata.map(|m| m.0))
-            .set_storage_class(self.upload_storage_class.clone())
             .content_length(from_size_bytes.try_into()?)
             .body(bytes_stream)
             .send();
@@ -636,7 +637,6 @@ impl RemoteStorage for S3Bucket {
             .copy_object()
             .bucket(self.bucket_name.clone())
             .key(self.relative_path_to_s3_object(to))
-            .set_storage_class(self.upload_storage_class.clone())
             .copy_source(copy_source)
             .send();
 
@@ -894,7 +894,6 @@ impl RemoteStorage for S3Bucket {
                                     .copy_object()
                                     .bucket(self.bucket_name.clone())
                                     .key(key)
-                                    .set_storage_class(self.upload_storage_class.clone())
                                     .copy_source(&source_id)
                                     .send();
 
@@ -1051,22 +1050,22 @@ mod tests {
             Some("/test/prefix/"),
         ];
         let expected_outputs = [
-            vec!["", "some/path", "some/path/"],
-            vec!["/", "/some/path", "/some/path/"],
+            vec!["", "some/path", "some/path"],
+            vec!["/", "/some/path", "/some/path"],
             vec![
                 "test/prefix/",
                 "test/prefix/some/path",
-                "test/prefix/some/path/",
+                "test/prefix/some/path",
             ],
             vec![
                 "test/prefix/",
                 "test/prefix/some/path",
-                "test/prefix/some/path/",
+                "test/prefix/some/path",
             ],
             vec![
                 "test/prefix/",
                 "test/prefix/some/path",
-                "test/prefix/some/path/",
+                "test/prefix/some/path",
             ],
         ];
 
@@ -1078,7 +1077,6 @@ mod tests {
                 endpoint: None,
                 concurrency_limit: NonZeroUsize::new(100).unwrap(),
                 max_keys_per_list_response: Some(5),
-                upload_storage_class: None,
             };
             let storage =
                 S3Bucket::new(&config, std::time::Duration::ZERO).expect("remote storage init");

libs/remote_storage/src/simulate_failures.rs

@@ -107,6 +107,27 @@ impl UnreliableWrapper {
 type VoidStorage = crate::LocalFs;
 
 impl RemoteStorage for UnreliableWrapper {
+    async fn list_prefixes(
+        &self,
+        prefix: Option<&RemotePath>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        self.attempt(RemoteOp::ListPrefixes(prefix.cloned()))
+            .map_err(DownloadError::Other)?;
+        self.inner.list_prefixes(prefix, cancel).await
+    }
+
+    async fn list_files(
+        &self,
+        folder: Option<&RemotePath>,
+        max_keys: Option<NonZeroU32>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        self.attempt(RemoteOp::ListPrefixes(folder.cloned()))
+            .map_err(DownloadError::Other)?;
+        self.inner.list_files(folder, max_keys, cancel).await
+    }
+
     async fn list(
         &self,
         prefix: Option<&RemotePath>,

libs/remote_storage/tests/common/tests.rs

@@ -1,6 +1,5 @@
 use anyhow::Context;
 use camino::Utf8Path;
-use remote_storage::ListingMode;
 use remote_storage::RemotePath;
 use std::sync::Arc;
 use std::{collections::HashSet, num::NonZeroU32};
@@ -55,9 +54,9 @@ async fn pagination_should_work(ctx: &mut MaybeEnabledStorageWithTestBlobs) -> a
     let base_prefix = RemotePath::new(Utf8Path::new(ctx.enabled.base_prefix))
         .context("common_prefix construction")?;
     let root_remote_prefixes = test_client
-        .list(None, ListingMode::WithDelimiter, None, &cancel)
-        .await?
-        .prefixes
+        .list_prefixes(None, &cancel)
+        .await
+        .context("client list root prefixes failure")?
         .into_iter()
         .collect::<HashSet<_>>();
     assert_eq!(
@@ -66,14 +65,9 @@ async fn pagination_should_work(ctx: &mut MaybeEnabledStorageWithTestBlobs) -> a
     );
 
     let nested_remote_prefixes = test_client
-        .list(
-            Some(&base_prefix.add_trailing_slash()),
-            ListingMode::WithDelimiter,
-            None,
-            &cancel,
-        )
-        .await?
-        .prefixes
+        .list_prefixes(Some(&base_prefix), &cancel)
+        .await
+        .context("client list nested prefixes failure")?
         .into_iter()
         .collect::<HashSet<_>>();
     let remote_only_prefixes = nested_remote_prefixes
@@ -96,13 +90,11 @@ async fn pagination_should_work(ctx: &mut MaybeEnabledStorageWithTestBlobs) -> a
 ///
 /// First, create a set of S3 objects with keys `random_prefix/folder{j}/blob_{i}.txt` in [`upload_remote_data`]
 /// Then performs the following queries:
-///    1. `list(None)`. This should return all files `random_prefix/folder{j}/blob_{i}.txt`
-///    2. `list("folder1")`.  This  should return all files `random_prefix/folder1/blob_{i}.txt`
+///    1. `list_files(None)`. This should return all files `random_prefix/folder{j}/blob_{i}.txt`
+///    2. `list_files("folder1")`.  This  should return all files `random_prefix/folder1/blob_{i}.txt`
 #[test_context(MaybeEnabledStorageWithSimpleTestBlobs)]
 #[tokio::test]
-async fn list_no_delimiter_works(
-    ctx: &mut MaybeEnabledStorageWithSimpleTestBlobs,
-) -> anyhow::Result<()> {
+async fn list_files_works(ctx: &mut MaybeEnabledStorageWithSimpleTestBlobs) -> anyhow::Result<()> {
     let ctx = match ctx {
         MaybeEnabledStorageWithSimpleTestBlobs::Enabled(ctx) => ctx,
         MaybeEnabledStorageWithSimpleTestBlobs::Disabled => return Ok(()),
@@ -115,36 +107,29 @@ async fn list_no_delimiter_works(
     let base_prefix =
         RemotePath::new(Utf8Path::new("folder1")).context("common_prefix construction")?;
     let root_files = test_client
-        .list(None, ListingMode::NoDelimiter, None, &cancel)
+        .list_files(None, None, &cancel)
         .await
         .context("client list root files failure")?
-        .keys
         .into_iter()
         .collect::<HashSet<_>>();
     assert_eq!(
         root_files,
         ctx.remote_blobs.clone(),
-        "remote storage list on root mismatches with the uploads."
+        "remote storage list_files on root mismatches with the uploads."
     );
 
     // Test that max_keys limit works. In total there are about 21 files (see
     // upload_simple_remote_data call in test_real_s3.rs).
     let limited_root_files = test_client
-        .list(
-            None,
-            ListingMode::NoDelimiter,
-            Some(NonZeroU32::new(2).unwrap()),
-            &cancel,
-        )
+        .list_files(None, Some(NonZeroU32::new(2).unwrap()), &cancel)
         .await
         .context("client list root files failure")?;
-    assert_eq!(limited_root_files.keys.len(), 2);
+    assert_eq!(limited_root_files.len(), 2);
 
     let nested_remote_files = test_client
-        .list(Some(&base_prefix), ListingMode::NoDelimiter, None, &cancel)
+        .list_files(Some(&base_prefix), None, &cancel)
         .await
         .context("client list nested files failure")?
-        .keys
         .into_iter()
         .collect::<HashSet<_>>();
     let trim_remote_blobs: HashSet<_> = ctx
@@ -156,7 +141,7 @@ async fn list_no_delimiter_works(
         .collect();
     assert_eq!(
         nested_remote_files, trim_remote_blobs,
-        "remote storage list on subdirrectory mismatches with the uploads."
+        "remote storage list_files on subdirrectory mismatches with the uploads."
     );
     Ok(())
 }
@@ -214,11 +199,7 @@ async fn delete_objects_works(ctx: &mut MaybeEnabledStorage) -> anyhow::Result<(
 
     ctx.client.delete_objects(&[path1, path2], &cancel).await?;
 
-    let prefixes = ctx
-        .client
-        .list(None, ListingMode::WithDelimiter, None, &cancel)
-        .await?
-        .prefixes;
+    let prefixes = ctx.client.list_prefixes(None, &cancel).await?;
 
     assert_eq!(prefixes.len(), 1);
 

libs/remote_storage/tests/test_real_azure.rs

@@ -132,6 +132,10 @@ impl AsyncTestContext for MaybeEnabledStorageWithTestBlobs {
     }
 }
 
+// NOTE: the setups for the list_prefixes test and the list_files test are very similar
+// However, they are not idential. The list_prefixes function is concerned with listing prefixes,
+// whereas the list_files function is concerned with listing files.
+// See `RemoteStorage::list_files` documentation for more details
 enum MaybeEnabledStorageWithSimpleTestBlobs {
     Enabled(AzureWithSimpleTestBlobs),
     Disabled,

libs/remote_storage/tests/test_real_s3.rs

@@ -12,8 +12,8 @@ use anyhow::Context;
 use camino::Utf8Path;
 use futures_util::StreamExt;
 use remote_storage::{
-    DownloadError, GenericRemoteStorage, ListingMode, RemotePath, RemoteStorageConfig,
-    RemoteStorageKind, S3Config,
+    DownloadError, GenericRemoteStorage, RemotePath, RemoteStorageConfig, RemoteStorageKind,
+    S3Config,
 };
 use test_context::test_context;
 use test_context::AsyncTestContext;
@@ -75,14 +75,11 @@ async fn s3_time_travel_recovery_works(ctx: &mut MaybeEnabledStorage) -> anyhow:
         client: &Arc<GenericRemoteStorage>,
         cancel: &CancellationToken,
     ) -> anyhow::Result<HashSet<RemotePath>> {
-        Ok(
-            retry(|| client.list(None, ListingMode::NoDelimiter, None, cancel))
-                .await
-                .context("list root files failure")?
-                .keys
-                .into_iter()
-                .collect::<HashSet<_>>(),
-        )
+        Ok(retry(|| client.list_files(None, None, cancel))
+            .await
+            .context("list root files failure")?
+            .into_iter()
+            .collect::<HashSet<_>>())
     }
 
     let cancel = CancellationToken::new();
@@ -297,6 +294,10 @@ impl AsyncTestContext for MaybeEnabledStorageWithTestBlobs {
     }
 }
 
+// NOTE: the setups for the list_prefixes test and the list_files test are very similar
+// However, they are not idential. The list_prefixes function is concerned with listing prefixes,
+// whereas the list_files function is concerned with listing files.
+// See `RemoteStorage::list_files` documentation for more details
 enum MaybeEnabledStorageWithSimpleTestBlobs {
     Enabled(S3WithSimpleTestBlobs),
     Disabled,
@@ -380,7 +381,6 @@ fn create_s3_client(
             endpoint: None,
             concurrency_limit: NonZeroUsize::new(100).unwrap(),
             max_keys_per_list_response,
-            upload_storage_class: None,
         }),
         timeout: RemoteStorageConfig::DEFAULT_TIMEOUT,
     };

libs/utils/src/rate_limit.rs

@@ -5,6 +5,7 @@ use std::time::{Duration, Instant};
 pub struct RateLimit {
     last: Option<Instant>,
     interval: Duration,
+    rate_limited: u32,
 }
 
 impl RateLimit {
@@ -12,9 +13,16 @@ impl RateLimit {
         Self {
             last: None,
             interval,
+            rate_limited: 0,
         }
     }
 
+    /// Returns the number of calls that were rate limited
+    /// since the last one was allowed.
+    pub fn rate_limited(&self) -> u32 {
+        self.rate_limited
+    }
+
     /// Call `f` if the rate limit allows.
     /// Don't call it otherwise.
     pub fn call<F: FnOnce()>(&mut self, f: F) {
@@ -22,9 +30,13 @@ impl RateLimit {
         match self.last {
             Some(last) if now - last <= self.interval => {
                 // ratelimit
+                if let Some(updated) = self.rate_limited.checked_add(1) {
+                    self.rate_limited = updated;
+                }
             }
             _ => {
                 self.last = Some(now);
+                self.rate_limited = 0;
                 f();
             }
         }
@@ -50,17 +62,24 @@ mod tests {
 
         f.call(cl);
         assert_eq!(called.load(Relaxed), 1);
+        assert_eq!(f.rate_limited(), 0);
         f.call(cl);
         assert_eq!(called.load(Relaxed), 1);
+        assert_eq!(f.rate_limited(), 1);
         f.call(cl);
         assert_eq!(called.load(Relaxed), 1);
+        assert_eq!(f.rate_limited(), 2);
         std::thread::sleep(Duration::from_millis(100));
         f.call(cl);
         assert_eq!(called.load(Relaxed), 2);
+        assert_eq!(f.rate_limited(), 0);
         f.call(cl);
         assert_eq!(called.load(Relaxed), 2);
+        assert_eq!(f.rate_limited(), 1);
+        f.call(cl);
         std::thread::sleep(Duration::from_millis(100));
         f.call(cl);
         assert_eq!(called.load(Relaxed), 3);
+        assert_eq!(f.rate_limited(), 0);
     }
 }

pageserver/Cargo.toml

@@ -70,7 +70,6 @@ tokio-stream.workspace = true
 tokio-util.workspace = true
 toml_edit = { workspace = true, features = [ "serde" ] }
 tracing.workspace = true
-twox-hash.workspace = true
 url.workspace = true
 walkdir.workspace = true
 metrics.workspace = true

pageserver/client/src/mgmt_api.rs

@@ -279,7 +279,7 @@ impl Client {
         lazy: bool,
     ) -> Result<()> {
         let req_body = TenantLocationConfigRequest {
-            tenant_id: None,
+            tenant_id: Some(tenant_shard_id),
             config,
         };
 

pageserver/client/src/page_service.rs

@@ -60,7 +60,7 @@ impl Client {
     ) -> anyhow::Result<PagestreamClient> {
         let copy_both: tokio_postgres::CopyBothDuplex<bytes::Bytes> = self
             .client
-            .copy_both_simple(&format!("pagestream_v2 {tenant_id} {timeline_id}"))
+            .copy_both_simple(&format!("pagestream {tenant_id} {timeline_id}"))
             .await?;
         let Client {
             cancel_on_client_drop,

pageserver/pagebench/src/cmd/getpage_latest_lsn.rs

@@ -312,12 +312,8 @@ async fn main_impl(
                     let (rel_tag, block_no) =
                         key_to_rel_block(key).expect("we filter non-rel-block keys out above");
                     PagestreamGetPageRequest {
-                        request_lsn: if rng.gen_bool(args.req_latest_probability) {
-                            Lsn::MAX
-                        } else {
-                            r.timeline_lsn
-                        },
-                        not_modified_since: r.timeline_lsn,
+                        latest: rng.gen_bool(args.req_latest_probability),
+                        lsn: r.timeline_lsn,
                         rel: rel_tag,
                         blkno: block_no,
                     }

pageserver/src/aux_file.rs

@@ -1,112 +0,0 @@
-use pageserver_api::key::{Key, AUX_KEY_PREFIX, METADATA_KEY_SIZE};
-use tracing::warn;
-
-/// Create a metadata key from a hash, encoded as [AUX_KEY_PREFIX, 2B directory prefix, first 13B of 128b xxhash].
-fn aux_hash_to_metadata_key(dir_level1: u8, dir_level2: u8, data: &[u8]) -> Key {
-    let mut key = [0; METADATA_KEY_SIZE];
-    let hash = twox_hash::xxh3::hash128(data).to_be_bytes();
-    key[0] = AUX_KEY_PREFIX;
-    key[1] = dir_level1;
-    key[2] = dir_level2;
-    key[3..16].copy_from_slice(&hash[0..13]);
-    Key::from_metadata_key_fixed_size(&key)
-}
-
-const AUX_DIR_PG_LOGICAL: u8 = 0x01;
-const AUX_DIR_PG_REPLSLOT: u8 = 0x02;
-const AUX_DIR_PG_UNKNOWN: u8 = 0xFF;
-
-/// Encode the aux file into a fixed-size key.
-///
-/// The first byte is the AUX key prefix. We use the next 2 bytes of the key for the directory / aux file type.
-/// We have one-to-one mapping for each of the aux file that we support. We hash the remaining part of the path
-/// (usually a single file name, or several components) into 13-byte hash. The way we determine the 2-byte prefix
-/// is roughly based on the first two components of the path, one unique number for one component.
-///
-/// * pg_logical/mappings -> 0x0101
-/// * pg_logical/snapshots -> 0x0102
-/// * pg_logical/replorigin_checkpoint -> 0x0103
-/// * pg_logical/others -> 0x01FF
-/// * pg_replslot/ -> 0x0201
-/// * others -> 0xFFFF
-///
-/// If you add new AUX files to this function, please also add a test case to `test_encoding_portable`.
-/// The new file type must have never been written to the storage before. Otherwise, there could be data
-/// corruptions as the new file belongs to a new prefix but it might have been stored under the `others` prefix.
-pub fn encode_aux_file_key(path: &str) -> Key {
-    if let Some(fname) = path.strip_prefix("pg_logical/mappings/") {
-        aux_hash_to_metadata_key(AUX_DIR_PG_LOGICAL, 0x01, fname.as_bytes())
-    } else if let Some(fname) = path.strip_prefix("pg_logical/snapshots/") {
-        aux_hash_to_metadata_key(AUX_DIR_PG_LOGICAL, 0x02, fname.as_bytes())
-    } else if path == "pg_logical/replorigin_checkpoint" {
-        aux_hash_to_metadata_key(AUX_DIR_PG_LOGICAL, 0x03, b"")
-    } else if let Some(fname) = path.strip_prefix("pg_logical/") {
-        if cfg!(debug_assertions) {
-            warn!(
-                "unsupported pg_logical aux file type: {}, putting to 0x01FF, would affect path scanning",
-                path
-            );
-        }
-        aux_hash_to_metadata_key(AUX_DIR_PG_LOGICAL, 0xFF, fname.as_bytes())
-    } else if let Some(fname) = path.strip_prefix("pg_replslot/") {
-        aux_hash_to_metadata_key(AUX_DIR_PG_REPLSLOT, 0x01, fname.as_bytes())
-    } else {
-        if cfg!(debug_assertions) {
-            warn!(
-                "unsupported aux file type: {}, putting to 0xFFFF, would affect path scanning",
-                path
-            );
-        }
-        aux_hash_to_metadata_key(AUX_DIR_PG_UNKNOWN, 0xFF, path.as_bytes())
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_hash_portable() {
-        // AUX file encoding requires the hash to be portable across all platforms. This test case checks
-        // if the algorithm produces the same hash across different environments.
-        assert_eq!(
-            305317690835051308206966631765527126151,
-            twox_hash::xxh3::hash128("test1".as_bytes())
-        );
-        assert_eq!(
-            85104974691013376326742244813280798847,
-            twox_hash::xxh3::hash128("test/test2".as_bytes())
-        );
-        assert_eq!(0, twox_hash::xxh3::hash128("".as_bytes()));
-    }
-
-    #[test]
-    fn test_encoding_portable() {
-        // To correct retrieve AUX files, the generated keys for the same file must be the same for all versions
-        // of the page server.
-        assert_eq!(
-            "6200000101E5B20C5F8DD5AA3289D6D9EAFA",
-            encode_aux_file_key("pg_logical/mappings/test1").to_string()
-        );
-        assert_eq!(
-            "620000010239AAC544893139B26F501B97E6",
-            encode_aux_file_key("pg_logical/snapshots/test2").to_string()
-        );
-        assert_eq!(
-            "620000010300000000000000000000000000",
-            encode_aux_file_key("pg_logical/replorigin_checkpoint").to_string()
-        );
-        assert_eq!(
-            "62000001FF8635AF2134B7266EC5B4189FD6",
-            encode_aux_file_key("pg_logical/unsupported").to_string()
-        );
-        assert_eq!(
-            "6200000201772D0E5D71DE14DA86142A1619",
-            encode_aux_file_key("pg_replslot/test3").to_string()
-        );
-        assert_eq!(
-            "620000FFFF1866EBEB53B807B26A2416F317",
-            encode_aux_file_key("other_file_not_supported").to_string()
-        );
-    }
-}

pageserver/src/basebackup.rs

@@ -376,7 +376,7 @@ where
     async fn add_rel(&mut self, src: RelTag, dst: RelTag) -> anyhow::Result<()> {
         let nblocks = self
             .timeline
-            .get_rel_size(src, Version::Lsn(self.lsn), self.ctx)
+            .get_rel_size(src, Version::Lsn(self.lsn), false, self.ctx)
             .await?;
 
         // If the relation is empty, create an empty file
@@ -397,7 +397,7 @@ where
             for blknum in startblk..endblk {
                 let img = self
                     .timeline
-                    .get_rel_page_at_lsn(src, blknum, Version::Lsn(self.lsn), self.ctx)
+                    .get_rel_page_at_lsn(src, blknum, Version::Lsn(self.lsn), false, self.ctx)
                     .await?;
                 segment_data.extend_from_slice(&img[..]);
             }

pageserver/src/bin/pageserver.rs

@@ -121,10 +121,8 @@ fn main() -> anyhow::Result<()> {
         &[("node_id", &conf.id.to_string())],
     );
 
-    // after setting up logging, log the effective IO engine choice and read path implementations
+    // after setting up logging, log the effective IO engine choice
     info!(?conf.virtual_file_io_engine, "starting with virtual_file IO engine");
-    info!(?conf.get_impl, "starting with get page implementation");
-    info!(?conf.get_vectored_impl, "starting with vectored get page implementation");
 
     let tenants_path = conf.tenants_path();
     if !tenants_path.exists() {

pageserver/src/config.rs

@@ -30,9 +30,9 @@ use utils::{
     logging::LogFormat,
 };
 
+use crate::tenant::config::TenantConfOpt;
 use crate::tenant::timeline::GetVectoredImpl;
 use crate::tenant::vectored_blob_io::MaxVectoredReadBytes;
-use crate::tenant::{config::TenantConfOpt, timeline::GetImpl};
 use crate::tenant::{
     TENANTS_SEGMENT_NAME, TENANT_DELETED_MARKER_FILE_NAME, TIMELINES_SEGMENT_NAME,
 };
@@ -91,8 +91,6 @@ pub mod defaults {
 
     pub const DEFAULT_GET_VECTORED_IMPL: &str = "sequential";
 
-    pub const DEFAULT_GET_IMPL: &str = "legacy";
-
     pub const DEFAULT_MAX_VECTORED_READ_BYTES: usize = 128 * 1024; // 128 KiB
 
     pub const DEFAULT_VALIDATE_VECTORED_GET: bool = true;
@@ -140,8 +138,6 @@ pub mod defaults {
 
 #get_vectored_impl = '{DEFAULT_GET_VECTORED_IMPL}'
 
-#get_impl = '{DEFAULT_GET_IMPL}'
-
 #max_vectored_read_bytes = '{DEFAULT_MAX_VECTORED_READ_BYTES}'
 
 #validate_vectored_get = '{DEFAULT_VALIDATE_VECTORED_GET}'
@@ -288,8 +284,6 @@ pub struct PageServerConf {
 
     pub get_vectored_impl: GetVectoredImpl,
 
-    pub get_impl: GetImpl,
-
     pub max_vectored_read_bytes: MaxVectoredReadBytes,
 
     pub validate_vectored_get: bool,
@@ -420,8 +414,6 @@ struct PageServerConfigBuilder {
 
     get_vectored_impl: BuilderValue<GetVectoredImpl>,
 
-    get_impl: BuilderValue<GetImpl>,
-
     max_vectored_read_bytes: BuilderValue<MaxVectoredReadBytes>,
 
     validate_vectored_get: BuilderValue<bool>,
@@ -511,7 +503,6 @@ impl PageServerConfigBuilder {
             virtual_file_io_engine: Set(DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap()),
 
             get_vectored_impl: Set(DEFAULT_GET_VECTORED_IMPL.parse().unwrap()),
-            get_impl: Set(DEFAULT_GET_IMPL.parse().unwrap()),
             max_vectored_read_bytes: Set(MaxVectoredReadBytes(
                 NonZeroUsize::new(DEFAULT_MAX_VECTORED_READ_BYTES).unwrap(),
             )),
@@ -690,10 +681,6 @@ impl PageServerConfigBuilder {
         self.get_vectored_impl = BuilderValue::Set(value);
     }
 
-    pub fn get_impl(&mut self, value: GetImpl) {
-        self.get_impl = BuilderValue::Set(value);
-    }
-
     pub fn get_max_vectored_read_bytes(&mut self, value: MaxVectoredReadBytes) {
         self.max_vectored_read_bytes = BuilderValue::Set(value);
     }
@@ -763,7 +750,6 @@ impl PageServerConfigBuilder {
                 secondary_download_concurrency,
                 ingest_batch_size,
                 get_vectored_impl,
-                get_impl,
                 max_vectored_read_bytes,
                 validate_vectored_get,
                 ephemeral_bytes_per_memory_kb,
@@ -1049,9 +1035,6 @@ impl PageServerConf {
                 "get_vectored_impl" => {
                     builder.get_vectored_impl(parse_toml_from_str("get_vectored_impl", item)?)
                 }
-                "get_impl" => {
-                    builder.get_impl(parse_toml_from_str("get_impl", item)?)
-                }
                 "max_vectored_read_bytes" => {
                     let bytes = parse_toml_u64("max_vectored_read_bytes", item)? as usize;
                     builder.get_max_vectored_read_bytes(
@@ -1143,7 +1126,6 @@ impl PageServerConf {
             ingest_batch_size: defaults::DEFAULT_INGEST_BATCH_SIZE,
             virtual_file_io_engine: DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap(),
             get_vectored_impl: defaults::DEFAULT_GET_VECTORED_IMPL.parse().unwrap(),
-            get_impl: defaults::DEFAULT_GET_IMPL.parse().unwrap(),
             max_vectored_read_bytes: MaxVectoredReadBytes(
                 NonZeroUsize::new(defaults::DEFAULT_MAX_VECTORED_READ_BYTES)
                     .expect("Invalid default constant"),
@@ -1383,7 +1365,6 @@ background_task_maximum_delay = '334 s'
                 ingest_batch_size: defaults::DEFAULT_INGEST_BATCH_SIZE,
                 virtual_file_io_engine: DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap(),
                 get_vectored_impl: defaults::DEFAULT_GET_VECTORED_IMPL.parse().unwrap(),
-                get_impl: defaults::DEFAULT_GET_IMPL.parse().unwrap(),
                 max_vectored_read_bytes: MaxVectoredReadBytes(
                     NonZeroUsize::new(defaults::DEFAULT_MAX_VECTORED_READ_BYTES)
                         .expect("Invalid default constant")
@@ -1457,7 +1438,6 @@ background_task_maximum_delay = '334 s'
                 ingest_batch_size: 100,
                 virtual_file_io_engine: DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap(),
                 get_vectored_impl: defaults::DEFAULT_GET_VECTORED_IMPL.parse().unwrap(),
-                get_impl: defaults::DEFAULT_GET_IMPL.parse().unwrap(),
                 max_vectored_read_bytes: MaxVectoredReadBytes(
                     NonZeroUsize::new(defaults::DEFAULT_MAX_VECTORED_READ_BYTES)
                         .expect("Invalid default constant")
@@ -1577,7 +1557,6 @@ broker_endpoint = '{broker_endpoint}'
                         endpoint: Some(endpoint.clone()),
                         concurrency_limit: s3_concurrency_limit,
                         max_keys_per_list_response: None,
-                        upload_storage_class: None,
                     }),
                     timeout: RemoteStorageConfig::DEFAULT_TIMEOUT,
                 },

pageserver/src/lib.rs

@@ -12,7 +12,6 @@ pub mod disk_usage_eviction_task;
 pub mod http;
 pub mod import_datadir;
 pub use pageserver_api::keyspace;
-pub mod aux_file;
 pub mod metrics;
 pub mod page_cache;
 pub mod page_service;

pageserver/src/metrics.rs

@@ -51,9 +51,6 @@ pub(crate) enum StorageTimeOperation {
     #[strum(serialize = "gc")]
     Gc,
 
-    #[strum(serialize = "update gc info")]
-    UpdateGcInfo,
-
     #[strum(serialize = "create tenant")]
     CreateTenant,
 }
@@ -108,39 +105,31 @@ pub(crate) static VEC_READ_NUM_LAYERS_VISITED: Lazy<Histogram> = Lazy::new(|| {
 });
 
 // Metrics collected on operations on the storage repository.
-#[derive(
-    Clone, Copy, enum_map::Enum, strum_macros::EnumString, strum_macros::Display, IntoStaticStr,
-)]
-pub(crate) enum GetKind {
-    Singular,
-    Vectored,
-}
 
 pub(crate) struct ReconstructTimeMetrics {
-    singular: Histogram,
-    vectored: Histogram,
+    ok: Histogram,
+    err: Histogram,
 }
 
 pub(crate) static RECONSTRUCT_TIME: Lazy<ReconstructTimeMetrics> = Lazy::new(|| {
     let inner = register_histogram_vec!(
         "pageserver_getpage_reconstruct_seconds",
         "Time spent in reconstruct_value (reconstruct a page from deltas)",
-        &["get_kind"],
+        &["result"],
         CRITICAL_OP_BUCKETS.into(),
     )
     .expect("failed to define a metric");
-
     ReconstructTimeMetrics {
-        singular: inner.with_label_values(&[GetKind::Singular.into()]),
-        vectored: inner.with_label_values(&[GetKind::Vectored.into()]),
+        ok: inner.get_metric_with_label_values(&["ok"]).unwrap(),
+        err: inner.get_metric_with_label_values(&["err"]).unwrap(),
     }
 });
 
 impl ReconstructTimeMetrics {
-    pub(crate) fn for_get_kind(&self, get_kind: GetKind) -> &Histogram {
-        match get_kind {
-            GetKind::Singular => &self.singular,
-            GetKind::Vectored => &self.vectored,
+    pub(crate) fn for_result<T, E>(&self, result: &Result<T, E>) -> &Histogram {
+        match result {
+            Ok(_) => &self.ok,
+            Err(_) => &self.err,
         }
     }
 }
@@ -153,33 +142,13 @@ pub(crate) static MATERIALIZED_PAGE_CACHE_HIT_DIRECT: Lazy<IntCounter> = Lazy::n
     .expect("failed to define a metric")
 });
 
-pub(crate) struct ReconstructDataTimeMetrics {
-    singular: Histogram,
-    vectored: Histogram,
-}
-
-impl ReconstructDataTimeMetrics {
-    pub(crate) fn for_get_kind(&self, get_kind: GetKind) -> &Histogram {
-        match get_kind {
-            GetKind::Singular => &self.singular,
-            GetKind::Vectored => &self.vectored,
-        }
-    }
-}
-
-pub(crate) static GET_RECONSTRUCT_DATA_TIME: Lazy<ReconstructDataTimeMetrics> = Lazy::new(|| {
-    let inner = register_histogram_vec!(
+pub(crate) static GET_RECONSTRUCT_DATA_TIME: Lazy<Histogram> = Lazy::new(|| {
+    register_histogram!(
         "pageserver_getpage_get_reconstruct_data_seconds",
         "Time spent in get_reconstruct_value_data",
-        &["get_kind"],
         CRITICAL_OP_BUCKETS.into(),
     )
-    .expect("failed to define a metric");
-
-    ReconstructDataTimeMetrics {
-        singular: inner.with_label_values(&[GetKind::Singular.into()]),
-        vectored: inner.with_label_values(&[GetKind::Vectored.into()]),
-    }
+    .expect("failed to define a metric")
 });
 
 pub(crate) static MATERIALIZED_PAGE_CACHE_HIT: Lazy<IntCounter> = Lazy::new(|| {
@@ -1522,6 +1491,35 @@ pub(crate) static DELETION_QUEUE: Lazy<DeletionQueueMetrics> = Lazy::new(|| {
 }
 });
 
+pub(crate) struct WalIngestMetrics {
+    pub(crate) bytes_received: IntCounter,
+    pub(crate) records_received: IntCounter,
+    pub(crate) records_committed: IntCounter,
+    pub(crate) records_filtered: IntCounter,
+}
+
+pub(crate) static WAL_INGEST: Lazy<WalIngestMetrics> = Lazy::new(|| WalIngestMetrics {
+    bytes_received: register_int_counter!(
+        "pageserver_wal_ingest_bytes_received",
+        "Bytes of WAL ingested from safekeepers",
+    )
+    .unwrap(),
+    records_received: register_int_counter!(
+        "pageserver_wal_ingest_records_received",
+        "Number of WAL records received from safekeepers"
+    )
+    .expect("failed to define a metric"),
+    records_committed: register_int_counter!(
+        "pageserver_wal_ingest_records_committed",
+        "Number of WAL records which resulted in writes to pageserver storage"
+    )
+    .expect("failed to define a metric"),
+    records_filtered: register_int_counter!(
+        "pageserver_wal_ingest_records_filtered",
+        "Number of WAL records filtered out due to sharding"
+    )
+    .expect("failed to define a metric"),
+});
 pub(crate) struct SecondaryModeMetrics {
     pub(crate) upload_heatmap: IntCounter,
     pub(crate) upload_heatmap_errors: IntCounter,
@@ -1723,43 +1721,6 @@ macro_rules! redo_bytes_histogram_count_buckets {
     };
 }
 
-pub(crate) struct WalIngestMetrics {
-    pub(crate) bytes_received: IntCounter,
-    pub(crate) records_received: IntCounter,
-    pub(crate) records_committed: IntCounter,
-    pub(crate) records_filtered: IntCounter,
-    pub(crate) time_spent_on_ingest: Histogram,
-}
-
-pub(crate) static WAL_INGEST: Lazy<WalIngestMetrics> = Lazy::new(|| WalIngestMetrics {
-    bytes_received: register_int_counter!(
-        "pageserver_wal_ingest_bytes_received",
-        "Bytes of WAL ingested from safekeepers",
-    )
-    .unwrap(),
-    records_received: register_int_counter!(
-        "pageserver_wal_ingest_records_received",
-        "Number of WAL records received from safekeepers"
-    )
-    .expect("failed to define a metric"),
-    records_committed: register_int_counter!(
-        "pageserver_wal_ingest_records_committed",
-        "Number of WAL records which resulted in writes to pageserver storage"
-    )
-    .expect("failed to define a metric"),
-    records_filtered: register_int_counter!(
-        "pageserver_wal_ingest_records_filtered",
-        "Number of WAL records filtered out due to sharding"
-    )
-    .expect("failed to define a metric"),
-    time_spent_on_ingest: register_histogram!(
-        "pageserver_wal_ingest_put_value_seconds",
-        "Actual time spent on ingesting a record",
-        redo_histogram_time_buckets!(),
-    )
-    .expect("failed to define a metric"),
-});
-
 pub(crate) static WAL_REDO_TIME: Lazy<Histogram> = Lazy::new(|| {
     register_histogram!(
         "pageserver_wal_redo_seconds",
@@ -1913,22 +1874,6 @@ impl StorageTimeMetricsTimer {
         self.metrics.timeline_count.inc();
         self.metrics.global_histogram.observe(duration);
     }
-
-    /// Turns this timer into a timer, which will always record -- usually this means recording
-    /// regardless an early `?` path was taken in a function.
-    pub(crate) fn record_on_drop(self) -> AlwaysRecordingStorageTimeMetricsTimer {
-        AlwaysRecordingStorageTimeMetricsTimer(Some(self))
-    }
-}
-
-pub(crate) struct AlwaysRecordingStorageTimeMetricsTimer(Option<StorageTimeMetricsTimer>);
-
-impl Drop for AlwaysRecordingStorageTimeMetricsTimer {
-    fn drop(&mut self) {
-        if let Some(inner) = self.0.take() {
-            inner.stop_and_record();
-        }
-    }
 }
 
 /// Timing facilities for an globally histogrammed metric, which is supported by per tenant and
@@ -1989,7 +1934,6 @@ pub(crate) struct TimelineMetrics {
     pub imitate_logical_size_histo: StorageTimeMetrics,
     pub load_layer_map_histo: StorageTimeMetrics,
     pub garbage_collect_histo: StorageTimeMetrics,
-    pub update_gc_info_histo: StorageTimeMetrics,
     pub last_record_gauge: IntGauge,
     resident_physical_size_gauge: UIntGauge,
     /// copy of LayeredTimeline.current_logical_size
@@ -2050,12 +1994,6 @@ impl TimelineMetrics {
             &shard_id,
             &timeline_id,
         );
-        let update_gc_info_histo = StorageTimeMetrics::new(
-            StorageTimeOperation::UpdateGcInfo,
-            &tenant_id,
-            &shard_id,
-            &timeline_id,
-        );
         let last_record_gauge = LAST_RECORD_LSN
             .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
             .unwrap();
@@ -2098,7 +2036,6 @@ impl TimelineMetrics {
             logical_size_histo,
             imitate_logical_size_histo,
             garbage_collect_histo,
-            update_gc_info_histo,
             load_layer_map_histo,
             last_record_gauge,
             resident_physical_size_gauge,

pageserver/src/page_service.rs

@@ -1,5 +1,13 @@
+//
 //! The Page Service listens for client connections and serves their GetPage@LSN
 //! requests.
+//
+//   It is possible to connect here using usual psql/pgbench/libpq. Following
+// commands are supported now:
+//     *status* -- show actual info about this pageserver,
+//     *pagestream* -- enter mode where smgr and pageserver talk with their
+//  custom protocol.
+//
 
 use anyhow::Context;
 use async_compression::tokio::write::GzipEncoder;
@@ -15,7 +23,7 @@ use pageserver_api::models::{
     PagestreamErrorResponse, PagestreamExistsRequest, PagestreamExistsResponse,
     PagestreamFeMessage, PagestreamGetPageRequest, PagestreamGetPageResponse,
     PagestreamGetSlruSegmentRequest, PagestreamGetSlruSegmentResponse, PagestreamNblocksRequest,
-    PagestreamNblocksResponse, PagestreamProtocolVersion,
+    PagestreamNblocksResponse,
 };
 use pageserver_api::shard::ShardIndex;
 use pageserver_api::shard::ShardNumber;
@@ -543,7 +551,6 @@ impl PageServerHandler {
         pgb: &mut PostgresBackend<IO>,
         tenant_id: TenantId,
         timeline_id: TimelineId,
-        protocol_version: PagestreamProtocolVersion,
         ctx: RequestContext,
     ) -> Result<(), QueryError>
     where
@@ -606,15 +613,14 @@ impl PageServerHandler {
                 t.trace(&copy_data_bytes)
             }
 
-            let neon_fe_msg =
-                PagestreamFeMessage::parse(&mut copy_data_bytes.reader(), protocol_version)?;
+            let neon_fe_msg = PagestreamFeMessage::parse(&mut copy_data_bytes.reader())?;
 
             // TODO: We could create a new per-request context here, with unique ID.
             // Currently we use the same per-timeline context for all requests
 
             let (response, span) = match neon_fe_msg {
                 PagestreamFeMessage::Exists(req) => {
-                    let span = tracing::info_span!("handle_get_rel_exists_request", rel = %req.rel, req_lsn = %req.request_lsn);
+                    let span = tracing::info_span!("handle_get_rel_exists_request", rel = %req.rel, req_lsn = %req.lsn);
                     (
                         self.handle_get_rel_exists_request(tenant_id, timeline_id, &req, &ctx)
                             .instrument(span.clone())
@@ -623,7 +629,7 @@ impl PageServerHandler {
                     )
                 }
                 PagestreamFeMessage::Nblocks(req) => {
-                    let span = tracing::info_span!("handle_get_nblocks_request", rel = %req.rel, req_lsn = %req.request_lsn);
+                    let span = tracing::info_span!("handle_get_nblocks_request", rel = %req.rel, req_lsn = %req.lsn);
                     (
                         self.handle_get_nblocks_request(tenant_id, timeline_id, &req, &ctx)
                             .instrument(span.clone())
@@ -633,7 +639,7 @@ impl PageServerHandler {
                 }
                 PagestreamFeMessage::GetPage(req) => {
                     // shard_id is filled in by the handler
-                    let span = tracing::info_span!("handle_get_page_at_lsn_request", rel = %req.rel, blkno = %req.blkno, req_lsn = %req.request_lsn);
+                    let span = tracing::info_span!("handle_get_page_at_lsn_request", rel = %req.rel, blkno = %req.blkno, req_lsn = %req.lsn);
                     (
                         self.handle_get_page_at_lsn_request(tenant_id, timeline_id, &req, &ctx)
                             .instrument(span.clone())
@@ -642,7 +648,7 @@ impl PageServerHandler {
                     )
                 }
                 PagestreamFeMessage::DbSize(req) => {
-                    let span = tracing::info_span!("handle_db_size_request", dbnode = %req.dbnode, req_lsn = %req.request_lsn);
+                    let span = tracing::info_span!("handle_db_size_request", dbnode = %req.dbnode, req_lsn = %req.lsn);
                     (
                         self.handle_db_size_request(tenant_id, timeline_id, &req, &ctx)
                             .instrument(span.clone())
@@ -651,7 +657,7 @@ impl PageServerHandler {
                     )
                 }
                 PagestreamFeMessage::GetSlruSegment(req) => {
-                    let span = tracing::info_span!("handle_get_slru_segment_request", kind = %req.kind, segno = %req.segno, req_lsn = %req.request_lsn);
+                    let span = tracing::info_span!("handle_get_slru_segment_request", kind = %req.kind, segno = %req.segno, req_lsn = %req.lsn);
                     (
                         self.handle_get_slru_segment_request(tenant_id, timeline_id, &req, &ctx)
                             .instrument(span.clone())
@@ -832,80 +838,78 @@ impl PageServerHandler {
     /// Helper function to handle the LSN from client request.
     ///
     /// Each GetPage (and Exists and Nblocks) request includes information about
-    /// which version of the page is being requested. The primary compute node
-    /// will always request the latest page version, by setting 'request_lsn' to
-    /// the last inserted or flushed WAL position, while a standby will request
-    /// a version at the LSN that it's currently caught up to.
+    /// which version of the page is being requested. The client can request the
+    /// latest version of the page, or the version that's valid at a particular
+    /// LSN. The primary compute node will always request the latest page
+    /// version, while a standby will request a version at the LSN that it's
+    /// currently caught up to.
     ///
     /// In either case, if the page server hasn't received the WAL up to the
     /// requested LSN yet, we will wait for it to arrive. The return value is
     /// the LSN that should be used to look up the page versions.
-    ///
-    /// In addition to the request LSN, each request carries another LSN,
-    /// 'not_modified_since', which is a hint to the pageserver that the client
-    /// knows that the page has not been modified between 'not_modified_since'
-    /// and the request LSN. This allows skipping the wait, as long as the WAL
-    /// up to 'not_modified_since' has arrived. If the client doesn't have any
-    /// information about when the page was modified, it will use
-    /// not_modified_since == lsn. If the client lies and sends a too low
-    /// not_modified_hint such that there are in fact later page versions, the
-    /// behavior is undefined: the pageserver may return any of the page versions
-    /// or an error.
     async fn wait_or_get_last_lsn(
         timeline: &Timeline,
-        request_lsn: Lsn,
-        not_modified_since: Lsn,
+        mut lsn: Lsn,
+        latest: bool,
         latest_gc_cutoff_lsn: &RcuReadGuard<Lsn>,
         ctx: &RequestContext,
     ) -> Result<Lsn, PageStreamError> {
-        let last_record_lsn = timeline.get_last_record_lsn();
+        if latest {
+            // Latest page version was requested. If LSN is given, it is a hint
+            // to the page server that there have been no modifications to the
+            // page after that LSN. If we haven't received WAL up to that point,
+            // wait until it arrives.
+            let last_record_lsn = timeline.get_last_record_lsn();
 
-        // Sanity check the request
-        if request_lsn < not_modified_since {
-            return Err(PageStreamError::BadRequest(
-                format!(
-                    "invalid request with request LSN {} and not_modified_since {}",
-                    request_lsn, not_modified_since,
-                )
-                .into(),
-            ));
-        }
-
-        if request_lsn < **latest_gc_cutoff_lsn {
-            // Check explicitly for INVALID just to get a less scary error message if the
-            // request is obviously bogus
-            return Err(if request_lsn == Lsn::INVALID {
-                PageStreamError::BadRequest("invalid LSN(0) in request".into())
+            // Note: this covers the special case that lsn == Lsn(0). That
+            // special case means "return the latest version whatever it is",
+            // and it's used for bootstrapping purposes, when the page server is
+            // connected directly to the compute node. That is needed because
+            // when you connect to the compute node, to receive the WAL, the
+            // walsender process will do a look up in the pg_authid catalog
+            // table for authentication. That poses a deadlock problem: the
+            // catalog table lookup will send a GetPage request, but the GetPage
+            // request will block in the page server because the recent WAL
+            // hasn't been received yet, and it cannot be received until the
+            // walsender completes the authentication and starts streaming the
+            // WAL.
+            if lsn <= last_record_lsn {
+                lsn = last_record_lsn;
             } else {
-                PageStreamError::BadRequest(format!(
-                        "tried to request a page version that was garbage collected. requested at {} gc cutoff {}",
-                        request_lsn, **latest_gc_cutoff_lsn
-                    ).into())
-            });
-        }
-
-        // Wait for WAL up to 'not_modified_since' to arrive, if necessary
-        if not_modified_since > last_record_lsn {
+                timeline
+                    .wait_lsn(
+                        lsn,
+                        crate::tenant::timeline::WaitLsnWaiter::PageService,
+                        ctx,
+                    )
+                    .await?;
+                // Since we waited for 'lsn' to arrive, that is now the last
+                // record LSN. (Or close enough for our purposes; the
+                // last-record LSN can advance immediately after we return
+                // anyway)
+            }
+        } else {
+            if lsn == Lsn(0) {
+                return Err(PageStreamError::BadRequest(
+                    "invalid LSN(0) in request".into(),
+                ));
+            }
             timeline
                 .wait_lsn(
-                    not_modified_since,
+                    lsn,
                     crate::tenant::timeline::WaitLsnWaiter::PageService,
                     ctx,
                 )
                 .await?;
-            // Since we waited for 'not_modified_since' to arrive, that is now the last
-            // record LSN. (Or close enough for our purposes; the last-record LSN can
-            // advance immediately after we return anyway)
-            Ok(not_modified_since)
-        } else {
-            // It might be better to use max(not_modified_since, latest_gc_cutoff_lsn)
-            // here instead. That would give the same result, since we know that there
-            // haven't been any modifications since 'not_modified_since'. Using an older
-            // LSN might be faster, because that could allow skipping recent layers when
-            // finding the page. However, we have historically used 'last_record_lsn', so
-            // stick to that for now.
-            Ok(std::cmp::min(last_record_lsn, request_lsn))
         }
+
+        if lsn < **latest_gc_cutoff_lsn {
+            return Err(PageStreamError::BadRequest(format!(
+                "tried to request a page version that was garbage collected. requested at {} gc cutoff {}",
+                lsn, **latest_gc_cutoff_lsn
+            ).into()));
+        }
+        Ok(lsn)
     }
 
     #[instrument(skip_all, fields(shard_id))]
@@ -922,17 +926,12 @@ impl PageServerHandler {
             .start_timer(metrics::SmgrQueryType::GetRelExists, ctx);
 
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
-        let lsn = Self::wait_or_get_last_lsn(
-            timeline,
-            req.request_lsn,
-            req.not_modified_since,
-            &latest_gc_cutoff_lsn,
-            ctx,
-        )
-        .await?;
+        let lsn =
+            Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
+                .await?;
 
         let exists = timeline
-            .get_rel_exists(req.rel, Version::Lsn(lsn), ctx)
+            .get_rel_exists(req.rel, Version::Lsn(lsn), req.latest, ctx)
             .await?;
 
         Ok(PagestreamBeMessage::Exists(PagestreamExistsResponse {
@@ -955,17 +954,12 @@ impl PageServerHandler {
             .start_timer(metrics::SmgrQueryType::GetRelSize, ctx);
 
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
-        let lsn = Self::wait_or_get_last_lsn(
-            timeline,
-            req.request_lsn,
-            req.not_modified_since,
-            &latest_gc_cutoff_lsn,
-            ctx,
-        )
-        .await?;
+        let lsn =
+            Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
+                .await?;
 
         let n_blocks = timeline
-            .get_rel_size(req.rel, Version::Lsn(lsn), ctx)
+            .get_rel_size(req.rel, Version::Lsn(lsn), req.latest, ctx)
             .await?;
 
         Ok(PagestreamBeMessage::Nblocks(PagestreamNblocksResponse {
@@ -988,17 +982,18 @@ impl PageServerHandler {
             .start_timer(metrics::SmgrQueryType::GetDbSize, ctx);
 
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
-        let lsn = Self::wait_or_get_last_lsn(
-            timeline,
-            req.request_lsn,
-            req.not_modified_since,
-            &latest_gc_cutoff_lsn,
-            ctx,
-        )
-        .await?;
+        let lsn =
+            Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
+                .await?;
 
         let total_blocks = timeline
-            .get_db_size(DEFAULTTABLESPACE_OID, req.dbnode, Version::Lsn(lsn), ctx)
+            .get_db_size(
+                DEFAULTTABLESPACE_OID,
+                req.dbnode,
+                Version::Lsn(lsn),
+                req.latest,
+                ctx,
+            )
             .await?;
         let db_size = total_blocks as i64 * BLCKSZ as i64;
 
@@ -1165,17 +1160,12 @@ impl PageServerHandler {
             .start_timer(metrics::SmgrQueryType::GetPageAtLsn, ctx);
 
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
-        let lsn = Self::wait_or_get_last_lsn(
-            timeline,
-            req.request_lsn,
-            req.not_modified_since,
-            &latest_gc_cutoff_lsn,
-            ctx,
-        )
-        .await?;
+        let lsn =
+            Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
+                .await?;
 
         let page = timeline
-            .get_rel_page_at_lsn(req.rel, req.blkno, Version::Lsn(lsn), ctx)
+            .get_rel_page_at_lsn(req.rel, req.blkno, Version::Lsn(lsn), req.latest, ctx)
             .await?;
 
         Ok(PagestreamBeMessage::GetPage(PagestreamGetPageResponse {
@@ -1198,14 +1188,9 @@ impl PageServerHandler {
             .start_timer(metrics::SmgrQueryType::GetSlruSegment, ctx);
 
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
-        let lsn = Self::wait_or_get_last_lsn(
-            timeline,
-            req.request_lsn,
-            req.not_modified_since,
-            &latest_gc_cutoff_lsn,
-            ctx,
-        )
-        .await?;
+        let lsn =
+            Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
+                .await?;
 
         let kind = SlruKind::from_repr(req.kind)
             .ok_or(PageStreamError::BadRequest("invalid SLRU kind".into()))?;
@@ -1216,10 +1201,6 @@ impl PageServerHandler {
         ))
     }
 
-    /// Note on "fullbackup":
-    /// Full basebackups should only be used for debugging purposes.
-    /// Originally, it was introduced to enable breaking storage format changes,
-    /// but that is not applicable anymore.
     #[allow(clippy::too_many_arguments)]
     #[instrument(skip_all, fields(shard_id, ?lsn, ?prev_lsn, %full_backup))]
     async fn handle_basebackup_request<IO>(
@@ -1423,34 +1404,7 @@ where
 
         let ctx = self.connection_ctx.attached_child();
         debug!("process query {query_string:?}");
-        if query_string.starts_with("pagestream_v2 ") {
-            let (_, params_raw) = query_string.split_at("pagestream_v2 ".len());
-            let params = params_raw.split(' ').collect::<Vec<_>>();
-            if params.len() != 2 {
-                return Err(QueryError::Other(anyhow::anyhow!(
-                    "invalid param number for pagestream command"
-                )));
-            }
-            let tenant_id = TenantId::from_str(params[0])
-                .with_context(|| format!("Failed to parse tenant id from {}", params[0]))?;
-            let timeline_id = TimelineId::from_str(params[1])
-                .with_context(|| format!("Failed to parse timeline id from {}", params[1]))?;
-
-            tracing::Span::current()
-                .record("tenant_id", field::display(tenant_id))
-                .record("timeline_id", field::display(timeline_id));
-
-            self.check_permission(Some(tenant_id))?;
-
-            self.handle_pagerequests(
-                pgb,
-                tenant_id,
-                timeline_id,
-                PagestreamProtocolVersion::V2,
-                ctx,
-            )
-            .await?;
-        } else if query_string.starts_with("pagestream ") {
+        if query_string.starts_with("pagestream ") {
             let (_, params_raw) = query_string.split_at("pagestream ".len());
             let params = params_raw.split(' ').collect::<Vec<_>>();
             if params.len() != 2 {
@@ -1469,14 +1423,8 @@ where
 
             self.check_permission(Some(tenant_id))?;
 
-            self.handle_pagerequests(
-                pgb,
-                tenant_id,
-                timeline_id,
-                PagestreamProtocolVersion::V1,
-                ctx,
-            )
-            .await?;
+            self.handle_pagerequests(pgb, tenant_id, timeline_id, ctx)
+                .await?;
         } else if query_string.starts_with("basebackup ") {
             let (_, params_raw) = query_string.split_at("basebackup ".len());
             let params = params_raw.split_whitespace().collect::<Vec<_>>();

pageserver/src/pgdatadir_mapping.rs

@@ -9,7 +9,6 @@
 use super::tenant::{PageReconstructError, Timeline};
 use crate::context::RequestContext;
 use crate::keyspace::{KeySpace, KeySpaceAccum};
-use crate::metrics::WAL_INGEST;
 use crate::repository::*;
 use crate::span::debug_assert_current_span_has_tenant_and_timeline_id_no_shard_id;
 use crate::walrecord::NeonWalRecord;
@@ -176,6 +175,7 @@ impl Timeline {
         tag: RelTag,
         blknum: BlockNumber,
         version: Version<'_>,
+        latest: bool,
         ctx: &RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         if tag.relnode == 0 {
@@ -184,7 +184,7 @@ impl Timeline {
             ));
         }
 
-        let nblocks = self.get_rel_size(tag, version, ctx).await?;
+        let nblocks = self.get_rel_size(tag, version, latest, ctx).await?;
         if blknum >= nblocks {
             debug!(
                 "read beyond EOF at {} blk {} at {}, size is {}: returning all-zeros page",
@@ -206,6 +206,7 @@ impl Timeline {
         spcnode: Oid,
         dbnode: Oid,
         version: Version<'_>,
+        latest: bool,
         ctx: &RequestContext,
     ) -> Result<usize, PageReconstructError> {
         let mut total_blocks = 0;
@@ -213,7 +214,7 @@ impl Timeline {
         let rels = self.list_rels(spcnode, dbnode, version, ctx).await?;
 
         for rel in rels {
-            let n_blocks = self.get_rel_size(rel, version, ctx).await?;
+            let n_blocks = self.get_rel_size(rel, version, latest, ctx).await?;
             total_blocks += n_blocks as usize;
         }
         Ok(total_blocks)
@@ -224,6 +225,7 @@ impl Timeline {
         &self,
         tag: RelTag,
         version: Version<'_>,
+        latest: bool,
         ctx: &RequestContext,
     ) -> Result<BlockNumber, PageReconstructError> {
         if tag.relnode == 0 {
@@ -237,7 +239,7 @@ impl Timeline {
         }
 
         if (tag.forknum == FSM_FORKNUM || tag.forknum == VISIBILITYMAP_FORKNUM)
-            && !self.get_rel_exists(tag, version, ctx).await?
+            && !self.get_rel_exists(tag, version, latest, ctx).await?
         {
             // FIXME: Postgres sometimes calls smgrcreate() to create
             // FSM, and smgrnblocks() on it immediately afterwards,
@@ -250,8 +252,16 @@ impl Timeline {
         let mut buf = version.get(self, key, ctx).await?;
         let nblocks = buf.get_u32_le();
 
-        self.update_cached_rel_size(tag, version.get_lsn(), nblocks);
-
+        if latest {
+            // Update relation size cache only if "latest" flag is set.
+            // This flag is set by compute when it is working with most recent version of relation.
+            // Typically master compute node always set latest=true.
+            // Please notice, that even if compute node "by mistake" specifies old LSN but set
+            // latest=true, then it can not cause cache corruption, because with latest=true
+            // pageserver choose max(request_lsn, last_written_lsn) and so cached value will be
+            // associated with most recent value of LSN.
+            self.update_cached_rel_size(tag, version.get_lsn(), nblocks);
+        }
         Ok(nblocks)
     }
 
@@ -260,6 +270,7 @@ impl Timeline {
         &self,
         tag: RelTag,
         version: Version<'_>,
+        _latest: bool,
         ctx: &RequestContext,
     ) -> Result<bool, PageReconstructError> {
         if tag.relnode == 0 {
@@ -455,12 +466,6 @@ impl Timeline {
                 // Didn't find any commit timestamps smaller than the request
                 Ok(LsnForTimestamp::Past(min_lsn))
             }
-            (true, _) if commit_lsn < min_lsn => {
-                // the search above did set found_smaller to true but it never increased the lsn.
-                // Then, low is still the old min_lsn, and the subtraction above gave a value
-                // below the min_lsn. We should never do that.
-                Ok(LsnForTimestamp::Past(min_lsn))
-            }
             (true, false) => {
                 // Only found commits with timestamps smaller than the request.
                 // It's still a valid case for branch creation, return it.
@@ -812,7 +817,7 @@ impl Timeline {
     /// Get cached size of relation if it not updated after specified LSN
     pub fn get_cached_rel_size(&self, tag: &RelTag, lsn: Lsn) -> Option<BlockNumber> {
         let rel_size_cache = self.rel_size_cache.read().unwrap();
-        if let Some((cached_lsn, nblocks)) = rel_size_cache.map.get(tag) {
+        if let Some((cached_lsn, nblocks)) = rel_size_cache.get(tag) {
             if lsn >= *cached_lsn {
                 return Some(*nblocks);
             }
@@ -823,16 +828,7 @@ impl Timeline {
     /// Update cached relation size if there is no more recent update
     pub fn update_cached_rel_size(&self, tag: RelTag, lsn: Lsn, nblocks: BlockNumber) {
         let mut rel_size_cache = self.rel_size_cache.write().unwrap();
-
-        if lsn < rel_size_cache.complete_as_of {
-            // Do not cache old values. It's safe to cache the size on read, as long as
-            // the read was at an LSN since we started the WAL ingestion. Reasoning: we
-            // never evict values from the cache, so if the relation size changed after
-            // 'lsn', the new value is already in the cache.
-            return;
-        }
-
-        match rel_size_cache.map.entry(tag) {
+        match rel_size_cache.entry(tag) {
             hash_map::Entry::Occupied(mut entry) => {
                 let cached_lsn = entry.get_mut();
                 if lsn >= cached_lsn.0 {
@@ -848,13 +844,13 @@ impl Timeline {
     /// Store cached relation size
     pub fn set_cached_rel_size(&self, tag: RelTag, lsn: Lsn, nblocks: BlockNumber) {
         let mut rel_size_cache = self.rel_size_cache.write().unwrap();
-        rel_size_cache.map.insert(tag, (lsn, nblocks));
+        rel_size_cache.insert(tag, (lsn, nblocks));
     }
 
     /// Remove cached relation size
     pub fn remove_cached_rel_size(&self, tag: &RelTag) {
         let mut rel_size_cache = self.rel_size_cache.write().unwrap();
-        rel_size_cache.map.remove(tag);
+        rel_size_cache.remove(tag);
     }
 }
 
@@ -1092,7 +1088,7 @@ impl<'a> DatadirModification<'a> {
     ) -> anyhow::Result<()> {
         let total_blocks = self
             .tline
-            .get_db_size(spcnode, dbnode, Version::Modified(self), ctx)
+            .get_db_size(spcnode, dbnode, Version::Modified(self), true, ctx)
             .await?;
 
         // Remove entry from dbdir
@@ -1191,7 +1187,7 @@ impl<'a> DatadirModification<'a> {
         anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
         if self
             .tline
-            .get_rel_exists(rel, Version::Modified(self), ctx)
+            .get_rel_exists(rel, Version::Modified(self), true, ctx)
             .await?
         {
             let size_key = rel_size_to_key(rel);
@@ -1405,7 +1401,7 @@ impl<'a> DatadirModification<'a> {
         let n_files;
         let mut aux_files = self.tline.aux_files.lock().await;
         if let Some(mut dir) = aux_files.dir.take() {
-            // We already updated aux files in `self`: emit a delta and update our latest value.
+            // We already updated aux files in `self`: emit a delta and update our latest value
             dir.upsert(file_path.clone(), content.clone());
             n_files = dir.files.len();
             if aux_files.n_deltas == MAX_AUX_FILE_DELTAS {
@@ -1549,8 +1545,6 @@ impl<'a> DatadirModification<'a> {
     pub async fn commit(&mut self, ctx: &RequestContext) -> anyhow::Result<()> {
         let mut writer = self.tline.writer().await;
 
-        let timer = WAL_INGEST.time_spent_on_ingest.start_timer();
-
         let pending_nblocks = self.pending_nblocks;
         self.pending_nblocks = 0;
 
@@ -1590,8 +1584,6 @@ impl<'a> DatadirModification<'a> {
             writer.update_directory_entries_count(kind, count as u64);
         }
 
-        timer.observe_duration();
-
         Ok(())
     }
 

pageserver/src/task_mgr.rs

@@ -361,8 +361,6 @@ pub enum TaskKind {
 
     DebugTool,
 
-    EphemeralFilePreWarmPageCache,
-
     #[cfg(test)]
     UnitTest,
 }

pageserver/src/tenant.rs

@@ -2870,23 +2870,20 @@ impl Tenant {
                 }
             }
 
-            let cutoff = timeline
-                .get_last_record_lsn()
-                .checked_sub(horizon)
-                .unwrap_or(Lsn(0));
+            if let Some(cutoff) = timeline.get_last_record_lsn().checked_sub(horizon) {
+                let branchpoints: Vec<Lsn> = all_branchpoints
+                    .range((
+                        Included((timeline_id, Lsn(0))),
+                        Included((timeline_id, Lsn(u64::MAX))),
+                    ))
+                    .map(|&x| x.1)
+                    .collect();
+                timeline
+                    .update_gc_info(branchpoints, cutoff, pitr, cancel, ctx)
+                    .await?;
 
-            let branchpoints: Vec<Lsn> = all_branchpoints
-                .range((
-                    Included((timeline_id, Lsn(0))),
-                    Included((timeline_id, Lsn(u64::MAX))),
-                ))
-                .map(|&x| x.1)
-                .collect();
-            timeline
-                .update_gc_info(branchpoints, cutoff, pitr, cancel, ctx)
-                .await?;
-
-            gc_timelines.push(timeline);
+                gc_timelines.push(timeline);
+            }
         }
         drop(gc_cs);
         Ok(gc_timelines)
@@ -3402,11 +3399,7 @@ impl Tenant {
         // is in progress (which is not a common case).
         //
         // See more for on the issue #2748 condenced out of the initial PR review.
-        let mut shared_cache = tokio::select! {
-            locked = self.cached_logical_sizes.lock() => locked,
-            _ = cancel.cancelled() => anyhow::bail!("cancelled"),
-            _ = self.cancel.cancelled() => anyhow::bail!("tenant is shutting down"),
-        };
+        let mut shared_cache = self.cached_logical_sizes.lock().await;
 
         size::gather_inputs(
             self,
@@ -3668,7 +3661,6 @@ pub(crate) mod harness {
                 image_layer_creation_check_threshold: Some(
                     tenant_conf.image_layer_creation_check_threshold,
                 ),
-                switch_to_aux_file_v2: Some(tenant_conf.switch_to_aux_file_v2),
             }
         }
     }
@@ -3867,10 +3859,8 @@ mod tests {
     use crate::DEFAULT_PG_VERSION;
     use bytes::BytesMut;
     use hex_literal::hex;
-    use pageserver_api::key::NON_INHERITED_RANGE;
     use pageserver_api::keyspace::KeySpace;
     use rand::{thread_rng, Rng};
-    use tests::storage_layer::ValuesReconstructState;
     use tests::timeline::{GetVectoredError, ShutdownMode};
 
     static TEST_KEY: Lazy<Key> =
@@ -4659,9 +4649,7 @@ mod tests {
         for read in reads {
             info!("Doing vectored read on {:?}", read);
 
-            let vectored_res = tline
-                .get_vectored_impl(read.clone(), reads_lsn, ValuesReconstructState::new(), &ctx)
-                .await;
+            let vectored_res = tline.get_vectored_impl(read.clone(), reads_lsn, &ctx).await;
             tline
                 .validate_get_vectored_impl(&vectored_res, read, reads_lsn, &ctx)
                 .await;
@@ -4670,67 +4658,6 @@ mod tests {
         Ok(())
     }
 
-    #[tokio::test]
-    async fn test_get_vectored_aux_files() -> anyhow::Result<()> {
-        let harness = TenantHarness::create("test_get_vectored_aux_files")?;
-
-        let (tenant, ctx) = harness.load().await;
-        let tline = tenant
-            .create_empty_timeline(TIMELINE_ID, Lsn(0), DEFAULT_PG_VERSION, &ctx)
-            .await?;
-        let tline = tline.raw_timeline().unwrap();
-
-        let mut modification = tline.begin_modification(Lsn(0x1000));
-        modification.put_file("foo/bar1", b"content1", &ctx).await?;
-        modification.set_lsn(Lsn(0x1008))?;
-        modification.put_file("foo/bar2", b"content2", &ctx).await?;
-        modification.commit(&ctx).await?;
-
-        let child_timeline_id = TimelineId::generate();
-        tenant
-            .branch_timeline_test(
-                tline,
-                child_timeline_id,
-                Some(tline.get_last_record_lsn()),
-                &ctx,
-            )
-            .await?;
-
-        let child_timeline = tenant
-            .get_timeline(child_timeline_id, true)
-            .expect("Should have the branched timeline");
-
-        let aux_keyspace = KeySpace {
-            ranges: vec![NON_INHERITED_RANGE],
-        };
-        let read_lsn = child_timeline.get_last_record_lsn();
-
-        let vectored_res = child_timeline
-            .get_vectored_impl(
-                aux_keyspace.clone(),
-                read_lsn,
-                ValuesReconstructState::new(),
-                &ctx,
-            )
-            .await;
-
-        child_timeline
-            .validate_get_vectored_impl(&vectored_res, aux_keyspace, read_lsn, &ctx)
-            .await;
-
-        let images = vectored_res?;
-        let mut key = NON_INHERITED_RANGE.start;
-        while key < NON_INHERITED_RANGE.end {
-            assert!(matches!(
-                images[&key],
-                Err(PageReconstructError::MissingKey(_))
-            ));
-            key = key.next();
-        }
-
-        Ok(())
-    }
-
     // Test that vectored get handles layer gaps correctly
     // by advancing into the next ancestor timeline if required.
     //
@@ -4859,12 +4786,7 @@ mod tests {
             ranges: vec![key_near_gap..gap_at_key.next(), key_near_end..current_key],
         };
         let results = child_timeline
-            .get_vectored_impl(
-                read.clone(),
-                current_lsn,
-                ValuesReconstructState::new(),
-                &ctx,
-            )
+            .get_vectored_impl(read.clone(), current_lsn, &ctx)
             .await?;
 
         for (key, img_res) in results {
@@ -4997,7 +4919,6 @@ mod tests {
                         ranges: vec![child_gap_at_key..child_gap_at_key.next()],
                     },
                     query_lsn,
-                    ValuesReconstructState::new(),
                     &ctx,
                 )
                 .await;

pageserver/src/tenant/blob_io.rs

@@ -121,7 +121,7 @@ impl<const BUFFERED: bool> BlobWriter<BUFFERED> {
         self.offset
     }
 
-    const CAPACITY: usize = if BUFFERED { 64 * 1024 } else { 0 };
+    const CAPACITY: usize = if BUFFERED { PAGE_SZ } else { 0 };
 
     /// Writes the given buffer directly to the underlying `VirtualFile`.
     /// You need to make sure that the internal buffer is empty, otherwise

pageserver/src/tenant/config.rs

@@ -369,10 +369,6 @@ pub struct TenantConf {
     // How much WAL must be ingested before checking again whether a new image layer is required.
     // Expresed in multiples of checkpoint distance.
     pub image_layer_creation_check_threshold: u8,
-
-    /// Switch to aux file v2. Switching this flag requires the user has not written any aux file into
-    /// the storage before, and this flag cannot be switched back. Otherwise there will be data corruptions.
-    pub switch_to_aux_file_v2: bool,
 }
 
 /// Same as TenantConf, but this struct preserves the information about
@@ -468,10 +464,6 @@ pub struct TenantConfOpt {
 
     #[serde(skip_serializing_if = "Option::is_none")]
     pub image_layer_creation_check_threshold: Option<u8>,
-
-    #[serde(skip_serializing_if = "Option::is_none")]
-    #[serde(default)]
-    pub switch_to_aux_file_v2: Option<bool>,
 }
 
 impl TenantConfOpt {
@@ -529,9 +521,6 @@ impl TenantConfOpt {
             image_layer_creation_check_threshold: self
                 .image_layer_creation_check_threshold
                 .unwrap_or(global_conf.image_layer_creation_check_threshold),
-            switch_to_aux_file_v2: self
-                .switch_to_aux_file_v2
-                .unwrap_or(global_conf.switch_to_aux_file_v2),
         }
     }
 }
@@ -573,7 +562,6 @@ impl Default for TenantConf {
             lazy_slru_download: false,
             timeline_get_throttle: crate::tenant::throttle::Config::disabled(),
             image_layer_creation_check_threshold: DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD,
-            switch_to_aux_file_v2: false,
         }
     }
 }
@@ -648,7 +636,6 @@ impl From<TenantConfOpt> for models::TenantConfig {
             lazy_slru_download: value.lazy_slru_download,
             timeline_get_throttle: value.timeline_get_throttle.map(ThrottleConfig::from),
             image_layer_creation_check_threshold: value.image_layer_creation_check_threshold,
-            switch_to_aux_file_v2: value.switch_to_aux_file_v2,
         }
     }
 }

pageserver/src/tenant/ephemeral_file.rs

@@ -3,26 +3,36 @@
 
 use crate::config::PageServerConf;
 use crate::context::RequestContext;
-use crate::page_cache;
+use crate::page_cache::{self, PAGE_SZ};
 use crate::tenant::block_io::{BlockCursor, BlockLease, BlockReader};
 use crate::virtual_file::{self, VirtualFile};
+use bytes::BytesMut;
 use camino::Utf8PathBuf;
 use pageserver_api::shard::TenantShardId;
+use std::cmp::min;
 
-use std::io;
+use std::io::{self, ErrorKind};
+use std::ops::DerefMut;
 use std::sync::atomic::AtomicU64;
+use tracing::*;
 use utils::id::TimelineId;
 
 pub struct EphemeralFile {
+    page_cache_file_id: page_cache::FileId,
+
     _tenant_shard_id: TenantShardId,
     _timeline_id: TimelineId,
-
-    rw: page_caching::RW,
+    file: VirtualFile,
+    len: u64,
+    /// An ephemeral file is append-only.
+    /// We keep the last page, which can still be modified, in [`Self::mutable_tail`].
+    /// The other pages, which can no longer be modified, are accessed through the page cache.
+    ///
+    /// None <=> IO is ongoing.
+    /// Size is fixed to PAGE_SZ at creation time and must not be changed.
+    mutable_tail: Option<BytesMut>,
 }
 
-mod page_caching;
-mod zero_padded_read_write;
-
 impl EphemeralFile {
     pub async fn create(
         conf: &PageServerConf,
@@ -49,18 +59,21 @@ impl EphemeralFile {
         .await?;
 
         Ok(EphemeralFile {
+            page_cache_file_id: page_cache::next_file_id(),
             _tenant_shard_id: tenant_shard_id,
             _timeline_id: timeline_id,
-            rw: page_caching::RW::new(file),
+            file,
+            len: 0,
+            mutable_tail: Some(BytesMut::zeroed(PAGE_SZ)),
         })
     }
 
     pub(crate) fn len(&self) -> u64 {
-        self.rw.bytes_written()
+        self.len
     }
 
-    pub(crate) fn page_cache_file_id(&self) -> page_cache::FileId {
-        self.rw.page_cache_file_id()
+    pub(crate) fn id(&self) -> page_cache::FileId {
+        self.page_cache_file_id
     }
 
     pub(crate) async fn read_blk(
@@ -68,30 +81,182 @@ impl EphemeralFile {
         blknum: u32,
         ctx: &RequestContext,
     ) -> Result<BlockLease, io::Error> {
-        self.rw.read_blk(blknum, ctx).await
+        let flushed_blknums = 0..self.len / PAGE_SZ as u64;
+        if flushed_blknums.contains(&(blknum as u64)) {
+            let cache = page_cache::get();
+            match cache
+                .read_immutable_buf(self.page_cache_file_id, blknum, ctx)
+                .await
+                .map_err(|e| {
+                    std::io::Error::new(
+                        std::io::ErrorKind::Other,
+                        // order path before error because error is anyhow::Error => might have many contexts
+                        format!(
+                            "ephemeral file: read immutable page #{}: {}: {:#}",
+                            blknum, self.file.path, e,
+                        ),
+                    )
+                })? {
+                page_cache::ReadBufResult::Found(guard) => {
+                    return Ok(BlockLease::PageReadGuard(guard))
+                }
+                page_cache::ReadBufResult::NotFound(write_guard) => {
+                    let write_guard = self
+                        .file
+                        .read_exact_at_page(write_guard, blknum as u64 * PAGE_SZ as u64)
+                        .await?;
+                    let read_guard = write_guard.mark_valid();
+                    return Ok(BlockLease::PageReadGuard(read_guard));
+                }
+            };
+        } else {
+            debug_assert_eq!(blknum as u64, self.len / PAGE_SZ as u64);
+            Ok(BlockLease::EphemeralFileMutableTail(
+                self.mutable_tail
+                    .as_deref()
+                    .expect("we're not doing IO, it must be Some()")
+                    .try_into()
+                    .expect("we ensure that it's always PAGE_SZ"),
+            ))
+        }
     }
 
     pub(crate) async fn write_blob(
         &mut self,
         srcbuf: &[u8],
-        _ctx: &RequestContext,
+        ctx: &RequestContext,
     ) -> Result<u64, io::Error> {
-        let pos = self.rw.bytes_written();
+        struct Writer<'a> {
+            ephemeral_file: &'a mut EphemeralFile,
+            /// The block to which the next [`push_bytes`] will write.
+            blknum: u32,
+            /// The offset inside the block identified by [`blknum`] to which [`push_bytes`] will write.
+            off: usize,
+        }
+        impl<'a> Writer<'a> {
+            fn new(ephemeral_file: &'a mut EphemeralFile) -> io::Result<Writer<'a>> {
+                Ok(Writer {
+                    blknum: (ephemeral_file.len / PAGE_SZ as u64) as u32,
+                    off: (ephemeral_file.len % PAGE_SZ as u64) as usize,
+                    ephemeral_file,
+                })
+            }
+            #[inline(always)]
+            async fn push_bytes(
+                &mut self,
+                src: &[u8],
+                ctx: &RequestContext,
+            ) -> Result<(), io::Error> {
+                let mut src_remaining = src;
+                while !src_remaining.is_empty() {
+                    let dst_remaining = &mut self
+                        .ephemeral_file
+                        .mutable_tail
+                        .as_deref_mut()
+                        .expect("IO is not yet ongoing")[self.off..];
+                    let n = min(dst_remaining.len(), src_remaining.len());
+                    dst_remaining[..n].copy_from_slice(&src_remaining[..n]);
+                    self.off += n;
+                    src_remaining = &src_remaining[n..];
+                    if self.off == PAGE_SZ {
+                        let mutable_tail = std::mem::take(&mut self.ephemeral_file.mutable_tail)
+                            .expect("IO is not yet ongoing");
+                        let (mutable_tail, res) = self
+                            .ephemeral_file
+                            .file
+                            .write_all_at(mutable_tail, self.blknum as u64 * PAGE_SZ as u64)
+                            .await;
+                        // TODO: If we panic before we can put the mutable_tail back, subsequent calls will fail.
+                        // I.e., the IO isn't retryable if we panic.
+                        self.ephemeral_file.mutable_tail = Some(mutable_tail);
+                        match res {
+                            Ok(_) => {
+                                // Pre-warm the page cache with what we just wrote.
+                                // This isn't necessary for coherency/correctness, but it's how we've always done it.
+                                let cache = page_cache::get();
+                                match cache
+                                    .read_immutable_buf(
+                                        self.ephemeral_file.page_cache_file_id,
+                                        self.blknum,
+                                        ctx,
+                                    )
+                                    .await
+                                {
+                                    Ok(page_cache::ReadBufResult::Found(_guard)) => {
+                                        // This function takes &mut self, so, it shouldn't be possible to reach this point.
+                                        unreachable!("we just wrote blknum {} and this function takes &mut self, so, no concurrent read_blk is possible", self.blknum);
+                                    }
+                                    Ok(page_cache::ReadBufResult::NotFound(mut write_guard)) => {
+                                        let buf: &mut [u8] = write_guard.deref_mut();
+                                        debug_assert_eq!(buf.len(), PAGE_SZ);
+                                        buf.copy_from_slice(
+                                            self.ephemeral_file
+                                                .mutable_tail
+                                                .as_deref()
+                                                .expect("IO is not ongoing"),
+                                        );
+                                        let _ = write_guard.mark_valid();
+                                        // pre-warm successful
+                                    }
+                                    Err(e) => {
+                                        error!("ephemeral_file write_blob failed to get immutable buf to pre-warm page cache: {e:?}");
+                                        // fail gracefully, it's not the end of the world if we can't pre-warm the cache here
+                                    }
+                                }
+                                // Zero the buffer for re-use.
+                                // Zeroing is critical for correcntess because the write_blob code below
+                                // and similarly read_blk expect zeroed pages.
+                                self.ephemeral_file
+                                    .mutable_tail
+                                    .as_deref_mut()
+                                    .expect("IO is not ongoing")
+                                    .fill(0);
+                                // This block is done, move to next one.
+                                self.blknum += 1;
+                                self.off = 0;
+                            }
+                            Err(e) => {
+                                return Err(std::io::Error::new(
+                                    ErrorKind::Other,
+                                    // order error before path because path is long and error is short
+                                    format!(
+                                        "ephemeral_file: write_blob: write-back full tail blk #{}: {:#}: {}",
+                                        self.blknum,
+                                        e,
+                                        self.ephemeral_file.file.path,
+                                    ),
+                                ));
+                            }
+                        }
+                    }
+                }
+                Ok(())
+            }
+        }
+
+        let pos = self.len;
+        let mut writer = Writer::new(self)?;
 
         // Write the length field
         if srcbuf.len() < 0x80 {
             // short one-byte length header
             let len_buf = [srcbuf.len() as u8];
-
-            self.rw.write_all_borrowed(&len_buf).await?;
+            writer.push_bytes(&len_buf, ctx).await?;
         } else {
             let mut len_buf = u32::to_be_bytes(srcbuf.len() as u32);
             len_buf[0] |= 0x80;
-            self.rw.write_all_borrowed(&len_buf).await?;
+            writer.push_bytes(&len_buf, ctx).await?;
         }
 
         // Write the payload
-        self.rw.write_all_borrowed(srcbuf).await?;
+        writer.push_bytes(srcbuf, ctx).await?;
+
+        if srcbuf.len() < 0x80 {
+            self.len += 1;
+        } else {
+            self.len += 4;
+        }
+        self.len += srcbuf.len() as u64;
 
         Ok(pos)
     }
@@ -106,6 +271,28 @@ pub fn is_ephemeral_file(filename: &str) -> bool {
     }
 }
 
+impl Drop for EphemeralFile {
+    fn drop(&mut self) {
+        // There might still be pages in the [`crate::page_cache`] for this file.
+        // We leave them there, [`crate::page_cache::PageCache::find_victim`] will evict them when needed.
+
+        // unlink the file
+        let res = std::fs::remove_file(&self.file.path);
+        if let Err(e) = res {
+            if e.kind() != std::io::ErrorKind::NotFound {
+                // just never log the not found errors, we cannot do anything for them; on detach
+                // the tenant directory is already gone.
+                //
+                // not found files might also be related to https://github.com/neondatabase/neon/issues/2442
+                error!(
+                    "could not remove ephemeral file '{}': {}",
+                    self.file.path, e
+                );
+            }
+        }
+    }
+}
+
 impl BlockReader for EphemeralFile {
     fn block_cursor(&self) -> super::block_io::BlockCursor<'_> {
         BlockCursor::new(super::block_io::BlockReaderRef::EphemeralFile(self))

pageserver/src/tenant/ephemeral_file/page_caching.rs

@@ -1,218 +0,0 @@
-//! Wrapper around [`super::zero_padded_read_write::RW`] that uses the
-//! [`crate::page_cache`] to serve reads that need to go to the underlying [`VirtualFile`].
-
-use crate::context::RequestContext;
-use crate::page_cache::{self, PAGE_SZ};
-use crate::tenant::block_io::BlockLease;
-use crate::virtual_file::VirtualFile;
-
-use once_cell::sync::Lazy;
-use std::io::{self, ErrorKind};
-use tokio_epoll_uring::BoundedBuf;
-use tracing::*;
-
-use super::zero_padded_read_write;
-
-/// See module-level comment.
-pub struct RW {
-    page_cache_file_id: page_cache::FileId,
-    rw: super::zero_padded_read_write::RW<PreWarmingWriter>,
-}
-
-impl RW {
-    pub fn new(file: VirtualFile) -> Self {
-        let page_cache_file_id = page_cache::next_file_id();
-        Self {
-            page_cache_file_id,
-            rw: super::zero_padded_read_write::RW::new(PreWarmingWriter::new(
-                page_cache_file_id,
-                file,
-            )),
-        }
-    }
-
-    pub fn page_cache_file_id(&self) -> page_cache::FileId {
-        self.page_cache_file_id
-    }
-
-    pub(crate) async fn write_all_borrowed(&mut self, srcbuf: &[u8]) -> Result<usize, io::Error> {
-        // It doesn't make sense to proactively fill the page cache on the Pageserver write path
-        // because Compute is unlikely to access recently written data.
-        self.rw.write_all_borrowed(srcbuf).await
-    }
-
-    pub(crate) fn bytes_written(&self) -> u64 {
-        self.rw.bytes_written()
-    }
-
-    pub(crate) async fn read_blk(
-        &self,
-        blknum: u32,
-        ctx: &RequestContext,
-    ) -> Result<BlockLease, io::Error> {
-        match self.rw.read_blk(blknum).await? {
-            zero_padded_read_write::ReadResult::NeedsReadFromWriter { writer } => {
-                let cache = page_cache::get();
-                match cache
-                    .read_immutable_buf(self.page_cache_file_id, blknum, ctx)
-                    .await
-                    .map_err(|e| {
-                        std::io::Error::new(
-                            std::io::ErrorKind::Other,
-                            // order path before error because error is anyhow::Error => might have many contexts
-                            format!(
-                                "ephemeral file: read immutable page #{}: {}: {:#}",
-                                blknum,
-                                self.rw.as_writer().file.path,
-                                e,
-                            ),
-                        )
-                    })? {
-                    page_cache::ReadBufResult::Found(guard) => {
-                        return Ok(BlockLease::PageReadGuard(guard))
-                    }
-                    page_cache::ReadBufResult::NotFound(write_guard) => {
-                        let write_guard = writer
-                            .file
-                            .read_exact_at_page(write_guard, blknum as u64 * PAGE_SZ as u64)
-                            .await?;
-                        let read_guard = write_guard.mark_valid();
-                        return Ok(BlockLease::PageReadGuard(read_guard));
-                    }
-                }
-            }
-            zero_padded_read_write::ReadResult::ServedFromZeroPaddedMutableTail { buffer } => {
-                Ok(BlockLease::EphemeralFileMutableTail(buffer))
-            }
-        }
-    }
-}
-
-impl Drop for RW {
-    fn drop(&mut self) {
-        // There might still be pages in the [`crate::page_cache`] for this file.
-        // We leave them there, [`crate::page_cache::PageCache::find_victim`] will evict them when needed.
-
-        // unlink the file
-        let res = std::fs::remove_file(&self.rw.as_writer().file.path);
-        if let Err(e) = res {
-            if e.kind() != std::io::ErrorKind::NotFound {
-                // just never log the not found errors, we cannot do anything for them; on detach
-                // the tenant directory is already gone.
-                //
-                // not found files might also be related to https://github.com/neondatabase/neon/issues/2442
-                error!(
-                    "could not remove ephemeral file '{}': {}",
-                    self.rw.as_writer().file.path,
-                    e
-                );
-            }
-        }
-    }
-}
-
-struct PreWarmingWriter {
-    nwritten_blocks: u32,
-    page_cache_file_id: page_cache::FileId,
-    file: VirtualFile,
-}
-
-impl PreWarmingWriter {
-    fn new(page_cache_file_id: page_cache::FileId, file: VirtualFile) -> Self {
-        Self {
-            nwritten_blocks: 0,
-            page_cache_file_id,
-            file,
-        }
-    }
-}
-
-impl crate::virtual_file::owned_buffers_io::write::OwnedAsyncWriter for PreWarmingWriter {
-    async fn write_all<
-        B: tokio_epoll_uring::BoundedBuf<Buf = Buf>,
-        Buf: tokio_epoll_uring::IoBuf + Send,
-    >(
-        &mut self,
-        buf: B,
-    ) -> std::io::Result<(usize, B::Buf)> {
-        let buf = buf.slice(..);
-        let saved_bounds = buf.bounds(); // save for reconstructing the Slice from iobuf after the IO is done
-        let check_bounds_stuff_works = if cfg!(test) && cfg!(debug_assertions) {
-            Some(buf.to_vec())
-        } else {
-            None
-        };
-        let buflen = buf.len();
-        assert_eq!(
-            buflen % PAGE_SZ,
-            0,
-            "{buflen} ; we know TAIL_SZ is a PAGE_SZ multiple, and write_buffered_borrowed is used"
-        );
-
-        // Do the IO.
-        let iobuf = match self.file.write_all(buf).await {
-            (iobuf, Ok(nwritten)) => {
-                assert_eq!(nwritten, buflen);
-                iobuf
-            }
-            (_, Err(e)) => {
-                return Err(std::io::Error::new(
-                    ErrorKind::Other,
-                    // order error before path because path is long and error is short
-                    format!(
-                        "ephemeral_file: write_blob: write-back tail self.nwritten_blocks={}, buflen={}, {:#}: {}",
-                        self.nwritten_blocks, buflen, e, self.file.path,
-                    ),
-                ));
-            }
-        };
-
-        // Reconstruct the Slice (the write path consumed the Slice and returned us the underlying IoBuf)
-        let buf = tokio_epoll_uring::Slice::from_buf_bounds(iobuf, saved_bounds);
-        if let Some(check_bounds_stuff_works) = check_bounds_stuff_works {
-            assert_eq!(&check_bounds_stuff_works, &*buf);
-        }
-
-        // Pre-warm page cache with the contents.
-        // At least in isolated bulk ingest benchmarks (test_bulk_insert.py), the pre-warming
-        // benefits the code that writes InMemoryLayer=>L0 layers.
-        let nblocks = buflen / PAGE_SZ;
-        let nblocks32 = u32::try_from(nblocks).unwrap();
-        let cache = page_cache::get();
-        static CTX: Lazy<RequestContext> = Lazy::new(|| {
-            RequestContext::new(
-                crate::task_mgr::TaskKind::EphemeralFilePreWarmPageCache,
-                crate::context::DownloadBehavior::Error,
-            )
-        });
-        for blknum_in_buffer in 0..nblocks {
-            let blk_in_buffer = &buf[blknum_in_buffer * PAGE_SZ..(blknum_in_buffer + 1) * PAGE_SZ];
-            let blknum = self
-                .nwritten_blocks
-                .checked_add(blknum_in_buffer as u32)
-                .unwrap();
-            match cache
-                .read_immutable_buf(self.page_cache_file_id, blknum, &CTX)
-                .await
-            {
-                Err(e) => {
-                    error!("ephemeral_file write_blob failed to get immutable buf to pre-warm page cache: {e:?}");
-                    // fail gracefully, it's not the end of the world if we can't pre-warm the cache here
-                }
-                Ok(v) => match v {
-                    page_cache::ReadBufResult::Found(_guard) => {
-                        // This function takes &mut self, so, it shouldn't be possible to reach this point.
-                        unreachable!("we just wrote block {blknum} to the VirtualFile, which is owned by Self, \
-                                      and this function takes &mut self, so, no concurrent read_blk is possible");
-                    }
-                    page_cache::ReadBufResult::NotFound(mut write_guard) => {
-                        write_guard.copy_from_slice(blk_in_buffer);
-                        let _ = write_guard.mark_valid();
-                    }
-                },
-            }
-        }
-        self.nwritten_blocks = self.nwritten_blocks.checked_add(nblocks32).unwrap();
-        Ok((buflen, buf.into_inner()))
-    }
-}

pageserver/src/tenant/ephemeral_file/zero_padded_read_write.rs

@@ -1,125 +0,0 @@
-//! The heart of how [`super::EphemeralFile`] does its reads and writes.
-//!
-//! # Writes
-//!
-//! [`super::EphemeralFile`] writes small, borrowed buffers using [`RW::write_all_borrowed`].
-//! The [`RW`] batches these into [`TAIL_SZ`] bigger writes, using [`owned_buffers_io::write::BufferedWriter`].
-//!
-//! # Reads
-//!
-//! [`super::EphemeralFile`] always reads full [`PAGE_SZ`]ed blocks using [`RW::read_blk`].
-//!
-//! The [`RW`] serves these reads either from the buffered writer's in-memory buffer
-//! or redirects the caller to read from the underlying [`OwnedAsyncWriter`]
-//! if the read is for the prefix that has already been flushed.
-//!
-//! # Current Usage
-//!
-//! The current user of this module is [`super::page_caching::RW`].
-
-mod zero_padded;
-
-use crate::{
-    page_cache::PAGE_SZ,
-    virtual_file::owned_buffers_io::{
-        self,
-        write::{Buffer, OwnedAsyncWriter},
-    },
-};
-
-const TAIL_SZ: usize = 64 * 1024;
-
-/// See module-level comment.
-pub struct RW<W: OwnedAsyncWriter> {
-    buffered_writer: owned_buffers_io::write::BufferedWriter<
-        zero_padded::Buffer<TAIL_SZ>,
-        owned_buffers_io::util::size_tracking_writer::Writer<W>,
-    >,
-}
-
-pub enum ReadResult<'a, W> {
-    NeedsReadFromWriter { writer: &'a W },
-    ServedFromZeroPaddedMutableTail { buffer: &'a [u8; PAGE_SZ] },
-}
-
-impl<W> RW<W>
-where
-    W: OwnedAsyncWriter,
-{
-    pub fn new(writer: W) -> Self {
-        let bytes_flushed_tracker =
-            owned_buffers_io::util::size_tracking_writer::Writer::new(writer);
-        let buffered_writer = owned_buffers_io::write::BufferedWriter::new(
-            bytes_flushed_tracker,
-            zero_padded::Buffer::default(),
-        );
-        Self { buffered_writer }
-    }
-
-    pub(crate) fn as_writer(&self) -> &W {
-        self.buffered_writer.as_inner().as_inner()
-    }
-
-    pub async fn write_all_borrowed(&mut self, buf: &[u8]) -> std::io::Result<usize> {
-        self.buffered_writer.write_buffered_borrowed(buf).await
-    }
-
-    pub fn bytes_written(&self) -> u64 {
-        let flushed_offset = self.buffered_writer.as_inner().bytes_written();
-        let buffer: &zero_padded::Buffer<TAIL_SZ> = self.buffered_writer.inspect_buffer();
-        flushed_offset + u64::try_from(buffer.pending()).unwrap()
-    }
-
-    pub(crate) async fn read_blk(&self, blknum: u32) -> Result<ReadResult<'_, W>, std::io::Error> {
-        let flushed_offset = self.buffered_writer.as_inner().bytes_written();
-        let buffer: &zero_padded::Buffer<TAIL_SZ> = self.buffered_writer.inspect_buffer();
-        let buffered_offset = flushed_offset + u64::try_from(buffer.pending()).unwrap();
-        let read_offset = (blknum as u64) * (PAGE_SZ as u64);
-
-        // The trailing page ("block") might only be partially filled,
-        // yet the blob_io code relies on us to return a full PAGE_SZed slice anyway.
-        // Moreover, it has to be zero-padded, because when we still had
-        // a write-back page cache, it provided pre-zeroed pages, and blob_io came to rely on it.
-        // DeltaLayer probably has the same issue, not sure why it needs no special treatment.
-        // => check here that the read doesn't go beyond this potentially trailing
-        // => the zero-padding is done in the `else` branch below
-        let blocks_written = if buffered_offset % (PAGE_SZ as u64) == 0 {
-            buffered_offset / (PAGE_SZ as u64)
-        } else {
-            (buffered_offset / (PAGE_SZ as u64)) + 1
-        };
-        if (blknum as u64) >= blocks_written {
-            return Err(std::io::Error::new(std::io::ErrorKind::Other, anyhow::anyhow!("read past end of ephemeral_file: read=0x{read_offset:x} buffered=0x{buffered_offset:x} flushed=0x{flushed_offset}")));
-        }
-
-        // assertions for the `if-else` below
-        assert_eq!(
-            flushed_offset % (TAIL_SZ as u64), 0,
-            "we only use write_buffered_borrowed to write to the buffered writer, so it's guaranteed that flushes happen buffer.cap()-sized chunks"
-        );
-        assert_eq!(
-            flushed_offset % (PAGE_SZ as u64),
-            0,
-            "the logic below can't handle if the page is spread across the flushed part and the buffer"
-        );
-
-        if read_offset < flushed_offset {
-            assert!(read_offset + (PAGE_SZ as u64) <= flushed_offset);
-            Ok(ReadResult::NeedsReadFromWriter {
-                writer: self.as_writer(),
-            })
-        } else {
-            let read_offset_in_buffer = read_offset
-                .checked_sub(flushed_offset)
-                .expect("would have taken `if` branch instead of this one");
-            let read_offset_in_buffer = usize::try_from(read_offset_in_buffer).unwrap();
-            let zero_padded_slice = buffer.as_zero_padded_slice();
-            let page = &zero_padded_slice[read_offset_in_buffer..(read_offset_in_buffer + PAGE_SZ)];
-            Ok(ReadResult::ServedFromZeroPaddedMutableTail {
-                buffer: page
-                    .try_into()
-                    .expect("the slice above got it as page-size slice"),
-            })
-        }
-    }
-}

pageserver/src/tenant/ephemeral_file/zero_padded_read_write/zero_padded.rs

@@ -1,108 +0,0 @@
-//! A [`crate::virtual_file::owned_buffers_io::write::Buffer`] whose
-//! unwritten range is guaranteed to be zero-initialized.
-//! This is used by [`crate::tenant::ephemeral_file::zero_padded_read_write::RW::read_blk`]
-//! to serve page-sized reads of the trailing page when the trailing page has only been partially filled.
-
-use std::mem::MaybeUninit;
-
-/// See module-level comment.
-pub struct Buffer<const N: usize> {
-    allocation: Box<[u8; N]>,
-    written: usize,
-}
-
-impl<const N: usize> Default for Buffer<N> {
-    fn default() -> Self {
-        Self {
-            allocation: Box::new(
-                // SAFETY: zeroed memory is a valid [u8; N]
-                unsafe { MaybeUninit::zeroed().assume_init() },
-            ),
-            written: 0,
-        }
-    }
-}
-
-impl<const N: usize> Buffer<N> {
-    #[inline(always)]
-    fn invariants(&self) {
-        // don't check by default, unoptimized is too expensive even for debug mode
-        if false {
-            debug_assert!(self.written <= N, "{}", self.written);
-            debug_assert!(self.allocation[self.written..N].iter().all(|v| *v == 0));
-        }
-    }
-
-    pub fn as_zero_padded_slice(&self) -> &[u8; N] {
-        &self.allocation
-    }
-}
-
-impl<const N: usize> crate::virtual_file::owned_buffers_io::write::Buffer for Buffer<N> {
-    type IoBuf = Self;
-
-    fn cap(&self) -> usize {
-        self.allocation.len()
-    }
-
-    fn extend_from_slice(&mut self, other: &[u8]) {
-        self.invariants();
-        let remaining = self.allocation.len() - self.written;
-        if other.len() > remaining {
-            panic!("calling extend_from_slice() with insufficient remaining capacity");
-        }
-        self.allocation[self.written..(self.written + other.len())].copy_from_slice(other);
-        self.written += other.len();
-        self.invariants();
-    }
-
-    fn pending(&self) -> usize {
-        self.written
-    }
-
-    fn flush(self) -> tokio_epoll_uring::Slice<Self> {
-        self.invariants();
-        let written = self.written;
-        tokio_epoll_uring::BoundedBuf::slice(self, 0..written)
-    }
-
-    fn reuse_after_flush(iobuf: Self::IoBuf) -> Self {
-        let Self {
-            mut allocation,
-            written,
-        } = iobuf;
-        allocation[0..written].fill(0);
-        let new = Self {
-            allocation,
-            written: 0,
-        };
-        new.invariants();
-        new
-    }
-}
-
-/// We have this trait impl so that the `flush` method in the `Buffer` impl above can produce a
-/// [`tokio_epoll_uring::BoundedBuf::slice`] of the [`Self::written`] range of the data.
-///
-/// Remember that bytes_init is generally _not_ a tracker of the amount
-/// of valid data in the io buffer; we use `Slice` for that.
-/// The `IoBuf` is _only_ for keeping track of uninitialized memory, a bit like MaybeUninit.
-///
-/// SAFETY:
-///
-/// The [`Self::allocation`] is stable becauses boxes are stable.
-/// The memory is zero-initialized, so, bytes_init is always N.
-unsafe impl<const N: usize> tokio_epoll_uring::IoBuf for Buffer<N> {
-    fn stable_ptr(&self) -> *const u8 {
-        self.allocation.as_ptr()
-    }
-
-    fn bytes_init(&self) -> usize {
-        // Yes, N, not self.written; Read the full comment of this impl block!
-        N
-    }
-
-    fn bytes_total(&self) -> usize {
-        N
-    }
-}

pageserver/src/tenant/remote_timeline_client.rs

@@ -202,9 +202,7 @@ use std::sync::atomic::{AtomicU32, Ordering};
 use std::sync::{Arc, Mutex};
 use std::time::Duration;
 
-use remote_storage::{
-    DownloadError, GenericRemoteStorage, ListingMode, RemotePath, TimeoutOrCancel,
-};
+use remote_storage::{DownloadError, GenericRemoteStorage, RemotePath, TimeoutOrCancel};
 use std::ops::DerefMut;
 use tracing::{debug, error, info, instrument, warn};
 use tracing::{info_span, Instrument};
@@ -1147,7 +1145,7 @@ impl RemoteTimelineClient {
         // and retry will arrive to different pageserver there wont be any traces of it on remote storage
         let timeline_storage_path = remote_timeline_path(&self.tenant_shard_id, &self.timeline_id);
 
-        // Execute all pending deletions, so that when we proceed to do a listing below, we aren't
+        // Execute all pending deletions, so that when we proceed to do a list_prefixes below, we aren't
         // taking the burden of listing all the layers that we already know we should delete.
         self.flush_deletion_queue().await?;
 
@@ -1156,20 +1154,14 @@ impl RemoteTimelineClient {
         let remaining = download_retry(
             || async {
                 self.storage_impl
-                    .list(
-                        Some(&timeline_storage_path),
-                        ListingMode::NoDelimiter,
-                        None,
-                        &cancel,
-                    )
+                    .list_files(Some(&timeline_storage_path), None, &cancel)
                     .await
             },
             "list remaining files",
             &cancel,
         )
         .await
-        .context("list files remaining files")?
-        .keys;
+        .context("list files remaining files")?;
 
         // We will delete the current index_part object last, since it acts as a deletion
         // marker via its deleted_at attribute

pageserver/src/tenant/remote_timeline_client/download.rs

@@ -182,7 +182,6 @@ async fn download_object<'a>(
         #[cfg(target_os = "linux")]
         crate::virtual_file::io_engine::IoEngine::TokioEpollUring => {
             use crate::virtual_file::owned_buffers_io::{self, util::size_tracking_writer};
-            use bytes::BytesMut;
             async {
                 let destination_file = VirtualFile::create(dst_path)
                     .await
@@ -195,10 +194,10 @@ async fn download_object<'a>(
                 // There's chunks_vectored() on the stream.
                 let (bytes_amount, destination_file) = async {
                     let size_tracking = size_tracking_writer::Writer::new(destination_file);
-                    let mut buffered = owned_buffers_io::write::BufferedWriter::<BytesMut, _>::new(
-                        size_tracking,
-                        BytesMut::with_capacity(super::BUFFER_SIZE),
-                    );
+                    let mut buffered = owned_buffers_io::write::BufferedWriter::<
+                        { super::BUFFER_SIZE },
+                        _,
+                    >::new(size_tracking);
                     while let Some(res) =
                         futures::StreamExt::next(&mut download.download_stream).await
                     {
@@ -259,7 +258,7 @@ pub async fn list_remote_timelines(
     tenant_shard_id: TenantShardId,
     cancel: CancellationToken,
 ) -> anyhow::Result<(HashSet<TimelineId>, HashSet<String>)> {
-    let remote_path = remote_timelines_path(&tenant_shard_id).add_trailing_slash();
+    let remote_path = remote_timelines_path(&tenant_shard_id);
 
     fail::fail_point!("storage-sync-list-remote-timelines", |_| {
         anyhow::bail!("storage-sync-list-remote-timelines");
@@ -418,16 +417,11 @@ pub(super) async fn download_index_part(
     let index_prefix = remote_index_path(tenant_shard_id, timeline_id, Generation::none());
 
     let indices = download_retry(
-        || async {
-            storage
-                .list(Some(&index_prefix), ListingMode::NoDelimiter, None, cancel)
-                .await
-        },
+        || async { storage.list_files(Some(&index_prefix), None, cancel).await },
         "list index_part files",
         cancel,
     )
-    .await?
-    .keys;
+    .await?;
 
     // General case logic for which index to use: the latest index whose generation
     // is <= our own.  See "Finding the remote indices for timelines" in docs/rfcs/025-generation-numbers.md

pageserver/src/tenant/size.rs

@@ -118,9 +118,6 @@ pub(super) async fn gather_inputs(
     ctx: &RequestContext,
 ) -> anyhow::Result<ModelInputs> {
     // refresh is needed to update gc related pitr_cutoff and horizon_cutoff
-    //
-    // FIXME: if a single timeline is deleted while refresh gc info is ongoing, we will fail the
-    // whole computation. It does not make sense from the billing perspective.
     tenant
         .refresh_gc_info(cancel, ctx)
         .await

pageserver/src/tenant/storage_layer.rs

@@ -148,29 +148,6 @@ impl ValuesReconstructState {
         self.layers_visited
     }
 
-    /// This function is called after reading a keyspace from a layer.
-    /// It checks if the read path has now moved past the cached Lsn for any keys.
-    ///
-    /// Implementation note: We intentionally iterate over the keys for which we've
-    /// already collected some reconstruct data. This avoids scaling complexity with
-    /// the size of the search space.
-    pub(crate) fn on_lsn_advanced(&mut self, keyspace: &KeySpace, advanced_to: Lsn) {
-        for (key, value) in self.keys.iter_mut() {
-            if !keyspace.contains(key) {
-                continue;
-            }
-
-            if let Ok(state) = value {
-                if state.situation != ValueReconstructSituation::Complete
-                    && state.get_cached_lsn() >= Some(advanced_to)
-                {
-                    state.situation = ValueReconstructSituation::Complete;
-                    self.keys_done.add_key(*key);
-                }
-            }
-        }
-    }
-
     /// Update the state collected for a given key.
     /// Returns true if this was the last value needed for the key and false otherwise.
     ///
@@ -195,18 +172,11 @@ impl ValuesReconstructState {
                         true
                     }
                     Value::WalRecord(rec) => {
-                        debug_assert!(
-                            Some(lsn) > state.get_cached_lsn(),
-                            "Attempt to collect a record below cached LSN for walredo: {} < {}",
-                            lsn,
-                            state
-                                .get_cached_lsn()
-                                .expect("Assertion can only fire if a cached lsn is present")
-                        );
-
+                        let reached_cache =
+                            state.get_cached_lsn().map(|clsn| clsn + 1) == Some(lsn);
                         let will_init = rec.will_init();
                         state.records.push((lsn, rec));
-                        will_init
+                        will_init || reached_cache
                     }
                 },
             };

pageserver/src/tenant/storage_layer/delta_layer.rs

@@ -217,7 +217,6 @@ pub struct DeltaLayerInner {
     // values copied from summary
     index_start_blk: u32,
     index_root_blk: u32,
-    lsn_range: Range<Lsn>,
 
     file: VirtualFile,
     file_id: FileId,
@@ -746,7 +745,6 @@ impl DeltaLayerInner {
             file_id,
             index_start_blk: actual_summary.index_start_blk,
             index_root_blk: actual_summary.index_root_blk,
-            lsn_range: actual_summary.lsn_range,
             max_vectored_read_bytes,
         }))
     }
@@ -871,7 +869,7 @@ impl DeltaLayerInner {
         let data_end_offset = self.index_start_offset();
 
         let reads = Self::plan_reads(
-            &keyspace,
+            keyspace,
             lsn_range,
             data_end_offset,
             index_reader,
@@ -885,13 +883,11 @@ impl DeltaLayerInner {
         self.do_reads_and_update_state(reads, reconstruct_state)
             .await;
 
-        reconstruct_state.on_lsn_advanced(&keyspace, self.lsn_range.start);
-
         Ok(())
     }
 
     async fn plan_reads<Reader>(
-        keyspace: &KeySpace,
+        keyspace: KeySpace,
         lsn_range: Range<Lsn>,
         data_end_offset: u64,
         index_reader: DiskBtreeReader<Reader, DELTA_KEY_SIZE>,
@@ -1539,7 +1535,7 @@ mod test {
 
         // Plan and validate
         let vectored_reads = DeltaLayerInner::plan_reads(
-            &keyspace,
+            keyspace.clone(),
             lsn_range.clone(),
             disk_offset,
             reader,
@@ -1791,7 +1787,7 @@ mod test {
             let data_end_offset = inner.index_start_blk as u64 * PAGE_SZ as u64;
 
             let vectored_reads = DeltaLayerInner::plan_reads(
-                &keyspace,
+                keyspace.clone(),
                 entries_meta.lsn_range.clone(),
                 data_end_offset,
                 index_reader,

pageserver/src/tenant/storage_layer/inmemory_layer.rs

@@ -438,8 +438,6 @@ impl InMemoryLayer {
             }
         }
 
-        reconstruct_state.on_lsn_advanced(&keyspace, self.start_lsn);
-
         Ok(())
     }
 }
@@ -482,7 +480,7 @@ impl InMemoryLayer {
         trace!("initializing new empty InMemoryLayer for writing on timeline {timeline_id} at {start_lsn}");
 
         let file = EphemeralFile::create(conf, tenant_shard_id, timeline_id).await?;
-        let key = InMemoryLayerFileId(file.page_cache_file_id());
+        let key = InMemoryLayerFileId(file.id());
 
         Ok(InMemoryLayer {
             file_id: key,

pageserver/src/tenant/storage_layer/layer.rs

@@ -336,12 +336,6 @@ impl Layer {
             .get_values_reconstruct_data(keyspace, lsn_range, reconstruct_data, &self.0, ctx)
             .instrument(tracing::debug_span!("get_values_reconstruct_data", layer=%self))
             .await
-            .map_err(|err| match err {
-                GetVectoredError::Other(err) => GetVectoredError::Other(
-                    err.context(format!("get_values_reconstruct_data for layer {self}")),
-                ),
-                err => err,
-            })
     }
 
     /// Download the layer if evicted.

pageserver/src/tenant/tasks.rs

@@ -62,7 +62,7 @@ impl BackgroundLoopKind {
 pub(crate) async fn concurrent_background_tasks_rate_limit_permit(
     loop_kind: BackgroundLoopKind,
     _ctx: &RequestContext,
-) -> tokio::sync::SemaphorePermit<'static> {
+) -> impl Drop {
     let _guard = crate::metrics::BACKGROUND_LOOP_SEMAPHORE_WAIT_GAUGE
         .with_label_values(&[loop_kind.as_static_str()])
         .guard();

pageserver/src/tenant/timeline.rs

@@ -16,7 +16,7 @@ use enumset::EnumSet;
 use fail::fail_point;
 use once_cell::sync::Lazy;
 use pageserver_api::{
-    key::{AUX_FILES_KEY, NON_INHERITED_RANGE},
+    key::AUX_FILES_KEY,
     keyspace::KeySpaceAccum,
     models::{
         CompactionAlgorithm, DownloadRemoteLayersTaskInfo, DownloadRemoteLayersTaskSpawnRequest,
@@ -86,7 +86,7 @@ use crate::{
 use crate::config::PageServerConf;
 use crate::keyspace::{KeyPartitioning, KeySpace};
 use crate::metrics::{
-    GetKind, TimelineMetrics, MATERIALIZED_PAGE_CACHE_HIT, MATERIALIZED_PAGE_CACHE_HIT_DIRECT,
+    TimelineMetrics, MATERIALIZED_PAGE_CACHE_HIT, MATERIALIZED_PAGE_CACHE_HIT_DIRECT,
 };
 use crate::pgdatadir_mapping::CalculateLogicalSizeError;
 use crate::tenant::config::TenantConfOpt;
@@ -119,8 +119,8 @@ use self::layer_manager::LayerManager;
 use self::logical_size::LogicalSize;
 use self::walreceiver::{WalReceiver, WalReceiverConf};
 
+use super::config::TenantConf;
 use super::secondary::heatmap::{HeatMapLayer, HeatMapTimeline};
-use super::{config::TenantConf, storage_layer::VectoredValueReconstructState};
 use super::{debug_assert_current_span_has_tenant_and_timeline_id, AttachedTenantConf};
 use super::{remote_timeline_client::index::IndexPart, storage_layer::LayerFringe};
 use super::{remote_timeline_client::RemoteTimelineClient, storage_layer::ReadableLayer};
@@ -182,16 +182,6 @@ pub(crate) struct AuxFilesState {
     pub(crate) n_deltas: usize,
 }
 
-/// The relation size cache caches relation sizes at the end of the timeline. It speeds up WAL
-/// ingestion considerably, because WAL ingestion needs to check on most records if the record
-/// implicitly extends the relation.  At startup, `complete_as_of` is initialized to the current end
-/// of the timeline (disk_consistent_lsn).  It's used on reads of relation sizes to check if the
-/// value can be used to also update the cache, see [`Timeline::update_cached_rel_size`].
-pub(crate) struct RelSizeCache {
-    pub(crate) complete_as_of: Lsn,
-    pub(crate) map: HashMap<RelTag, (Lsn, BlockNumber)>,
-}
-
 pub struct Timeline {
     conf: &'static PageServerConf,
     tenant_conf: Arc<ArcSwap<AttachedTenantConf>>,
@@ -334,7 +324,7 @@ pub struct Timeline {
     pub walreceiver: Mutex<Option<WalReceiver>>,
 
     /// Relation size cache
-    pub(crate) rel_size_cache: RwLock<RelSizeCache>,
+    pub rel_size_cache: RwLock<HashMap<RelTag, (Lsn, BlockNumber)>>,
 
     download_all_remote_layers_task_info: RwLock<Option<DownloadRemoteLayersTaskInfo>>,
 
@@ -653,19 +643,6 @@ impl From<GetVectoredError> for CreateImageLayersError {
     }
 }
 
-impl From<GetVectoredError> for PageReconstructError {
-    fn from(e: GetVectoredError) -> Self {
-        match e {
-            GetVectoredError::Cancelled => PageReconstructError::Cancelled,
-            GetVectoredError::InvalidLsn(_) => PageReconstructError::Other(anyhow!("Invalid LSN")),
-            err @ GetVectoredError::Oversized(_) => PageReconstructError::Other(err.into()),
-            err @ GetVectoredError::MissingKey(_) => PageReconstructError::Other(err.into()),
-            GetVectoredError::GetReadyAncestorError(err) => PageReconstructError::from(err),
-            GetVectoredError::Other(err) => PageReconstructError::Other(err),
-        }
-    }
-}
-
 impl From<GetReadyAncestorError> for PageReconstructError {
     fn from(e: GetReadyAncestorError) -> Self {
         use GetReadyAncestorError::*;
@@ -695,23 +672,6 @@ pub enum GetVectoredImpl {
     Vectored,
 }
 
-#[derive(
-    Eq,
-    PartialEq,
-    Debug,
-    Copy,
-    Clone,
-    strum_macros::EnumString,
-    strum_macros::Display,
-    serde_with::DeserializeFromStr,
-    serde_with::SerializeDisplay,
-)]
-#[strum(serialize_all = "kebab-case")]
-pub enum GetImpl {
-    Legacy,
-    Vectored,
-}
-
 pub(crate) enum WaitLsnWaiter<'a> {
     Timeline(&'a Timeline),
     Tenant,
@@ -773,6 +733,16 @@ impl Timeline {
         key: Key,
         lsn: Lsn,
         ctx: &RequestContext,
+    ) -> Result<Bytes, PageReconstructError> {
+        self.timeline_get_throttle.throttle(ctx, 1).await;
+        self.get_impl(key, lsn, ctx).await
+    }
+    /// Not subject to [`Self::timeline_get_throttle`].
+    async fn get_impl(
+        &self,
+        key: Key,
+        lsn: Lsn,
+        ctx: &RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         if !lsn.is_valid() {
             return Err(PageReconstructError::Other(anyhow::anyhow!("Invalid LSN")));
@@ -783,7 +753,13 @@ impl Timeline {
         // page_service.
         debug_assert!(!self.shard_identity.is_key_disposable(&key));
 
-        self.timeline_get_throttle.throttle(ctx, 1).await;
+        // XXX: structured stats collection for layer eviction here.
+        trace!(
+            "get page request for {}@{} from task kind {:?}",
+            key,
+            lsn,
+            ctx.task_kind()
+        );
 
         // Check the page cache. We will get back the most recent page with lsn <= `lsn`.
         // The cached image can be returned directly if there is no WAL between the cached image
@@ -806,85 +782,12 @@ impl Timeline {
             None => None,
         };
 
-        match self.conf.get_impl {
-            GetImpl::Legacy => {
-                let reconstruct_state = ValueReconstructState {
-                    records: Vec::new(),
-                    img: cached_page_img,
-                };
+        let mut reconstruct_state = ValueReconstructState {
+            records: Vec::new(),
+            img: cached_page_img,
+        };
 
-                self.get_impl(key, lsn, reconstruct_state, ctx).await
-            }
-            GetImpl::Vectored => {
-                let keyspace = KeySpace {
-                    ranges: vec![key..key.next()],
-                };
-
-                // Initialise the reconstruct state for the key with the cache
-                // entry returned above.
-                let mut reconstruct_state = ValuesReconstructState::new();
-                let mut key_state = VectoredValueReconstructState::default();
-                key_state.img = cached_page_img;
-                reconstruct_state.keys.insert(key, Ok(key_state));
-
-                let vectored_res = self
-                    .get_vectored_impl(keyspace.clone(), lsn, reconstruct_state, ctx)
-                    .await;
-
-                if self.conf.validate_vectored_get {
-                    self.validate_get_vectored_impl(&vectored_res, keyspace, lsn, ctx)
-                        .await;
-                }
-
-                let key_value = vectored_res?.pop_first();
-                match key_value {
-                    Some((got_key, value)) => {
-                        if got_key != key {
-                            error!(
-                                "Expected {}, but singular vectored get returned {}",
-                                key, got_key
-                            );
-                            Err(PageReconstructError::Other(anyhow!(
-                                "Singular vectored get returned wrong key"
-                            )))
-                        } else {
-                            value
-                        }
-                    }
-                    None => {
-                        error!(
-                            "Expected {}, but singular vectored get returned nothing",
-                            key
-                        );
-                        Err(PageReconstructError::Other(anyhow!(
-                            "Singular vectored get did not return a value for {}",
-                            key
-                        )))
-                    }
-                }
-            }
-        }
-    }
-
-    /// Not subject to [`Self::timeline_get_throttle`].
-    async fn get_impl(
-        &self,
-        key: Key,
-        lsn: Lsn,
-        mut reconstruct_state: ValueReconstructState,
-        ctx: &RequestContext,
-    ) -> Result<Bytes, PageReconstructError> {
-        // XXX: structured stats collection for layer eviction here.
-        trace!(
-            "get page request for {}@{} from task kind {:?}",
-            key,
-            lsn,
-            ctx.task_kind()
-        );
-
-        let timer = crate::metrics::GET_RECONSTRUCT_DATA_TIME
-            .for_get_kind(GetKind::Singular)
-            .start_timer();
+        let timer = crate::metrics::GET_RECONSTRUCT_DATA_TIME.start_timer();
         let path = self
             .get_reconstruct_data(key, lsn, &mut reconstruct_state, ctx)
             .await?;
@@ -894,7 +797,7 @@ impl Timeline {
         let res = self.reconstruct_value(key, lsn, reconstruct_state).await;
         let elapsed = start.elapsed();
         crate::metrics::RECONSTRUCT_TIME
-            .for_get_kind(GetKind::Singular)
+            .for_result(&res)
             .observe(elapsed.as_secs_f64());
 
         if cfg!(feature = "testing") && res.is_err() {
@@ -921,6 +824,10 @@ impl Timeline {
     }
 
     pub(crate) const MAX_GET_VECTORED_KEYS: u64 = 32;
+    // If we are visiting more than 50 layers to reconstruct one value,
+    // it's likely that something is wrong with compaction. We print a
+    // rate limited warning in that case.
+    pub(crate) const LAYERS_VISITED_WARN_THRESHOLD: u64 = 50;
 
     /// Look up multiple page versions at a given LSN
     ///
@@ -973,9 +880,7 @@ impl Timeline {
                 self.get_vectored_sequential_impl(keyspace, lsn, ctx).await
             }
             GetVectoredImpl::Vectored => {
-                let vectored_res = self
-                    .get_vectored_impl(keyspace.clone(), lsn, ValuesReconstructState::new(), ctx)
-                    .await;
+                let vectored_res = self.get_vectored_impl(keyspace.clone(), lsn, ctx).await;
 
                 if self.conf.validate_vectored_get {
                     self.validate_get_vectored_impl(&vectored_res, keyspace, lsn, ctx)
@@ -1021,9 +926,7 @@ impl Timeline {
         for range in keyspace.ranges {
             let mut key = range.start;
             while key != range.end {
-                let block = self
-                    .get_impl(key, lsn, ValueReconstructState::default(), ctx)
-                    .await;
+                let block = self.get_impl(key, lsn, ctx).await;
 
                 use PageReconstructError::*;
                 match block {
@@ -1034,30 +937,7 @@ impl Timeline {
                     Err(MissingKey(MissingKeyError {
                         stuck_at_lsn: false,
                         ..
-                    })) if !NON_INHERITED_RANGE.contains(&key) => {
-                        // The vectored read path handles non inherited keys specially.
-                        // If such a a key cannot be reconstructed from the current timeline,
-                        // the vectored read path returns a key level error as opposed to a top
-                        // level error.
-                        return Err(GetVectoredError::MissingKey(key));
-                    }
-                    Err(Other(err))
-                        if err
-                            .to_string()
-                            .contains("downloading evicted layer file failed") =>
-                    {
-                        return Err(GetVectoredError::Other(err))
-                    }
-                    Err(Other(err))
-                        if err
-                            .chain()
-                            .any(|cause| cause.to_string().contains("layer loading failed")) =>
-                    {
-                        // The intent here is to achieve error parity with the vectored read path.
-                        // When vectored read fails to load a layer it fails the whole read, hence
-                        // we mimic this behaviour here to keep the validation happy.
-                        return Err(GetVectoredError::Other(err));
-                    }
+                    })) => return Err(GetVectoredError::MissingKey(key)),
                     _ => {
                         values.insert(key, block);
                         key = key.next();
@@ -1073,25 +953,13 @@ impl Timeline {
         &self,
         keyspace: KeySpace,
         lsn: Lsn,
-        mut reconstruct_state: ValuesReconstructState,
         ctx: &RequestContext,
     ) -> Result<BTreeMap<Key, Result<Bytes, PageReconstructError>>, GetVectoredError> {
-        let get_kind = if keyspace.total_size() == 1 {
-            GetKind::Singular
-        } else {
-            GetKind::Vectored
-        };
+        let mut reconstruct_state = ValuesReconstructState::new();
 
-        let get_data_timer = crate::metrics::GET_RECONSTRUCT_DATA_TIME
-            .for_get_kind(get_kind)
-            .start_timer();
         self.get_vectored_reconstruct_data(keyspace, lsn, &mut reconstruct_state, ctx)
             .await?;
-        get_data_timer.stop_and_record();
 
-        let reconstruct_timer = crate::metrics::RECONSTRUCT_TIME
-            .for_get_kind(get_kind)
-            .start_timer();
         let mut results: BTreeMap<Key, Result<Bytes, PageReconstructError>> = BTreeMap::new();
         let layers_visited = reconstruct_state.get_layers_visited();
         for (key, res) in reconstruct_state.keys {
@@ -1107,13 +975,25 @@ impl Timeline {
                 }
             }
         }
-        reconstruct_timer.stop_and_record();
 
         // Note that this is an approximation. Tracking the exact number of layers visited
         // per key requires virtually unbounded memory usage and is inefficient
         // (i.e. segment tree tracking each range queried from a layer)
-        crate::metrics::VEC_READ_NUM_LAYERS_VISITED
-            .observe(layers_visited as f64 / results.len() as f64);
+        let average_layers_visited = layers_visited as f64 / results.len() as f64;
+
+        crate::metrics::VEC_READ_NUM_LAYERS_VISITED.observe(average_layers_visited);
+
+        if average_layers_visited >= Self::LAYERS_VISITED_WARN_THRESHOLD as f64 {
+            use utils::rate_limit::RateLimit;
+            static LOGGED: Lazy<Mutex<RateLimit>> =
+                Lazy::new(|| Mutex::new(RateLimit::new(Duration::from_secs(10))));
+            let mut rate_limit = LOGGED.lock().unwrap();
+            let rate_limited = rate_limit.rate_limited();
+            rate_limit.call(|| {
+                warn!("Too many layers visited by vectored read: {} >= {}; rate limited {} occurences",
+                      average_layers_visited, Self::LAYERS_VISITED_WARN_THRESHOLD, rate_limited);
+            });
+        }
 
         Ok(results)
     }
@@ -1871,15 +1751,6 @@ const REPARTITION_FREQ_IN_CHECKPOINT_DISTANCE: u64 = 10;
 
 // Private functions
 impl Timeline {
-    #[allow(dead_code)]
-    pub(crate) fn get_switch_to_aux_file_v2(&self) -> bool {
-        let tenant_conf = self.tenant_conf.load();
-        tenant_conf
-            .tenant_conf
-            .switch_to_aux_file_v2
-            .unwrap_or(self.conf.default_tenant_conf.switch_to_aux_file_v2)
-    }
-
     pub(crate) fn get_lazy_slru_download(&self) -> bool {
         let tenant_conf = self.tenant_conf.load();
         tenant_conf
@@ -2104,10 +1975,7 @@ impl Timeline {
                 last_image_layer_creation_check_at: AtomicLsn::new(0),
 
                 last_received_wal: Mutex::new(None),
-                rel_size_cache: RwLock::new(RelSizeCache {
-                    complete_as_of: disk_consistent_lsn,
-                    map: HashMap::new(),
-                }),
+                rel_size_cache: RwLock::new(HashMap::new()),
 
                 download_all_remote_layers_task_info: RwLock::new(None),
 
@@ -2950,7 +2818,18 @@ impl Timeline {
         let mut timeline = self;
 
         let mut read_count = scopeguard::guard(0, |cnt| {
-            crate::metrics::READ_NUM_LAYERS_VISITED.observe(cnt as f64)
+            crate::metrics::READ_NUM_LAYERS_VISITED.observe(cnt as f64);
+            if cnt >= Self::LAYERS_VISITED_WARN_THRESHOLD {
+                use utils::rate_limit::RateLimit;
+                static LOGGED: Lazy<Mutex<RateLimit>> =
+                    Lazy::new(|| Mutex::new(RateLimit::new(Duration::from_secs(10))));
+                let mut rate_limit = LOGGED.lock().unwrap();
+                let rate_limited = rate_limit.rate_limited();
+                rate_limit.call(|| {
+                    warn!("Too many layers visited by non-vectored read: {} >= {}; rate limited {} occurences",
+                          cnt, Self::LAYERS_VISITED_WARN_THRESHOLD, rate_limited);
+                });
+            }
         });
 
         // For debugging purposes, collect the path of layers that we traversed
@@ -3167,41 +3046,6 @@ impl Timeline {
             .await?;
 
             keyspace.remove_overlapping_with(&completed);
-
-            // Do not descend into the ancestor timeline for aux files.
-            // We don't return a blanket [`GetVectoredError::MissingKey`] to avoid
-            // stalling compaction.
-            // TODO(chi): this will need to be updated for aux files v2 storage
-            if keyspace.overlaps(&NON_INHERITED_RANGE) {
-                let removed = keyspace.remove_overlapping_with(&KeySpace {
-                    ranges: vec![NON_INHERITED_RANGE],
-                });
-
-                for range in removed.ranges {
-                    let mut key = range.start;
-                    while key < range.end {
-                        reconstruct_state.on_key_error(
-                            key,
-                            PageReconstructError::MissingKey(MissingKeyError {
-                                stuck_at_lsn: false,
-                                key,
-                                shard: self.shard_identity.get_shard_number(&key),
-                                cont_lsn,
-                                request_lsn,
-                                ancestor_lsn: None,
-                                traversal_path: Vec::default(),
-                                backtrace: if cfg!(test) {
-                                    Some(std::backtrace::Backtrace::force_capture())
-                                } else {
-                                    None
-                                },
-                            }),
-                        );
-                        key = key.next();
-                    }
-                }
-            }
-
             if keyspace.total_size() == 0 || timeline.ancestor_timeline.is_none() {
                 break;
             }
@@ -3257,62 +3101,56 @@ impl Timeline {
             unmapped_keyspace.remove_overlapping_with(&keys_done_last_step);
             completed_keyspace.merge(&keys_done_last_step);
 
-            // Do not descent any further if the last layer we visited
-            // completed all keys in the keyspace it inspected. This is not
-            // required for correctness, but avoids visiting extra layers
-            // which turns out to be a perf bottleneck in some cases.
-            if !unmapped_keyspace.is_empty() {
-                let guard = timeline.layers.read().await;
-                let layers = guard.layer_map();
+            let guard = timeline.layers.read().await;
+            let layers = guard.layer_map();
 
-                let in_memory_layer = layers.find_in_memory_layer(|l| {
-                    let start_lsn = l.get_lsn_range().start;
-                    cont_lsn > start_lsn
-                });
+            let in_memory_layer = layers.find_in_memory_layer(|l| {
+                let start_lsn = l.get_lsn_range().start;
+                cont_lsn > start_lsn
+            });
 
-                match in_memory_layer {
-                    Some(l) => {
-                        let lsn_range = l.get_lsn_range().start..cont_lsn;
-                        fringe.update(
-                            ReadableLayer::InMemoryLayer(l),
-                            unmapped_keyspace.clone(),
-                            lsn_range,
-                        );
-                    }
-                    None => {
-                        for range in unmapped_keyspace.ranges.iter() {
-                            let results = layers.range_search(range.clone(), cont_lsn);
+            match in_memory_layer {
+                Some(l) => {
+                    let lsn_range = l.get_lsn_range().start..cont_lsn;
+                    fringe.update(
+                        ReadableLayer::InMemoryLayer(l),
+                        unmapped_keyspace.clone(),
+                        lsn_range,
+                    );
+                }
+                None => {
+                    for range in unmapped_keyspace.ranges.iter() {
+                        let results = layers.range_search(range.clone(), cont_lsn);
 
-                            results
-                                .found
-                                .into_iter()
-                                .map(|(SearchResult { layer, lsn_floor }, keyspace_accum)| {
-                                    (
-                                        ReadableLayer::PersistentLayer(guard.get_from_desc(&layer)),
-                                        keyspace_accum.to_keyspace(),
-                                        lsn_floor..cont_lsn,
-                                    )
-                                })
-                                .for_each(|(layer, keyspace, lsn_range)| {
-                                    fringe.update(layer, keyspace, lsn_range)
-                                });
-                        }
+                        results
+                            .found
+                            .into_iter()
+                            .map(|(SearchResult { layer, lsn_floor }, keyspace_accum)| {
+                                (
+                                    ReadableLayer::PersistentLayer(guard.get_from_desc(&layer)),
+                                    keyspace_accum.to_keyspace(),
+                                    lsn_floor..cont_lsn,
+                                )
+                            })
+                            .for_each(|(layer, keyspace, lsn_range)| {
+                                fringe.update(layer, keyspace, lsn_range)
+                            });
                     }
                 }
-
-                // It's safe to drop the layer map lock after planning the next round of reads.
-                // The fringe keeps readable handles for the layers which are safe to read even
-                // if layers were compacted or flushed.
-                //
-                // The more interesting consideration is: "Why is the read algorithm still correct
-                // if the layer map changes while it is operating?". Doing a vectored read on a
-                // timeline boils down to pushing an imaginary lsn boundary downwards for each range
-                // covered by the read. The layer map tells us how to move the lsn downwards for a
-                // range at *a particular point in time*. It is fine for the answer to be different
-                // at two different time points.
-                drop(guard);
             }
 
+            // It's safe to drop the layer map lock after planning the next round of reads.
+            // The fringe keeps readable handles for the layers which are safe to read even
+            // if layers were compacted or flushed.
+            //
+            // The more interesting consideration is: "Why is the read algorithm still correct
+            // if the layer map changes while it is operating?". Doing a vectored read on a
+            // timeline boils down to pushing an imaginary lsn boundary downwards for each range
+            // covered by the read. The layer map tells us how to move the lsn downwards for a
+            // range at *a particular point in time*. It is fine for the answer to be different
+            // at two different time points.
+            drop(guard);
+
             if let Some((layer_to_read, keyspace_to_read, lsn_range)) = fringe.next_layer() {
                 let next_cont_lsn = lsn_range.start;
                 layer_to_read
@@ -3915,6 +3753,24 @@ impl Timeline {
 
     // Is it time to create a new image layer for the given partition?
     async fn time_for_new_image_layer(&self, partition: &KeySpace, lsn: Lsn) -> bool {
+        let last = self.last_image_layer_creation_check_at.load();
+        if lsn != Lsn(0) {
+            let distance = lsn
+                .checked_sub(last)
+                .expect("Attempt to compact with LSN going backwards");
+
+            let min_distance = self.get_image_layer_creation_check_threshold() as u64
+                * self.get_checkpoint_distance();
+
+            // Skip the expensive delta layer counting below if we've not ingested
+            // sufficient WAL since the last check.
+            if distance.0 < min_distance {
+                return false;
+            }
+        }
+
+        self.last_image_layer_creation_check_at.store(lsn);
+
         let threshold = self.get_image_creation_threshold();
 
         let guard = self.layers.read().await;
@@ -3986,37 +3842,9 @@ impl Timeline {
         // image layers  <100000000..100000099> and <200000000..200000199> are not completely covering it.
         let mut start = Key::MIN;
 
-        let check_for_image_layers = {
-            let last_checks_at = self.last_image_layer_creation_check_at.load();
-            let distance = lsn
-                .checked_sub(last_checks_at)
-                .expect("Attempt to compact with LSN going backwards");
-            let min_distance = self.get_image_layer_creation_check_threshold() as u64
-                * self.get_checkpoint_distance();
-
-            // Skip the expensive delta layer counting if this timeline has not ingested sufficient
-            // WAL since the last check.
-            distance.0 >= min_distance
-        };
-
-        if check_for_image_layers {
-            self.last_image_layer_creation_check_at.store(lsn);
-        }
-
         for partition in partitioning.parts.iter() {
             let img_range = start..partition.ranges.last().unwrap().end;
-
-            let do_it = if force {
-                true
-            } else if check_for_image_layers {
-                // [`Self::time_for_new_image_layer`] is CPU expensive,
-                // so skip if we've not collected enough WAL since the last time
-                self.time_for_new_image_layer(partition, lsn).await
-            } else {
-                false
-            };
-
-            if !do_it {
+            if !force && !self.time_for_new_image_layer(partition, lsn).await {
                 start = img_range.end;
                 continue;
             }
@@ -4345,12 +4173,6 @@ impl Timeline {
         cancel: &CancellationToken,
         ctx: &RequestContext,
     ) -> anyhow::Result<()> {
-        let _timer = self
-            .metrics
-            .update_gc_info_histo
-            .start_timer()
-            .record_on_drop();
-
         // First, calculate pitr_cutoff_timestamp and then convert it to LSN.
         //
         // Some unit tests depend on garbage-collection working even when
@@ -4396,8 +4218,9 @@ impl Timeline {
                 *self.get_latest_gc_cutoff_lsn()
             }
         } else {
-            // No time-based retention was configured. Interpret this as "keep no history".
-            self.get_last_record_lsn()
+            // No time-based retention was configured. Set time-based cutoff to
+            // same as LSN based.
+            cutoff_horizon
         };
 
         // Grab the lock and update the values

pageserver/src/tenant/timeline/eviction_task.rs

@@ -188,10 +188,24 @@ impl Timeline {
     ) -> ControlFlow<()> {
         let now = SystemTime::now();
 
-        let permit = self.acquire_imitation_permit(cancel, ctx).await?;
+        let acquire_permit = crate::tenant::tasks::concurrent_background_tasks_rate_limit_permit(
+            BackgroundLoopKind::Eviction,
+            ctx,
+        );
 
-        self.imitate_layer_accesses(tenant, p, cancel, gate, permit, ctx)
-            .await?;
+        let _permit = tokio::select! {
+            permit = acquire_permit => permit,
+            _ = cancel.cancelled() => return ControlFlow::Break(()),
+            _ = self.cancel.cancelled() => return ControlFlow::Break(()),
+        };
+
+        match self
+            .imitate_layer_accesses(tenant, p, cancel, gate, ctx)
+            .await
+        {
+            ControlFlow::Break(()) => return ControlFlow::Break(()),
+            ControlFlow::Continue(()) => (),
+        }
 
         #[derive(Debug, Default)]
         struct EvictionStats {
@@ -316,27 +330,19 @@ impl Timeline {
         gate: &GateGuard,
         ctx: &RequestContext,
     ) -> ControlFlow<()> {
-        let permit = self.acquire_imitation_permit(cancel, ctx).await?;
-
-        self.imitate_layer_accesses(tenant, p, cancel, gate, permit, ctx)
-            .await
-    }
-
-    async fn acquire_imitation_permit(
-        &self,
-        cancel: &CancellationToken,
-        ctx: &RequestContext,
-    ) -> ControlFlow<(), tokio::sync::SemaphorePermit<'static>> {
         let acquire_permit = crate::tenant::tasks::concurrent_background_tasks_rate_limit_permit(
             BackgroundLoopKind::Eviction,
             ctx,
         );
 
-        tokio::select! {
-            permit = acquire_permit => ControlFlow::Continue(permit),
-            _ = cancel.cancelled() => ControlFlow::Break(()),
-            _ = self.cancel.cancelled() => ControlFlow::Break(()),
-        }
+        let _permit = tokio::select! {
+            permit = acquire_permit => permit,
+            _ = cancel.cancelled() => return ControlFlow::Break(()),
+            _ = self.cancel.cancelled() => return ControlFlow::Break(()),
+        };
+
+        self.imitate_layer_accesses(tenant, p, cancel, gate, ctx)
+            .await
     }
 
     /// If we evict layers but keep cached values derived from those layers, then
@@ -370,7 +376,6 @@ impl Timeline {
         p: &EvictionPolicyLayerAccessThreshold,
         cancel: &CancellationToken,
         gate: &GateGuard,
-        permit: tokio::sync::SemaphorePermit<'static>,
         ctx: &RequestContext,
     ) -> ControlFlow<()> {
         if !self.tenant_shard_id.is_shard_zero() {
@@ -403,28 +408,7 @@ impl Timeline {
         // Make one of the tenant's timelines draw the short straw and run the calculation.
         // The others wait until the calculation is done so that they take into account the
         // imitated accesses that the winner made.
-        let (mut state, _permit) = {
-            if let Ok(locked) = tenant.eviction_task_tenant_state.try_lock() {
-                (locked, permit)
-            } else {
-                // we might need to wait for a long time here in case of pathological synthetic
-                // size calculation performance
-                drop(permit);
-                let locked = tokio::select! {
-                    locked = tenant.eviction_task_tenant_state.lock() => locked,
-                    _ = self.cancel.cancelled() => {
-                        return ControlFlow::Break(())
-                    },
-                    _ = cancel.cancelled() => {
-                        return ControlFlow::Break(())
-                    }
-                };
-                // then reacquire -- this will be bad if there is a lot of traffic, but because we
-                // released the permit, the overall latency will be much better.
-                let permit = self.acquire_imitation_permit(cancel, ctx).await?;
-                (locked, permit)
-            }
-        };
+        let mut state = tenant.eviction_task_tenant_state.lock().await;
         match state.last_layer_access_imitation {
             Some(ts) if ts.elapsed() < inter_imitate_period => { /* no need to run */ }
             _ => {

pageserver/src/virtual_file.rs

@@ -32,11 +32,11 @@ pub use io_engine::feature_test as io_engine_feature_test;
 pub use io_engine::FeatureTestResult as IoEngineFeatureTestResult;
 mod metadata;
 mod open_options;
-use self::owned_buffers_io::write::OwnedAsyncWriter;
 pub(crate) use io_engine::IoEngineKind;
 pub(crate) use metadata::Metadata;
 pub(crate) use open_options::*;
 
+#[cfg_attr(not(target_os = "linux"), allow(dead_code))]
 pub(crate) mod owned_buffers_io {
     //! Abstractions for IO with owned buffers.
     //!
@@ -1083,17 +1083,6 @@ impl Drop for VirtualFile {
     }
 }
 
-impl OwnedAsyncWriter for VirtualFile {
-    #[inline(always)]
-    async fn write_all<B: BoundedBuf<Buf = Buf>, Buf: IoBuf + Send>(
-        &mut self,
-        buf: B,
-    ) -> std::io::Result<(usize, B::Buf)> {
-        let (buf, res) = VirtualFile::write_all(self, buf).await;
-        res.map(move |v| (v, buf))
-    }
-}
-
 impl OpenFiles {
     fn new(num_slots: usize) -> OpenFiles {
         let mut slots = Box::new(Vec::with_capacity(num_slots));

pageserver/src/virtual_file/owned_buffers_io/util/size_tracking_writer.rs

@@ -1,45 +1,33 @@
-use crate::virtual_file::owned_buffers_io::write::OwnedAsyncWriter;
+use crate::virtual_file::{owned_buffers_io::write::OwnedAsyncWriter, VirtualFile};
 use tokio_epoll_uring::{BoundedBuf, IoBuf};
 
-pub struct Writer<W> {
-    dst: W,
+pub struct Writer {
+    dst: VirtualFile,
     bytes_amount: u64,
 }
 
-impl<W> Writer<W> {
-    pub fn new(dst: W) -> Self {
+impl Writer {
+    pub fn new(dst: VirtualFile) -> Self {
         Self {
             dst,
             bytes_amount: 0,
         }
     }
-
-    pub fn bytes_written(&self) -> u64 {
-        self.bytes_amount
-    }
-
-    pub fn as_inner(&self) -> &W {
-        &self.dst
-    }
-
     /// Returns the wrapped `VirtualFile` object as well as the number
     /// of bytes that were written to it through this object.
-    #[cfg_attr(target_os = "macos", allow(dead_code))]
-    pub fn into_inner(self) -> (u64, W) {
+    pub fn into_inner(self) -> (u64, VirtualFile) {
         (self.bytes_amount, self.dst)
     }
 }
 
-impl<W> OwnedAsyncWriter for Writer<W>
-where
-    W: OwnedAsyncWriter,
-{
+impl OwnedAsyncWriter for Writer {
     #[inline(always)]
     async fn write_all<B: BoundedBuf<Buf = Buf>, Buf: IoBuf + Send>(
         &mut self,
         buf: B,
     ) -> std::io::Result<(usize, B::Buf)> {
-        let (nwritten, buf) = self.dst.write_all(buf).await?;
+        let (buf, res) = self.dst.write_all(buf).await;
+        let nwritten = res?;
         self.bytes_amount += u64::try_from(nwritten).unwrap();
         Ok((nwritten, buf))
     }

pageserver/src/virtual_file/owned_buffers_io/write.rs

@@ -10,14 +10,14 @@ pub trait OwnedAsyncWriter {
     ) -> std::io::Result<(usize, B::Buf)>;
 }
 
-/// A wrapper aorund an [`OwnedAsyncWriter`] that uses a [`Buffer`] to batch
-/// small writes into larger writes of size [`Buffer::cap`].
+/// A wrapper aorund an [`OwnedAsyncWriter`] that batches smaller writers
+/// into `BUFFER_SIZE`-sized writes.
 ///
 /// # Passthrough Of Large Writers
 ///
-/// Calls to [`BufferedWriter::write_buffered`] that are larger than [`Buffer::cap`]
-/// cause the internal buffer to be flushed prematurely so that the large
-/// buffered write is passed through to the underlying [`OwnedAsyncWriter`].
+/// Buffered writes larger than the `BUFFER_SIZE` cause the internal
+/// buffer to be flushed, even if it is not full yet. Then, the large
+/// buffered write is passed through to the unerlying [`OwnedAsyncWriter`].
 ///
 /// This pass-through is generally beneficial for throughput, but if
 /// the storage backend of the [`OwnedAsyncWriter`] is a shared resource,
@@ -25,38 +25,27 @@ pub trait OwnedAsyncWriter {
 ///
 /// In such cases, a different implementation that always buffers in memory
 /// may be preferable.
-pub struct BufferedWriter<B, W> {
+pub struct BufferedWriter<const BUFFER_SIZE: usize, W> {
     writer: W,
-    /// invariant: always remains Some(buf) except
-    /// - while IO is ongoing => goes back to Some() once the IO completed successfully
-    /// - after an IO error => stays `None` forever
-    /// In these exceptional cases, it's `None`.
-    buf: Option<B>,
+    // invariant: always remains Some(buf)
+    // with buf.capacity() == BUFFER_SIZE except
+    // - while IO is ongoing => goes back to Some() once the IO completed successfully
+    // - after an IO error => stays `None` forever
+    // In these exceptional cases, it's `None`.
+    buf: Option<BytesMut>,
 }
 
-impl<B, Buf, W> BufferedWriter<B, W>
+impl<const BUFFER_SIZE: usize, W> BufferedWriter<BUFFER_SIZE, W>
 where
-    B: Buffer<IoBuf = Buf> + Send,
-    Buf: IoBuf + Send,
     W: OwnedAsyncWriter,
 {
-    pub fn new(writer: W, buf: B) -> Self {
+    pub fn new(writer: W) -> Self {
         Self {
             writer,
-            buf: Some(buf),
+            buf: Some(BytesMut::with_capacity(BUFFER_SIZE)),
         }
     }
 
-    pub fn as_inner(&self) -> &W {
-        &self.writer
-    }
-
-    /// Panics if used after any of the write paths returned an error
-    pub fn inspect_buffer(&self) -> &B {
-        self.buf()
-    }
-
-    #[cfg_attr(target_os = "macos", allow(dead_code))]
     pub async fn flush_and_into_inner(mut self) -> std::io::Result<W> {
         self.flush().await?;
         let Self { buf, writer } = self;
@@ -64,144 +53,61 @@ where
         Ok(writer)
     }
 
-    #[inline(always)]
-    fn buf(&self) -> &B {
-        self.buf
-            .as_ref()
-            .expect("must not use after we returned an error")
-    }
-
-    #[cfg_attr(target_os = "macos", allow(dead_code))]
-    pub async fn write_buffered<S: IoBuf>(&mut self, chunk: Slice<S>) -> std::io::Result<(usize, S)>
+    pub async fn write_buffered<B: IoBuf>(&mut self, chunk: Slice<B>) -> std::io::Result<()>
     where
-        S: IoBuf + Send,
+        B: IoBuf + Send,
     {
-        let chunk_len = chunk.len();
         // avoid memcpy for the middle of the chunk
-        if chunk.len() >= self.buf().cap() {
+        if chunk.len() >= BUFFER_SIZE {
             self.flush().await?;
             // do a big write, bypassing `buf`
             assert_eq!(
                 self.buf
                     .as_ref()
                     .expect("must not use after an error")
-                    .pending(),
+                    .len(),
                 0
             );
+            let chunk_len = chunk.len();
             let (nwritten, chunk) = self.writer.write_all(chunk).await?;
             assert_eq!(nwritten, chunk_len);
-            return Ok((nwritten, chunk));
+            drop(chunk);
+            return Ok(());
         }
         // in-memory copy the < BUFFER_SIZED tail of the chunk
-        assert!(chunk.len() < self.buf().cap());
-        let mut slice = &chunk[..];
-        while !slice.is_empty() {
-            let buf = self.buf.as_mut().expect("must not use after an error");
-            let need = buf.cap() - buf.pending();
-            let have = slice.len();
-            let n = std::cmp::min(need, have);
-            buf.extend_from_slice(&slice[..n]);
-            slice = &slice[n..];
-            if buf.pending() >= buf.cap() {
-                assert_eq!(buf.pending(), buf.cap());
-                self.flush().await?;
-            }
-        }
-        assert!(slice.is_empty(), "by now we should have drained the chunk");
-        Ok((chunk_len, chunk.into_inner()))
-    }
-
-    /// Strictly less performant variant of [`Self::write_buffered`] that allows writing borrowed data.
-    ///
-    /// It is less performant because we always have to copy the borrowed data into the internal buffer
-    /// before we can do the IO. The [`Self::write_buffered`] can avoid this, which is more performant
-    /// for large writes.
-    pub async fn write_buffered_borrowed(&mut self, mut chunk: &[u8]) -> std::io::Result<usize> {
-        let chunk_len = chunk.len();
+        assert!(chunk.len() < BUFFER_SIZE);
+        let mut chunk = &chunk[..];
         while !chunk.is_empty() {
             let buf = self.buf.as_mut().expect("must not use after an error");
-            let need = buf.cap() - buf.pending();
+            let need = BUFFER_SIZE - buf.len();
             let have = chunk.len();
             let n = std::cmp::min(need, have);
             buf.extend_from_slice(&chunk[..n]);
             chunk = &chunk[n..];
-            if buf.pending() >= buf.cap() {
-                assert_eq!(buf.pending(), buf.cap());
+            if buf.len() >= BUFFER_SIZE {
+                assert_eq!(buf.len(), BUFFER_SIZE);
                 self.flush().await?;
             }
         }
-        Ok(chunk_len)
+        assert!(chunk.is_empty(), "by now we should have drained the chunk");
+        Ok(())
     }
 
     async fn flush(&mut self) -> std::io::Result<()> {
         let buf = self.buf.take().expect("must not use after an error");
-        let buf_len = buf.pending();
-        if buf_len == 0 {
+        if buf.is_empty() {
             self.buf = Some(buf);
-            return Ok(());
+            return std::io::Result::Ok(());
         }
-        let (nwritten, io_buf) = self.writer.write_all(buf.flush()).await?;
+        let buf_len = buf.len();
+        let (nwritten, mut buf) = self.writer.write_all(buf).await?;
         assert_eq!(nwritten, buf_len);
-        self.buf = Some(Buffer::reuse_after_flush(io_buf));
+        buf.clear();
+        self.buf = Some(buf);
         Ok(())
     }
 }
 
-/// A [`Buffer`] is used by [`BufferedWriter`] to batch smaller writes into larger ones.
-pub trait Buffer {
-    type IoBuf: IoBuf;
-
-    /// Capacity of the buffer. Must not change over the lifetime `self`.`
-    fn cap(&self) -> usize;
-
-    /// Add data to the buffer.
-    /// Panics if there is not enough room to accomodate `other`'s content, i.e.,
-    /// panics if `other.len() > self.cap() - self.pending()`.
-    fn extend_from_slice(&mut self, other: &[u8]);
-
-    /// Number of bytes in the buffer.
-    fn pending(&self) -> usize;
-
-    /// Turns `self` into a [`tokio_epoll_uring::Slice`] of the pending data
-    /// so we can use [`tokio_epoll_uring`] to write it to disk.
-    fn flush(self) -> Slice<Self::IoBuf>;
-
-    /// After the write to disk is done and we have gotten back the slice,
-    /// [`BufferedWriter`] uses this method to re-use the io buffer.
-    fn reuse_after_flush(iobuf: Self::IoBuf) -> Self;
-}
-
-impl Buffer for BytesMut {
-    type IoBuf = BytesMut;
-
-    #[inline(always)]
-    fn cap(&self) -> usize {
-        self.capacity()
-    }
-
-    fn extend_from_slice(&mut self, other: &[u8]) {
-        BytesMut::extend_from_slice(self, other)
-    }
-
-    #[inline(always)]
-    fn pending(&self) -> usize {
-        self.len()
-    }
-
-    fn flush(self) -> Slice<BytesMut> {
-        if self.is_empty() {
-            return self.slice_full();
-        }
-        let len = self.len();
-        self.slice(0..len)
-    }
-
-    fn reuse_after_flush(mut iobuf: BytesMut) -> Self {
-        iobuf.clear();
-        iobuf
-    }
-}
-
 impl OwnedAsyncWriter for Vec<u8> {
     async fn write_all<B: BoundedBuf<Buf = Buf>, Buf: IoBuf + Send>(
         &mut self,
@@ -219,8 +125,6 @@ impl OwnedAsyncWriter for Vec<u8> {
 
 #[cfg(test)]
 mod tests {
-    use bytes::BytesMut;
-
     use super::*;
 
     #[derive(Default)]
@@ -254,7 +158,7 @@ mod tests {
     #[tokio::test]
     async fn test_buffered_writes_only() -> std::io::Result<()> {
         let recorder = RecorderWriter::default();
-        let mut writer = BufferedWriter::new(recorder, BytesMut::with_capacity(2));
+        let mut writer = BufferedWriter::<2, _>::new(recorder);
         write!(writer, b"a");
         write!(writer, b"b");
         write!(writer, b"c");
@@ -271,7 +175,7 @@ mod tests {
     #[tokio::test]
     async fn test_passthrough_writes_only() -> std::io::Result<()> {
         let recorder = RecorderWriter::default();
-        let mut writer = BufferedWriter::new(recorder, BytesMut::with_capacity(2));
+        let mut writer = BufferedWriter::<2, _>::new(recorder);
         write!(writer, b"abc");
         write!(writer, b"de");
         write!(writer, b"");
@@ -287,7 +191,7 @@ mod tests {
     #[tokio::test]
     async fn test_passthrough_write_with_nonempty_buffer() -> std::io::Result<()> {
         let recorder = RecorderWriter::default();
-        let mut writer = BufferedWriter::new(recorder, BytesMut::with_capacity(2));
+        let mut writer = BufferedWriter::<2, _>::new(recorder);
         write!(writer, b"a");
         write!(writer, b"bc");
         write!(writer, b"d");
@@ -299,31 +203,4 @@ mod tests {
         );
         Ok(())
     }
-
-    #[tokio::test]
-    async fn test_write_all_borrowed_always_goes_through_buffer() -> std::io::Result<()> {
-        let recorder = RecorderWriter::default();
-        let mut writer = BufferedWriter::new(recorder, BytesMut::with_capacity(2));
-
-        writer.write_buffered_borrowed(b"abc").await?;
-        writer.write_buffered_borrowed(b"d").await?;
-        writer.write_buffered_borrowed(b"e").await?;
-        writer.write_buffered_borrowed(b"fg").await?;
-        writer.write_buffered_borrowed(b"hi").await?;
-        writer.write_buffered_borrowed(b"j").await?;
-        writer.write_buffered_borrowed(b"klmno").await?;
-
-        let recorder = writer.flush_and_into_inner().await?;
-        assert_eq!(
-            recorder.writes,
-            {
-                let expect: &[&[u8]] = &[b"ab", b"cd", b"ef", b"gh", b"ij", b"kl", b"mn", b"o"];
-                expect
-            }
-            .iter()
-            .map(|v| v[..].to_vec())
-            .collect::<Vec<_>>()
-        );
-        Ok(())
-    }
 }

pageserver/src/walingest.rs

@@ -1034,7 +1034,7 @@ impl WalIngest {
 
             let nblocks = modification
                 .tline
-                .get_rel_size(src_rel, Version::Modified(modification), ctx)
+                .get_rel_size(src_rel, Version::Modified(modification), true, ctx)
                 .await?;
             let dst_rel = RelTag {
                 spcnode: tablespace_id,
@@ -1068,7 +1068,13 @@ impl WalIngest {
 
                 let content = modification
                     .tline
-                    .get_rel_page_at_lsn(src_rel, blknum, Version::Modified(modification), ctx)
+                    .get_rel_page_at_lsn(
+                        src_rel,
+                        blknum,
+                        Version::Modified(modification),
+                        true,
+                        ctx,
+                    )
                     .await?;
                 modification.put_rel_page_image(dst_rel, blknum, content)?;
                 num_blocks_copied += 1;
@@ -1236,7 +1242,7 @@ impl WalIngest {
                 };
                 if modification
                     .tline
-                    .get_rel_exists(rel, Version::Modified(modification), ctx)
+                    .get_rel_exists(rel, Version::Modified(modification), true, ctx)
                     .await?
                 {
                     self.put_rel_drop(modification, rel, ctx).await?;
@@ -1535,7 +1541,7 @@ impl WalIngest {
             nblocks
         } else if !modification
             .tline
-            .get_rel_exists(rel, Version::Modified(modification), ctx)
+            .get_rel_exists(rel, Version::Modified(modification), true, ctx)
             .await?
         {
             // create it with 0 size initially, the logic below will extend it
@@ -1547,7 +1553,7 @@ impl WalIngest {
         } else {
             modification
                 .tline
-                .get_rel_size(rel, Version::Modified(modification), ctx)
+                .get_rel_size(rel, Version::Modified(modification), true, ctx)
                 .await?
         };
 
@@ -1644,14 +1650,14 @@ async fn get_relsize(
 ) -> anyhow::Result<BlockNumber> {
     let nblocks = if !modification
         .tline
-        .get_rel_exists(rel, Version::Modified(modification), ctx)
+        .get_rel_exists(rel, Version::Modified(modification), true, ctx)
         .await?
     {
         0
     } else {
         modification
             .tline
-            .get_rel_size(rel, Version::Modified(modification), ctx)
+            .get_rel_size(rel, Version::Modified(modification), true, ctx)
             .await?
     };
     Ok(nblocks)
@@ -1726,29 +1732,29 @@ mod tests {
         // The relation was created at LSN 2, not visible at LSN 1 yet.
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x10)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x10)), false, &ctx)
                 .await?,
             false
         );
         assert!(tline
-            .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x10)), &ctx)
+            .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x10)), false, &ctx)
             .await
             .is_err());
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x20)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x20)), false, &ctx)
                 .await?,
             true
         );
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x20)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x20)), false, &ctx)
                 .await?,
             1
         );
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x50)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x50)), false, &ctx)
                 .await?,
             3
         );
@@ -1756,46 +1762,46 @@ mod tests {
         // Check page contents at each LSN
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x20)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x20)), false, &ctx)
                 .await?,
             test_img("foo blk 0 at 2")
         );
 
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x30)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x30)), false, &ctx)
                 .await?,
             test_img("foo blk 0 at 3")
         );
 
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x40)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x40)), false, &ctx)
                 .await?,
             test_img("foo blk 0 at 3")
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x40)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x40)), false, &ctx)
                 .await?,
             test_img("foo blk 1 at 4")
         );
 
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x50)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x50)), false, &ctx)
                 .await?,
             test_img("foo blk 0 at 3")
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x50)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x50)), false, &ctx)
                 .await?,
             test_img("foo blk 1 at 4")
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 2, Version::Lsn(Lsn(0x50)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 2, Version::Lsn(Lsn(0x50)), false, &ctx)
                 .await?,
             test_img("foo blk 2 at 5")
         );
@@ -1811,19 +1817,19 @@ mod tests {
         // Check reported size and contents after truncation
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x60)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x60)), false, &ctx)
                 .await?,
             2
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x60)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x60)), false, &ctx)
                 .await?,
             test_img("foo blk 0 at 3")
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x60)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x60)), false, &ctx)
                 .await?,
             test_img("foo blk 1 at 4")
         );
@@ -1831,13 +1837,13 @@ mod tests {
         // should still see the truncated block with older LSN
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x50)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x50)), false, &ctx)
                 .await?,
             3
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 2, Version::Lsn(Lsn(0x50)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 2, Version::Lsn(Lsn(0x50)), false, &ctx)
                 .await?,
             test_img("foo blk 2 at 5")
         );
@@ -1850,7 +1856,7 @@ mod tests {
         m.commit(&ctx).await?;
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x68)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x68)), false, &ctx)
                 .await?,
             0
         );
@@ -1863,19 +1869,19 @@ mod tests {
         m.commit(&ctx).await?;
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x70)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x70)), false, &ctx)
                 .await?,
             2
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x70)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 0, Version::Lsn(Lsn(0x70)), false, &ctx)
                 .await?,
             ZERO_PAGE
         );
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x70)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 1, Version::Lsn(Lsn(0x70)), false, &ctx)
                 .await?,
             test_img("foo blk 1")
         );
@@ -1888,21 +1894,21 @@ mod tests {
         m.commit(&ctx).await?;
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x80)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x80)), false, &ctx)
                 .await?,
             1501
         );
         for blk in 2..1500 {
             assert_eq!(
                 tline
-                    .get_rel_page_at_lsn(TESTREL_A, blk, Version::Lsn(Lsn(0x80)), &ctx)
+                    .get_rel_page_at_lsn(TESTREL_A, blk, Version::Lsn(Lsn(0x80)), false, &ctx)
                     .await?,
                 ZERO_PAGE
             );
         }
         assert_eq!(
             tline
-                .get_rel_page_at_lsn(TESTREL_A, 1500, Version::Lsn(Lsn(0x80)), &ctx)
+                .get_rel_page_at_lsn(TESTREL_A, 1500, Version::Lsn(Lsn(0x80)), false, &ctx)
                 .await?,
             test_img("foo blk 1500")
         );
@@ -1929,13 +1935,13 @@ mod tests {
         // Check that rel exists and size is correct
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x20)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x20)), false, &ctx)
                 .await?,
             true
         );
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x20)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x20)), false, &ctx)
                 .await?,
             1
         );
@@ -1948,7 +1954,7 @@ mod tests {
         // Check that rel is not visible anymore
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x30)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x30)), false, &ctx)
                 .await?,
             false
         );
@@ -1966,13 +1972,13 @@ mod tests {
         // Check that rel exists and size is correct
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x40)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x40)), false, &ctx)
                 .await?,
             true
         );
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x40)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x40)), false, &ctx)
                 .await?,
             1
         );
@@ -2005,24 +2011,24 @@ mod tests {
         // The relation was created at LSN 20, not visible at LSN 1 yet.
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x10)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x10)), false, &ctx)
                 .await?,
             false
         );
         assert!(tline
-            .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x10)), &ctx)
+            .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x10)), false, &ctx)
             .await
             .is_err());
 
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x20)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x20)), false, &ctx)
                 .await?,
             true
         );
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x20)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x20)), false, &ctx)
                 .await?,
             relsize
         );
@@ -2033,7 +2039,7 @@ mod tests {
             let data = format!("foo blk {} at {}", blkno, lsn);
             assert_eq!(
                 tline
-                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(lsn), &ctx)
+                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(lsn), false, &ctx)
                     .await?,
                 test_img(&data)
             );
@@ -2050,7 +2056,7 @@ mod tests {
         // Check reported size and contents after truncation
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x60)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x60)), false, &ctx)
                 .await?,
             1
         );
@@ -2060,7 +2066,7 @@ mod tests {
             let data = format!("foo blk {} at {}", blkno, lsn);
             assert_eq!(
                 tline
-                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(Lsn(0x60)), &ctx)
+                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(Lsn(0x60)), false, &ctx)
                     .await?,
                 test_img(&data)
             );
@@ -2069,7 +2075,7 @@ mod tests {
         // should still see all blocks with older LSN
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x50)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x50)), false, &ctx)
                 .await?,
             relsize
         );
@@ -2078,7 +2084,7 @@ mod tests {
             let data = format!("foo blk {} at {}", blkno, lsn);
             assert_eq!(
                 tline
-                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(Lsn(0x50)), &ctx)
+                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(Lsn(0x50)), false, &ctx)
                     .await?,
                 test_img(&data)
             );
@@ -2098,13 +2104,13 @@ mod tests {
 
         assert_eq!(
             tline
-                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x80)), &ctx)
+                .get_rel_exists(TESTREL_A, Version::Lsn(Lsn(0x80)), false, &ctx)
                 .await?,
             true
         );
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x80)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(0x80)), false, &ctx)
                 .await?,
             relsize
         );
@@ -2114,7 +2120,7 @@ mod tests {
             let data = format!("foo blk {} at {}", blkno, lsn);
             assert_eq!(
                 tline
-                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(Lsn(0x80)), &ctx)
+                    .get_rel_page_at_lsn(TESTREL_A, blkno, Version::Lsn(Lsn(0x80)), false, &ctx)
                     .await?,
                 test_img(&data)
             );
@@ -2148,7 +2154,7 @@ mod tests {
 
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), false, &ctx)
                 .await?,
             RELSEG_SIZE + 1
         );
@@ -2162,7 +2168,7 @@ mod tests {
         m.commit(&ctx).await?;
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), false, &ctx)
                 .await?,
             RELSEG_SIZE
         );
@@ -2177,7 +2183,7 @@ mod tests {
         m.commit(&ctx).await?;
         assert_eq!(
             tline
-                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), &ctx)
+                .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), false, &ctx)
                 .await?,
             RELSEG_SIZE - 1
         );
@@ -2195,7 +2201,7 @@ mod tests {
             m.commit(&ctx).await?;
             assert_eq!(
                 tline
-                    .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), &ctx)
+                    .get_rel_size(TESTREL_A, Version::Lsn(Lsn(lsn)), false, &ctx)
                     .await?,
                 size as BlockNumber
             );

pgxn/neon/libpagestore.c

@@ -49,8 +49,6 @@ char	   *neon_auth_token;
 int			readahead_buffer_size = 128;
 int			flush_every_n_requests = 8;
 
-int         neon_protocol_version = 1;
-
 static int	n_reconnect_attempts = 0;
 static int	max_reconnect_attempts = 60;
 static int	stripe_size;
@@ -381,17 +379,7 @@ pageserver_connect(shardno_t shard_no, int elevel)
 		pfree(msg);
 		return false;
 	}
-	switch (neon_protocol_version)
-	{
-		case 2:
-			query = psprintf("pagestream_v2 %s %s", neon_tenant, neon_timeline);
-			break;
-		case 1:
-			query = psprintf("pagestream %s %s", neon_tenant, neon_timeline);
-			break;
-		default:
-			elog(ERROR, "unexpected neon_protocol_version %d", neon_protocol_version);
-	}
+	query = psprintf("pagestream %s %s", neon_tenant, neon_timeline);
 	ret = PQsendQuery(conn, query);
 	pfree(query);
 	if (ret != 1)
@@ -452,7 +440,7 @@ pageserver_connect(shardno_t shard_no, int elevel)
 		return false;
 	}
 
-	neon_shard_log(shard_no, LOG, "libpagestore: connected to '%s' with protocol version %d", connstr, neon_protocol_version);
+	neon_shard_log(shard_no, LOG, "libpagestore: connected to '%s'", connstr);
 	page_servers[shard_no].conn = conn;
 	page_servers[shard_no].wes = wes;
 
@@ -856,16 +844,6 @@ pg_init_libpagestore(void)
 							PGC_USERSET,
 							0,	/* no flags required */
 							NULL, (GucIntAssignHook) &readahead_buffer_resize, NULL);
-	DefineCustomIntVariable("neon.protocol_version",
-							"Version of compute<->page server protocol",
-							NULL,
-							&neon_protocol_version,
-							1, /* default to old protocol for now */
-							1, /* min */
-							2, /* max */
-							PGC_SU_BACKEND,
-							0,	/* no flags required */
-							NULL, NULL, NULL);
 
 	relsize_hash_init();
 

pgxn/neon/pagestore_client.h

@@ -69,33 +69,18 @@ typedef enum {
 	SLRU_MULTIXACT_OFFSETS
 } SlruKind;
 
-/*--
- * supertype of all the Neon*Request structs below.
+/*
+ * supertype of all the Neon*Request structs below
  *
- * All requests contain two LSNs:
- *
- * lsn:                request page (or relation size, etc) at this LSN
- * not_modified_since: Hint that the page hasn't been modified between
- *                     this LSN and the request LSN (`lsn`).
- *
- * To request the latest version of a page, you can use MAX_LSN as the request
- * LSN.
- *
- * If you don't know any better, you can always set 'not_modified_since' equal
- * to 'lsn', but providing a lower value can speed up processing the request
- * in the pageserver, as it doesn't need to wait for the WAL to arrive, and it
- * can skip traversing through recent layers which we know to not contain any
- * versions for the requested page.
- *
- * These structs describe the V2 of these requests. The old V1 protocol contained
- * just one LSN and a boolean 'latest' flag. If the neon_protocol_version GUC is
- * set to 1, we will convert these to the V1 requests before sending.
+ * If 'latest' is true, we are requesting the latest page version, and 'lsn'
+ * is just a hint to the server that we know there are no versions of the page
+ * (or relation size, for exists/nblocks requests) later than the 'lsn'.
  */
 typedef struct
 {
 	NeonMessageTag tag;
-	XLogRecPtr	lsn;
-	XLogRecPtr	not_modified_since;
+	bool		latest;			/* if true, request latest page version */
+	XLogRecPtr	lsn;			/* request page version @ this LSN */
 } NeonRequest;
 
 typedef struct
@@ -208,7 +193,6 @@ extern int	readahead_buffer_size;
 extern char *neon_timeline;
 extern char *neon_tenant;
 extern int32 max_cluster_size;
-extern int  neon_protocol_version;
 
 extern shardno_t get_shard_number(BufferTag* tag);
 
@@ -241,14 +225,14 @@ extern bool neon_prefetch(SMgrRelation reln, ForkNumber forknum,
 extern void neon_read(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
 					  char *buffer);
 extern PGDLLEXPORT void neon_read_at_lsn(NRelFileInfo rnode, ForkNumber forkNum, BlockNumber blkno,
-										 XLogRecPtr request_lsn, XLogRecPtr not_modified_since, char *buffer);
+										 XLogRecPtr request_lsn, bool request_latest, char *buffer);
 extern void neon_write(SMgrRelation reln, ForkNumber forknum,
 					   BlockNumber blocknum, char *buffer, bool skipFsync);
 #else
 extern void neon_read(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
 					  void *buffer);
 extern PGDLLEXPORT void neon_read_at_lsn(NRelFileInfo rnode, ForkNumber forkNum, BlockNumber blkno,
-										 XLogRecPtr request_lsn, XLogRecPtr not_modified_since, void *buffer);
+										 XLogRecPtr request_lsn, bool request_latest, void *buffer);
 extern void neon_write(SMgrRelation reln, ForkNumber forknum,
 					   BlockNumber blocknum, const void *buffer, bool skipFsync);
 #endif

pgxn/neon/pagestore_smgr.c

@@ -168,8 +168,8 @@ typedef enum PrefetchStatus
 typedef struct PrefetchRequest
 {
 	BufferTag	buftag;			/* must be first entry in the struct */
-	XLogRecPtr	request_lsn;
-	XLogRecPtr	not_modified_since;
+	XLogRecPtr	effective_request_lsn;
+	XLogRecPtr	actual_request_lsn;
 	NeonResponse *response;		/* may be null */
 	PrefetchStatus status;
 	shardno_t   shard_no;
@@ -269,19 +269,19 @@ static PrefetchState *MyPState;
 	) \
 )
 
+static XLogRecPtr prefetch_lsn = 0;
+
 static bool compact_prefetch_buffers(void);
 static void consume_prefetch_responses(void);
-static uint64 prefetch_register_buffer(BufferTag tag, XLogRecPtr *force_request_lsn, XLogRecPtr *force_not_modified_since);
+static uint64 prefetch_register_buffer(BufferTag tag, bool *force_latest, XLogRecPtr *force_lsn);
 static bool prefetch_read(PrefetchRequest *slot);
-static void prefetch_do_request(PrefetchRequest *slot, XLogRecPtr *force_request_lsn, XLogRecPtr *force_not_modified_since);
+static void prefetch_do_request(PrefetchRequest *slot, bool *force_latest, XLogRecPtr *force_lsn);
 static bool prefetch_wait_for(uint64 ring_index);
 static void prefetch_cleanup_trailing_unused(void);
 static inline void prefetch_set_unused(uint64 ring_index);
 
-static void neon_get_request_lsn(NRelFileInfo rinfo, ForkNumber forknum, BlockNumber blkno,
-								 XLogRecPtr *request_lsn, XLogRecPtr *not_modified_since);
-static bool neon_prefetch_response_usable(XLogRecPtr request_lsn, XLogRecPtr not_modified_since,
-										  PrefetchRequest *slot);
+static XLogRecPtr neon_get_request_lsn(bool *latest, NRelFileInfo rinfo,
+									   ForkNumber forknum, BlockNumber blkno);
 
 static bool
 compact_prefetch_buffers(void)
@@ -338,8 +338,8 @@ compact_prefetch_buffers(void)
 		target_slot->shard_no = source_slot->shard_no;
 		target_slot->status = source_slot->status;
 		target_slot->response = source_slot->response;
-		target_slot->request_lsn = source_slot->request_lsn;
-		target_slot->not_modified_since = source_slot->not_modified_since;
+		target_slot->effective_request_lsn = source_slot->effective_request_lsn;
+		target_slot->actual_request_lsn = source_slot->actual_request_lsn;
 		target_slot->my_ring_index = empty_ring_index;
 
 		prfh_delete(MyPState->prf_hash, source_slot);
@@ -358,8 +358,7 @@ compact_prefetch_buffers(void)
 		};
 		source_slot->response = NULL;
 		source_slot->my_ring_index = 0;
-		source_slot->request_lsn = InvalidXLogRecPtr;
-		source_slot->not_modified_since = InvalidXLogRecPtr;
+		source_slot->effective_request_lsn = 0;
 
 		/* update bookkeeping */
 		n_moved++;
@@ -684,39 +683,56 @@ prefetch_set_unused(uint64 ring_index)
 		compact_prefetch_buffers();
 }
 
-/*
- * Send one prefetch request to the pageserver. To wait for the response, call
- * prefetch_wait_for().
- */
 static void
-prefetch_do_request(PrefetchRequest *slot, XLogRecPtr *force_request_lsn, XLogRecPtr *force_not_modified_since)
+prefetch_do_request(PrefetchRequest *slot, bool *force_latest, XLogRecPtr *force_lsn)
 {
 	bool		found;
 	NeonGetPageRequest request = {
 		.req.tag = T_NeonGetPageRequest,
-		/* lsn and not_modified_since are filled in below */
+		.req.latest = false,
+		.req.lsn = 0,
 		.rinfo = BufTagGetNRelFileInfo(slot->buftag),
 		.forknum = slot->buftag.forkNum,
 		.blkno = slot->buftag.blockNum,
 	};
 
-	Assert(((force_request_lsn != NULL) == (force_not_modified_since != NULL)));
-
-	if (force_request_lsn)
+	if (force_lsn && force_latest)
 	{
-		request.req.lsn = *force_request_lsn;
-		request.req.not_modified_since = *force_not_modified_since;
+		request.req.lsn = *force_lsn;
+		request.req.latest = *force_latest;
+		slot->actual_request_lsn = slot->effective_request_lsn = *force_lsn;
 	}
 	else
 	{
-		neon_get_request_lsn(BufTagGetNRelFileInfo(slot->buftag),
-							 slot->buftag.forkNum,
-							 slot->buftag.blockNum,
-							 &request.req.lsn,
-							 &request.req.not_modified_since);
+		XLogRecPtr	lsn = neon_get_request_lsn(
+											   &request.req.latest,
+											   BufTagGetNRelFileInfo(slot->buftag),
+											   slot->buftag.forkNum,
+											   slot->buftag.blockNum
+			);
+
+		/*
+		 * Note: effective_request_lsn is potentially higher than the
+		 * requested LSN, but still correct:
+		 *
+		 * We know there are no changes between the actual requested LSN and
+		 * the value of effective_request_lsn: If there were, the page would
+		 * have been in cache and evicted between those LSN values, which then
+		 * would have had to result in a larger request LSN for this page.
+		 *
+		 * It is possible that a concurrent backend loads the page, modifies
+		 * it and then evicts it again, but the LSN of that eviction cannot be
+		 * smaller than the current WAL insert/redo pointer, which is already
+		 * larger than this prefetch_lsn. So in any case, that would
+		 * invalidate this cache.
+		 *
+		 * The best LSN to use for effective_request_lsn would be
+		 * XLogCtl->Insert.RedoRecPtr, but that's expensive to access.
+		 */
+		slot->actual_request_lsn = request.req.lsn = lsn;
+		prefetch_lsn = Max(prefetch_lsn, lsn);
+		slot->effective_request_lsn = prefetch_lsn;
 	}
-	slot->request_lsn = request.req.lsn;
-	slot->not_modified_since = request.req.not_modified_since;
 
 	Assert(slot->response == NULL);
 	Assert(slot->my_ring_index == MyPState->ring_unused);
@@ -733,6 +749,7 @@ prefetch_do_request(PrefetchRequest *slot, XLogRecPtr *force_request_lsn, XLogRe
 	/* update slot state */
 	slot->status = PRFS_REQUESTED;
 
+
 	prfh_insert(MyPState->prf_hash, slot, &found);
 	Assert(!found);
 }
@@ -742,25 +759,22 @@ prefetch_do_request(PrefetchRequest *slot, XLogRecPtr *force_request_lsn, XLogRe
  *
  * Register that we may want the contents of BufferTag in the near future.
  *
- * If force_request_lsn and force_not_modified_since are not NULL, those
- * values are sent to the pageserver. If they are NULL, we utilize the
- * lastWrittenLsn -infrastructure to fill them in.
+ * If force_latest and force_lsn are not NULL, those values are sent to the
+ * pageserver. If they are NULL, we utilize the lastWrittenLsn -infrastructure
+ * to fill in these values manually.
  *
  * NOTE: this function may indirectly update MyPState->pfs_hash; which
  * invalidates any active pointers into the hash table.
  */
 
 static uint64
-prefetch_register_buffer(BufferTag tag, XLogRecPtr *force_request_lsn,
-						 XLogRecPtr *force_not_modified_since)
+prefetch_register_buffer(BufferTag tag, bool *force_latest, XLogRecPtr *force_lsn)
 {
 	uint64		ring_index;
 	PrefetchRequest req;
 	PrefetchRequest *slot;
 	PrfHashEntry *entry;
 
-	Assert(((force_request_lsn != NULL) == (force_not_modified_since != NULL)));
-
 	/* use an intermediate PrefetchRequest struct to ensure correct alignment */
 	req.buftag = tag;
 Retry:
@@ -778,19 +792,40 @@ Retry:
 		Assert(BUFFERTAGS_EQUAL(slot->buftag, tag));
 
 		/*
-		 * If the caller specified a request LSN to use, only accept prefetch
-		 * responses that satisfy that request.
+		 * If we want a specific lsn, we do not accept requests that were made
+		 * with a potentially different LSN.
 		 */
-		if (force_request_lsn)
+		if (force_latest && force_lsn)
 		{
-			if (!neon_prefetch_response_usable(*force_request_lsn,
-											   *force_not_modified_since, slot))
+			/*
+			 * if we want the latest version, any effective_request_lsn <
+			 * request lsn is OK
+			 */
+			if (*force_latest)
 			{
-				/* Wait for the old request to finish and discard it */
-				if (!prefetch_wait_for(ring_index))
-					goto Retry;
-				prefetch_set_unused(ring_index);
-				entry = NULL;
+				if (*force_lsn > slot->effective_request_lsn)
+				{
+					if (!prefetch_wait_for(ring_index))
+						goto Retry;
+					prefetch_set_unused(ring_index);
+					entry = NULL;
+				}
+
+			}
+
+			/*
+			 * if we don't want the latest version, only accept requests with
+			 * the exact same LSN
+			 */
+			else
+			{
+				if (*force_lsn != slot->effective_request_lsn)
+				{
+					if (!prefetch_wait_for(ring_index))
+						goto Retry;
+					prefetch_set_unused(ring_index);
+					entry = NULL;
+				}
 			}
 		}
 
@@ -886,7 +921,7 @@ Retry:
 	slot->shard_no = get_shard_number(&tag);
 	slot->my_ring_index = ring_index;
 
-	prefetch_do_request(slot, force_request_lsn, force_not_modified_since);
+	prefetch_do_request(slot, force_latest, force_lsn);
 	Assert(slot->status == PRFS_REQUESTED);
 	Assert(MyPState->ring_last <= ring_index &&
 		   ring_index < MyPState->ring_unused);
@@ -915,7 +950,7 @@ page_server_request(void const *req)
 	BufferTag tag = {0};
 	shardno_t shard_no;
 
-	switch (messageTag(req))
+	switch (((NeonRequest *) req)->tag)
 	{
 		case T_NeonExistsRequest:
 			CopyNRelFileInfoToBufTag(tag, ((NeonExistsRequest *) req)->rinfo);
@@ -931,10 +966,11 @@ page_server_request(void const *req)
 			tag.blockNum = ((NeonGetPageRequest *) req)->blkno;
 			break;
 		default:
-			neon_log(ERROR, "Unexpected request tag: %d", messageTag(req));
+			neon_log(ERROR, "Unexpected request tag: %d", ((NeonRequest *) req)->tag);
 	}
 	shard_no = get_shard_number(&tag);
 
+
 	/*
 	 * Current sharding model assumes that all metadata is present only at shard 0.
 	 * We still need to call get_shard_no() to check if shard map is up-to-date.
@@ -961,52 +997,8 @@ nm_pack_request(NeonRequest *msg)
 	StringInfoData s;
 
 	initStringInfo(&s);
+	pq_sendbyte(&s, msg->tag);
 
-	if (neon_protocol_version >= 2)
-	{
-		pq_sendbyte(&s, msg->tag);
-		pq_sendint64(&s, msg->lsn);
-		pq_sendint64(&s, msg->not_modified_since);
-	}
-	else
-	{
-		bool		latest;
-		XLogRecPtr	lsn;
-
-		/*
-		 * In primary, we always request the latest page version.
-		 */
-		if (!RecoveryInProgress())
-		{
-			latest = true;
-			lsn = msg->not_modified_since;
-		}
-		else
-		{
-			/*
-			 * In the protocol V1, we cannot represent that we want to read
-			 * page at LSN X, and we know that it hasn't been modified since
-			 * Y. We can either use 'not_modified_lsn' as the request LSN, and
-			 * risk getting an error if that LSN is too old and has already
-			 * fallen out of the pageserver's GC horizon, or we can send
-			 * 'request_lsn', causing the pageserver to possibly wait for the
-			 * recent WAL to arrive unnecessarily. Or something in between. We
-			 * choose to use the old LSN and risk GC errors, because that's
-			 * what we've done historically.
-			 */
-			latest = false;
-			lsn = msg->not_modified_since;
-		}
-
-		pq_sendbyte(&s, msg->tag);
-		pq_sendbyte(&s, latest);
-		pq_sendint64(&s, lsn);
-	}
-
-	/*
-	 * The rest of the request messages are the same between protocol V1 and
-	 * V2
-	 */
 	switch (messageTag(msg))
 	{
 			/* pagestore_client -> pagestore */
@@ -1014,6 +1006,8 @@ nm_pack_request(NeonRequest *msg)
 			{
 				NeonExistsRequest *msg_req = (NeonExistsRequest *) msg;
 
+				pq_sendbyte(&s, msg_req->req.latest);
+				pq_sendint64(&s, msg_req->req.lsn);
 				pq_sendint32(&s, NInfoGetSpcOid(msg_req->rinfo));
 				pq_sendint32(&s, NInfoGetDbOid(msg_req->rinfo));
 				pq_sendint32(&s, NInfoGetRelNumber(msg_req->rinfo));
@@ -1025,6 +1019,8 @@ nm_pack_request(NeonRequest *msg)
 			{
 				NeonNblocksRequest *msg_req = (NeonNblocksRequest *) msg;
 
+				pq_sendbyte(&s, msg_req->req.latest);
+				pq_sendint64(&s, msg_req->req.lsn);
 				pq_sendint32(&s, NInfoGetSpcOid(msg_req->rinfo));
 				pq_sendint32(&s, NInfoGetDbOid(msg_req->rinfo));
 				pq_sendint32(&s, NInfoGetRelNumber(msg_req->rinfo));
@@ -1036,6 +1032,8 @@ nm_pack_request(NeonRequest *msg)
 			{
 				NeonDbSizeRequest *msg_req = (NeonDbSizeRequest *) msg;
 
+				pq_sendbyte(&s, msg_req->req.latest);
+				pq_sendint64(&s, msg_req->req.lsn);
 				pq_sendint32(&s, msg_req->dbNode);
 
 				break;
@@ -1044,6 +1042,8 @@ nm_pack_request(NeonRequest *msg)
 			{
 				NeonGetPageRequest *msg_req = (NeonGetPageRequest *) msg;
 
+				pq_sendbyte(&s, msg_req->req.latest);
+				pq_sendint64(&s, msg_req->req.lsn);
 				pq_sendint32(&s, NInfoGetSpcOid(msg_req->rinfo));
 				pq_sendint32(&s, NInfoGetDbOid(msg_req->rinfo));
 				pq_sendint32(&s, NInfoGetRelNumber(msg_req->rinfo));
@@ -1057,6 +1057,8 @@ nm_pack_request(NeonRequest *msg)
 			{
 				NeonGetSlruSegmentRequest *msg_req = (NeonGetSlruSegmentRequest *) msg;
 
+				pq_sendbyte(&s, msg_req->req.latest);
+				pq_sendint64(&s, msg_req->req.lsn);
 				pq_sendbyte(&s, msg_req->kind);
 				pq_sendint32(&s, msg_req->segno);
 
@@ -1207,7 +1209,7 @@ nm_to_string(NeonMessage *msg)
 				appendStringInfo(&s, ", \"rinfo\": \"%u/%u/%u\"", RelFileInfoFmt(msg_req->rinfo));
 				appendStringInfo(&s, ", \"forknum\": %d", msg_req->forknum);
 				appendStringInfo(&s, ", \"lsn\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.lsn));
-				appendStringInfo(&s, ", \"not_modified_since\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.not_modified_since));
+				appendStringInfo(&s, ", \"latest\": %d", msg_req->req.latest);
 				appendStringInfoChar(&s, '}');
 				break;
 			}
@@ -1220,7 +1222,7 @@ nm_to_string(NeonMessage *msg)
 				appendStringInfo(&s, ", \"rinfo\": \"%u/%u/%u\"", RelFileInfoFmt(msg_req->rinfo));
 				appendStringInfo(&s, ", \"forknum\": %d", msg_req->forknum);
 				appendStringInfo(&s, ", \"lsn\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.lsn));
-				appendStringInfo(&s, ", \"not_modified_since\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.not_modified_since));
+				appendStringInfo(&s, ", \"latest\": %d", msg_req->req.latest);
 				appendStringInfoChar(&s, '}');
 				break;
 			}
@@ -1234,7 +1236,7 @@ nm_to_string(NeonMessage *msg)
 				appendStringInfo(&s, ", \"forknum\": %d", msg_req->forknum);
 				appendStringInfo(&s, ", \"blkno\": %u", msg_req->blkno);
 				appendStringInfo(&s, ", \"lsn\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.lsn));
-				appendStringInfo(&s, ", \"not_modified_since\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.not_modified_since));
+				appendStringInfo(&s, ", \"latest\": %d", msg_req->req.latest);
 				appendStringInfoChar(&s, '}');
 				break;
 			}
@@ -1245,7 +1247,7 @@ nm_to_string(NeonMessage *msg)
 				appendStringInfoString(&s, "{\"type\": \"NeonDbSizeRequest\"");
 				appendStringInfo(&s, ", \"dbnode\": \"%u\"", msg_req->dbNode);
 				appendStringInfo(&s, ", \"lsn\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.lsn));
-				appendStringInfo(&s, ", \"not_modified_since\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.not_modified_since));
+				appendStringInfo(&s, ", \"latest\": %d", msg_req->req.latest);
 				appendStringInfoChar(&s, '}');
 				break;
 			}
@@ -1257,7 +1259,7 @@ nm_to_string(NeonMessage *msg)
 				appendStringInfo(&s, ", \"kind\": %u", msg_req->kind);
 				appendStringInfo(&s, ", \"segno\": %u", msg_req->segno);
 				appendStringInfo(&s, ", \"lsn\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.lsn));
-				appendStringInfo(&s, ", \"not_modified_since\": \"%X/%X\"", LSN_FORMAT_ARGS(msg_req->req.not_modified_since));
+				appendStringInfo(&s, ", \"latest\": %d", msg_req->req.latest);
 				appendStringInfoChar(&s, '}');
 				break;
 			}
@@ -1529,38 +1531,44 @@ nm_adjust_lsn(XLogRecPtr lsn)
 /*
  * Return LSN for requesting pages and number of blocks from page server
  */
-static void
-neon_get_request_lsn(NRelFileInfo rinfo, ForkNumber forknum, BlockNumber blkno,
-					 XLogRecPtr *request_lsn, XLogRecPtr *not_modified_since)
+static XLogRecPtr
+neon_get_request_lsn(bool *latest, NRelFileInfo rinfo, ForkNumber forknum, BlockNumber blkno)
 {
-	XLogRecPtr	last_written_lsn;
-
-	last_written_lsn = GetLastWrittenLSN(rinfo, forknum, blkno);
-	last_written_lsn = nm_adjust_lsn(last_written_lsn);
-	Assert(last_written_lsn != InvalidXLogRecPtr);
+	XLogRecPtr	lsn;
 
 	if (RecoveryInProgress())
 	{
-		/* Request the page at the last replayed LSN. */
-		*request_lsn = GetXLogReplayRecPtr(NULL);
-		*not_modified_since = last_written_lsn;
-		Assert(last_written_lsn <= *request_lsn);
+		/*
+		 * We don't know if WAL has been generated but not yet replayed, so
+		 * we're conservative in our estimates about latest pages.
+		 */
+		*latest = false;
 
-		neon_log(DEBUG1, "neon_get_request_lsn request lsn %X/%X, not_modified_since %X/%X",
-				 LSN_FORMAT_ARGS(*request_lsn), LSN_FORMAT_ARGS(*not_modified_since));
+		/*
+		 * Get the last written LSN of this page.
+		 */
+		lsn = GetLastWrittenLSN(rinfo, forknum, blkno);
+		lsn = nm_adjust_lsn(lsn);
+
+		neon_log(DEBUG1, "neon_get_request_lsn GetXLogReplayRecPtr %X/%X request lsn 0 ",
+			 (uint32) ((lsn) >> 32), (uint32) (lsn));
 	}
 	else
 	{
 		XLogRecPtr	flushlsn;
 
 		/*
-		 * Use the latest LSN that was evicted from the buffer cache as the
-		 * 'not_modified_since' hint. Any pages modified by later WAL records
-		 * must still in the buffer cache, so our request cannot concern
-		 * those.
+		 * Use the latest LSN that was evicted from the buffer cache. Any
+		 * pages modified by later WAL records must still in the buffer cache,
+		 * so our request cannot concern those.
 		 */
+		*latest = true;
+		lsn = GetLastWrittenLSN(rinfo, forknum, blkno);
+		Assert(lsn != InvalidXLogRecPtr);
 		neon_log(DEBUG1, "neon_get_request_lsn GetLastWrittenLSN lsn %X/%X ",
-				 LSN_FORMAT_ARGS(last_written_lsn));
+			 (uint32) ((lsn) >> 32), (uint32) (lsn));
+
+		lsn = nm_adjust_lsn(lsn);
 
 		/*
 		 * Is it possible that the last-written LSN is ahead of last flush
@@ -1575,109 +1583,16 @@ neon_get_request_lsn(NRelFileInfo rinfo, ForkNumber forknum, BlockNumber blkno,
 #else
 		flushlsn = GetFlushRecPtr();
 #endif
-		if (last_written_lsn > flushlsn)
+		if (lsn > flushlsn)
 		{
 			neon_log(DEBUG5, "last-written LSN %X/%X is ahead of last flushed LSN %X/%X",
-					 LSN_FORMAT_ARGS(last_written_lsn),
-					 LSN_FORMAT_ARGS(flushlsn));
-			XLogFlush(last_written_lsn);
-			flushlsn = last_written_lsn;
+				 (uint32) (lsn >> 32), (uint32) lsn,
+				 (uint32) (flushlsn >> 32), (uint32) flushlsn);
+			XLogFlush(lsn);
 		}
-
-		/*
-		 * Request the latest version of the page. The most up-to-date request
-		 * LSN we could use would be the current insert LSN, but to avoid the
-		 * overhead of looking it up, use 'flushlsn' instead. This relies on
-		 * the assumption that if the page was modified since the last WAL
-		 * flush, it should still be in the buffer cache, and we wouldn't be
-		 * requesting it.
-		 */
-		*request_lsn = flushlsn;
-		*not_modified_since = last_written_lsn;
-	}
-}
-
-/*
- *  neon_prefetch_response_usable -- Can a new request be satisfied by old one?
- *
- * This is used to check if the response to a prefetch request can be used to
- * satisfy a page read now.
- */
-static bool
-neon_prefetch_response_usable(XLogRecPtr request_lsn, XLogRecPtr not_modified_since,
-							  PrefetchRequest *slot)
-{
-	/* sanity check the LSN's on the old and the new request */
-	Assert(request_lsn >= not_modified_since);
-	Assert(slot->request_lsn >= slot->not_modified_since);
-	Assert(slot->status != PRFS_UNUSED);
-
-	/*
-	 * The new request's LSN should never be older than the old one.  This
-	 * could be an Assert, except that for testing purposes, we do provide an
-	 * interface in neon_test_utils to fetch pages at arbitary LSNs, which
-	 * violates this.
-	 *
-	 * Similarly, the not_modified_since value calculated for a page should
-	 * never move backwards. This assumption is a bit fragile; if we updated
-	 * the last-written cache when we read in a page, for example, then it
-	 * might. But as the code stands, it should not.
-	 *
-	 * (If two backends issue a request at the same time, they might race and
-	 * calculate LSNs "out of order" with each other, but the prefetch queue
-	 * is backend-private at the moment.)
-	 */
-	if (request_lsn < slot->request_lsn || not_modified_since < slot->not_modified_since)
-	{
-		ereport(LOG,
-				(errcode(ERRCODE_IO_ERROR),
-				 errmsg(NEON_TAG "request with unexpected LSN after prefetch"),
-				 errdetail("Request %X/%X not_modified_since %X/%X, prefetch %X/%X not_modified_since %X/%X)",
-						   LSN_FORMAT_ARGS(request_lsn), LSN_FORMAT_ARGS(not_modified_since),
-						   LSN_FORMAT_ARGS(slot->request_lsn), LSN_FORMAT_ARGS(slot->not_modified_since))));
-		return false;
 	}
 
-	/*---
-	 * Each request to the pageserver carries two LSN values:
-	 * `not_modified_since` and `request_lsn`. The (not_modified_since,
-	 * request_lsn] range of each request is effectively a claim that the page
-	 * has not been modified between those LSNs.  If the range of the old
-	 * request in the queue overlaps with the new request, we know that the
-	 * page hasn't been modified in the union of the ranges. We can use the
-	 * response to old request to satisfy the new request in that case. For
-	 * example:
-	 *
-	 *              100      500
-	 * Old request:  +--------+
-	 *
-	 *                     400      800
-	 * New request:         +--------+
-	 *
-	 * The old request claims that the page was not modified between LSNs 100
-	 * and 500, and the second claims that it was not modified between 400 and
-	 * 800. Together they mean that the page was not modified between 100 and
-	 * 800. Therefore the response to the old request is also valid for the
-	 * new request.
-	 *
-	 * This logic also holds at the boundary case that the old request's LSN
-	 * matches the new request's not_modified_since LSN exactly:
-	 *
-	 *              100      500
-	 * Old request:  +--------+
-	 *
-	 *                       500      900
-	 * New request:           +--------+
-	 *
-	 * The response to the old request is the page as it was at LSN 500, and
-	 * the page hasn't been changed in the range (500, 900], therefore the
-	 * response is valid also for the new request.
-	 */
-
-	/* this follows from the checks above */
-	Assert(request_lsn >= slot->not_modified_since);
-
-	return not_modified_since <= slot->request_lsn;
+	return lsn;
 }
 
 /*
@@ -1689,8 +1604,8 @@ neon_exists(SMgrRelation reln, ForkNumber forkNum)
 	bool		exists;
 	NeonResponse *resp;
 	BlockNumber n_blocks;
+	bool		latest;
 	XLogRecPtr	request_lsn;
-	XLogRecPtr	not_modified_since;
 
 	switch (reln->smgr_relpersistence)
 	{
@@ -1745,13 +1660,12 @@ neon_exists(SMgrRelation reln, ForkNumber forkNum)
 		return false;
 	}
 
-	neon_get_request_lsn(InfoFromSMgrRel(reln), forkNum, REL_METADATA_PSEUDO_BLOCKNO,
-						 &request_lsn, &not_modified_since);
+	request_lsn = neon_get_request_lsn(&latest, InfoFromSMgrRel(reln), forkNum, REL_METADATA_PSEUDO_BLOCKNO);
 	{
 		NeonExistsRequest request = {
 			.req.tag = T_NeonExistsRequest,
+			.req.latest = latest,
 			.req.lsn = request_lsn,
-			.req.not_modified_since = not_modified_since,
 			.rinfo = InfoFromSMgrRel(reln),
 		.forknum = forkNum};
 
@@ -2188,10 +2102,10 @@ neon_writeback(SMgrRelation reln, ForkNumber forknum,
 void
 #if PG_MAJORVERSION_NUM < 16
 neon_read_at_lsn(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
-				 XLogRecPtr request_lsn, XLogRecPtr not_modified_since, char *buffer)
+				 XLogRecPtr request_lsn, bool request_latest, char *buffer)
 #else
 neon_read_at_lsn(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
-				 XLogRecPtr request_lsn, XLogRecPtr not_modified_since, void *buffer)
+				 XLogRecPtr request_lsn, bool request_latest, void *buffer)
 #endif
 {
 	NeonResponse *resp;
@@ -2234,16 +2148,15 @@ neon_read_at_lsn(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
 	if (entry != NULL)
 	{
 		slot = entry->slot;
-		if (neon_prefetch_response_usable(request_lsn, not_modified_since, slot))
+		if (slot->effective_request_lsn >= request_lsn)
 		{
 			ring_index = slot->my_ring_index;
 			pgBufferUsage.prefetch.hits += 1;
 		}
-		else
+		else					/* the current prefetch LSN is not large
+								 * enough, so drop the prefetch */
 		{
 			/*
-			 * Cannot use this prefetch, discard it
-			 *
 			 * We can't drop cache for not-yet-received requested items. It is
 			 * unlikely this happens, but it can happen if prefetch distance
 			 * is large enough and a backend didn't consume all prefetch
@@ -2268,8 +2181,8 @@ neon_read_at_lsn(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
 		{
 			pgBufferUsage.prefetch.misses += 1;
 
-			ring_index = prefetch_register_buffer(buftag, &request_lsn,
-												  &not_modified_since);
+			ring_index = prefetch_register_buffer(buftag, &request_latest,
+												  &request_lsn);
 			slot = GetPrfSlot(ring_index);
 		}
 		else
@@ -2333,8 +2246,8 @@ neon_read(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno, char *buffer
 neon_read(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno, void *buffer)
 #endif
 {
+	bool		latest;
 	XLogRecPtr	request_lsn;
-	XLogRecPtr	not_modified_since;
 
 	switch (reln->smgr_relpersistence)
 	{
@@ -2359,9 +2272,8 @@ neon_read(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno, void *buffer
 		return;
 	}
 
-	neon_get_request_lsn(InfoFromSMgrRel(reln), forkNum, blkno,
-						 &request_lsn, &not_modified_since);
-	neon_read_at_lsn(InfoFromSMgrRel(reln), forkNum, blkno, request_lsn, not_modified_since, buffer);
+	request_lsn = neon_get_request_lsn(&latest, InfoFromSMgrRel(reln), forkNum, blkno);
+	neon_read_at_lsn(InfoFromSMgrRel(reln), forkNum, blkno, request_lsn, latest, buffer);
 
 #ifdef DEBUG_COMPARE_LOCAL
 	if (forkNum == MAIN_FORKNUM && IS_LOCAL_REL(reln))
@@ -2530,8 +2442,8 @@ neon_nblocks(SMgrRelation reln, ForkNumber forknum)
 {
 	NeonResponse *resp;
 	BlockNumber n_blocks;
+	bool		latest;
 	XLogRecPtr	request_lsn;
-	XLogRecPtr	not_modified_since;
 
 	switch (reln->smgr_relpersistence)
 	{
@@ -2558,13 +2470,12 @@ neon_nblocks(SMgrRelation reln, ForkNumber forknum)
 		return n_blocks;
 	}
 
-	neon_get_request_lsn(InfoFromSMgrRel(reln), forknum, REL_METADATA_PSEUDO_BLOCKNO,
-						 &request_lsn, &not_modified_since);
+	request_lsn = neon_get_request_lsn(&latest, InfoFromSMgrRel(reln), forknum, REL_METADATA_PSEUDO_BLOCKNO);
 	{
 		NeonNblocksRequest request = {
 			.req.tag = T_NeonNblocksRequest,
+			.req.latest = latest,
 			.req.lsn = request_lsn,
-			.req.not_modified_since = not_modified_since,
 			.rinfo = InfoFromSMgrRel(reln),
 			.forknum = forknum,
 		};
@@ -2612,17 +2523,16 @@ neon_dbsize(Oid dbNode)
 {
 	NeonResponse *resp;
 	int64		db_size;
-	XLogRecPtr	request_lsn,
-				not_modified_since;
+	XLogRecPtr	request_lsn;
+	bool		latest;
 	NRelFileInfo dummy_node = {0};
 
-	neon_get_request_lsn(dummy_node, MAIN_FORKNUM, REL_METADATA_PSEUDO_BLOCKNO,
-						 &request_lsn, &not_modified_since);
+	request_lsn = neon_get_request_lsn(&latest, dummy_node, MAIN_FORKNUM, REL_METADATA_PSEUDO_BLOCKNO);
 	{
 		NeonDbSizeRequest request = {
 			.req.tag = T_NeonDbSizeRequest,
+			.req.latest = latest,
 			.req.lsn = request_lsn,
-			.req.not_modified_since = not_modified_since,
 			.dbNode = dbNode,
 		};
 
@@ -2695,6 +2605,7 @@ neon_truncate(SMgrRelation reln, ForkNumber forknum, BlockNumber nblocks)
 	 * the most recently inserted WAL record's LSN.
 	 */
 	lsn = GetXLogInsertRecPtr();
+
 	lsn = nm_adjust_lsn(lsn);
 
 	/*
@@ -2894,33 +2805,14 @@ neon_end_unlogged_build(SMgrRelation reln)
 static int
 neon_read_slru_segment(SMgrRelation reln, const char* path, int segno, void* buffer)
 {
-	XLogRecPtr request_lsn,
-		not_modified_since;
-
-	if (RecoveryInProgress())
-	{
-		request_lsn = GetXLogReplayRecPtr(NULL);
-		if (request_lsn == InvalidXLogRecPtr)
-		{
-			/*
-			 * This happens in neon startup, we start up without replaying any
-			 * records.
-			 */
-			request_lsn = GetRedoStartLsn();
-		}
-	}
-	else
-		request_lsn = GetXLogInsertRecPtr();
-	request_lsn = nm_adjust_lsn(request_lsn);
-
+	XLogRecPtr request_lsn;
 	/*
-	 * GetRedoStartLsn() returns LSN of basebackup. We know that the SLRU
-	 * segment has not changed since the basebackup, because in order to
-	 * modify it, we would have had to download it already. And once
-	 * downloaded, we never evict SLRU segments from local disk.
+	 * GetRedoStartLsn() returns LSN of basebackup.
+	 * We need to download SLRU segments only once after node startup,
+	 * then SLRUs are maintained locally.
 	 */
-	not_modified_since = GetRedoStartLsn();
-
+	request_lsn = GetRedoStartLsn();
+	request_lsn = nm_adjust_lsn(request_lsn);
 	SlruKind kind;
 
     if (STRPREFIX(path, "pg_xact"))
@@ -2935,8 +2827,8 @@ neon_read_slru_segment(SMgrRelation reln, const char* path, int segno, void* buf
 	NeonResponse *resp;
 	NeonGetSlruSegmentRequest request = {
 		.req.tag = T_NeonGetSlruSegmentRequest,
+		.req.latest = false,
 		.req.lsn = request_lsn,
-		.req.not_modified_since = not_modified_since,
 
 		.kind = kind,
 		.segno = segno
@@ -3064,9 +2956,6 @@ neon_extend_rel_size(NRelFileInfo rinfo, ForkNumber forknum, BlockNumber blkno,
 {
 	BlockNumber relsize;
 
-	/* This is only used in WAL replay */
-	Assert(RecoveryInProgress());
-
 	/* Extend the relation if we know its size */
 	if (get_cached_relsize(rinfo, forknum, &relsize))
 	{
@@ -3085,13 +2974,14 @@ neon_extend_rel_size(NRelFileInfo rinfo, ForkNumber forknum, BlockNumber blkno,
 		 * This length is later reused when we open the smgr to read the
 		 * block, which is fine and expected.
 		 */
+
 		NeonResponse *response;
 		NeonNblocksResponse *nbresponse;
 		NeonNblocksRequest request = {
 			.req = (NeonRequest) {
-				.tag = T_NeonNblocksRequest,
 				.lsn = end_recptr,
-				.not_modified_since = end_recptr,
+				.latest = false,
+				.tag = T_NeonNblocksRequest,
 			},
 			.rinfo = rinfo,
 			.forknum = forknum,

pgxn/neon_test_utils/Makefile

@@ -7,7 +7,7 @@ OBJS = \
 	neontest.o
 
 EXTENSION = neon_test_utils
-DATA = neon_test_utils--1.1.sql
+DATA = neon_test_utils--1.0.sql
 PGFILEDESC = "neon_test_utils - helpers for neon testing and debugging"
 
 PG_CONFIG = pg_config

pgxn/neon_test_utils/neon_test_utils--1.0.sql

@@ -31,12 +31,12 @@ AS 'MODULE_PATHNAME', 'clear_buffer_cache'
 LANGUAGE C STRICT
 PARALLEL UNSAFE;
 
-CREATE FUNCTION get_raw_page_at_lsn(relname text, forkname text, blocknum int8, request_lsn pg_lsn, not_modified_since pg_lsn)
+CREATE FUNCTION get_raw_page_at_lsn(relname text, forkname text, blocknum int8, lsn pg_lsn)
 RETURNS bytea
 AS 'MODULE_PATHNAME', 'get_raw_page_at_lsn'
 LANGUAGE C PARALLEL UNSAFE;
 
-CREATE FUNCTION get_raw_page_at_lsn(tbspc oid, db oid, relfilenode oid, forknum int8, blocknum int8, request_lsn pg_lsn, not_modified_since pg_lsn)
+CREATE FUNCTION get_raw_page_at_lsn(tbspc oid, db oid, relfilenode oid, forknum int8, blocknum int8, lsn pg_lsn)
 RETURNS bytea
 AS 'MODULE_PATHNAME', 'get_raw_page_at_lsn_ex'
 LANGUAGE C PARALLEL UNSAFE;

pgxn/neon_test_utils/neon_test_utils.control

@@ -1,6 +1,6 @@
 # neon_test_utils extension
 comment = 'helpers for neon testing and debugging'
-default_version = '1.1'
+default_version = '1.0'
 module_pathname = '$libdir/neon_test_utils'
 relocatable = true
 trusted = true

pgxn/neon_test_utils/neontest.c

@@ -48,10 +48,10 @@ PG_FUNCTION_INFO_V1(neon_xlogflush);
  */
 #if PG_MAJORVERSION_NUM < 16
 typedef void (*neon_read_at_lsn_type) (NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
-									   XLogRecPtr request_lsn, XLogRecPtr not_modified_since, char *buffer);
+									   XLogRecPtr request_lsn, bool request_latest, char *buffer);
 #else
 typedef void (*neon_read_at_lsn_type) (NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
-									   XLogRecPtr request_lsn, XLogRecPtr not_modified_since, void *buffer);
+									   XLogRecPtr request_lsn, bool request_latest, void *buffer);
 #endif
 
 static neon_read_at_lsn_type neon_read_at_lsn_ptr;
@@ -299,11 +299,8 @@ get_raw_page_at_lsn(PG_FUNCTION_ARGS)
 	text	   *forkname;
 	uint32		blkno;
 
-	XLogRecPtr	request_lsn;
-	XLogRecPtr	not_modified_since;
-
-	if (PG_NARGS() != 5)
-		elog(ERROR, "unexpected number of arguments in SQL function signature");
+	bool		request_latest = PG_ARGISNULL(3);
+	uint64		read_lsn = request_latest ? GetXLogInsertRecPtr() : PG_GETARG_INT64(3);
 
 	if (PG_ARGISNULL(0) || PG_ARGISNULL(1) || PG_ARGISNULL(2))
 		PG_RETURN_NULL();
@@ -312,9 +309,6 @@ get_raw_page_at_lsn(PG_FUNCTION_ARGS)
 	forkname = PG_GETARG_TEXT_PP(1);
 	blkno = PG_GETARG_UINT32(2);
 
-	request_lsn = PG_ARGISNULL(3) ? GetXLogInsertRecPtr() : PG_GETARG_LSN(3);
-	not_modified_since = PG_ARGISNULL(4) ? request_lsn : PG_GETARG_LSN(4);
-
 	if (!superuser())
 		ereport(ERROR,
 				(errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
@@ -367,7 +361,7 @@ get_raw_page_at_lsn(PG_FUNCTION_ARGS)
 	SET_VARSIZE(raw_page, BLCKSZ + VARHDRSZ);
 	raw_page_data = VARDATA(raw_page);
 
-	neon_read_at_lsn(InfoFromRelation(rel), forknum, blkno, request_lsn, not_modified_since, raw_page_data);
+	neon_read_at_lsn(InfoFromRelation(rel), forknum, blkno, read_lsn, request_latest, raw_page_data);
 
 	relation_close(rel, AccessShareLock);
 
@@ -386,9 +380,6 @@ get_raw_page_at_lsn_ex(PG_FUNCTION_ARGS)
 {
 	char	   *raw_page_data;
 
-	if (PG_NARGS() != 7)
-		elog(ERROR, "unexpected number of arguments in SQL function signature");
-
 	if (!superuser())
 		ereport(ERROR,
 				(errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
@@ -412,20 +403,18 @@ get_raw_page_at_lsn_ex(PG_FUNCTION_ARGS)
 		};
 
 		ForkNumber	forknum = PG_GETARG_UINT32(3);
+
 		uint32		blkno = PG_GETARG_UINT32(4);
-		XLogRecPtr	request_lsn;
-		XLogRecPtr	not_modified_since;
+		bool		request_latest = PG_ARGISNULL(5);
+		uint64		read_lsn = request_latest ? GetXLogInsertRecPtr() : PG_GETARG_INT64(5);
 
 		/* Initialize buffer to copy to */
 		bytea	   *raw_page = (bytea *) palloc(BLCKSZ + VARHDRSZ);
 
-		request_lsn = PG_ARGISNULL(5) ? GetXLogInsertRecPtr() : PG_GETARG_LSN(5);
-		not_modified_since = PG_ARGISNULL(6) ? request_lsn : PG_GETARG_LSN(6);
-
 		SET_VARSIZE(raw_page, BLCKSZ + VARHDRSZ);
 		raw_page_data = VARDATA(raw_page);
 
-		neon_read_at_lsn(rinfo, forknum, blkno, request_lsn, not_modified_since, raw_page_data);
+		neon_read_at_lsn(rinfo, forknum, blkno, read_lsn, request_latest, raw_page_data);
 		PG_RETURN_BYTEA_P(raw_page);
 	}
 }

proxy/src/bin/proxy.rs

@@ -403,43 +403,27 @@ async fn main() -> anyhow::Result<()> {
         maintenance_tasks.spawn(usage_metrics::task_main(metrics_config));
         client_tasks.spawn(usage_metrics::task_backup(
             &metrics_config.backup_metric_collection_config,
-            cancellation_token.clone(),
+            cancellation_token,
         ));
     }
 
     if let auth::BackendType::Console(api, _) = &config.auth_backend {
         if let proxy::console::provider::ConsoleBackend::Console(api) = &**api {
-            match (redis_notifications_client, regional_redis_client.clone()) {
-                (None, None) => {}
-                (client1, client2) => {
-                    let cache = api.caches.project_info.clone();
-                    if let Some(client) = client1 {
-                        maintenance_tasks.spawn(notifications::task_main(
-                            client,
-                            cache.clone(),
-                            cancel_map.clone(),
-                            args.region.clone(),
-                        ));
-                    }
-                    if let Some(client) = client2 {
-                        maintenance_tasks.spawn(notifications::task_main(
-                            client,
-                            cache.clone(),
-                            cancel_map.clone(),
-                            args.region.clone(),
-                        ));
-                    }
-                    maintenance_tasks.spawn(async move { cache.clone().gc_worker().await });
-                }
+            if let Some(redis_notifications_client) = redis_notifications_client {
+                let cache = api.caches.project_info.clone();
+                maintenance_tasks.spawn(notifications::task_main(
+                    redis_notifications_client,
+                    cache.clone(),
+                    cancel_map.clone(),
+                    args.region.clone(),
+                ));
+                maintenance_tasks.spawn(async move { cache.clone().gc_worker().await });
             }
             if let Some(regional_redis_client) = regional_redis_client {
                 let cache = api.caches.endpoints_cache.clone();
                 let con = regional_redis_client;
                 let span = tracing::info_span!("endpoints_cache");
-                maintenance_tasks.spawn(
-                    async move { cache.do_read(con, cancellation_token.clone()).await }
-                        .instrument(span),
-                );
+                maintenance_tasks.spawn(async move { cache.do_read(con).await }.instrument(span));
             }
         }
     }

proxy/src/cache/endpoints.rs

@@ -4,7 +4,6 @@ use std::{
         atomic::{AtomicBool, Ordering},
         Arc,
     },
-    time::Duration,
 };
 
 use dashmap::DashSet;
@@ -14,7 +13,6 @@ use redis::{
 };
 use serde::Deserialize;
 use tokio::sync::Mutex;
-use tokio_util::sync::CancellationToken;
 use tracing::info;
 
 use crate::{
@@ -113,22 +111,16 @@ impl EndpointsCache {
     pub async fn do_read(
         &self,
         mut con: ConnectionWithCredentialsProvider,
-        cancellation_token: CancellationToken,
     ) -> anyhow::Result<Infallible> {
         let mut last_id = "0-0".to_string();
         loop {
+            self.ready.store(false, Ordering::Release);
             if let Err(e) = con.connect().await {
                 tracing::error!("error connecting to redis: {:?}", e);
-                self.ready.store(false, Ordering::Release);
+                continue;
             }
             if let Err(e) = self.read_from_stream(&mut con, &mut last_id).await {
                 tracing::error!("error reading from redis: {:?}", e);
-                self.ready.store(false, Ordering::Release);
-            }
-            if cancellation_token.is_cancelled() {
-                info!("cancellation token is cancelled, exiting");
-                tokio::time::sleep(Duration::from_secs(60 * 60 * 24 * 7)).await;
-                // 1 week.
             }
             tokio::time::sleep(self.config.retry_interval).await;
         }

proxy/src/config.rs

@@ -533,13 +533,13 @@ pub struct RetryConfig {
 impl RetryConfig {
     /// Default options for RetryConfig.
 
-    /// Total delay for 8 retries with 100ms base delay and 1.6 backoff factor is about 7s.
+    /// Total delay for 4 retries with 1s base delay and 2.0 backoff factor is 7s.
     pub const CONNECT_TO_COMPUTE_DEFAULT_VALUES: &'static str =
-        "num_retries=8,base_retry_wait_duration=100ms,retry_wait_exponent_base=1.6";
-    /// Total delay for 8 retries with 100ms base delay and 1.6 backoff factor is about 7s.
-    /// Cplane has timeout of 60s on each request. 8m7s in total.
+        "num_retries=4,base_retry_wait_duration=1s,retry_wait_exponent_base=2.0";
+    /// Total delay for 4 retries with 1s base delay and 2.0 backoff factor is 7s.
+    /// Cplane has timeout of 60s on each request.
     pub const WAKE_COMPUTE_DEFAULT_VALUES: &'static str =
-        "num_retries=8,base_retry_wait_duration=100ms,retry_wait_exponent_base=1.6";
+        "num_retries=4,base_retry_wait_duration=1s,retry_wait_exponent_base=2.0";
 
     /// Parse retry options passed via cmdline.
     /// Example: [`Self::CONNECT_TO_COMPUTE_DEFAULT_VALUES`].

proxy/src/context/parquet.rs

@@ -413,7 +413,6 @@ mod tests {
                     )
                     .unwrap(),
                     max_keys_per_list_response: DEFAULT_MAX_KEYS_PER_LIST_RESPONSE,
-                    upload_storage_class: None,
                 }),
                 timeout: RemoteStorageConfig::DEFAULT_TIMEOUT,
             })

proxy/src/protocol2.rs

@@ -1,26 +1,42 @@
 //! Proxy Protocol V2 implementation
 
 use std::{
+    future::{poll_fn, Future},
     io,
     net::SocketAddr,
-    pin::Pin,
-    task::{Context, Poll},
+    pin::{pin, Pin},
+    task::{ready, Context, Poll},
 };
 
-use bytes::BytesMut;
+use bytes::{Buf, BytesMut};
+use hyper::server::conn::AddrIncoming;
 use pin_project_lite::pin_project;
 use tokio::io::{AsyncRead, AsyncReadExt, AsyncWrite, ReadBuf};
 
+pub struct ProxyProtocolAccept {
+    pub incoming: AddrIncoming,
+    pub protocol: &'static str,
+}
+
 pin_project! {
-    /// A chained [`AsyncRead`] with [`AsyncWrite`] passthrough
-    pub struct ChainRW<T> {
+    pub struct WithClientIp<T> {
         #[pin]
         pub inner: T,
         buf: BytesMut,
+        tlv_bytes: u16,
+        state: ProxyParse,
     }
 }
 
-impl<T: AsyncWrite> AsyncWrite for ChainRW<T> {
+#[derive(Clone, PartialEq, Debug)]
+enum ProxyParse {
+    NotStarted,
+
+    Finished(SocketAddr),
+    None,
+}
+
+impl<T: AsyncWrite> AsyncWrite for WithClientIp<T> {
     #[inline]
     fn poll_write(
         self: Pin<&mut Self>,
@@ -55,174 +71,267 @@ impl<T: AsyncWrite> AsyncWrite for ChainRW<T> {
     }
 }
 
+impl<T> WithClientIp<T> {
+    pub fn new(inner: T) -> Self {
+        WithClientIp {
+            inner,
+            buf: BytesMut::with_capacity(128),
+            tlv_bytes: 0,
+            state: ProxyParse::NotStarted,
+        }
+    }
+
+    pub fn client_addr(&self) -> Option<SocketAddr> {
+        match self.state {
+            ProxyParse::Finished(socket) => Some(socket),
+            _ => None,
+        }
+    }
+}
+
+impl<T: AsyncRead + Unpin> WithClientIp<T> {
+    pub async fn wait_for_addr(&mut self) -> io::Result<Option<SocketAddr>> {
+        match self.state {
+            ProxyParse::NotStarted => {
+                let mut pin = Pin::new(&mut *self);
+                let addr = poll_fn(|cx| pin.as_mut().poll_client_ip(cx)).await?;
+                match addr {
+                    Some(addr) => self.state = ProxyParse::Finished(addr),
+                    None => self.state = ProxyParse::None,
+                }
+                Ok(addr)
+            }
+            ProxyParse::Finished(addr) => Ok(Some(addr)),
+            ProxyParse::None => Ok(None),
+        }
+    }
+}
+
 /// Proxy Protocol Version 2 Header
 const HEADER: [u8; 12] = [
     0x0D, 0x0A, 0x0D, 0x0A, 0x00, 0x0D, 0x0A, 0x51, 0x55, 0x49, 0x54, 0x0A,
 ];
 
-pub async fn read_proxy_protocol<T: AsyncRead + Unpin>(
-    mut read: T,
-) -> std::io::Result<(ChainRW<T>, Option<SocketAddr>)> {
-    let mut buf = BytesMut::with_capacity(128);
-    while buf.len() < 16 {
-        let bytes_read = read.read_buf(&mut buf).await?;
+impl<T: AsyncRead> WithClientIp<T> {
+    /// implementation of <https://www.haproxy.org/download/2.4/doc/proxy-protocol.txt>
+    /// Version 2 (Binary Format)
+    fn poll_client_ip(
+        mut self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+    ) -> Poll<io::Result<Option<SocketAddr>>> {
+        // The binary header format starts with a constant 12 bytes block containing the protocol signature :
+        //    \x0D \x0A \x0D \x0A \x00 \x0D \x0A \x51 \x55 \x49 \x54 \x0A
+        while self.buf.len() < 16 {
+            let mut this = self.as_mut().project();
+            let bytes_read = pin!(this.inner.read_buf(this.buf)).poll(cx)?;
 
-        // exit for bad header
-        let len = usize::min(buf.len(), HEADER.len());
-        if buf[..len] != HEADER[..len] {
-            return Ok((ChainRW { inner: read, buf }, None));
+            // exit for bad header
+            let len = usize::min(self.buf.len(), HEADER.len());
+            if self.buf[..len] != HEADER[..len] {
+                return Poll::Ready(Ok(None));
+            }
+
+            // if no more bytes available then exit
+            if ready!(bytes_read) == 0 {
+                return Poll::Ready(Ok(None));
+            };
         }
 
-        // if no more bytes available then exit
-        if bytes_read == 0 {
-            return Ok((ChainRW { inner: read, buf }, None));
-        };
-    }
-
-    let header = buf.split_to(16);
-
-    // The next byte (the 13th one) is the protocol version and command.
-    // The highest four bits contains the version. As of this specification, it must
-    // always be sent as \x2 and the receiver must only accept this value.
-    let vc = header[12];
-    let version = vc >> 4;
-    let command = vc & 0b1111;
-    if version != 2 {
-        return Err(io::Error::new(
-            io::ErrorKind::Other,
-            "invalid proxy protocol version. expected version 2",
-        ));
-    }
-    match command {
-        // the connection was established on purpose by the proxy
-        // without being relayed. The connection endpoints are the sender and the
-        // receiver. Such connections exist when the proxy sends health-checks to the
-        // server. The receiver must accept this connection as valid and must use the
-        // real connection endpoints and discard the protocol block including the
-        // family which is ignored.
-        0 => {}
-        // the connection was established on behalf of another node,
-        // and reflects the original connection endpoints. The receiver must then use
-        // the information provided in the protocol block to get original the address.
-        1 => {}
-        // other values are unassigned and must not be emitted by senders. Receivers
-        // must drop connections presenting unexpected values here.
-        _ => {
-            return Err(io::Error::new(
+        // The next byte (the 13th one) is the protocol version and command.
+        // The highest four bits contains the version. As of this specification, it must
+        // always be sent as \x2 and the receiver must only accept this value.
+        let vc = self.buf[12];
+        let version = vc >> 4;
+        let command = vc & 0b1111;
+        if version != 2 {
+            return Poll::Ready(Err(io::Error::new(
                 io::ErrorKind::Other,
-                "invalid proxy protocol command. expected local (0) or proxy (1)",
-            ))
+                "invalid proxy protocol version. expected version 2",
+            )));
         }
-    };
+        match command {
+            // the connection was established on purpose by the proxy
+            // without being relayed. The connection endpoints are the sender and the
+            // receiver. Such connections exist when the proxy sends health-checks to the
+            // server. The receiver must accept this connection as valid and must use the
+            // real connection endpoints and discard the protocol block including the
+            // family which is ignored.
+            0 => {}
+            // the connection was established on behalf of another node,
+            // and reflects the original connection endpoints. The receiver must then use
+            // the information provided in the protocol block to get original the address.
+            1 => {}
+            // other values are unassigned and must not be emitted by senders. Receivers
+            // must drop connections presenting unexpected values here.
+            _ => {
+                return Poll::Ready(Err(io::Error::new(
+                    io::ErrorKind::Other,
+                    "invalid proxy protocol command. expected local (0) or proxy (1)",
+                )))
+            }
+        };
 
-    // The 14th byte contains the transport protocol and address family. The highest 4
-    // bits contain the address family, the lowest 4 bits contain the protocol.
-    let ft = header[13];
-    let address_length = match ft {
-        // - \x11 : TCP over IPv4 : the forwarded connection uses TCP over the AF_INET
-        //   protocol family. Address length is 2*4 + 2*2 = 12 bytes.
-        // - \x12 : UDP over IPv4 : the forwarded connection uses UDP over the AF_INET
-        //   protocol family. Address length is 2*4 + 2*2 = 12 bytes.
-        0x11 | 0x12 => 12,
-        // - \x21 : TCP over IPv6 : the forwarded connection uses TCP over the AF_INET6
-        //   protocol family. Address length is 2*16 + 2*2 = 36 bytes.
-        // - \x22 : UDP over IPv6 : the forwarded connection uses UDP over the AF_INET6
-        //   protocol family. Address length is 2*16 + 2*2 = 36 bytes.
-        0x21 | 0x22 => 36,
-        // unspecified or unix stream. ignore the addresses
-        _ => 0,
-    };
+        // The 14th byte contains the transport protocol and address family. The highest 4
+        // bits contain the address family, the lowest 4 bits contain the protocol.
+        let ft = self.buf[13];
+        let address_length = match ft {
+            // - \x11 : TCP over IPv4 : the forwarded connection uses TCP over the AF_INET
+            //   protocol family. Address length is 2*4 + 2*2 = 12 bytes.
+            // - \x12 : UDP over IPv4 : the forwarded connection uses UDP over the AF_INET
+            //   protocol family. Address length is 2*4 + 2*2 = 12 bytes.
+            0x11 | 0x12 => 12,
+            // - \x21 : TCP over IPv6 : the forwarded connection uses TCP over the AF_INET6
+            //   protocol family. Address length is 2*16 + 2*2 = 36 bytes.
+            // - \x22 : UDP over IPv6 : the forwarded connection uses UDP over the AF_INET6
+            //   protocol family. Address length is 2*16 + 2*2 = 36 bytes.
+            0x21 | 0x22 => 36,
+            // unspecified or unix stream. ignore the addresses
+            _ => 0,
+        };
 
-    // The 15th and 16th bytes is the address length in bytes in network endian order.
-    // It is used so that the receiver knows how many address bytes to skip even when
-    // it does not implement the presented protocol. Thus the length of the protocol
-    // header in bytes is always exactly 16 + this value. When a sender presents a
-    // LOCAL connection, it should not present any address so it sets this field to
-    // zero. Receivers MUST always consider this field to skip the appropriate number
-    // of bytes and must not assume zero is presented for LOCAL connections. When a
-    // receiver accepts an incoming connection showing an UNSPEC address family or
-    // protocol, it may or may not decide to log the address information if present.
-    let remaining_length = u16::from_be_bytes(header[14..16].try_into().unwrap());
-    if remaining_length < address_length {
-        return Err(io::Error::new(
-            io::ErrorKind::Other,
-            "invalid proxy protocol length. not enough to fit requested IP addresses",
-        ));
-    }
-    drop(header);
-
-    while buf.len() < remaining_length as usize {
-        if read.read_buf(&mut buf).await? == 0 {
-            return Err(io::Error::new(
-                io::ErrorKind::UnexpectedEof,
-                "stream closed while waiting for proxy protocol addresses",
-            ));
+        // The 15th and 16th bytes is the address length in bytes in network endian order.
+        // It is used so that the receiver knows how many address bytes to skip even when
+        // it does not implement the presented protocol. Thus the length of the protocol
+        // header in bytes is always exactly 16 + this value. When a sender presents a
+        // LOCAL connection, it should not present any address so it sets this field to
+        // zero. Receivers MUST always consider this field to skip the appropriate number
+        // of bytes and must not assume zero is presented for LOCAL connections. When a
+        // receiver accepts an incoming connection showing an UNSPEC address family or
+        // protocol, it may or may not decide to log the address information if present.
+        let remaining_length = u16::from_be_bytes(self.buf[14..16].try_into().unwrap());
+        if remaining_length < address_length {
+            return Poll::Ready(Err(io::Error::new(
+                io::ErrorKind::Other,
+                "invalid proxy protocol length. not enough to fit requested IP addresses",
+            )));
         }
+
+        while self.buf.len() < 16 + address_length as usize {
+            let mut this = self.as_mut().project();
+            if ready!(pin!(this.inner.read_buf(this.buf)).poll(cx)?) == 0 {
+                return Poll::Ready(Err(io::Error::new(
+                    io::ErrorKind::UnexpectedEof,
+                    "stream closed while waiting for proxy protocol addresses",
+                )));
+            }
+        }
+
+        let this = self.as_mut().project();
+
+        // we are sure this is a proxy protocol v2 entry and we have read all the bytes we need
+        // discard the header we have parsed
+        this.buf.advance(16);
+
+        // Starting from the 17th byte, addresses are presented in network byte order.
+        // The address order is always the same :
+        //   - source layer 3 address in network byte order
+        //   - destination layer 3 address in network byte order
+        //   - source layer 4 address if any, in network byte order (port)
+        //   - destination layer 4 address if any, in network byte order (port)
+        let addresses = this.buf.split_to(address_length as usize);
+        let socket = match address_length {
+            12 => {
+                let src_addr: [u8; 4] = addresses[0..4].try_into().unwrap();
+                let src_port = u16::from_be_bytes(addresses[8..10].try_into().unwrap());
+                Some(SocketAddr::from((src_addr, src_port)))
+            }
+            36 => {
+                let src_addr: [u8; 16] = addresses[0..16].try_into().unwrap();
+                let src_port = u16::from_be_bytes(addresses[32..34].try_into().unwrap());
+                Some(SocketAddr::from((src_addr, src_port)))
+            }
+            _ => None,
+        };
+
+        *this.tlv_bytes = remaining_length - address_length;
+        self.as_mut().skip_tlv_inner();
+
+        Poll::Ready(Ok(socket))
     }
 
-    // Starting from the 17th byte, addresses are presented in network byte order.
-    // The address order is always the same :
-    //   - source layer 3 address in network byte order
-    //   - destination layer 3 address in network byte order
-    //   - source layer 4 address if any, in network byte order (port)
-    //   - destination layer 4 address if any, in network byte order (port)
-    let addresses = buf.split_to(remaining_length as usize);
-    let socket = match address_length {
-        12 => {
-            let src_addr: [u8; 4] = addresses[0..4].try_into().unwrap();
-            let src_port = u16::from_be_bytes(addresses[8..10].try_into().unwrap());
-            Some(SocketAddr::from((src_addr, src_port)))
+    #[cold]
+    fn read_ip(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
+        let ip = ready!(self.as_mut().poll_client_ip(cx)?);
+        match ip {
+            Some(x) => *self.as_mut().project().state = ProxyParse::Finished(x),
+            None => *self.as_mut().project().state = ProxyParse::None,
         }
-        36 => {
-            let src_addr: [u8; 16] = addresses[0..16].try_into().unwrap();
-            let src_port = u16::from_be_bytes(addresses[32..34].try_into().unwrap());
-            Some(SocketAddr::from((src_addr, src_port)))
-        }
-        _ => None,
-    };
+        Poll::Ready(Ok(()))
+    }
 
-    Ok((ChainRW { inner: read, buf }, socket))
+    #[cold]
+    fn skip_tlv(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
+        let mut this = self.as_mut().project();
+        // we know that this.buf is empty
+        debug_assert_eq!(this.buf.len(), 0);
+
+        this.buf.reserve((*this.tlv_bytes).clamp(0, 1024) as usize);
+        ready!(pin!(this.inner.read_buf(this.buf)).poll(cx)?);
+        self.skip_tlv_inner();
+
+        Poll::Ready(Ok(()))
+    }
+
+    fn skip_tlv_inner(self: Pin<&mut Self>) {
+        let tlv_bytes_read = match u16::try_from(self.buf.len()) {
+            // we read more than u16::MAX therefore we must have read the full tlv_bytes
+            Err(_) => self.tlv_bytes,
+            // we might not have read the full tlv bytes yet
+            Ok(n) => u16::min(n, self.tlv_bytes),
+        };
+        let this = self.project();
+        *this.tlv_bytes -= tlv_bytes_read;
+        this.buf.advance(tlv_bytes_read as usize);
+    }
 }
 
-impl<T: AsyncRead> AsyncRead for ChainRW<T> {
+impl<T: AsyncRead> AsyncRead for WithClientIp<T> {
     #[inline]
     fn poll_read(
-        self: Pin<&mut Self>,
+        mut self: Pin<&mut Self>,
         cx: &mut Context<'_>,
         buf: &mut ReadBuf<'_>,
     ) -> Poll<io::Result<()>> {
-        if self.buf.is_empty() {
-            self.project().inner.poll_read(cx, buf)
-        } else {
-            self.read_from_buf(buf)
-        }
-    }
-}
+        // I'm assuming these 3 comparisons will be easy to branch predict.
+        // especially with the cold attributes
+        // which should make this read wrapper almost invisible
+
+        if let ProxyParse::NotStarted = self.state {
+            ready!(self.as_mut().read_ip(cx)?);
+        }
+
+        while self.tlv_bytes > 0 {
+            ready!(self.as_mut().skip_tlv(cx)?)
+        }
 
-impl<T: AsyncRead> ChainRW<T> {
-    #[cold]
-    fn read_from_buf(self: Pin<&mut Self>, buf: &mut ReadBuf<'_>) -> Poll<io::Result<()>> {
-        debug_assert!(!self.buf.is_empty());
         let this = self.project();
-
-        let write = usize::min(this.buf.len(), buf.remaining());
-        let slice = this.buf.split_to(write).freeze();
-        buf.put_slice(&slice);
-
-        // reset the allocation so it can be freed
         if this.buf.is_empty() {
-            *this.buf = BytesMut::new();
-        }
+            this.inner.poll_read(cx, buf)
+        } else {
+            // we know that tlv_bytes is 0
+            debug_assert_eq!(*this.tlv_bytes, 0);
 
-        Poll::Ready(Ok(()))
+            let write = usize::min(this.buf.len(), buf.remaining());
+            let slice = this.buf.split_to(write).freeze();
+            buf.put_slice(&slice);
+
+            // reset the allocation so it can be freed
+            if this.buf.is_empty() {
+                *this.buf = BytesMut::new();
+            }
+
+            Poll::Ready(Ok(()))
+        }
     }
 }
 
 #[cfg(test)]
 mod tests {
+    use std::pin::pin;
+
     use tokio::io::AsyncReadExt;
 
-    use crate::protocol2::read_proxy_protocol;
+    use crate::protocol2::{ProxyParse, WithClientIp};
 
     #[tokio::test]
     async fn test_ipv4() {
@@ -244,15 +353,16 @@ mod tests {
 
         let extra_data = [0x55; 256];
 
-        let (mut read, addr) = read_proxy_protocol(header.chain(extra_data.as_slice()))
-            .await
-            .unwrap();
+        let mut read = pin!(WithClientIp::new(header.chain(extra_data.as_slice())));
 
         let mut bytes = vec![];
         read.read_to_end(&mut bytes).await.unwrap();
 
         assert_eq!(bytes, extra_data);
-        assert_eq!(addr, Some(([127, 0, 0, 1], 65535).into()));
+        assert_eq!(
+            read.state,
+            ProxyParse::Finished(([127, 0, 0, 1], 65535).into())
+        );
     }
 
     #[tokio::test]
@@ -275,17 +385,17 @@ mod tests {
 
         let extra_data = [0x55; 256];
 
-        let (mut read, addr) = read_proxy_protocol(header.chain(extra_data.as_slice()))
-            .await
-            .unwrap();
+        let mut read = pin!(WithClientIp::new(header.chain(extra_data.as_slice())));
 
         let mut bytes = vec![];
         read.read_to_end(&mut bytes).await.unwrap();
 
         assert_eq!(bytes, extra_data);
         assert_eq!(
-            addr,
-            Some(([15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0], 257).into())
+            read.state,
+            ProxyParse::Finished(
+                ([15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0], 257).into()
+            )
         );
     }
 
@@ -293,24 +403,24 @@ mod tests {
     async fn test_invalid() {
         let data = [0x55; 256];
 
-        let (mut read, addr) = read_proxy_protocol(data.as_slice()).await.unwrap();
+        let mut read = pin!(WithClientIp::new(data.as_slice()));
 
         let mut bytes = vec![];
         read.read_to_end(&mut bytes).await.unwrap();
         assert_eq!(bytes, data);
-        assert_eq!(addr, None);
+        assert_eq!(read.state, ProxyParse::None);
     }
 
     #[tokio::test]
     async fn test_short() {
         let data = [0x55; 10];
 
-        let (mut read, addr) = read_proxy_protocol(data.as_slice()).await.unwrap();
+        let mut read = pin!(WithClientIp::new(data.as_slice()));
 
         let mut bytes = vec![];
         read.read_to_end(&mut bytes).await.unwrap();
         assert_eq!(bytes, data);
-        assert_eq!(addr, None);
+        assert_eq!(read.state, ProxyParse::None);
     }
 
     #[tokio::test]
@@ -336,14 +446,15 @@ mod tests {
 
         let extra_data = [0xaa; 256];
 
-        let (mut read, addr) = read_proxy_protocol(header.chain(extra_data.as_slice()))
-            .await
-            .unwrap();
+        let mut read = pin!(WithClientIp::new(header.chain(extra_data.as_slice())));
 
         let mut bytes = vec![];
         read.read_to_end(&mut bytes).await.unwrap();
 
         assert_eq!(bytes, extra_data);
-        assert_eq!(addr, Some(([55, 56, 57, 58], 65535).into()));
+        assert_eq!(
+            read.state,
+            ProxyParse::Finished(([55, 56, 57, 58], 65535).into())
+        );
     }
 }

proxy/src/proxy.rs

@@ -17,7 +17,7 @@ use crate::{
     context::RequestMonitoring,
     error::ReportableError,
     metrics::{Metrics, NumClientConnectionsGuard},
-    protocol2::read_proxy_protocol,
+    protocol2::WithClientIp,
     proxy::handshake::{handshake, HandshakeData},
     stream::{PqStream, Stream},
     EndpointCacheKey,
@@ -88,18 +88,20 @@ pub async fn task_main(
         tracing::info!(protocol = "tcp", %session_id, "accepted new TCP connection");
 
         connections.spawn(async move {
-            let (socket, peer_addr) = match read_proxy_protocol(socket).await{
-                Ok((socket, Some(addr))) => (socket, addr.ip()),
+            let mut socket = WithClientIp::new(socket);
+            let mut peer_addr = peer_addr.ip();
+            match socket.wait_for_addr().await {
+                Ok(Some(addr)) => peer_addr = addr.ip(),
                 Err(e) => {
                     error!("per-client task finished with an error: {e:#}");
                     return;
                 }
-                Ok((_socket, None)) if config.require_client_ip => {
+                Ok(None) if config.require_client_ip => {
                     error!("missing required client IP");
                     return;
                 }
-                Ok((socket, None)) => (socket, peer_addr.ip())
-            };
+                Ok(None) => {}
+            }
 
             match socket.inner.set_nodelay(true) {
                 Ok(()) => {},

proxy/src/proxy/tests.rs

@@ -174,7 +174,7 @@ async fn dummy_proxy(
     tls: Option<TlsConfig>,
     auth: impl TestAuth + Send,
 ) -> anyhow::Result<()> {
-    let (client, _) = read_proxy_protocol(client).await?;
+    let client = WithClientIp::new(client);
     let mut stream = match handshake(client, tls.as_ref(), false).await? {
         HandshakeData::Startup(stream, _) => stream,
         HandshakeData::Cancel(_) => bail!("cancellation not supported"),

proxy/src/serverless.rs

@@ -33,7 +33,7 @@ use crate::cancellation::CancellationHandlerMain;
 use crate::config::ProxyConfig;
 use crate::context::RequestMonitoring;
 use crate::metrics::Metrics;
-use crate::protocol2::read_proxy_protocol;
+use crate::protocol2::WithClientIp;
 use crate::proxy::run_until_cancelled;
 use crate::serverless::backend::PoolingBackend;
 use crate::serverless::http_util::{api_error_into_response, json_response};
@@ -158,8 +158,9 @@ async fn connection_handler(
         .guard(crate::metrics::Protocol::Http);
 
     // handle PROXY protocol
-    let (conn, peer) = match read_proxy_protocol(conn).await {
-        Ok(c) => c,
+    let mut conn = WithClientIp::new(conn);
+    let peer = match conn.wait_for_addr().await {
+        Ok(peer) => peer,
         Err(e) => {
             tracing::error!(?session_id, %peer_addr, "failed to accept TCP connection: invalid PROXY protocol V2 header: {e:#}");
             return;

safekeeper/src/wal_backup.rs

@@ -18,7 +18,7 @@ use std::time::Duration;
 use postgres_ffi::v14::xlog_utils::XLogSegNoOffsetToRecPtr;
 use postgres_ffi::XLogFileName;
 use postgres_ffi::{XLogSegNo, PG_TLI};
-use remote_storage::{GenericRemoteStorage, ListingMode, RemotePath, StorageMetadata};
+use remote_storage::{GenericRemoteStorage, RemotePath, StorageMetadata};
 use tokio::fs::File;
 
 use tokio::select;
@@ -601,18 +601,12 @@ pub async fn delete_timeline(ttid: &TenantTimelineId) -> Result<()> {
     backoff::retry(
         || async {
             // Do list-delete in batch_size batches to make progress even if there a lot of files.
-            // Alternatively we could make remote storage list return iterator, but it is more complicated and
+            // Alternatively we could make list_files return iterator, but it is more complicated and
             // I'm not sure deleting while iterating is expected in s3.
             loop {
                 let files = storage
-                    .list(
-                        Some(&remote_path),
-                        ListingMode::NoDelimiter,
-                        Some(batch_size),
-                        &cancel,
-                    )
-                    .await?
-                    .keys;
+                    .list_files(Some(&remote_path), Some(batch_size), &cancel)
+                    .await?;
                 if files.is_empty() {
                     return Ok(()); // done
                 }
@@ -672,9 +666,8 @@ pub async fn copy_s3_segments(
     let cancel = CancellationToken::new();
 
     let files = storage
-        .list(Some(&remote_path), ListingMode::NoDelimiter, None, &cancel)
-        .await?
-        .keys;
+        .list_files(Some(&remote_path), None, &cancel)
+        .await?;
 
     let uploaded_segments = &files
         .iter()

scripts/export_import_between_pageservers.py

@@ -0,0 +1,730 @@
+#
+# Script to export tenants from one pageserver and import them into another page server.
+#
+# Outline of steps:
+# 1. Get `(last_lsn, prev_lsn)` from old pageserver
+# 2. Get `fullbackup` from old pageserver, which creates a basebackup tar file
+# 3. This tar file might be missing relation files for empty relations, if the pageserver
+#    is old enough (we didn't always store those). So to recreate them, we start a local
+#    vanilla postgres on this basebackup and ask it what relations should exist, then touch
+#    any missing files and re-pack the tar.
+#    TODO This functionality is no longer needed, so we can delete it later if we don't
+#         end up using the same utils for the pg 15 upgrade. Not sure.
+# 4. We import the patched basebackup into a new pageserver
+# 5. We export again via fullbackup, now from the new pageserver and compare the returned
+#    tar file with the one we imported. This confirms that we imported everything that was
+#    exported, but doesn't guarantee correctness (what if we didn't **export** everything
+#    initially?)
+# 6. We wait for the new pageserver's remote_consistent_lsn to catch up
+#
+# For more context on how to use this, see:
+# https://www.notion.so/neondatabase/Storage-format-migration-9a8eba33ccf8417ea8cf50e6a0c542cf
+
+import argparse
+import os
+import shutil
+import subprocess
+import tempfile
+import time
+import uuid
+from contextlib import closing
+from pathlib import Path
+from typing import Any, Dict, List, Optional, Tuple, cast
+
+import psycopg2
+import requests
+from psycopg2.extensions import connection as PgConnection
+from psycopg2.extensions import parse_dsn
+
+###############################################
+### client-side utils copied from test fixtures
+###############################################
+
+Env = Dict[str, str]
+
+_global_counter = 0
+
+
+def global_counter() -> int:
+    """A really dumb global counter.
+    This is useful for giving output files a unique number, so if we run the
+    same command multiple times we can keep their output separate.
+    """
+    global _global_counter
+    _global_counter += 1
+    return _global_counter
+
+
+def subprocess_capture(capture_dir: str, cmd: List[str], **kwargs: Any) -> str:
+    """Run a process and capture its output
+    Output will go to files named "cmd_NNN.stdout" and "cmd_NNN.stderr"
+    where "cmd" is the name of the program and NNN is an incrementing
+    counter.
+    If those files already exist, we will overwrite them.
+    Returns basepath for files with captured output.
+    """
+    assert isinstance(cmd, list)
+    base = f"{os.path.basename(cmd[0])}_{global_counter()}"
+    basepath = os.path.join(capture_dir, base)
+    stdout_filename = basepath + ".stdout"
+    stderr_filename = basepath + ".stderr"
+
+    with open(stdout_filename, "w") as stdout_f:
+        with open(stderr_filename, "w") as stderr_f:
+            print(f'(capturing output to "{base}.stdout")')
+            subprocess.run(cmd, **kwargs, stdout=stdout_f, stderr=stderr_f)
+
+    return basepath
+
+
+class PgBin:
+    """A helper class for executing postgres binaries"""
+
+    def __init__(self, log_dir: Path, pg_distrib_dir, pg_version):
+        self.log_dir = log_dir
+        self.pg_bin_path = os.path.join(str(pg_distrib_dir), f"v{pg_version}", "bin")
+        self.env = os.environ.copy()
+        self.env["LD_LIBRARY_PATH"] = os.path.join(str(pg_distrib_dir), f"v{pg_version}", "lib")
+
+    def _fixpath(self, command: List[str]):
+        if "/" not in command[0]:
+            command[0] = os.path.join(self.pg_bin_path, command[0])
+
+    def _build_env(self, env_add: Optional[Env]) -> Env:
+        if env_add is None:
+            return self.env
+        env = self.env.copy()
+        env.update(env_add)
+        return env
+
+    def run(self, command: List[str], env: Optional[Env] = None, cwd: Optional[str] = None):
+        """
+        Run one of the postgres binaries.
+        The command should be in list form, e.g. ['pgbench', '-p', '55432']
+        All the necessary environment variables will be set.
+        If the first argument (the command name) doesn't include a path (no '/'
+        characters present), then it will be edited to include the correct path.
+        If you want stdout/stderr captured to files, use `run_capture` instead.
+        """
+
+        self._fixpath(command)
+        print(f'Running command "{" ".join(command)}"')
+        env = self._build_env(env)
+        subprocess.run(command, env=env, cwd=cwd, check=True)
+
+    def run_capture(
+        self,
+        command: List[str],
+        env: Optional[Env] = None,
+        cwd: Optional[str] = None,
+        **kwargs: Any,
+    ) -> str:
+        """
+        Run one of the postgres binaries, with stderr and stdout redirected to a file.
+        This is just like `run`, but for chatty programs. Returns basepath for files
+        with captured output.
+        """
+
+        self._fixpath(command)
+        print(f'Running command "{" ".join(command)}"')
+        env = self._build_env(env)
+        return subprocess_capture(
+            str(self.log_dir), command, env=env, cwd=cwd, check=True, **kwargs
+        )
+
+
+class PgProtocol:
+    """Reusable connection logic"""
+
+    def __init__(self, **kwargs):
+        self.default_options = kwargs
+
+    def conn_options(self, **kwargs):
+        conn_options = self.default_options.copy()
+        if "dsn" in kwargs:
+            conn_options.update(parse_dsn(kwargs["dsn"]))
+        conn_options.update(kwargs)
+
+        # Individual statement timeout in seconds. 2 minutes should be
+        # enough for our tests, but if you need a longer, you can
+        # change it by calling "SET statement_timeout" after
+        # connecting.
+        conn_options["options"] = f"-cstatement_timeout=120s {conn_options.get('options', '')}"
+
+        return conn_options
+
+    # autocommit=True here by default because that's what we need most of the time
+    def connect(self, autocommit=True, **kwargs) -> PgConnection:
+        """
+        Connect to the node.
+        Returns psycopg2's connection object.
+        This method passes all extra params to connstr.
+        """
+        conn: PgConnection = psycopg2.connect(**self.conn_options(**kwargs))
+
+        # WARNING: this setting affects *all* tests!
+        conn.autocommit = autocommit
+        return conn
+
+    def safe_psql(self, query: str, **kwargs: Any) -> List[Tuple[Any, ...]]:
+        """
+        Execute query against the node and return all rows.
+        This method passes all extra params to connstr.
+        """
+        return self.safe_psql_many([query], **kwargs)[0]
+
+    def safe_psql_many(self, queries: List[str], **kwargs: Any) -> List[List[Tuple[Any, ...]]]:
+        """
+        Execute queries against the node and return all rows.
+        This method passes all extra params to connstr.
+        """
+        result: List[List[Any]] = []
+        with closing(self.connect(**kwargs)) as conn:
+            with conn.cursor() as cur:
+                for query in queries:
+                    print(f"Executing query: {query}")
+                    cur.execute(query)
+
+                    if cur.description is None:
+                        result.append([])  # query didn't return data
+                    else:
+                        result.append(cast(List[Any], cur.fetchall()))
+        return result
+
+
+class VanillaPostgres(PgProtocol):
+    def __init__(self, pgdatadir: Path, pg_bin: PgBin, port: int, init=True):
+        super().__init__(host="localhost", port=port, dbname="postgres")
+        self.pgdatadir = pgdatadir
+        self.pg_bin = pg_bin
+        self.running = False
+        if init:
+            self.pg_bin.run_capture(["initdb", "-D", str(pgdatadir)])
+        self.configure([f"port = {port}\n"])
+
+    def configure(self, options: List[str]):
+        """Append lines into postgresql.conf file."""
+        assert not self.running
+        with open(os.path.join(self.pgdatadir, "postgresql.conf"), "a") as conf_file:
+            conf_file.write("\n".join(options))
+
+    def start(self, log_path: Optional[str] = None):
+        assert not self.running
+        self.running = True
+
+        log_path = log_path or os.path.join(self.pgdatadir, "pg.log")
+
+        self.pg_bin.run_capture(
+            ["pg_ctl", "-w", "-D", str(self.pgdatadir), "-l", log_path, "start"]
+        )
+
+    def stop(self):
+        assert self.running
+        self.running = False
+        self.pg_bin.run_capture(["pg_ctl", "-w", "-D", str(self.pgdatadir), "stop"])
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, exc_type, exc, tb):
+        if self.running:
+            self.stop()
+
+
+class NeonPageserverApiException(Exception):
+    pass
+
+
+class NeonPageserverHttpClient(requests.Session):
+    def __init__(self, host, port):
+        super().__init__()
+        self.host = host
+        self.port = port
+
+    def verbose_error(self, res: requests.Response):
+        try:
+            res.raise_for_status()
+        except requests.RequestException as e:
+            try:
+                msg = res.json()["msg"]
+            except:  # noqa: E722
+                msg = ""
+            raise NeonPageserverApiException(msg) from e
+
+    def check_status(self):
+        self.get(f"http://{self.host}:{self.port}/v1/status").raise_for_status()
+
+    def tenant_list(self):
+        res = self.get(f"http://{self.host}:{self.port}/v1/tenant")
+        self.verbose_error(res)
+        res_json = res.json()
+        assert isinstance(res_json, list)
+        return res_json
+
+    def tenant_create(self, new_tenant_id: uuid.UUID, ok_if_exists):
+        res = self.post(
+            f"http://{self.host}:{self.port}/v1/tenant",
+            json={"new_tenant_id": new_tenant_id.hex, "generation": 1},
+        )
+
+        if res.status_code == 409:
+            if ok_if_exists:
+                print(f"could not create tenant: already exists for id {new_tenant_id}")
+            else:
+                res.raise_for_status()
+        elif res.status_code == 201:
+            print(f"created tenant {new_tenant_id}")
+        else:
+            self.verbose_error(res)
+
+        return new_tenant_id
+
+    def timeline_list(self, tenant_id: uuid.UUID):
+        res = self.get(f"http://{self.host}:{self.port}/v1/tenant/{tenant_id.hex}/timeline")
+        self.verbose_error(res)
+        res_json = res.json()
+        assert isinstance(res_json, list)
+        return res_json
+
+    def timeline_detail(self, tenant_id: uuid.UUID, timeline_id: uuid.UUID) -> Dict[Any, Any]:
+        res = self.get(
+            f"http://localhost:{self.port}/v1/tenant/{tenant_id.hex}/timeline/{timeline_id.hex}?include-non-incremental-logical-size=true"
+        )
+        self.verbose_error(res)
+        res_json = res.json()
+        assert isinstance(res_json, dict)
+        return res_json
+
+
+def lsn_to_hex(num: int) -> str:
+    """Convert lsn from int to standard hex notation."""
+    return f"{num >> 32:X}/{num & 0xFFFFFFFF:X}"
+
+
+def lsn_from_hex(lsn_hex: str) -> int:
+    """Convert lsn from hex notation to int."""
+    left, right = lsn_hex.split("/")
+    return (int(left, 16) << 32) + int(right, 16)
+
+
+def remote_consistent_lsn(
+    pageserver_http_client: NeonPageserverHttpClient, tenant: uuid.UUID, timeline: uuid.UUID
+) -> int:
+    detail = pageserver_http_client.timeline_detail(tenant, timeline)
+
+    lsn_str = detail["remote_consistent_lsn"]
+    assert isinstance(lsn_str, str)
+    return lsn_from_hex(lsn_str)
+
+
+def wait_for_upload(
+    pageserver_http_client: NeonPageserverHttpClient,
+    tenant: uuid.UUID,
+    timeline: uuid.UUID,
+    lsn: int,
+):
+    """waits for local timeline upload up to specified lsn"""
+    for i in range(10):
+        current_lsn = remote_consistent_lsn(pageserver_http_client, tenant, timeline)
+        if current_lsn >= lsn:
+            return
+        print(
+            f"waiting for remote_consistent_lsn to reach {lsn_to_hex(lsn)}, now {lsn_to_hex(current_lsn)}, iteration {i + 1}"
+        )
+        time.sleep(1)
+
+    raise Exception(
+        f"timed out while waiting for remote_consistent_lsn to reach {lsn_to_hex(lsn)}, was {lsn_to_hex(current_lsn)}"
+    )
+
+
+##############
+# End of utils
+##############
+
+
+def pack_base(log_dir, restored_dir, output_tar):
+    """Create tar file from basebackup, being careful to produce relative filenames."""
+    tmp_tar_name = "tmp.tar"
+    tmp_tar_path = os.path.join(restored_dir, tmp_tar_name)
+    cmd = ["tar", "-cf", tmp_tar_name] + os.listdir(restored_dir)
+    # We actually cd into the dir and call tar from there. If we call tar from
+    # outside we won't encode filenames as relative, and they won't parse well
+    # on import.
+    subprocess_capture(log_dir, cmd, cwd=restored_dir)
+    shutil.move(tmp_tar_path, output_tar)
+
+
+def reconstruct_paths(log_dir, pg_bin, base_tar, port: int):
+    """Reconstruct what relation files should exist in the datadir by querying postgres."""
+    with tempfile.TemporaryDirectory() as restored_dir:
+        # Unpack the base tar
+        subprocess_capture(log_dir, ["tar", "-xf", base_tar, "-C", restored_dir])
+
+        # Start a vanilla postgres from the given datadir and query it to find
+        # what relfiles should exist, but possibly don't.
+        with VanillaPostgres(Path(restored_dir), pg_bin, port, init=False) as vanilla_pg:
+            vanilla_pg.configure([f"port={port}"])
+            vanilla_pg.start(log_path=os.path.join(log_dir, "tmp_pg.log"))
+
+            # Create database based on template0 because we can't connect to template0
+            query = "create database template0copy template template0"
+            vanilla_pg.safe_psql(query, user="cloud_admin")
+            vanilla_pg.safe_psql("CHECKPOINT", user="cloud_admin")
+
+            # Get all databases
+            query = "select oid, datname from pg_database"
+            oid_dbname_pairs = vanilla_pg.safe_psql(query, user="cloud_admin")
+            template0_oid = [
+                oid for (oid, database) in oid_dbname_pairs if database == "template0"
+            ][0]
+
+            # Get rel paths for each database
+            for oid, database in oid_dbname_pairs:
+                if database == "template0":
+                    # We can't connect to template0
+                    continue
+
+                query = "select relname, pg_relation_filepath(oid) from pg_class"
+                result = vanilla_pg.safe_psql(query, user="cloud_admin", dbname=database)
+                for _relname, filepath in result:
+                    if filepath is not None:
+                        if database == "template0copy":
+                            # Add all template0copy paths to template0
+                            prefix = f"base/{oid}/"
+                            if filepath.startswith(prefix):
+                                suffix = filepath[len(prefix) :]
+                                yield f"base/{template0_oid}/{suffix}"
+                            elif filepath.startswith("global"):
+                                print(f"skipping {database} global file {filepath}")
+                            else:
+                                raise AssertionError
+                        else:
+                            yield filepath
+
+
+def touch_missing_rels(log_dir, corrupt_tar, output_tar, paths):
+    """Add the appropriate empty files to a basebadkup tar."""
+    with tempfile.TemporaryDirectory() as restored_dir:
+        # Unpack the base tar
+        subprocess_capture(log_dir, ["tar", "-xf", corrupt_tar, "-C", restored_dir])
+
+        # Touch files that don't exist
+        for path in paths:
+            absolute_path = os.path.join(restored_dir, path)
+            exists = os.path.exists(absolute_path)
+            if not exists:
+                print(f"File {absolute_path} didn't exist. Creating..")
+                Path(absolute_path).touch()
+
+        # Repackage
+        pack_base(log_dir, restored_dir, output_tar)
+
+
+# HACK This is a workaround for exporting from old pageservers that
+#      can't export empty relations. In this case we need to start
+#      a vanilla postgres from the exported datadir, and query it
+#      to see what empty relations are missing, and then create
+#      those empty files before importing.
+def add_missing_rels(base_tar, output_tar, log_dir, pg_bin, tmp_pg_port: int):
+    reconstructed_paths = set(reconstruct_paths(log_dir, pg_bin, base_tar, tmp_pg_port))
+    touch_missing_rels(log_dir, base_tar, output_tar, reconstructed_paths)
+
+
+def get_rlsn(pageserver_connstr, tenant_id, timeline_id):
+    with closing(psycopg2.connect(pageserver_connstr)) as conn:
+        conn.autocommit = True
+        with conn.cursor() as cur:
+            cmd = f"get_last_record_rlsn {tenant_id} {timeline_id}"
+            cur.execute(cmd)
+            res = cur.fetchone()
+            assert res is not None
+            prev_lsn = res[0]
+            last_lsn = res[1]
+
+    return last_lsn, prev_lsn
+
+
+def import_timeline(
+    args,
+    psql_path,
+    pageserver_connstr,
+    pageserver_http,
+    tenant_id,
+    timeline_id,
+    last_lsn,
+    prev_lsn,
+    tar_filename,
+    pg_version,
+):
+    # Import timelines to new pageserver
+    import_cmd = f"import basebackup {tenant_id} {timeline_id} {last_lsn} {last_lsn} {pg_version}"
+    full_cmd = rf"""cat {tar_filename} | {psql_path} {pageserver_connstr} -c '{import_cmd}' """
+
+    stderr_filename2 = os.path.join(args.work_dir, f"import_{tenant_id}_{timeline_id}.stderr")
+    stdout_filename = os.path.join(args.work_dir, f"import_{tenant_id}_{timeline_id}.stdout")
+
+    print(f"Running: {full_cmd}")
+
+    with open(stdout_filename, "w") as stdout_f:
+        with open(stderr_filename2, "w") as stderr_f:
+            print(f"(capturing output to {stdout_filename})")
+            pg_bin = PgBin(args.work_dir, args.pg_distrib_dir, pg_version)
+            subprocess.run(
+                full_cmd,
+                stdout=stdout_f,
+                stderr=stderr_f,
+                env=pg_bin._build_env(None),
+                shell=True,
+                check=True,
+            )
+
+            print("Done import")
+
+    # Wait until pageserver persists the files
+    wait_for_upload(
+        pageserver_http, uuid.UUID(tenant_id), uuid.UUID(timeline_id), lsn_from_hex(last_lsn)
+    )
+
+
+def export_timeline(
+    args,
+    psql_path,
+    pageserver_connstr,
+    tenant_id,
+    timeline_id,
+    last_lsn,
+    prev_lsn,
+    tar_filename,
+    pg_version,
+):
+    # Choose filenames
+    incomplete_filename = tar_filename + ".incomplete"
+    stderr_filename = os.path.join(args.work_dir, f"{tenant_id}_{timeline_id}.stderr")
+
+    # Construct export command
+    query = f"fullbackup {tenant_id} {timeline_id} {last_lsn} {prev_lsn}"
+    cmd = [psql_path, "--no-psqlrc", pageserver_connstr, "-c", query]
+
+    # Run export command
+    print(f"Running: {cmd}")
+    with open(incomplete_filename, "w") as stdout_f:
+        with open(stderr_filename, "w") as stderr_f:
+            print(f"(capturing output to {incomplete_filename})")
+            pg_bin = PgBin(args.work_dir, args.pg_distrib_dir, pg_version)
+            subprocess.run(
+                cmd, stdout=stdout_f, stderr=stderr_f, env=pg_bin._build_env(None), check=True
+            )
+
+    # Add missing rels
+    pg_bin = PgBin(args.work_dir, args.pg_distrib_dir, pg_version)
+    add_missing_rels(incomplete_filename, tar_filename, args.work_dir, pg_bin, args.tmp_pg_port)
+
+    # Log more info
+    file_size = os.path.getsize(tar_filename)
+    print(f"Done export: {tar_filename}, size {file_size}")
+
+
+def main(args: argparse.Namespace):
+    # any psql version will do here. use current DEFAULT_PG_VERSION = 15
+    psql_path = str(Path(args.pg_distrib_dir) / "v15" / "bin" / "psql")
+
+    old_pageserver_host = args.old_pageserver_host
+    new_pageserver_host = args.new_pageserver_host
+
+    old_http_client = NeonPageserverHttpClient(old_pageserver_host, args.old_pageserver_http_port)
+    old_http_client.check_status()
+    old_pageserver_connstr = f"postgresql://{old_pageserver_host}:{args.old_pageserver_pg_port}"
+
+    new_http_client = NeonPageserverHttpClient(new_pageserver_host, args.new_pageserver_http_port)
+    new_http_client.check_status()
+    new_pageserver_connstr = f"postgresql://{new_pageserver_host}:{args.new_pageserver_pg_port}"
+
+    for tenant_id in args.tenants:
+        print(f"Tenant: {tenant_id}")
+        timelines = old_http_client.timeline_list(uuid.UUID(tenant_id))
+        print(f"Timelines: {timelines}")
+
+        # Create tenant in new pageserver
+        if args.only_import is False and not args.timelines:
+            new_http_client.tenant_create(uuid.UUID(tenant_id), args.ok_if_exists)
+
+        for timeline in timelines:
+            # Skip timelines we don't need to export
+            if args.timelines and timeline["timeline_id"] not in args.timelines:
+                print(f"Skipping timeline {timeline['timeline_id']}")
+                continue
+
+            # Choose filenames
+            tar_filename = os.path.join(
+                args.work_dir, f"{timeline['tenant_id']}_{timeline['timeline_id']}.tar"
+            )
+
+            pg_version = timeline["pg_version"]
+
+            # Export timeline from old pageserver
+            if args.only_import is False:
+                last_lsn, prev_lsn = get_rlsn(
+                    old_pageserver_connstr,
+                    timeline["tenant_id"],
+                    timeline["timeline_id"],
+                )
+                export_timeline(
+                    args,
+                    psql_path,
+                    old_pageserver_connstr,
+                    timeline["tenant_id"],
+                    timeline["timeline_id"],
+                    last_lsn,
+                    prev_lsn,
+                    tar_filename,
+                    pg_version,
+                )
+
+            # Import into new pageserver
+            import_timeline(
+                args,
+                psql_path,
+                new_pageserver_connstr,
+                new_http_client,
+                timeline["tenant_id"],
+                timeline["timeline_id"],
+                last_lsn,
+                prev_lsn,
+                tar_filename,
+                pg_version,
+            )
+
+            # Re-export and compare
+            re_export_filename = tar_filename + ".reexport"
+            export_timeline(
+                args,
+                psql_path,
+                new_pageserver_connstr,
+                timeline["tenant_id"],
+                timeline["timeline_id"],
+                last_lsn,
+                prev_lsn,
+                re_export_filename,
+                pg_version,
+            )
+
+            # Check the size is the same
+            old_size = (os.path.getsize(tar_filename),)
+            new_size = (os.path.getsize(re_export_filename),)
+            if old_size != new_size:
+                raise AssertionError(f"Sizes don't match old: {old_size} new: {new_size}")
+
+
+def non_zero_tcp_port(arg: Any):
+    port = int(arg)
+    if port < 1 or port > 65535:
+        raise argparse.ArgumentTypeError(f"invalid tcp port: {arg}")
+    return port
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--tenant-id",
+        dest="tenants",
+        required=True,
+        nargs="+",
+        help="Id of the tenant to migrate. You can pass multiple arguments",
+    )
+    parser.add_argument(
+        "--timeline-id",
+        dest="timelines",
+        required=False,
+        nargs="+",
+        help="Id of the timeline to migrate. You can pass multiple arguments",
+    )
+    parser.add_argument(
+        "--from-host",
+        dest="old_pageserver_host",
+        required=True,
+        help="Host of the pageserver to migrate data from",
+    )
+    parser.add_argument(
+        "--from-http-port",
+        dest="old_pageserver_http_port",
+        required=False,
+        type=int,
+        default=9898,
+        help="HTTP port of the pageserver to migrate data from. Default: 9898",
+    )
+    parser.add_argument(
+        "--from-pg-port",
+        dest="old_pageserver_pg_port",
+        required=False,
+        type=int,
+        default=6400,
+        help="pg port of the pageserver to migrate data from. Default: 6400",
+    )
+    parser.add_argument(
+        "--to-host",
+        dest="new_pageserver_host",
+        required=True,
+        help="Host of the pageserver to migrate data to",
+    )
+    parser.add_argument(
+        "--to-http-port",
+        dest="new_pageserver_http_port",
+        required=False,
+        default=9898,
+        type=int,
+        help="HTTP port of the pageserver to migrate data to. Default: 9898",
+    )
+    parser.add_argument(
+        "--to-pg-port",
+        dest="new_pageserver_pg_port",
+        required=False,
+        default=6400,
+        type=int,
+        help="pg port of the pageserver to migrate data to. Default: 6400",
+    )
+    parser.add_argument(
+        "--ignore-tenant-exists",
+        dest="ok_if_exists",
+        required=False,
+        help="Ignore error if we are trying to create the tenant that already exists. It can be dangerous if existing tenant already contains some data.",
+    )
+    parser.add_argument(
+        "--pg-distrib-dir",
+        dest="pg_distrib_dir",
+        required=False,
+        default="/usr/local/",
+        help="Path where postgres binaries are installed. Default: /usr/local/",
+    )
+    parser.add_argument(
+        "--psql-path",
+        dest="psql_path",
+        required=False,
+        default="/usr/local/v14/bin/psql",
+        help="Path to the psql binary. Default: /usr/local/v14/bin/psql",
+    )
+    parser.add_argument(
+        "--only-import",
+        dest="only_import",
+        required=False,
+        default=False,
+        action="store_true",
+        help="Skip export and tenant creation part",
+    )
+    parser.add_argument(
+        "--work-dir",
+        dest="work_dir",
+        required=True,
+        default=False,
+        help="directory where temporary tar files are stored",
+    )
+    parser.add_argument(
+        "--tmp-pg-port",
+        dest="tmp_pg_port",
+        required=False,
+        default=55439,
+        type=non_zero_tcp_port,
+        help="localhost port to use for temporary postgres instance",
+    )
+    args = parser.parse_args()
+    main(args)

storage_controller/src/heartbeater.rs

@@ -184,19 +184,6 @@ impl HeartbeaterTask {
                 }
             }
         }
-        tracing::info!(
-            "Heartbeat round complete for {} nodes, {} offline",
-            new_state.len(),
-            new_state
-                .values()
-                .filter(|s| match s {
-                    PageserverState::Available { .. } => {
-                        false
-                    }
-                    PageserverState::Offline => true,
-                })
-                .count()
-        );
 
         let mut deltas = Vec::new();
         let now = Instant::now();

storage_controller/src/main.rs

@@ -5,13 +5,10 @@ use diesel::Connection;
 use metrics::launch_timestamp::LaunchTimestamp;
 use metrics::BuildInfo;
 use std::sync::Arc;
-use std::time::Duration;
 use storage_controller::http::make_router;
 use storage_controller::metrics::preinitialize_metrics;
 use storage_controller::persistence::Persistence;
-use storage_controller::service::{
-    Config, Service, MAX_UNAVAILABLE_INTERVAL_DEFAULT, RECONCILER_CONCURRENCY_DEFAULT,
-};
+use storage_controller::service::{Config, Service, MAX_UNAVAILABLE_INTERVAL_DEFAULT};
 use tokio::signal::unix::SignalKind;
 use tokio_util::sync::CancellationToken;
 use utils::auth::{JwtAuth, SwappableJwtAuth};
@@ -66,10 +63,6 @@ struct Cli {
     /// Grace period before marking unresponsive pageserver offline
     #[arg(long)]
     max_unavailable_interval: Option<humantime::Duration>,
-
-    /// Maximum number of reconcilers that may run in parallel
-    #[arg(long)]
-    reconciler_concurrency: Option<usize>,
 }
 
 enum StrictMode {
@@ -249,14 +242,9 @@ async fn async_main() -> anyhow::Result<()> {
             .max_unavailable_interval
             .map(humantime::Duration::into)
             .unwrap_or(MAX_UNAVAILABLE_INTERVAL_DEFAULT),
-        reconciler_concurrency: args
-            .reconciler_concurrency
-            .unwrap_or(RECONCILER_CONCURRENCY_DEFAULT),
     };
 
     // After loading secrets & config, but before starting anything else, apply database migrations
-    Persistence::await_connection(&secrets.database_url, Duration::from_secs(5)).await?;
-
     migration_run(&secrets.database_url)
         .await
         .context("Running database migrations")?;

storage_controller/src/persistence.rs

@@ -2,7 +2,6 @@ pub(crate) mod split_state;
 use std::collections::HashMap;
 use std::str::FromStr;
 use std::time::Duration;
-use std::time::Instant;
 
 use self::split_state::SplitState;
 use camino::Utf8Path;
@@ -145,31 +144,6 @@ impl Persistence {
         }
     }
 
-    /// A helper for use during startup, where we would like to tolerate concurrent restarts of the
-    /// database and the storage controller, therefore the database might not be available right away
-    pub async fn await_connection(
-        database_url: &str,
-        timeout: Duration,
-    ) -> Result<(), diesel::ConnectionError> {
-        let started_at = Instant::now();
-        loop {
-            match PgConnection::establish(database_url) {
-                Ok(_) => {
-                    tracing::info!("Connected to database.");
-                    return Ok(());
-                }
-                Err(e) => {
-                    if started_at.elapsed() > timeout {
-                        return Err(e);
-                    } else {
-                        tracing::info!("Database not yet available, waiting... ({e})");
-                        tokio::time::sleep(Duration::from_millis(100)).await;
-                    }
-                }
-            }
-        }
-    }
-
     /// Wraps `with_conn` in order to collect latency and error metrics
     async fn with_measured_conn<F, R>(&self, op: DatabaseOperation, func: F) -> DatabaseResult<R>
     where

storage_controller/src/reconciler.rs

@@ -51,10 +51,6 @@ pub(super) struct Reconciler {
     /// so that we can set [`crate::tenant_shard::TenantShard::pending_compute_notification`] to ensure a later retry.
     pub(crate) compute_notify_failure: bool,
 
-    /// Reconciler is responsible for keeping alive semaphore units that limit concurrency on how many
-    /// we will spawn.
-    pub(crate) _resource_units: ReconcileUnits,
-
     /// A means to abort background reconciliation: it is essential to
     /// call this when something changes in the original TenantShard that
     /// will make this reconciliation impossible or unnecessary, for
@@ -70,19 +66,6 @@ pub(super) struct Reconciler {
     pub(crate) persistence: Arc<Persistence>,
 }
 
-/// RAII resource units granted to a Reconciler, which it should keep alive until it finishes doing I/O
-pub(crate) struct ReconcileUnits {
-    _sem_units: tokio::sync::OwnedSemaphorePermit,
-}
-
-impl ReconcileUnits {
-    pub(crate) fn new(sem_units: tokio::sync::OwnedSemaphorePermit) -> Self {
-        Self {
-            _sem_units: sem_units,
-        }
-    }
-}
-
 /// This is a snapshot of [`crate::tenant_shard::IntentState`], but it does not do any
 /// reference counting for Scheduler.  The IntentState is what the scheduler works with,
 /// and the TargetState is just the instruction for a particular Reconciler run.
@@ -767,10 +750,7 @@ impl Reconciler {
                 // It is up to the caller whether they want to drop out on this error, but they don't have to:
                 // in general we should avoid letting unavailability of the cloud control plane stop us from
                 // making progress.
-                if !matches!(e, NotifyError::ShuttingDown) {
-                    tracing::warn!("Failed to notify compute of attached pageserver {node}: {e}");
-                }
-
+                tracing::warn!("Failed to notify compute of attached pageserver {node}: {e}");
                 // Set this flag so that in our ReconcileResult we will set the flag on the shard that it
                 // needs to retry at some point.
                 self.compute_notify_failure = true;

storage_controller/src/service.rs

@@ -10,9 +10,8 @@ use std::{
 use crate::{
     id_lock_map::IdLockMap,
     persistence::{AbortShardSplitStatus, TenantFilter},
-    reconciler::{ReconcileError, ReconcileUnits},
+    reconciler::ReconcileError,
     scheduler::{ScheduleContext, ScheduleMode},
-    tenant_shard::ReconcileNeeded,
 };
 use anyhow::Context;
 use control_plane::storage_controller::{
@@ -49,7 +48,7 @@ use pageserver_api::{
     },
 };
 use pageserver_client::mgmt_api;
-use tokio::sync::{mpsc::error::TrySendError, OwnedRwLockWriteGuard};
+use tokio::sync::OwnedRwLockWriteGuard;
 use tokio_util::sync::CancellationToken;
 use tracing::instrument;
 use utils::{
@@ -91,13 +90,6 @@ pub(crate) const STARTUP_RECONCILE_TIMEOUT: Duration = Duration::from_secs(30);
 
 pub const MAX_UNAVAILABLE_INTERVAL_DEFAULT: Duration = Duration::from_secs(30);
 
-pub const RECONCILER_CONCURRENCY_DEFAULT: usize = 128;
-
-// Depth of the channel used to enqueue shards for reconciliation when they can't do it immediately.
-// This channel is finite-size to avoid using excessive memory if we get into a state where reconciles are finishing more slowly
-// than they're being pushed onto the queue.
-const MAX_DELAYED_RECONCILES: usize = 10000;
-
 // Top level state available to all HTTP handlers
 struct ServiceState {
     tenants: BTreeMap<TenantShardId, TenantShard>,
@@ -105,9 +97,6 @@ struct ServiceState {
     nodes: Arc<HashMap<NodeId, Node>>,
 
     scheduler: Scheduler,
-
-    /// Queue of tenants who are waiting for concurrency limits to permit them to reconcile
-    delayed_reconcile_rx: tokio::sync::mpsc::Receiver<TenantShardId>,
 }
 
 impl ServiceState {
@@ -115,13 +104,11 @@ impl ServiceState {
         nodes: HashMap<NodeId, Node>,
         tenants: BTreeMap<TenantShardId, TenantShard>,
         scheduler: Scheduler,
-        delayed_reconcile_rx: tokio::sync::mpsc::Receiver<TenantShardId>,
     ) -> Self {
         Self {
             tenants,
             nodes: Arc::new(nodes),
             scheduler,
-            delayed_reconcile_rx,
         }
     }
 
@@ -155,9 +142,6 @@ pub struct Config {
     /// considered active. Once the grace period elapses, the next heartbeat failure will
     /// mark the pagseserver offline.
     pub max_unavailable_interval: Duration,
-
-    /// How many Reconcilers may be spawned concurrently
-    pub reconciler_concurrency: usize,
 }
 
 impl From<DatabaseError> for ApiError {
@@ -196,17 +180,6 @@ pub struct Service {
     // that transition it to/from Active.
     node_op_locks: IdLockMap<NodeId>,
 
-    // Limit how many Reconcilers we will spawn concurrently
-    reconciler_concurrency: Arc<tokio::sync::Semaphore>,
-
-    /// Queue of tenants who are waiting for concurrency limits to permit them to reconcile
-    /// Send into this queue to promptly attempt to reconcile this shard next time units are available.
-    ///
-    /// Note that this state logically lives inside ServiceInner, but carrying Sender here makes the code simpler
-    /// by avoiding needing a &mut ref to something inside the ServiceInner.  This could be optimized to
-    /// use a VecDeque instead of a channel to reduce synchronization overhead, at the cost of some code complexity.
-    delayed_reconcile_tx: tokio::sync::mpsc::Sender<TenantShardId>,
-
     // Process shutdown will fire this token
     cancel: CancellationToken,
 
@@ -769,9 +742,8 @@ impl Service {
     }
 
     /// Apply the contents of a [`ReconcileResult`] to our in-memory state: if the reconciliation
-    /// was successful and intent hasn't changed since the Reconciler was spawned, this will update
-    /// the observed state of the tenant such that subsequent calls to [`TenantShard::get_reconcile_needed`]
-    /// will indicate that reconciliation is not needed.
+    /// was successful, this will update the observed state of the tenant such that subsequent
+    /// calls to [`TenantShard::maybe_reconcile`] will do nothing.
     #[instrument(skip_all, fields(
         tenant_id=%result.tenant_shard_id.tenant_id, shard_id=%result.tenant_shard_id.shard_slug(),
         sequence=%result.sequence
@@ -824,28 +796,14 @@ impl Service {
 
                 // Ordering: populate last_error before advancing error_seq,
                 // so that waiters will see the correct error after waiting.
-                tenant.set_last_error(result.sequence, e);
+                *(tenant.last_error.lock().unwrap()) = format!("{e}");
+                tenant.error_waiter.advance(result.sequence);
 
                 for (node_id, o) in result.observed.locations {
                     tenant.observed.locations.insert(node_id, o);
                 }
             }
         }
-
-        // Maybe some other work can proceed now that this job finished.
-        if self.reconciler_concurrency.available_permits() > 0 {
-            while let Ok(tenant_shard_id) = locked.delayed_reconcile_rx.try_recv() {
-                let (nodes, tenants, _scheduler) = locked.parts_mut();
-                if let Some(shard) = tenants.get_mut(&tenant_shard_id) {
-                    shard.delayed_reconcile = false;
-                    self.maybe_reconcile_shard(shard, nodes);
-                }
-
-                if self.reconciler_concurrency.available_permits() == 0 {
-                    break;
-                }
-            }
-        }
     }
 
     async fn process_results(
@@ -1028,9 +986,6 @@ impl Service {
 
         let (startup_completion, startup_complete) = utils::completion::channel();
 
-        let (delayed_reconcile_tx, delayed_reconcile_rx) =
-            tokio::sync::mpsc::channel(MAX_DELAYED_RECONCILES);
-
         let cancel = CancellationToken::new();
         let heartbeater = Heartbeater::new(
             config.jwt_token.clone(),
@@ -1039,20 +994,13 @@ impl Service {
         );
         let this = Arc::new(Self {
             inner: Arc::new(std::sync::RwLock::new(ServiceState::new(
-                nodes,
-                tenants,
-                scheduler,
-                delayed_reconcile_rx,
+                nodes, tenants, scheduler,
             ))),
             config: config.clone(),
             persistence,
-            compute_hook: Arc::new(ComputeHook::new(config.clone())),
+            compute_hook: Arc::new(ComputeHook::new(config)),
             result_tx,
             heartbeater,
-            reconciler_concurrency: Arc::new(tokio::sync::Semaphore::new(
-                config.reconciler_concurrency,
-            )),
-            delayed_reconcile_tx,
             abort_tx,
             startup_complete: startup_complete.clone(),
             cancel,
@@ -1587,7 +1535,7 @@ impl Service {
 
         let (response, waiters) = self.do_tenant_create(create_req).await?;
 
-        if let Err(e) = self.await_waiters(waiters, RECONCILE_TIMEOUT).await {
+        if let Err(e) = self.await_waiters(waiters, SHORT_RECONCILE_TIMEOUT).await {
             // Avoid deadlock: reconcile may fail while notifying compute, if the cloud control plane refuses to
             // accept compute notifications while it is in the process of creating.  Reconciliation will
             // be retried in the background.
@@ -2804,14 +2752,7 @@ impl Service {
                 tenant_shard_id: shard.tenant_shard_id,
                 node_attached: *shard.intent.get_attached(),
                 node_secondary: shard.intent.get_secondary().to_vec(),
-                last_error: shard
-                    .last_error
-                    .lock()
-                    .unwrap()
-                    .as_ref()
-                    .map(|e| format!("{e}"))
-                    .unwrap_or("".to_string())
-                    .clone(),
+                last_error: shard.last_error.lock().unwrap().clone(),
                 is_reconciling: shard.reconciler.is_some(),
                 is_pending_compute_notification: shard.pending_compute_notification,
                 is_splitting: matches!(shard.splitting, SplitState::Splitting),
@@ -4037,7 +3978,7 @@ impl Service {
                 // TODO: in the background, we should balance work back onto this pageserver
             }
             AvailabilityTransition::Unchanged => {
-                tracing::debug!("Node {} no change during config", node_id);
+                tracing::info!("Node {} no change during config", node_id);
             }
         }
 
@@ -4112,64 +4053,20 @@ impl Service {
         Ok(())
     }
 
-    /// Wrap [`TenantShard`] reconciliation methods with acquisition of [`Gate`] and [`ReconcileUnits`],
+    /// Convenience wrapper around [`TenantShard::maybe_reconcile`] that provides
+    /// all the references to parts of Self that are needed
     fn maybe_reconcile_shard(
         &self,
         shard: &mut TenantShard,
         nodes: &Arc<HashMap<NodeId, Node>>,
     ) -> Option<ReconcilerWaiter> {
-        let reconcile_needed = shard.get_reconcile_needed(nodes);
-
-        match reconcile_needed {
-            ReconcileNeeded::No => return None,
-            ReconcileNeeded::WaitExisting(waiter) => return Some(waiter),
-            ReconcileNeeded::Yes => {
-                // Fall through to try and acquire units for spawning reconciler
-            }
-        };
-
-        let units = match self.reconciler_concurrency.clone().try_acquire_owned() {
-            Ok(u) => ReconcileUnits::new(u),
-            Err(_) => {
-                tracing::info!(tenant_id=%shard.tenant_shard_id.tenant_id, shard_id=%shard.tenant_shard_id.shard_slug(),
-                    "Concurrency limited: enqueued for reconcile later");
-                if !shard.delayed_reconcile {
-                    match self.delayed_reconcile_tx.try_send(shard.tenant_shard_id) {
-                        Err(TrySendError::Closed(_)) => {
-                            // Weird mid-shutdown case?
-                        }
-                        Err(TrySendError::Full(_)) => {
-                            // It is safe to skip sending our ID in the channel: we will eventually get retried by the background reconcile task.
-                            tracing::warn!(
-                                "Many shards are waiting to reconcile: delayed_reconcile queue is full"
-                            );
-                        }
-                        Ok(()) => {
-                            shard.delayed_reconcile = true;
-                        }
-                    }
-                }
-
-                // We won't spawn a reconciler, but we will construct a waiter that waits for the shard's sequence
-                // number to advance.  When this function is eventually called again and succeeds in getting units,
-                // it will spawn a reconciler that makes this waiter complete.
-                return Some(shard.future_reconcile_waiter());
-            }
-        };
-
-        let Ok(gate_guard) = self.gate.enter() else {
-            // Gate closed: we're shutting down, drop out.
-            return None;
-        };
-
-        shard.spawn_reconciler(
+        shard.maybe_reconcile(
             &self.result_tx,
             nodes,
             &self.compute_hook,
             &self.config,
             &self.persistence,
-            units,
-            gate_guard,
+            &self.gate,
             &self.cancel,
         )
     }
@@ -4191,11 +4088,6 @@ impl Service {
                 schedule_context = ScheduleContext::default();
             }
 
-            // Skip checking if this shard is already enqueued for reconciliation
-            if shard.delayed_reconcile {
-                continue;
-            }
-
             // Eventual consistency: if an earlier reconcile job failed, and the shard is still
             // dirty, spawn another rone
             if self.maybe_reconcile_shard(shard, &pageservers).is_some() {
@@ -4357,26 +4249,7 @@ impl Service {
         };
 
         let waiter_count = waiters.len();
-        match self.await_waiters(waiters, RECONCILE_TIMEOUT).await {
-            Ok(()) => {}
-            Err(ReconcileWaitError::Failed(_, reconcile_error))
-                if matches!(*reconcile_error, ReconcileError::Cancel) =>
-            {
-                // Ignore reconciler cancel errors: this reconciler might have shut down
-                // because some other change superceded it.  We will return a nonzero number,
-                // so the caller knows they might have to call again to quiesce the system.
-            }
-            Err(e) => {
-                return Err(e);
-            }
-        };
-
-        tracing::info!(
-            "{} reconciles in reconcile_all, {} waiters",
-            reconciles_spawned,
-            waiter_count
-        );
-
+        self.await_waiters(waiters, RECONCILE_TIMEOUT).await?;
         Ok(waiter_count)
     }
 

storage_controller/src/tenant_shard.rs

@@ -7,7 +7,6 @@ use std::{
 use crate::{
     metrics::{self, ReconcileCompleteLabelGroup, ReconcileOutcome},
     persistence::TenantShardPersistence,
-    reconciler::ReconcileUnits,
     scheduler::{AffinityScore, MaySchedule, ScheduleContext},
 };
 use pageserver_api::controller_api::{PlacementPolicy, ShardSchedulingPolicy};
@@ -23,7 +22,7 @@ use utils::{
     generation::Generation,
     id::NodeId,
     seqwait::{SeqWait, SeqWaitError},
-    sync::gate::GateGuard,
+    sync::gate::Gate,
 };
 
 use crate::{
@@ -38,18 +37,12 @@ use crate::{
 };
 
 /// Serialization helper
-fn read_last_error<S, T>(v: &std::sync::Mutex<Option<T>>, serializer: S) -> Result<S::Ok, S::Error>
+fn read_mutex_content<S, T>(v: &std::sync::Mutex<T>, serializer: S) -> Result<S::Ok, S::Error>
 where
     S: serde::ser::Serializer,
-    T: std::fmt::Display,
+    T: Clone + std::fmt::Display,
 {
-    serializer.collect_str(
-        &v.lock()
-            .unwrap()
-            .as_ref()
-            .map(|e| format!("{e}"))
-            .unwrap_or("".to_string()),
-    )
+    serializer.collect_str(&v.lock().unwrap())
 }
 
 /// In-memory state for a particular tenant shard.
@@ -102,10 +95,6 @@ pub(crate) struct TenantShard {
     /// reconciliation, and timeline creation.
     pub(crate) splitting: SplitState,
 
-    /// If a tenant was enqueued for later reconcile due to hitting concurrency limit, this flag
-    /// is set. This flag is cleared when the tenant is popped off the delay queue.
-    pub(crate) delayed_reconcile: bool,
-
     /// Optionally wait for reconciliation to complete up to a particular
     /// sequence number.
     #[serde(skip)]
@@ -117,19 +106,15 @@ pub(crate) struct TenantShard {
     #[serde(skip)]
     pub(crate) error_waiter: std::sync::Arc<SeqWait<Sequence, Sequence>>,
 
-    /// The most recent error from a reconcile on this tenant.  This is a nested Arc
-    /// because:
-    ///  - ReconcileWaiters need to Arc-clone the overall object to read it later
-    ///  - ReconcileWaitError needs to use an `Arc<ReconcileError>` because we can construct
-    ///    many waiters for one shard, and the underlying error types are not Clone.
+    /// The most recent error from a reconcile on this tenant
     /// TODO: generalize to an array of recent events
     /// TOOD: use a ArcSwap instead of mutex for faster reads?
-    #[serde(serialize_with = "read_last_error")]
-    pub(crate) last_error: std::sync::Arc<std::sync::Mutex<Option<Arc<ReconcileError>>>>,
+    #[serde(serialize_with = "read_mutex_content")]
+    pub(crate) last_error: std::sync::Arc<std::sync::Mutex<String>>,
 
     /// If we have a pending compute notification that for some reason we weren't able to send,
-    /// set this to true. If this is set, calls to [`Self::get_reconcile_needed`] will return Yes
-    /// and trigger a Reconciler run.  This is the mechanism by which compute notifications are included in the scope
+    /// set this to true. If this is set, calls to [`Self::maybe_reconcile`] will run a task to retry
+    /// sending it.  This is the mechanism by which compute notifications are included in the scope
     /// of state that we publish externally in an eventually consistent way.
     pub(crate) pending_compute_notification: bool,
 
@@ -303,18 +288,18 @@ pub(crate) struct ReconcilerWaiter {
 
     seq_wait: std::sync::Arc<SeqWait<Sequence, Sequence>>,
     error_seq_wait: std::sync::Arc<SeqWait<Sequence, Sequence>>,
-    error: std::sync::Arc<std::sync::Mutex<Option<Arc<ReconcileError>>>>,
+    error: std::sync::Arc<std::sync::Mutex<String>>,
     seq: Sequence,
 }
 
 #[derive(thiserror::Error, Debug)]
-pub(crate) enum ReconcileWaitError {
+pub enum ReconcileWaitError {
     #[error("Timeout waiting for shard {0}")]
     Timeout(TenantShardId),
     #[error("shutting down")]
     Shutdown,
     #[error("Reconcile error on shard {0}: {1}")]
-    Failed(TenantShardId, Arc<ReconcileError>),
+    Failed(TenantShardId, String),
 }
 
 #[derive(Eq, PartialEq, Debug)]
@@ -352,8 +337,7 @@ impl ReconcilerWaiter {
                     SeqWaitError::Timeout => unreachable!()
                 })?;
 
-                return Err(ReconcileWaitError::Failed(self.tenant_shard_id,
-                    self.error.lock().unwrap().clone().expect("If error_seq_wait was advanced error was set").clone()))
+                return Err(ReconcileWaitError::Failed(self.tenant_shard_id, self.error.lock().unwrap().clone()))
             }
         }
 
@@ -369,17 +353,6 @@ pub(crate) struct ReconcilerHandle {
     cancel: CancellationToken,
 }
 
-pub(crate) enum ReconcileNeeded {
-    /// shard either doesn't need reconciliation, or is forbidden from spawning a reconciler
-    /// in its current state (e.g. shard split in progress, or ShardSchedulingPolicy forbids it)
-    No,
-    /// shard has a reconciler running, and its intent hasn't changed since that one was
-    /// spawned: wait for the existing reconciler rather than spawning a new one.
-    WaitExisting(ReconcilerWaiter),
-    /// shard needs reconciliation: call into [`TenantShard::spawn_reconciler`]
-    Yes,
-}
-
 /// When a reconcile task completes, it sends this result object
 /// to be applied to the primary TenantShard.
 pub(crate) struct ReconcileResult {
@@ -423,7 +396,6 @@ impl TenantShard {
             reconciler: None,
             splitting: SplitState::Idle,
             sequence: Sequence(1),
-            delayed_reconcile: false,
             waiter: Arc::new(SeqWait::new(Sequence(0))),
             error_waiter: Arc::new(SeqWait::new(Sequence(0))),
             last_error: Arc::default(),
@@ -859,10 +831,16 @@ impl TenantShard {
 
     #[allow(clippy::too_many_arguments)]
     #[instrument(skip_all, fields(tenant_id=%self.tenant_shard_id.tenant_id, shard_id=%self.tenant_shard_id.shard_slug()))]
-    pub(crate) fn get_reconcile_needed(
+    pub(crate) fn maybe_reconcile(
         &mut self,
+        result_tx: &tokio::sync::mpsc::UnboundedSender<ReconcileResult>,
         pageservers: &Arc<HashMap<NodeId, Node>>,
-    ) -> ReconcileNeeded {
+        compute_hook: &Arc<ComputeHook>,
+        service_config: &service::Config,
+        persistence: &Arc<Persistence>,
+        gate: &Gate,
+        cancel: &CancellationToken,
+    ) -> Option<ReconcilerWaiter> {
         // If there are any ambiguous observed states, and the nodes they refer to are available,
         // we should reconcile to clean them up.
         let mut dirty_observed = false;
@@ -884,8 +862,8 @@ impl TenantShard {
             active_nodes_dirty || dirty_observed || self.pending_compute_notification;
 
         if !do_reconcile {
-            tracing::debug!("Not dirty, no reconciliation needed.");
-            return ReconcileNeeded::No;
+            tracing::info!("Not dirty, no reconciliation needed.");
+            return None;
         }
 
         // If we are currently splitting, then never start a reconciler task: the splitting logic
@@ -893,7 +871,7 @@ impl TenantShard {
         // up top, so that we only log this message if we would otherwise have done a reconciliation.
         if !matches!(self.splitting, SplitState::Idle) {
             tracing::info!("Refusing to reconcile, splitting in progress");
-            return ReconcileNeeded::No;
+            return None;
         }
 
         // Reconcile already in flight for the current sequence?
@@ -903,7 +881,7 @@ impl TenantShard {
                     "Reconciliation already in progress for sequence {:?}",
                     self.sequence,
                 );
-                return ReconcileNeeded::WaitExisting(ReconcilerWaiter {
+                return Some(ReconcilerWaiter {
                     tenant_shard_id: self.tenant_shard_id,
                     seq_wait: self.waiter.clone(),
                     error_seq_wait: self.error_waiter.clone(),
@@ -922,67 +900,10 @@ impl TenantShard {
                 // We only reach this point if there is work to do and we're going to skip
                 // doing it: warn it obvious why this tenant isn't doing what it ought to.
                 tracing::warn!("Skipping reconcile for policy {:?}", self.scheduling_policy);
-                return ReconcileNeeded::No;
+                return None;
             }
         }
 
-        ReconcileNeeded::Yes
-    }
-
-    /// Ensure the sequence number is set to a value where waiting for this value will make us wait
-    /// for the next reconcile: i.e. it is ahead of all completed or running reconcilers.
-    ///
-    /// Constructing a ReconcilerWaiter with the resulting sequence number gives the property
-    /// that the waiter will not complete until some future Reconciler is constructed and run.
-    fn ensure_sequence_ahead(&mut self) {
-        // Find the highest sequence for which a Reconciler has previously run or is currently
-        // running
-        let max_seen = std::cmp::max(
-            self.reconciler
-                .as_ref()
-                .map(|r| r.sequence)
-                .unwrap_or(Sequence(0)),
-            std::cmp::max(self.waiter.load(), self.error_waiter.load()),
-        );
-
-        if self.sequence <= max_seen {
-            self.sequence = max_seen.next();
-        }
-    }
-
-    /// Create a waiter that will wait for some future Reconciler that hasn't been spawned yet.
-    ///
-    /// This is appropriate when you can't spawn a recociler (e.g. due to resource limits), but
-    /// you would like to wait until one gets spawned in the background.
-    pub(crate) fn future_reconcile_waiter(&mut self) -> ReconcilerWaiter {
-        self.ensure_sequence_ahead();
-
-        ReconcilerWaiter {
-            tenant_shard_id: self.tenant_shard_id,
-            seq_wait: self.waiter.clone(),
-            error_seq_wait: self.error_waiter.clone(),
-            error: self.last_error.clone(),
-            seq: self.sequence,
-        }
-    }
-
-    #[allow(clippy::too_many_arguments)]
-    #[instrument(skip_all, fields(tenant_id=%self.tenant_shard_id.tenant_id, shard_id=%self.tenant_shard_id.shard_slug()))]
-    pub(crate) fn spawn_reconciler(
-        &mut self,
-        result_tx: &tokio::sync::mpsc::UnboundedSender<ReconcileResult>,
-        pageservers: &Arc<HashMap<NodeId, Node>>,
-        compute_hook: &Arc<ComputeHook>,
-        service_config: &service::Config,
-        persistence: &Arc<Persistence>,
-        units: ReconcileUnits,
-        gate_guard: GateGuard,
-        cancel: &CancellationToken,
-    ) -> Option<ReconcilerWaiter> {
-        // Reconcile in flight for a stale sequence?  Our sequence's task will wait for it before
-        // doing our sequence's work.
-        let old_handle = self.reconciler.take();
-
         // Build list of nodes from which the reconciler should detach
         let mut detach = Vec::new();
         for node_id in self.observed.locations.keys() {
@@ -998,9 +919,18 @@ impl TenantShard {
             }
         }
 
+        // Reconcile in flight for a stale sequence?  Our sequence's task will wait for it before
+        // doing our sequence's work.
+        let old_handle = self.reconciler.take();
+
+        let Ok(gate_guard) = gate.enter() else {
+            // Shutting down, don't start a reconciler
+            return None;
+        };
+
         // Advance the sequence before spawning a reconciler, so that sequence waiters
         // can distinguish between before+after the reconcile completes.
-        self.ensure_sequence_ahead();
+        self.sequence = self.sequence.next();
 
         let reconciler_cancel = cancel.child_token();
         let reconciler_intent = TargetState::from_intent(pageservers, &self.intent);
@@ -1015,7 +945,6 @@ impl TenantShard {
             compute_hook: compute_hook.clone(),
             service_config: service_config.clone(),
             _gate_guard: gate_guard,
-            _resource_units: units,
             cancel: reconciler_cancel.clone(),
             persistence: persistence.clone(),
             compute_notify_failure: false,
@@ -1082,18 +1011,16 @@ impl TenantShard {
                         status: outcome_label,
                     });
 
-                // Constructing result implicitly drops Reconciler, freeing any ReconcileUnits before the Service might
-                // try and schedule more work in response to our result.
-                let result = ReconcileResult {
-                    sequence: reconcile_seq,
-                    result,
-                    tenant_shard_id: reconciler.tenant_shard_id,
-                    generation: reconciler.generation,
-                    observed: reconciler.observed,
-                    pending_compute_notification: reconciler.compute_notify_failure,
-                };
-
-                result_tx.send(result).ok();
+                result_tx
+                    .send(ReconcileResult {
+                        sequence: reconcile_seq,
+                        result,
+                        tenant_shard_id: reconciler.tenant_shard_id,
+                        generation: reconciler.generation,
+                        observed: reconciler.observed,
+                        pending_compute_notification: reconciler.compute_notify_failure,
+                    })
+                    .ok();
             }
             .instrument(reconciler_span),
         );
@@ -1162,13 +1089,6 @@ impl TenantShard {
         &self.scheduling_policy
     }
 
-    pub(crate) fn set_last_error(&mut self, sequence: Sequence, error: ReconcileError) {
-        // Ordering: always set last_error before advancing sequence, so that sequence
-        // waiters are guaranteed to see a Some value when they see an error.
-        *(self.last_error.lock().unwrap()) = Some(Arc::new(error));
-        self.error_waiter.advance(sequence);
-    }
-
     pub(crate) fn from_persistent(
         tsp: TenantShardPersistence,
         intent: IntentState,
@@ -1191,7 +1111,6 @@ impl TenantShard {
             error_waiter: Arc::new(SeqWait::new(Sequence::initial())),
             last_error: Arc::default(),
             pending_compute_notification: false,
-            delayed_reconcile: false,
             scheduling_policy: serde_json::from_str(&tsp.scheduling_policy).unwrap(),
         })
     }

test_runner/fixtures/neon_fixtures.py

@@ -507,11 +507,6 @@ class NeonEnvBuilder:
             self.pageserver_get_vectored_impl = "vectored"
             log.debug('Overriding pageserver get_vectored_impl config to "vectored"')
 
-        self.pageserver_get_impl: Optional[str] = None
-        if os.getenv("PAGESERVER_GET_IMPL", "") == "vectored":
-            self.pageserver_get_impl = "vectored"
-            log.debug('Overriding pageserver get_impl config to "vectored"')
-
         assert test_name.startswith(
             "test_"
         ), "Unexpectedly instantiated from outside a test function"
@@ -1083,8 +1078,6 @@ class NeonEnv:
                 ps_cfg["virtual_file_io_engine"] = self.pageserver_virtual_file_io_engine
             if config.pageserver_get_vectored_impl is not None:
                 ps_cfg["get_vectored_impl"] = config.pageserver_get_vectored_impl
-            if config.pageserver_get_impl is not None:
-                ps_cfg["get_impl"] = config.pageserver_get_impl
 
             # Create a corresponding NeonPageserver object
             self.pageservers.append(

test_runner/regress/test_attach_tenant_config.py

@@ -190,7 +190,6 @@ def test_fully_custom_config(positive_env: NeonEnv):
         "trace_read_requests": True,
         "walreceiver_connect_timeout": "13m",
         "image_layer_creation_check_threshold": 1,
-        "switch_to_aux_file_v2": True,
     }
 
     ps_http = env.pageserver.http_client()

test_runner/regress/test_broken_timeline.py

@@ -17,16 +17,11 @@ from fixtures.types import TenantId, TimelineId
 # Test restarting page server, while safekeeper and compute node keep
 # running.
 def test_local_corruption(neon_env_builder: NeonEnvBuilder):
-    if neon_env_builder.pageserver_get_impl == "vectored":
-        reconstruct_function_name = "get_values_reconstruct_data"
-    else:
-        reconstruct_function_name = "get_value_reconstruct_data"
-
     env = neon_env_builder.init_start()
 
     env.pageserver.allowed_errors.extend(
         [
-            f".*{reconstruct_function_name} for layer .*",
+            ".*get_value_reconstruct_data for layer .*",
             ".*could not find data for key.*",
             ".*is not active. Current state: Broken.*",
             ".*will not become active. Current state: Broken.*",
@@ -89,7 +84,7 @@ def test_local_corruption(neon_env_builder: NeonEnvBuilder):
     # (We don't check layer file contents on startup, when loading the timeline)
     #
     # This will change when we implement checksums for layers
-    with pytest.raises(Exception, match=f"{reconstruct_function_name} for layer ") as err:
+    with pytest.raises(Exception, match="get_value_reconstruct_data for layer ") as err:
         pg2.start()
     log.info(
         f"As expected, compute startup failed for timeline {tenant2}/{timeline2} with corrupt layers: {err}"

test_runner/regress/test_compaction.py

@@ -14,6 +14,8 @@ AGGRESIVE_COMPACTION_TENANT_CONF = {
     # Compact small layers
     "compaction_target_size": 1024**2,
     "image_creation_threshold": 2,
+    # INC-186: remove when merging the fix
+    "image_layer_creation_check_threshold": 0,
 }
 
 
@@ -77,11 +79,11 @@ page_cache_size=10
 
     non_vectored_sum = metrics.query_one("pageserver_layers_visited_per_read_global_sum")
     non_vectored_count = metrics.query_one("pageserver_layers_visited_per_read_global_count")
-    non_vectored_average = non_vectored_sum.value / non_vectored_count.value
+    non_vectored_average = non_vectored_sum / non_vectored_count
 
     vectored_sum = metrics.query_one("pageserver_layers_visited_per_vectored_read_global_sum")
     vectored_count = metrics.query_one("pageserver_layers_visited_per_vectored_read_global_count")
-    vectored_average = vectored_sum.value / vectored_count.value
+    vectored_average = vectored_sum / vectored_count
 
     log.info(f"{non_vectored_average=} {vectored_average=}")
 

test_runner/regress/test_compatibility.py

@@ -226,11 +226,6 @@ def test_forward_compatibility(
     )
 
     try:
-        # Previous version neon_local and pageserver are not aware
-        # of the new config.
-        # TODO: remove this once the code reaches main
-        neon_env_builder.pageserver_get_impl = None
-
         neon_env_builder.num_safekeepers = 3
         neon_local_binpath = neon_env_builder.neon_binpath
         env = neon_env_builder.from_repo_dir(

test_runner/regress/test_hot_standby.py

@@ -3,7 +3,7 @@ import re
 import time
 
 from fixtures.log_helper import log
-from fixtures.neon_fixtures import NeonEnv, NeonEnvBuilder, tenant_get_shards, wait_replica_caughtup
+from fixtures.neon_fixtures import NeonEnv, wait_replica_caughtup
 
 
 # Check for corrupted WAL messages which might otherwise go unnoticed if
@@ -102,80 +102,3 @@ def test_2_replicas_start(neon_simple_env: NeonEnv):
             ) as secondary2:
                 wait_replica_caughtup(primary, secondary1)
                 wait_replica_caughtup(primary, secondary2)
-
-
-# We had an issue that a standby server made GetPage requests with an
-# old LSN, based on the last-written LSN cache, to avoid waits in the
-# pageserver.  However, requesting a page with a very old LSN, such
-# that the GC horizon has already advanced past it, results in an
-# error from the pageserver:
-# "Bad request: tried to request a page version that was garbage collected"
-#
-# To avoid that, the compute<-> pageserver protocol was updated so
-# that that the standby now sends two LSNs, the old last-written LSN
-# and the current replay LSN.
-#
-# https://github.com/neondatabase/neon/issues/6211
-def test_hot_standby_gc(neon_env_builder: NeonEnvBuilder):
-    tenant_conf = {
-        # set PITR interval to be small, so we can do GC
-        "pitr_interval": "0 s",
-    }
-    env = neon_env_builder.init_start(initial_tenant_conf=tenant_conf)
-    timeline_id = env.initial_timeline
-    tenant_id = env.initial_tenant
-
-    with env.endpoints.create_start(
-        branch_name="main",
-        endpoint_id="primary",
-    ) as primary:
-        with env.endpoints.new_replica_start(
-            origin=primary,
-            endpoint_id="secondary",
-            # Protocol version 2 was introduced to fix the issue
-            # that this test exercises. With protocol version 1 it
-            # fails.
-            config_lines=["neon.protocol_version=2"],
-        ) as secondary:
-            p_cur = primary.connect().cursor()
-            p_cur.execute("CREATE EXTENSION neon_test_utils")
-            p_cur.execute("CREATE TABLE test (id int primary key) WITH (autovacuum_enabled=false)")
-            p_cur.execute("INSERT INTO test SELECT generate_series(1, 10000) AS g")
-
-            wait_replica_caughtup(primary, secondary)
-
-            s_cur = secondary.connect().cursor()
-
-            s_cur.execute("SELECT 1 WHERE pg_is_in_recovery()")
-            res = s_cur.fetchone()
-            assert res is not None
-
-            s_cur.execute("SELECT COUNT(*) FROM test")
-            res = s_cur.fetchone()
-            assert res[0] == 10000
-
-            # Clear the cache in the standby, so that when we
-            # re-execute the query, it will make GetPage
-            # requests. This does not clear the last-written LSN cache
-            # so we still remember the LSNs of the pages.
-            s_cur.execute("SELECT clear_buffer_cache()")
-
-            # Do other stuff on the primary, to advance the WAL
-            p_cur.execute("CREATE TABLE test2 AS SELECT generate_series(1, 1000000) AS g")
-
-            # Run GC. The PITR interval is very small, so this advances the GC cutoff LSN
-            # very close to the primary's current insert LSN.
-            shards = tenant_get_shards(env, tenant_id, None)
-            for tenant_shard_id, pageserver in shards:
-                client = pageserver.http_client()
-                client.timeline_checkpoint(tenant_shard_id, timeline_id)
-                client.timeline_compact(tenant_shard_id, timeline_id)
-                client.timeline_gc(tenant_shard_id, timeline_id, 0)
-
-            # Re-execute the query. The GetPage requests that this
-            # generates use old not_modified_since LSNs, older than
-            # the GC cutoff, but new request LSNs. (In protocol
-            # version 1 there was only one LSN, and this failed.)
-            s_cur.execute("SELECT COUNT(*) FROM test")
-            res = s_cur.fetchone()
-            assert res[0] == 10000

test_runner/regress/test_local_file_cache.py

@@ -4,21 +4,16 @@ import threading
 import time
 from typing import List
 
-from fixtures.neon_fixtures import DEFAULT_BRANCH_NAME, NeonEnvBuilder
+from fixtures.neon_fixtures import NeonEnv
 from fixtures.utils import query_scalar
 
 
-def test_local_file_cache_unlink(neon_env_builder: NeonEnvBuilder, build_type: str):
-    if build_type == "debug":
-        # Disable vectored read path cross validation since it makes the test time out.
-        neon_env_builder.pageserver_config_override = "validate_vectored_get=false"
-
-    env = neon_env_builder.init_start()
+def test_local_file_cache_unlink(neon_simple_env: NeonEnv):
+    env = neon_simple_env
 
     cache_dir = os.path.join(env.repo_dir, "file_cache")
     os.mkdir(cache_dir)
 
-    env.neon_cli.create_branch("empty", ancestor_branch_name=DEFAULT_BRANCH_NAME)
     env.neon_cli.create_branch("test_local_file_cache_unlink", "empty")
 
     endpoint = env.endpoints.create_start(

test_runner/regress/test_lsn_mapping.py

@@ -1,4 +1,3 @@
-import re
 import time
 from datetime import datetime, timedelta, timezone
 
@@ -110,11 +109,6 @@ def test_lsn_mapping(neon_env_builder: NeonEnvBuilder):
 
 # Test pageserver get_timestamp_of_lsn API
 def test_ts_of_lsn_api(neon_env_builder: NeonEnvBuilder):
-    if neon_env_builder.pageserver_get_impl == "vectored":
-        key_not_found_error = r".*Requested key.*not found,*"
-    else:
-        key_not_found_error = r".*could not find data for key.*"
-
     env = neon_env_builder.init_start()
 
     new_timeline_id = env.neon_cli.create_branch("test_ts_of_lsn_api")
@@ -183,8 +177,8 @@ def test_ts_of_lsn_api(neon_env_builder: NeonEnvBuilder):
             raise RuntimeError("there should have been an 'could not find data for key' error")
         except PageserverApiException as error:
             assert error.status_code == 500
-            assert re.match(key_not_found_error, str(error))
-            env.pageserver.allowed_errors.append(key_not_found_error)
+            assert str(error).startswith("could not find data for key")
+            env.pageserver.allowed_errors.append(".*could not find data for key.*")
 
         # Probe a bunch of timestamps in the valid range
         step_size = 100

test_runner/regress/test_ondemand_slru_download.py

@@ -1,131 +0,0 @@
-from typing import Optional
-
-import pytest
-from fixtures.log_helper import log
-from fixtures.neon_fixtures import NeonEnvBuilder, tenant_get_shards
-from fixtures.types import Lsn
-from fixtures.utils import query_scalar
-
-
-#
-# Test on-demand download of the pg_xact SLRUs
-#
-@pytest.mark.parametrize("shard_count", [None, 4])
-def test_ondemand_download_pg_xact(neon_env_builder: NeonEnvBuilder, shard_count: Optional[int]):
-    if shard_count is not None:
-        neon_env_builder.num_pageservers = shard_count
-
-    tenant_conf = {
-        "lazy_slru_download": "true",
-        # set PITR interval to be small, so we can do GC
-        "pitr_interval": "0 s",
-    }
-    env = neon_env_builder.init_start(
-        initial_tenant_conf=tenant_conf, initial_tenant_shard_count=shard_count
-    )
-
-    timeline_id = env.initial_timeline
-    tenant_id = env.initial_tenant
-    endpoint = env.endpoints.create_start("main")
-
-    pg_conn = endpoint.connect()
-    cur = pg_conn.cursor()
-
-    cur.execute("CREATE EXTENSION neon_test_utils")
-
-    # Create a test table
-    cur.execute("CREATE TABLE clogtest (id integer)")
-    cur.execute("INSERT INTO clogtest VALUES (1)")
-
-    # Consume a lot of XIDs, to create more pg_xact segments
-    for _ in range(1000):
-        cur.execute("select test_consume_xids(10000);")
-    cur.execute("INSERT INTO clogtest VALUES (2)")
-    for _ in range(1000):
-        cur.execute("select test_consume_xids(10000);")
-    cur.execute("INSERT INTO clogtest VALUES (2)")
-    for _ in range(1000):
-        cur.execute("select test_consume_xids(10000);")
-    cur.execute("INSERT INTO clogtest VALUES (3)")
-
-    # Restart postgres. After restart, the new instance will download the
-    # pg_xact segments lazily.
-    endpoint.stop()
-    endpoint.start()
-    pg_conn = endpoint.connect()
-    cur = pg_conn.cursor()
-
-    # Consume more WAL, so that the pageserver can compact and GC older data,
-    # including the LSN that we started the new endpoint at,
-    cur.execute("CREATE TABLE anothertable (i int, t text)")
-    cur.execute(
-        "INSERT INTO anothertable SELECT g, 'long string to consume some space' || g FROM generate_series(1, 10000) g"
-    )
-
-    # Run GC
-    shards = tenant_get_shards(env, tenant_id, None)
-    for tenant_shard_id, pageserver in shards:
-        client = pageserver.http_client()
-        client.timeline_checkpoint(tenant_shard_id, timeline_id)
-        client.timeline_compact(tenant_shard_id, timeline_id)
-        client.timeline_gc(tenant_shard_id, timeline_id, 0)
-
-    # Test that this can still on-demand download the old pg_xact segments
-    cur.execute("select xmin, xmax, * from clogtest")
-    tup = cur.fetchall()
-    log.info(f"tuples = {tup}")
-
-
-@pytest.mark.parametrize("shard_count", [None, 4])
-def test_ondemand_download_replica(neon_env_builder: NeonEnvBuilder, shard_count: Optional[int]):
-    if shard_count is not None:
-        neon_env_builder.num_pageservers = shard_count
-
-    tenant_conf = {
-        "lazy_slru_download": "true",
-    }
-    env = neon_env_builder.init_start(
-        initial_tenant_conf=tenant_conf, initial_tenant_shard_count=shard_count
-    )
-
-    endpoint = env.endpoints.create_start("main")
-
-    pg_conn = endpoint.connect()
-    cur = pg_conn.cursor()
-
-    cur.execute("CREATE EXTENSION neon_test_utils")
-
-    # Create a test table
-    cur.execute("CREATE TABLE clogtest (id integer)")
-    cur.execute("INSERT INTO clogtest VALUES (1)")
-
-    # Consume a lot of XIDs, to create more pg_xact segments
-    for _ in range(1000):
-        cur.execute("select test_consume_xids(10000);")
-
-    # Open a new connection and insert another row, but leave
-    # the transaction open
-    pg_conn2 = endpoint.connect()
-    cur2 = pg_conn2.cursor()
-    cur2.execute("BEGIN")
-    cur2.execute("INSERT INTO clogtest VALUES (2)")
-
-    # Another insert on the first connection, which is committed.
-    for _ in range(1000):
-        cur.execute("select test_consume_xids(10000);")
-    cur.execute("INSERT INTO clogtest VALUES (3)")
-
-    # Start standby at this point in time
-    lsn = Lsn(query_scalar(cur, "SELECT pg_current_wal_insert_lsn()"))
-    endpoint_at_lsn = env.endpoints.create_start(
-        branch_name="main", endpoint_id="ep-at-lsn", lsn=lsn
-    )
-
-    # Commit transaction 2, after the standby was launched.
-    cur2.execute("COMMIT")
-
-    # The replica should not see transaction 2 as committed.
-    conn_replica = endpoint_at_lsn.connect()
-    cur_replica = conn_replica.cursor()
-    cur_replica.execute("SELECT * FROM clogtest")
-    assert cur_replica.fetchall() == [(1,), (3,)]

test_runner/regress/test_pg_regress.py

@@ -18,7 +18,6 @@ from fixtures.remote_storage import s3_storage
 def test_pg_regress(
     neon_env_builder: NeonEnvBuilder,
     test_output_dir: Path,
-    build_type: str,
     pg_bin,
     capsys,
     base_dir: Path,
@@ -31,11 +30,6 @@ def test_pg_regress(
     """
     if shard_count is not None:
         neon_env_builder.num_pageservers = shard_count
-
-    if build_type == "debug":
-        # Disable vectored read path cross validation since it makes the test time out.
-        neon_env_builder.pageserver_config_override = "validate_vectored_get=false"
-
     neon_env_builder.enable_pageserver_remote_storage(s3_storage())
     neon_env_builder.enable_scrub_on_exit()
     env = neon_env_builder.init_start(initial_tenant_shard_count=shard_count)

test_runner/regress/test_read_validation.py

@@ -17,14 +17,7 @@ def test_read_validation(neon_simple_env: NeonEnv):
     env = neon_simple_env
     env.neon_cli.create_branch("test_read_validation", "empty")
 
-    endpoint = env.endpoints.create_start(
-        "test_read_validation",
-        # Use protocol version 2, because the code that constructs the V1 messages
-        # assumes that a primary always wants to read the latest version of a page,
-        # and therefore doesn't work with the test functions below to read an older
-        # page version.
-        config_lines=["neon.protocol_version=2"],
-    )
+    endpoint = env.endpoints.create_start("test_read_validation")
 
     with closing(endpoint.connect()) as con:
         with con.cursor() as c:
@@ -71,7 +64,7 @@ def test_read_validation(neon_simple_env: NeonEnv):
             log.info("Cache is clear, reading stale page version")
 
             c.execute(
-                f"select lsn, lower, upper from page_header(get_raw_page_at_lsn('foo', 'main', 0, '{first[0]}', NULL))"
+                f"select lsn, lower, upper from page_header(get_raw_page_at_lsn('foo', 'main', 0, '{first[0]}'))"
             )
             direct_first = c.fetchone()
             assert first == direct_first, "Failed fetch page at historic lsn"
@@ -84,7 +77,7 @@ def test_read_validation(neon_simple_env: NeonEnv):
             log.info("Cache is clear, reading latest page version without cache")
 
             c.execute(
-                "select lsn, lower, upper from page_header(get_raw_page_at_lsn('foo', 'main', 0, NULL, NULL))"
+                "select lsn, lower, upper from page_header(get_raw_page_at_lsn('foo', 'main', 0, NULL))"
             )
             direct_latest = c.fetchone()
             assert second == direct_latest, "Failed fetch page at latest lsn"
@@ -99,7 +92,7 @@ def test_read_validation(neon_simple_env: NeonEnv):
             )
 
             c.execute(
-                f"select lsn, lower, upper from page_header(get_raw_page_at_lsn({reln[0]}, {reln[1]}, {reln[2]}, 0, 0, '{first[0]}', NULL))"
+                f"select lsn, lower, upper from page_header(get_raw_page_at_lsn({reln[0]}, {reln[1]}, {reln[2]}, 0, 0, '{first[0]}'))"
             )
             direct_first = c.fetchone()
             assert first == direct_first, "Failed fetch page at historic lsn using oid"
@@ -109,7 +102,7 @@ def test_read_validation(neon_simple_env: NeonEnv):
             )
 
             c.execute(
-                f"select lsn, lower, upper from page_header(get_raw_page_at_lsn({reln[0]}, {reln[1]}, {reln[2]}, 0, 0, NULL, NULL))"
+                f"select lsn, lower, upper from page_header(get_raw_page_at_lsn({reln[0]}, {reln[1]}, {reln[2]}, 0, 0, NULL))"
             )
             direct_latest = c.fetchone()
             assert second == direct_latest, "Failed fetch page at latest lsn"
@@ -121,7 +114,7 @@ def test_read_validation(neon_simple_env: NeonEnv):
             )
 
             c.execute(
-                f"select lsn, lower, upper from page_header(get_raw_page_at_lsn({reln[0]}, {reln[1]}, {reln[2]}, 0, 0, '{first[0]}', NULL))"
+                f"select lsn, lower, upper from page_header(get_raw_page_at_lsn({reln[0]}, {reln[1]}, {reln[2]}, 0, 0, '{first[0]}'))"
             )
             direct_first = c.fetchone()
             assert first == direct_first, "Failed fetch page at historic lsn using oid"
@@ -140,14 +133,7 @@ def test_read_validation_neg(neon_simple_env: NeonEnv):
 
     env.pageserver.allowed_errors.append(".*invalid LSN\\(0\\) in request.*")
 
-    endpoint = env.endpoints.create_start(
-        "test_read_validation_neg",
-        # Use protocol version 2, because the code that constructs the V1 messages
-        # assumes that a primary always wants to read the latest version of a page,
-        # and therefore doesn't work with the test functions below to read an older
-        # page version.
-        config_lines=["neon.protocol_version=2"],
-    )
+    endpoint = env.endpoints.create_start("test_read_validation_neg")
 
     with closing(endpoint.connect()) as con:
         with con.cursor() as c:
@@ -157,7 +143,7 @@ def test_read_validation_neg(neon_simple_env: NeonEnv):
             log.info("read a page of a missing relation")
             try:
                 c.execute(
-                    "select lsn, lower, upper from page_header(get_raw_page_at_lsn('Unknown', 'main', 0, '0/0', NULL))"
+                    "select lsn, lower, upper from page_header(get_raw_page_at_lsn('Unknown', 'main', 0, '0/0'))"
                 )
                 raise AssertionError("query should have failed")
             except UndefinedTable as e:
@@ -169,7 +155,7 @@ def test_read_validation_neg(neon_simple_env: NeonEnv):
             log.info("read a page at lsn 0")
             try:
                 c.execute(
-                    "select lsn, lower, upper from page_header(get_raw_page_at_lsn('foo', 'main', 0, '0/0', NULL))"
+                    "select lsn, lower, upper from page_header(get_raw_page_at_lsn('foo', 'main', 0, '0/0'))"
                 )
                 raise AssertionError("query should have failed")
             except IoError as e:
@@ -178,22 +164,22 @@ def test_read_validation_neg(neon_simple_env: NeonEnv):
             log.info("Pass NULL as an input")
             expected = (None, None, None)
             c.execute(
-                "select lsn, lower, upper from page_header(get_raw_page_at_lsn(NULL, 'main', 0, '0/0', NULL))"
+                "select lsn, lower, upper from page_header(get_raw_page_at_lsn(NULL, 'main', 0, '0/0'))"
             )
             assert c.fetchone() == expected, "Expected null output"
 
             c.execute(
-                "select lsn, lower, upper from page_header(get_raw_page_at_lsn('foo', NULL, 0, '0/0', NULL))"
+                "select lsn, lower, upper from page_header(get_raw_page_at_lsn('foo', NULL, 0, '0/0'))"
             )
             assert c.fetchone() == expected, "Expected null output"
 
             c.execute(
-                "select lsn, lower, upper from page_header(get_raw_page_at_lsn('foo', 'main', NULL, '0/0', NULL))"
+                "select lsn, lower, upper from page_header(get_raw_page_at_lsn('foo', 'main', NULL, '0/0'))"
             )
             assert c.fetchone() == expected, "Expected null output"
 
             # This check is currently failing, reading beyond EOF is returning a 0-page
             log.info("Read beyond EOF")
             c.execute(
-                "select lsn, lower, upper from page_header(get_raw_page_at_lsn('foo', 'main', 1, NULL, NULL))"
+                "select lsn, lower, upper from page_header(get_raw_page_at_lsn('foo', 'main', 1, NULL))"
             )

test_runner/regress/test_tenant_relocation.py

@@ -16,6 +16,7 @@ from fixtures.pageserver.utils import (
     wait_for_upload,
     wait_tenant_status_404,
 )
+from fixtures.port_distributor import PortDistributor
 from fixtures.remote_storage import (
     LocalFsStorage,
     RemoteStorageKind,
@@ -23,6 +24,7 @@ from fixtures.remote_storage import (
 from fixtures.types import Lsn, TenantId, TimelineId
 from fixtures.utils import (
     query_scalar,
+    subprocess_capture,
     wait_until,
 )
 
@@ -182,14 +184,20 @@ def post_migration_check(endpoint: Endpoint, sum_before_migration: int, old_loca
         # A minor migration involves no storage breaking changes.
         # It is done by attaching the tenant to a new pageserver.
         "minor",
-        # In the unlikely and unfortunate event that we have to break
-        # the storage format, extend this test with the param below.
-        # "major",
+        # A major migration involves exporting a postgres datadir
+        # basebackup and importing it into the new pageserver.
+        # This kind of migration can tolerate breaking changes
+        # to storage format
+        "major",
     ],
 )
 @pytest.mark.parametrize("with_load", ["with_load", "without_load"])
 def test_tenant_relocation(
     neon_env_builder: NeonEnvBuilder,
+    port_distributor: PortDistributor,
+    test_output_dir: Path,
+    neon_binpath: Path,
+    base_dir: Path,
     method: str,
     with_load: str,
 ):
@@ -291,7 +299,40 @@ def test_tenant_relocation(
         current_lsn=current_lsn_second,
     )
 
-    if method == "minor":
+    # Migrate either by attaching from s3 or import/export basebackup
+    if method == "major":
+        cmd = [
+            "poetry",
+            "run",
+            "python",
+            str(base_dir / "scripts/export_import_between_pageservers.py"),
+            "--tenant-id",
+            str(tenant_id),
+            "--from-host",
+            "localhost",
+            "--from-http-port",
+            str(origin_http.port),
+            "--from-pg-port",
+            str(origin_ps.service_port.pg),
+            "--to-host",
+            "localhost",
+            "--to-http-port",
+            str(destination_http.port),
+            "--to-pg-port",
+            str(destination_ps.service_port.pg),
+            "--pg-distrib-dir",
+            str(neon_env_builder.pg_distrib_dir),
+            "--work-dir",
+            str(test_output_dir),
+            "--tmp-pg-port",
+            str(port_distributor.get_port()),
+        ]
+        subprocess_capture(test_output_dir, cmd, check=True)
+
+        destination_ps.allowed_errors.append(
+            ".*ignored .* unexpected bytes after the tar archive.*"
+        )
+    elif method == "minor":
         # call to attach timeline to new pageserver
         destination_ps.tenant_attach(tenant_id)
 

test_runner/regress/test_tenant_size.py

@@ -292,12 +292,33 @@ def test_single_branch_get_tenant_size_grows(
     Operate on single branch reading the tenants size after each transaction.
     """
 
-    # Disable automatic compaction and GC, and set a long PITR interval: we will expect
-    # size to always increase with writes as all writes remain within the PITR
+    # Disable automatic gc and compaction.
+    # The pitr_interval here is quite problematic, so we cannot really use it.
+    # it'd have to be calibrated per test executing env.
+
+    # there was a bug which was hidden if the create table and first batch of
+    # inserts is larger than gc_horizon. for example 0x20000 here hid the fact
+    # that there next_gc_cutoff could be smaller than initdb_lsn, which will
+    # obviously lead to issues when calculating the size.
+    gc_horizon = 0x3BA00
+
+    # it's a bit of a hack, but different versions of postgres have different
+    # amount of WAL generated for the same amount of data. so we need to
+    # adjust the gc_horizon accordingly.
+    if pg_version == PgVersion.V14:
+        gc_horizon = 0x4A000
+    elif pg_version == PgVersion.V15:
+        gc_horizon = 0x3BA00
+    elif pg_version == PgVersion.V16:
+        gc_horizon = 210000
+    else:
+        raise NotImplementedError(pg_version)
+
     tenant_config = {
         "compaction_period": "0s",
         "gc_period": "0s",
-        "pitr_interval": "3600s",
+        "pitr_interval": "0s",
+        "gc_horizon": gc_horizon,
     }
 
     env = neon_env_builder.init_start(initial_tenant_conf=tenant_config)
@@ -311,6 +332,18 @@ def test_single_branch_get_tenant_size_grows(
 
     size_debug_file = open(test_output_dir / "size_debug.html", "w")
 
+    def check_size_change(
+        current_lsn: Lsn, initdb_lsn: Lsn, gc_horizon: int, size: int, prev_size: int
+    ):
+        if current_lsn - initdb_lsn >= gc_horizon:
+            assert (
+                size >= prev_size
+            ), f"tenant_size may grow or not grow, because we only add gc_horizon amount of WAL to initial snapshot size (Currently at: {current_lsn}, Init at: {initdb_lsn})"
+        else:
+            assert (
+                size > prev_size
+            ), f"tenant_size should grow, because we continue to add WAL to initial snapshot size (Currently at: {current_lsn}, Init at: {initdb_lsn})"
+
     def get_current_consistent_size(
         env: NeonEnv,
         endpoint: Endpoint,
@@ -379,6 +412,14 @@ def test_single_branch_get_tenant_size_grows(
             )
 
             prev_size = collected_responses[-1][2]
+
+            # branch start shouldn't be past gc_horizon yet
+            # thus the size should grow as we insert more data
+            # "gc_horizon" is tuned so that it kicks in _after_ the
+            # insert phase, but before the update phase ends.
+            assert (
+                current_lsn - initdb_lsn <= gc_horizon
+            ), "Tuning of GC window is likely out-of-date"
             assert size > prev_size
 
             collected_responses.append(("INSERT", current_lsn, size))
@@ -398,7 +439,8 @@ def test_single_branch_get_tenant_size_grows(
             )
 
             prev_size = collected_responses[-1][2]
-            assert size > prev_size
+
+            check_size_change(current_lsn, initdb_lsn, gc_horizon, size, prev_size)
 
             collected_responses.append(("UPDATE", current_lsn, size))
 
@@ -415,7 +457,8 @@ def test_single_branch_get_tenant_size_grows(
             )
 
             prev_size = collected_responses[-1][2]
-            assert size > prev_size
+
+            check_size_change(current_lsn, initdb_lsn, gc_horizon, size, prev_size)
 
             collected_responses.append(("DELETE", current_lsn, size))
 
@@ -426,20 +469,20 @@ def test_single_branch_get_tenant_size_grows(
         with endpoint.cursor() as cur:
             cur.execute("DROP TABLE t0")
 
-        # Dropping the table doesn't reclaim any space
-        # from the user's point of view, because the DROP transaction is still
-        # within pitr_interval.
+        # Without setting a PITR interval, dropping the table doesn't reclaim any space
+        # from the user's point of view, because the DROP transaction is too small
+        # to fall out of gc_horizon.
         (current_lsn, size) = get_current_consistent_size(
             env, endpoint, size_debug_file, http_client, tenant_id, timeline_id
         )
-        assert size >= prev_size
-        prev_size = size
+        prev_size = collected_responses[-1][2]
+        check_size_change(current_lsn, initdb_lsn, gc_horizon, size, prev_size)
 
-        # Set a zero PITR interval to allow the DROP to impact the synthetic size
+        # Set a tiny PITR interval to allow the DROP to impact the synthetic size
         # Because synthetic size calculation uses pitr interval when available,
         # when our tenant is configured with a tiny pitr interval, dropping a table should
         # cause synthetic size to go down immediately
-        tenant_config["pitr_interval"] = "0s"
+        tenant_config["pitr_interval"] = "1ms"
         env.pageserver.http_client().set_tenant_config(tenant_id, tenant_config)
         (current_lsn, size) = get_current_consistent_size(
             env, endpoint, size_debug_file, http_client, tenant_id, timeline_id
@@ -451,6 +494,10 @@ def test_single_branch_get_tenant_size_grows(
         # defined by gc_horizon.
         collected_responses.append(("DROP", current_lsn, size))
 
+    # Should have gone past gc_horizon, otherwise gc_horizon is too large
+    bytes_written = current_lsn - initdb_lsn
+    assert bytes_written > gc_horizon
+
     # this isn't too many lines to forget for a while. observed while
     # developing these tests that locally the value is a bit more than what we
     # get in the ci.

test_runner/regress/test_vm_bits.py

@@ -173,9 +173,7 @@ def test_vm_bit_clear_on_heap_lock(neon_env_builder: NeonEnvBuilder):
     # which changes the LSN on the page.
     cur.execute("select get_raw_page( 'vmtest_lock', 'vm', 0 )")
     vm_page_in_cache = (cur.fetchall()[0][0])[8:100].hex()
-    cur.execute(
-        "select get_raw_page_at_lsn( 'vmtest_lock', 'vm', 0, pg_current_wal_insert_lsn(), NULL )"
-    )
+    cur.execute("select get_raw_page_at_lsn( 'vmtest_lock', 'vm', 0, pg_current_wal_insert_lsn() )")
     vm_page_at_pageserver = (cur.fetchall()[0][0])[8:100].hex()
 
     assert vm_page_at_pageserver == vm_page_in_cache

trace/src/main.rs

@@ -7,9 +7,7 @@ use std::{
     io::BufReader,
 };
 
-use pageserver_api::models::{
-    PagestreamFeMessage, PagestreamGetPageRequest, PagestreamProtocolVersion,
-};
+use pageserver_api::models::{PagestreamFeMessage, PagestreamGetPageRequest};
 use utils::id::{ConnectionId, TenantId, TimelineId};
 
 use clap::{Parser, Subcommand};
@@ -58,7 +56,7 @@ fn analyze_trace<R: std::io::Read>(mut reader: R) {
     let mut prev: Option<PagestreamGetPageRequest> = None;
 
     // Compute stats
-    while let Ok(msg) = PagestreamFeMessage::parse(&mut reader, PagestreamProtocolVersion::V2) {
+    while let Ok(msg) = PagestreamFeMessage::parse(&mut reader) {
         match msg {
             PagestreamFeMessage::Exists(_) => {}
             PagestreamFeMessage::Nblocks(_) => {}
@@ -91,7 +89,7 @@ fn analyze_trace<R: std::io::Read>(mut reader: R) {
 }
 
 fn dump_trace<R: std::io::Read>(mut reader: R) {
-    while let Ok(msg) = PagestreamFeMessage::parse(&mut reader, PagestreamProtocolVersion::V2) {
+    while let Ok(msg) = PagestreamFeMessage::parse(&mut reader) {
         println!("{msg:?}");
     }
 }