Revert postgres prewarming

## Problem

Compatibility tests fail from time to time due to `pg_tenant_only_port`
port collision (added in https://github.com/neondatabase/neon/pull/4731)

## Summary of changes
- replace `pg_tenant_only_port` value in config with new port
- remove old logic, than we don't need anymore
- unify config overrides

Dockerfile.compute-node

@@ -551,10 +551,8 @@ FROM build-deps AS pg-embedding-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 ENV PATH "/usr/local/pgsql/bin/:$PATH"
-# eeb3ba7c3a60c95b2604dd543c64b2f1bb4a3703 made on 15/07/2023
-# There is no release tag yet
-RUN wget https://github.com/neondatabase/pg_embedding/archive/eeb3ba7c3a60c95b2604dd543c64b2f1bb4a3703.tar.gz -O pg_embedding.tar.gz && \
-    echo "030846df723652f99a8689ce63b66fa0c23477a7fd723533ab8a6b28ab70730f pg_embedding.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/neondatabase/pg_embedding/archive/refs/tags/0.3.1.tar.gz -O pg_embedding.tar.gz && \
+    echo "c4ae84eef36fa8ec5868f6e061f39812f19ee5ba3604d428d40935685c7be512 pg_embedding.tar.gz" | sha256sum --check && \
     mkdir pg_embedding-src && cd pg_embedding-src && tar xvzf ../pg_embedding.tar.gz --strip-components=1 -C . && \
     make -j $(getconf _NPROCESSORS_ONLN) && \
     make -j $(getconf _NPROCESSORS_ONLN) install && \

compute_tools/src/bin/compute_ctl.rs

@@ -193,6 +193,13 @@ fn main() -> Result<()> {
     if !spec_set {
         // No spec provided, hang waiting for it.
         info!("no compute spec provided, waiting");
+
+        // TODO this can stall startups in the unlikely event that we bind
+        //      this compute node while it's busy prewarming. It's not too
+        //      bad because it's just 100ms and unlikely, but it's an
+        //      avoidable problem.
+        // compute.prewarm_postgres()?;
+
         let mut state = compute.state.lock().unwrap();
         while state.status != ComputeStatus::ConfigurationPending {
             state = compute.state_changed.wait(state).unwrap();

compute_tools/src/compute.rs

@@ -532,6 +532,50 @@ impl ComputeNode {
         Ok(())
     }
 
+    /// Start and stop a postgres process to warm up the VM for startup.
+    pub fn prewarm_postgres(&self) -> Result<()> {
+        info!("prewarming");
+
+        // Create pgdata
+        let pgdata = &format!("{}.warmup", self.pgdata);
+        create_pgdata(pgdata)?;
+
+        // Run initdb to completion
+        info!("running initdb");
+        let initdb_bin = Path::new(&self.pgbin).parent().unwrap().join("initdb");
+        Command::new(initdb_bin)
+            .args(["-D", pgdata])
+            .output()
+            .expect("cannot start initdb process");
+
+        // Write conf
+        use std::io::Write;
+        let conf_path = Path::new(pgdata).join("postgresql.conf");
+        let mut file = std::fs::File::create(conf_path)?;
+        writeln!(file, "shared_buffers=65536")?;
+        writeln!(file, "port=51055")?; // Nobody should be connecting
+        writeln!(file, "shared_preload_libraries = 'neon'")?;
+
+        // Start postgres
+        info!("starting postgres");
+        let mut pg = Command::new(&self.pgbin)
+            .args(["-D", pgdata])
+            .spawn()
+            .expect("cannot start postgres process");
+
+        // Stop it when it's ready
+        info!("waiting for postgres");
+        wait_for_postgres(&mut pg, Path::new(pgdata))?;
+        pg.kill()?;
+        info!("sent kill signal");
+        pg.wait()?;
+        info!("done prewarming");
+
+        // clean up
+        let _ok = fs::remove_dir_all(pgdata);
+        Ok(())
+    }
+
     /// Start Postgres as a child process and manage DBs/roles.
     /// After that this will hang waiting on the postmaster process to exit.
     #[instrument(skip_all)]

pageserver/src/tenant/disk_btree.rs

@@ -390,39 +390,42 @@ where
     }
 
     #[allow(dead_code)]
-    pub fn dump(&self) -> Result<()> {
-        self.dump_recurse(self.root_blk, &[], 0)
-    }
+    pub async fn dump(&self) -> Result<()> {
+        let mut stack = Vec::new();
 
-    fn dump_recurse(&self, blknum: u32, path: &[u8], depth: usize) -> Result<()> {
-        let blk = self.reader.read_blk(self.start_blk + blknum)?;
-        let buf: &[u8] = blk.as_ref();
+        stack.push((self.root_blk, String::new(), 0, 0, 0));
 
-        let node = OnDiskNode::<L>::deparse(buf)?;
+        while let Some((blknum, path, depth, child_idx, key_off)) = stack.pop() {
+            let blk = self.reader.read_blk(self.start_blk + blknum)?;
+            let buf: &[u8] = blk.as_ref();
+            let node = OnDiskNode::<L>::deparse(buf)?;
 
-        print!("{:indent$}", "", indent = depth * 2);
-        println!(
-            "blk #{}: path {}: prefix {}, suffix_len {}",
-            blknum,
-            hex::encode(path),
-            hex::encode(node.prefix),
-            node.suffix_len
-        );
+            if child_idx == 0 {
+                print!("{:indent$}", "", indent = depth * 2);
+                let path_prefix = stack
+                    .iter()
+                    .map(|(_blknum, path, ..)| path.as_str())
+                    .collect::<String>();
+                println!(
+                    "blk #{blknum}: path {path_prefix}{path}: prefix {}, suffix_len {}",
+                    hex::encode(node.prefix),
+                    node.suffix_len
+                );
+            }
 
-        let mut idx = 0;
-        let mut key_off = 0;
-        while idx < node.num_children {
+            if child_idx + 1 < node.num_children {
+                let key_off = key_off + node.suffix_len as usize;
+                stack.push((blknum, path.clone(), depth, child_idx + 1, key_off));
+            }
             let key = &node.keys[key_off..key_off + node.suffix_len as usize];
-            let val = node.value(idx as usize);
+            let val = node.value(child_idx as usize);
+
             print!("{:indent$}", "", indent = depth * 2 + 2);
             println!("{}: {}", hex::encode(key), hex::encode(val.0));
 
             if node.level > 0 {
-                let child_path = [path, node.prefix].concat();
-                self.dump_recurse(val.to_blknum(), &child_path, depth + 1)?;
+                stack.push((val.to_blknum(), hex::encode(node.prefix), depth + 1, 0, 0));
             }
-            idx += 1;
-            key_off += node.suffix_len as usize;
         }
         Ok(())
     }
@@ -754,8 +757,8 @@ mod tests {
         }
     }
 
-    #[test]
-    fn basic() -> Result<()> {
+    #[tokio::test]
+    async fn basic() -> Result<()> {
         let mut disk = TestDisk::new();
         let mut writer = DiskBtreeBuilder::<_, 6>::new(&mut disk);
 
@@ -775,7 +778,7 @@ mod tests {
 
         let reader = DiskBtreeReader::new(0, root_offset, disk);
 
-        reader.dump()?;
+        reader.dump().await?;
 
         // Test the `get` function on all the keys.
         for (key, val) in all_data.iter() {
@@ -835,8 +838,8 @@ mod tests {
         Ok(())
     }
 
-    #[test]
-    fn lots_of_keys() -> Result<()> {
+    #[tokio::test]
+    async fn lots_of_keys() -> Result<()> {
         let mut disk = TestDisk::new();
         let mut writer = DiskBtreeBuilder::<_, 8>::new(&mut disk);
 
@@ -856,7 +859,7 @@ mod tests {
 
         let reader = DiskBtreeReader::new(0, root_offset, disk);
 
-        reader.dump()?;
+        reader.dump().await?;
 
         use std::sync::Mutex;
 
@@ -994,8 +997,8 @@ mod tests {
     ///
     /// This test contains a particular data set, see disk_btree_test_data.rs
     ///
-    #[test]
-    fn particular_data() -> Result<()> {
+    #[tokio::test]
+    async fn particular_data() -> Result<()> {
         // Build a tree from it
         let mut disk = TestDisk::new();
         let mut writer = DiskBtreeBuilder::<_, 26>::new(&mut disk);
@@ -1022,7 +1025,7 @@ mod tests {
         })?;
         assert_eq!(count, disk_btree_test_data::TEST_DATA.len());
 
-        reader.dump()?;
+        reader.dump().await?;
 
         Ok(())
     }

pageserver/src/tenant/remote_timeline_client/index.rs

@@ -223,6 +223,45 @@ mod tests {
         assert_eq!(part, expected);
     }
 
+    #[test]
+    fn v2_indexpart_is_parsed_with_deleted_at() {
+        let example = r#"{
+            "version":2,
+            "timeline_layers":["000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9"],
+            "missing_layers":["This shouldn't fail deserialization"],
+            "layer_metadata":{
+                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9": { "file_size": 25600000 },
+                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51": { "file_size": 9007199254741001 }
+            },
+            "disk_consistent_lsn":"0/16960E8",
+            "metadata_bytes":[112,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],
+            "deleted_at": "2023-07-31T09:00:00.123"
+        }"#;
+
+        let expected = IndexPart {
+            // note this is not verified, could be anything, but exists for humans debugging.. could be the git version instead?
+            version: 2,
+            timeline_layers: HashSet::from(["000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9".parse().unwrap()]),
+            layer_metadata: HashMap::from([
+                ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9".parse().unwrap(), IndexLayerMetadata {
+                    file_size: 25600000,
+                }),
+                ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), IndexLayerMetadata {
+                    // serde_json should always parse this but this might be a double with jq for
+                    // example.
+                    file_size: 9007199254741001,
+                })
+            ]),
+            disk_consistent_lsn: "0/16960E8".parse::<Lsn>().unwrap(),
+            metadata_bytes: [112,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0].to_vec(),
+            deleted_at: Some(chrono::NaiveDateTime::parse_from_str(
+                "2023-07-31T09:00:00.123000000", "%Y-%m-%dT%H:%M:%S.%f").unwrap())
+        };
+
+        let part = serde_json::from_str::<IndexPart>(example).unwrap();
+        assert_eq!(part, expected);
+    }
+
     #[test]
     fn empty_layers_are_parsed() {
         let empty_layers_json = r#"{

pageserver/src/tenant/storage_layer/delta_layer.rs

@@ -256,7 +256,7 @@ impl Layer for DeltaLayer {
             file,
         );
 
-        tree_reader.dump()?;
+        tree_reader.dump().await?;
 
         let mut cursor = file.block_cursor();
 

pageserver/src/tenant/storage_layer/image_layer.rs

@@ -175,7 +175,7 @@ impl Layer for ImageLayer {
         let tree_reader =
             DiskBtreeReader::<_, KEY_SIZE>::new(inner.index_start_blk, inner.index_root_blk, file);
 
-        tree_reader.dump()?;
+        tree_reader.dump().await?;
 
         tree_reader.visit(&[0u8; KEY_SIZE], VisitDirection::Forwards, |key, value| {
             println!("key: {} offset {}", hex::encode(key), value);

pageserver/src/tenant/timeline.rs

@@ -1600,7 +1600,7 @@ impl Timeline {
             if let Some(imgfilename) = ImageFileName::parse_str(&fname) {
                 // create an ImageLayer struct for each image file.
                 if imgfilename.lsn > disk_consistent_lsn {
-                    warn!(
+                    info!(
                         "found future image layer {} on timeline {} disk_consistent_lsn is {}",
                         imgfilename, self.timeline_id, disk_consistent_lsn
                     );
@@ -1632,7 +1632,7 @@ impl Timeline {
                 // is 102, then it might not have been fully flushed to disk
                 // before crash.
                 if deltafilename.lsn_range.end > disk_consistent_lsn + 1 {
-                    warn!(
+                    info!(
                         "found future delta layer {} on timeline {} disk_consistent_lsn is {}",
                         deltafilename, self.timeline_id, disk_consistent_lsn
                     );
@@ -1774,7 +1774,7 @@ impl Timeline {
             match remote_layer_name {
                 LayerFileName::Image(imgfilename) => {
                     if imgfilename.lsn > up_to_date_disk_consistent_lsn {
-                        warn!(
+                        info!(
                         "found future image layer {} on timeline {} remote_consistent_lsn is {}",
                         imgfilename, self.timeline_id, up_to_date_disk_consistent_lsn
                     );
@@ -1799,7 +1799,7 @@ impl Timeline {
                     // is 102, then it might not have been fully flushed to disk
                     // before crash.
                     if deltafilename.lsn_range.end > up_to_date_disk_consistent_lsn + 1 {
-                        warn!(
+                        info!(
                             "found future delta layer {} on timeline {} remote_consistent_lsn is {}",
                             deltafilename, self.timeline_id, up_to_date_disk_consistent_lsn
                         );

pgxn/neon/file_cache.c

@@ -88,50 +88,16 @@ static LWLockId lfc_lock;
 static int   lfc_max_size;
 static int   lfc_size_limit;
 static int   lfc_free_space_watermark;
-static int   lfc_free_memory_watermark;
 static char* lfc_path;
 static  FileCacheControl* lfc_ctl;
 static shmem_startup_hook_type prev_shmem_startup_hook;
 #if PG_VERSION_NUM>=150000
 static shmem_request_hook_type prev_shmem_request_hook;
 #endif
-static int   lfc_shrinking_factor; /* power of two by which local cache size will be shrinked when lfc_free_space_watermark or lfc_free_memory_watermak are reached */
+static int   lfc_shrinking_factor; /* power of two by which local cache size will be shrinked when lfc_free_space_watermark is reached */
 
 void FileCacheMonitorMain(Datum main_arg);
 
-#ifdef __APPLE__
-
-#include <sys/types.h>
-#include <sys/sysctl.h>
-
-static size_t
-get_available_memory(void)
-{
-	size_t total;
-	size_t sizeof_total = sizeof(total);
-	if (sysctlbyname("hw.memsize", &total, &sizeof_total, NULL, 0) < 0)
-		elog(ERROR, "Failed to get amount of RAM: %m");
-
-	return total;
-}
-
-#else
-
-#include <sys/sysinfo.h>
-
-static size_t
-get_available_memory(void)
-{
-	struct sysinfo si;
-	if (sysinfo(&si) < 0)
-		elog(ERROR, "Failed to get amount of RAM: %m");
-
-	return si.totalram*si.mem_unit;
-}
-
-#endif
-
-
 static void
 lfc_shmem_startup(void)
 {
@@ -229,11 +195,10 @@ lfc_change_limit_hook(int newval, void *extra)
 }
 
 /*
- * Local file system state monitor check available free space and memory.
- * If available disk space is lower than lfc_free_space_watermark or
- * available memory is lower than lfc_free_memory_watermark then we shrink size of local cache
+ * Local file system state monitor check available free space.
+ * If it is lower than lfc_free_space_watermark then we shrink size of local cache
  * but throwing away least recently accessed chunks.
- * First time the watermark is reached cache size is divided by two,
+ * First time low space watermark is reached cache size is divided by two,
  * second time by four,... Finally we remove all chunks from local cache.
  *
  * Please notice that we are not changing lfc_cache_size: it is used to be adjusted by autoscaler.
@@ -263,27 +228,23 @@ FileCacheMonitorMain(Datum main_arg)
 	{
 		if (lfc_size_limit != 0)
 		{
-			bool shrink_cache = false;
-			if (lfc_free_space_watermark != 0)
+			struct statvfs sfs;
+			if (statvfs(lfc_path, &sfs) < 0)
 			{
-				struct statvfs sfs;
-				if (statvfs(lfc_path, &sfs) < 0)
-					elog(WARNING, "Failed to obtain status of %s: %m", lfc_path);
-				else
-					shrink_cache |= sfs.f_bavail*sfs.f_bsize < lfc_free_space_watermark*MB;
-			}
-			if (lfc_free_memory_watermark != 0)
-				shrink_cache |= get_available_memory() < lfc_free_memory_watermark*MB;
-
-			if (shrink_cache)
-			{
-				if (lfc_shrinking_factor < 31) {
-					lfc_shrinking_factor += 1;
-				}
-				lfc_change_limit_hook(lfc_size_limit >> lfc_shrinking_factor, NULL);
+				elog(WARNING, "Failed to obtain status of %s: %m", lfc_path);
 			}
 			else
-				lfc_shrinking_factor = 0; /* reset to initial value */
+			{
+				if (sfs.f_bavail*sfs.f_bsize < lfc_free_space_watermark*MB)
+				{
+					if (lfc_shrinking_factor < 31) {
+						lfc_shrinking_factor += 1;
+					}
+					lfc_change_limit_hook(lfc_size_limit >> lfc_shrinking_factor, NULL);
+				}
+				else
+					lfc_shrinking_factor = 0; /* reset to initial value */
+			}
 		}
 		pg_usleep(monitor_interval);
 	}
@@ -356,19 +317,6 @@ lfc_init(void)
 							NULL,
 							NULL);
 
-	DefineCustomIntVariable("neon.free_memory_watermark",
-							"Minimal free memory in system after reaching which local file cache will be truncated",
-							NULL,
-							&lfc_free_memory_watermark,
-							0, /* disabled by default, because iurt makes sense only when local file cache is located i tmpfs  */
-							0,
-							INT_MAX,
-							PGC_SIGHUP,
-							GUC_UNIT_MB,
-							NULL,
-							NULL,
-							NULL);
-
 	DefineCustomStringVariable("neon.file_cache_path",
 							   "Path to local file cache (can be raw device)",
 							   NULL,

proxy/src/auth/backend/classic.rs

@@ -5,7 +5,7 @@ use crate::{
     auth::{self, AuthFlow, ClientCredentials},
     compute,
     console::{self, AuthInfo, CachedNodeInfo, ConsoleReqExtra},
-    proxy::{try_wake, NUM_RETRIES_CONNECT},
+    proxy::handle_try_wake,
     sasl, scram,
     stream::PqStream,
 };
@@ -51,14 +51,15 @@ pub(super) async fn authenticate(
         }
     };
 
+    info!("compute node's state has likely changed; requesting a wake-up");
     let mut num_retries = 0;
     let mut node = loop {
-        num_retries += 1;
-        match try_wake(api, extra, creds).await? {
+        let wake_res = api.wake_compute(extra, creds).await;
+        match handle_try_wake(wake_res, num_retries)? {
+            ControlFlow::Continue(_) => num_retries += 1,
             ControlFlow::Break(n) => break n,
-            ControlFlow::Continue(_) if num_retries < NUM_RETRIES_CONNECT => continue,
-            ControlFlow::Continue(e) => return Err(e.into()),
         }
+        info!(num_retries, "retrying wake compute");
     };
     if let Some(keys) = scram_keys {
         use tokio_postgres::config::AuthKeys;

proxy/src/proxy.rs

@@ -347,11 +347,6 @@ async fn connect_to_compute_once(
         .await
 }
 
-enum ConnectionState<E> {
-    Cached(console::CachedNodeInfo),
-    Invalid(compute::ConnCfg, E),
-}
-
 #[async_trait]
 pub trait ConnectMechanism {
     type Connection;
@@ -407,70 +402,67 @@ where
 
     mechanism.update_connect_config(&mut node_info.config);
 
-    let mut num_retries = 0;
-    let mut state = ConnectionState::<M::ConnectError>::Cached(node_info);
+    // try once
+    let (config, err) = match mechanism.connect_once(&node_info, CONNECT_TIMEOUT).await {
+        Ok(res) => return Ok(res),
+        Err(e) => {
+            error!(error = ?e, "could not connect to compute node");
+            (invalidate_cache(node_info), e)
+        }
+    };
 
-    loop {
-        match state {
-            ConnectionState::Invalid(config, err) => {
-                info!("compute node's state has likely changed; requesting a wake-up");
+    let mut num_retries = 1;
 
-                let wake_res = match creds {
-                    auth::BackendType::Console(api, creds) => api.wake_compute(extra, creds).await,
-                    auth::BackendType::Postgres(api, creds) => api.wake_compute(extra, creds).await,
-                    // nothing to do?
-                    auth::BackendType::Link(_) => return Err(err.into()),
-                    // test backend
-                    auth::BackendType::Test(x) => x.wake_compute(),
-                };
+    // if we failed to connect, it's likely that the compute node was suspended, wake a new compute node
+    info!("compute node's state has likely changed; requesting a wake-up");
+    let node_info = loop {
+        let wake_res = match creds {
+            auth::BackendType::Console(api, creds) => api.wake_compute(extra, creds).await,
+            auth::BackendType::Postgres(api, creds) => api.wake_compute(extra, creds).await,
+            // nothing to do?
+            auth::BackendType::Link(_) => return Err(err.into()),
+            // test backend
+            auth::BackendType::Test(x) => x.wake_compute(),
+        };
 
-                match handle_try_wake(wake_res) {
-                    // there was an error communicating with the control plane
-                    Err(e) => return Err(e.into()),
-                    // failed to wake up but we can continue to retry
-                    Ok(ControlFlow::Continue(_)) => {
-                        state = ConnectionState::Invalid(config, err);
-                        let wait_duration = retry_after(num_retries);
-                        num_retries += 1;
-
-                        info!(num_retries, "retrying wake compute");
-                        time::sleep(wait_duration).await;
-                        continue;
-                    }
-                    // successfully woke up a compute node and can break the wakeup loop
-                    Ok(ControlFlow::Break(mut node_info)) => {
-                        node_info.config.reuse_password(&config);
-                        mechanism.update_connect_config(&mut node_info.config);
-                        state = ConnectionState::Cached(node_info)
-                    }
-                }
+        match handle_try_wake(wake_res, num_retries)? {
+            // failed to wake up but we can continue to retry
+            ControlFlow::Continue(_) => {}
+            // successfully woke up a compute node and can break the wakeup loop
+            ControlFlow::Break(mut node_info) => {
+                node_info.config.reuse_password(&config);
+                mechanism.update_connect_config(&mut node_info.config);
+                break node_info;
             }
-            ConnectionState::Cached(node_info) => {
-                match mechanism.connect_once(&node_info, CONNECT_TIMEOUT).await {
-                    Ok(res) => return Ok(res),
-                    Err(e) => {
-                        error!(error = ?e, "could not connect to compute node");
-                        if !e.should_retry(num_retries) {
-                            return Err(e.into());
-                        }
+        }
 
-                        // after the first connect failure,
-                        // we should invalidate the cache and wake up a new compute node
-                        if num_retries == 0 {
-                            state = ConnectionState::Invalid(invalidate_cache(node_info), e);
-                        } else {
-                            state = ConnectionState::Cached(node_info);
-                        }
+        let wait_duration = retry_after(num_retries);
+        num_retries += 1;
 
-                        let wait_duration = retry_after(num_retries);
-                        num_retries += 1;
+        time::sleep(wait_duration).await;
+        info!(num_retries, "retrying wake compute");
+    };
 
-                        info!(num_retries, "retrying wake compute");
-                        time::sleep(wait_duration).await;
-                    }
+    // now that we have a new node, try connect to it repeatedly.
+    // this can error for a few reasons, for instance:
+    // * DNS connection settings haven't quite propagated yet
+    info!("wake_compute success. attempting to connect");
+    loop {
+        match mechanism.connect_once(&node_info, CONNECT_TIMEOUT).await {
+            Ok(res) => return Ok(res),
+            Err(e) => {
+                error!(error = ?e, "could not connect to compute node");
+                if !e.should_retry(num_retries) {
+                    return Err(e.into());
                 }
             }
         }
+
+        let wait_duration = retry_after(num_retries);
+        num_retries += 1;
+
+        time::sleep(wait_duration).await;
+        info!(num_retries, "retrying connect_once");
     }
 }
 
@@ -478,12 +470,15 @@ where
 /// * Returns Ok(Continue(e)) if there was an error waking but retries are acceptable
 /// * Returns Ok(Break(node)) if the wakeup succeeded
 /// * Returns Err(e) if there was an error
-fn handle_try_wake(
+pub fn handle_try_wake(
     result: Result<console::CachedNodeInfo, WakeComputeError>,
+    num_retries: u32,
 ) -> Result<ControlFlow<console::CachedNodeInfo, WakeComputeError>, WakeComputeError> {
     match result {
         Err(err) => match &err {
-            WakeComputeError::ApiError(api) if api.could_retry() => Ok(ControlFlow::Continue(err)),
+            WakeComputeError::ApiError(api) if api.should_retry(num_retries) => {
+                Ok(ControlFlow::Continue(err))
+            }
             _ => Err(err),
         },
         // Ready to try again.
@@ -491,22 +486,10 @@ fn handle_try_wake(
     }
 }
 
-/// Attempts to wake up the compute node.
-pub async fn try_wake(
-    api: &impl console::Api,
-    extra: &console::ConsoleReqExtra<'_>,
-    creds: &auth::ClientCredentials<'_>,
-) -> Result<ControlFlow<console::CachedNodeInfo, WakeComputeError>, WakeComputeError> {
-    info!("compute node's state has likely changed; requesting a wake-up");
-    handle_try_wake(api.wake_compute(extra, creds).await)
-}
-
 pub trait ShouldRetry {
     fn could_retry(&self) -> bool;
     fn should_retry(&self, num_retries: u32) -> bool {
         match self {
-            // retry all errors at least once
-            _ if num_retries == 0 => true,
             _ if num_retries >= NUM_RETRIES_CONNECT => false,
             err => err.could_retry(),
         }
@@ -558,14 +541,9 @@ impl ShouldRetry for compute::ConnectionError {
     }
 }
 
-pub fn retry_after(num_retries: u32) -> time::Duration {
-    match num_retries {
-        0 => time::Duration::ZERO,
-        _ => {
-            // 3/2 = 1.5 which seems to be an ok growth factor heuristic
-            BASE_RETRY_WAIT_DURATION * 3_u32.pow(num_retries) / 2_u32.pow(num_retries)
-        }
-    }
+fn retry_after(num_retries: u32) -> time::Duration {
+    // 1.5 seems to be an ok growth factor heuristic
+    BASE_RETRY_WAIT_DURATION.mul_f64(1.5_f64.powi(num_retries as i32))
 }
 
 /// Finish client connection initialization: confirm auth success, send params, etc.

proxy/src/proxy/tests.rs

@@ -302,7 +302,7 @@ async fn scram_auth_mock() -> anyhow::Result<()> {
 #[test]
 fn connect_compute_total_wait() {
     let mut total_wait = tokio::time::Duration::ZERO;
-    for num_retries in 0..10 {
+    for num_retries in 1..10 {
         total_wait += retry_after(num_retries);
     }
     assert!(total_wait < tokio::time::Duration::from_secs(12));

safekeeper/src/bin/safekeeper.rs

@@ -234,7 +234,10 @@ async fn start_safekeeper(conf: SafeKeeperConf) -> Result<()> {
                 listen_pg_addr_tenant_only
             );
             let listener = tcp_listener::bind(listen_pg_addr_tenant_only.clone()).map_err(|e| {
-                error!("failed to bind to address {}: {}", conf.listen_pg_addr, e);
+                error!(
+                    "failed to bind to address {}: {}",
+                    listen_pg_addr_tenant_only, e
+                );
                 e
             })?;
             Some(listener)

test_runner/regress/test_compatibility.py

@@ -257,28 +257,15 @@ def prepare_snapshot(
         shutil.rmtree(repo_dir / "pgdatadirs")
     os.mkdir(repo_dir / "endpoints")
 
-    # Remove wal-redo temp directory if it exists. Newer pageserver versions don't create
-    # them anymore, but old versions did.
-    for tenant in (repo_dir / "tenants").glob("*"):
-        wal_redo_dir = tenant / "wal-redo-datadir.___temp"
-        if wal_redo_dir.exists() and wal_redo_dir.is_dir():
-            shutil.rmtree(wal_redo_dir)
-
     # Update paths and ports in config files
     pageserver_toml = repo_dir / "pageserver.toml"
     pageserver_config = toml.load(pageserver_toml)
     pageserver_config["remote_storage"]["local_path"] = str(repo_dir / "local_fs_remote_storage")
-    pageserver_config["listen_http_addr"] = port_distributor.replace_with_new_port(
-        pageserver_config["listen_http_addr"]
-    )
-    pageserver_config["listen_pg_addr"] = port_distributor.replace_with_new_port(
-        pageserver_config["listen_pg_addr"]
-    )
+    for param in ("listen_http_addr", "listen_pg_addr", "broker_endpoint"):
+        pageserver_config[param] = port_distributor.replace_with_new_port(pageserver_config[param])
 
-    # Older pageserver versions had just one `auth_type` setting. Now there
-    # are separate settings for pg and http ports. We don't use authentication
-    # in compatibility tests so just remove authentication related settings.
-    pageserver_config.pop("auth_type", None)
+    # We don't use authentication in compatibility tests
+    # so just remove authentication related settings.
     pageserver_config.pop("pg_auth_type", None)
     pageserver_config.pop("http_auth_type", None)
 
@@ -290,19 +277,16 @@ def prepare_snapshot(
 
     snapshot_config_toml = repo_dir / "config"
     snapshot_config = toml.load(snapshot_config_toml)
-
-    broker_listen_addr = f"127.0.0.1:{port_distributor.get_port()}"
-    snapshot_config["broker"] = {"listen_addr": broker_listen_addr}
-
-    snapshot_config["pageserver"]["listen_http_addr"] = port_distributor.replace_with_new_port(
-        snapshot_config["pageserver"]["listen_http_addr"]
-    )
-    snapshot_config["pageserver"]["listen_pg_addr"] = port_distributor.replace_with_new_port(
-        snapshot_config["pageserver"]["listen_pg_addr"]
+    for param in ("listen_http_addr", "listen_pg_addr"):
+        snapshot_config["pageserver"][param] = port_distributor.replace_with_new_port(
+            snapshot_config["pageserver"][param]
+        )
+    snapshot_config["broker"]["listen_addr"] = port_distributor.replace_with_new_port(
+        snapshot_config["broker"]["listen_addr"]
     )
     for sk in snapshot_config["safekeepers"]:
-        sk["http_port"] = port_distributor.replace_with_new_port(sk["http_port"])
-        sk["pg_port"] = port_distributor.replace_with_new_port(sk["pg_port"])
+        for param in ("http_port", "pg_port", "pg_tenant_only_port"):
+            sk[param] = port_distributor.replace_with_new_port(sk[param])
 
     if pg_distrib_dir:
         snapshot_config["pg_distrib_dir"] = str(pg_distrib_dir)

test_runner/regress/test_gc_cutoff.py

@@ -14,10 +14,6 @@ from fixtures.neon_fixtures import NeonEnvBuilder, PgBin
 def test_gc_cutoff(neon_env_builder: NeonEnvBuilder, pg_bin: PgBin):
     env = neon_env_builder.init_start()
 
-    # These warnings are expected, when the pageserver is restarted abruptly
-    env.pageserver.allowed_errors.append(".*found future image layer.*")
-    env.pageserver.allowed_errors.append(".*found future delta layer.*")
-
     pageserver_http = env.pageserver.http_client()
 
     # Use aggressive GC and checkpoint settings, so that we also exercise GC during the test

test_runner/regress/test_pageserver_restart.py

@@ -72,10 +72,6 @@ def test_pageserver_restart(neon_env_builder: NeonEnvBuilder):
 def test_pageserver_chaos(neon_env_builder: NeonEnvBuilder):
     env = neon_env_builder.init_start()
 
-    # These warnings are expected, when the pageserver is restarted abruptly
-    env.pageserver.allowed_errors.append(".*found future image layer.*")
-    env.pageserver.allowed_errors.append(".*found future delta layer.*")
-
     # Use a tiny checkpoint distance, to create a lot of layers quickly.
     # That allows us to stress the compaction and layer flushing logic more.
     tenant, _ = env.neon_cli.create_tenant(

test_runner/regress/test_recovery.py

@@ -15,10 +15,6 @@ def test_pageserver_recovery(neon_env_builder: NeonEnvBuilder):
     env = neon_env_builder.init_start()
     env.pageserver.is_testing_enabled_or_skip()
 
-    # These warnings are expected, when the pageserver is restarted abruptly
-    env.pageserver.allowed_errors.append(".*found future delta layer.*")
-    env.pageserver.allowed_errors.append(".*found future image layer.*")
-
     # Create a branch for us
     env.neon_cli.create_branch("test_pageserver_recovery", "main")
 

test_runner/regress/test_remote_storage.py

@@ -348,9 +348,6 @@ def test_remote_storage_upload_queue_retries(
     # XXX: should vary this test to selectively fail just layer uploads, index uploads, deletions
     #      but how do we validate the result after restore?
 
-    # these are always possible when we do an immediate stop. perhaps something with compacting has changed since.
-    env.pageserver.allowed_errors.append(r".*found future (delta|image) layer.*")
-
     env.pageserver.stop(immediate=True)
     env.endpoints.stop_all()