Merge remote-tracking branch 'origin/main' into arpad/retain_lsn_mana…

…gement

Cargo.toml

@@ -203,21 +203,10 @@ env_logger = "0.10"
 log = "0.4"
 
 ## Libraries from neondatabase/ git forks, ideally with changes to be upstreamed
-
-# We want to use the 'neon' branch for these, but there's currently one
-# incompatible change on the branch. See:
-#
-# - PR #8076 which contained changes that depended on the new changes in
-#   the rust-postgres crate, and
-# - PR #8654 which reverted those changes and made the code in proxy incompatible
-#   with the tip of the 'neon' branch again.
-#
-# When those proxy changes are re-applied (see PR #8747), we can switch using
-# the tip of the 'neon' branch again.
-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev = "20031d7a9ee1addeae6e0968e3899ae6bf01cee2" }
-postgres-protocol = { git = "https://github.com/neondatabase/rust-postgres.git", rev = "20031d7a9ee1addeae6e0968e3899ae6bf01cee2" }
-postgres-types = { git = "https://github.com/neondatabase/rust-postgres.git", rev = "20031d7a9ee1addeae6e0968e3899ae6bf01cee2" }
-tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev = "20031d7a9ee1addeae6e0968e3899ae6bf01cee2" }
+postgres = { git = "https://github.com/neondatabase/rust-postgres.git", branch = "neon" }
+postgres-protocol = { git = "https://github.com/neondatabase/rust-postgres.git", branch = "neon" }
+postgres-types = { git = "https://github.com/neondatabase/rust-postgres.git", branch = "neon" }
+tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", branch = "neon" }
 
 ## Local libraries
 compute_api = { version = "0.1", path = "./libs/compute_api/" }
@@ -255,7 +244,7 @@ tonic-build = "0.12"
 [patch.crates-io]
 
 # Needed to get `tokio-postgres-rustls` to depend on our fork.
-tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev = "20031d7a9ee1addeae6e0968e3899ae6bf01cee2" }
+tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", branch = "neon" }
 
 ################# Binary contents sections
 

libs/pageserver_api/src/record.rs

@@ -41,6 +41,11 @@ pub enum NeonWalRecord {
         file_path: String,
         content: Option<Bytes>,
     },
+    // Truncate visibility map page
+    TruncateVisibilityMap {
+        trunc_byte: usize,
+        trunc_offs: usize,
+    },
 
     /// A testing record for unit testing purposes. It supports append data to an existing image, or clear it.
     #[cfg(feature = "testing")]

libs/postgres_ffi/src/pg_constants.rs

@@ -243,8 +243,11 @@ const FSM_LEAF_NODES_PER_PAGE: usize = FSM_NODES_PER_PAGE - FSM_NON_LEAF_NODES_P
 pub const SLOTS_PER_FSM_PAGE: u32 = FSM_LEAF_NODES_PER_PAGE as u32;
 
 /* From visibilitymap.c */
-pub const VM_HEAPBLOCKS_PER_PAGE: u32 =
-    (BLCKSZ as usize - SIZEOF_PAGE_HEADER_DATA) as u32 * (8 / 2); // MAPSIZE * (BITS_PER_BYTE / BITS_PER_HEAPBLOCK)
+
+pub const VM_MAPSIZE: usize = BLCKSZ as usize - MAXALIGN_SIZE_OF_PAGE_HEADER_DATA;
+pub const VM_BITS_PER_HEAPBLOCK: usize = 2;
+pub const VM_HEAPBLOCKS_PER_BYTE: usize = 8 / VM_BITS_PER_HEAPBLOCK;
+pub const VM_HEAPBLOCKS_PER_PAGE: usize = VM_MAPSIZE * VM_HEAPBLOCKS_PER_BYTE;
 
 /* From origin.c */
 pub const REPLICATION_STATE_MAGIC: u32 = 0x1257DADE;

pageserver/src/http/routes.rs

@@ -324,6 +324,7 @@ impl From<crate::tenant::DeleteTimelineError> for ApiError {
                     .into_boxed_str(),
             ),
             a @ AlreadyInProgress(_) => ApiError::Conflict(a.to_string()),
+            Cancelled => ApiError::ResourceUnavailable("shutting down".into()),
             Other(e) => ApiError::InternalServerError(e),
         }
     }

pageserver/src/tenant.rs

@@ -759,6 +759,9 @@ pub enum DeleteTimelineError {
     #[error("Timeline deletion is already in progress")]
     AlreadyInProgress(Arc<tokio::sync::Mutex<DeleteTimelineFlow>>),
 
+    #[error("Cancelled")]
+    Cancelled,
+
     #[error(transparent)]
     Other(#[from] anyhow::Error),
 }
@@ -769,6 +772,7 @@ impl Debug for DeleteTimelineError {
             Self::NotFound => write!(f, "NotFound"),
             Self::HasChildren(c) => f.debug_tuple("HasChildren").field(c).finish(),
             Self::AlreadyInProgress(_) => f.debug_tuple("AlreadyInProgress").finish(),
+            Self::Cancelled => f.debug_tuple("Cancelled").finish(),
             Self::Other(e) => f.debug_tuple("Other").field(e).finish(),
         }
     }

pageserver/src/tenant/remote_timeline_client.rs

@@ -243,7 +243,7 @@ use self::index::IndexPart;
 use super::metadata::MetadataUpdate;
 use super::storage_layer::{Layer, LayerName, ResidentLayer};
 use super::upload_queue::{NotInitialized, SetDeletedFlagProgress};
-use super::Generation;
+use super::{DeleteTimelineError, Generation};
 
 pub(crate) use download::{
     download_index_part, download_tenant_manifest, is_temp_download_file,
@@ -1550,15 +1550,17 @@ impl RemoteTimelineClient {
     /// Prerequisites: UploadQueue should be in stopped state and deleted_at should be successfuly set.
     /// The function deletes layer files one by one, then lists the prefix to see if we leaked something
     /// deletes leaked files if any and proceeds with deletion of index file at the end.
-    pub(crate) async fn delete_all(self: &Arc<Self>) -> anyhow::Result<()> {
+    pub(crate) async fn delete_all(self: &Arc<Self>) -> Result<(), DeleteTimelineError> {
         debug_assert_current_span_has_tenant_and_timeline_id();
 
         let layers: Vec<RemotePath> = {
             let mut locked = self.upload_queue.lock().unwrap();
-            let stopped = locked.stopped_mut()?;
+            let stopped = locked.stopped_mut().map_err(DeleteTimelineError::Other)?;
 
             if !matches!(stopped.deleted_at, SetDeletedFlagProgress::Successful(_)) {
-                anyhow::bail!("deleted_at is not set")
+                return Err(DeleteTimelineError::Other(anyhow::anyhow!(
+                    "deleted_at is not set"
+                )));
             }
 
             debug_assert!(stopped.upload_queue_for_deletion.no_pending_work());
@@ -1593,23 +1595,29 @@ impl RemoteTimelineClient {
         };
 
         let layer_deletion_count = layers.len();
-        self.deletion_queue_client.push_immediate(layers).await?;
+        self.deletion_queue_client
+            .push_immediate(layers)
+            .await
+            .map_err(|_| DeleteTimelineError::Cancelled)?;
 
         // Delete the initdb.tar.zst, which is not always present, but deletion attempts of
         // inexistant objects are not considered errors.
         let initdb_path =
             remote_initdb_archive_path(&self.tenant_shard_id.tenant_id, &self.timeline_id);
         self.deletion_queue_client
             .push_immediate(vec![initdb_path])
-            .await?;
+            .await
+            .map_err(|_| DeleteTimelineError::Cancelled)?;
 
         // Do not delete index part yet, it is needed for possible retry. If we remove it first
         // 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
         // taking the burden of listing all the layers that we already know we should delete.
-        self.flush_deletion_queue().await?;
+        self.flush_deletion_queue()
+            .await
+            .map_err(|_| DeleteTimelineError::Cancelled)?;
 
         let cancel = shutdown_token();
 
@@ -1672,28 +1680,32 @@ impl RemoteTimelineClient {
         if !remaining_layers.is_empty() {
             self.deletion_queue_client
                 .push_immediate(remaining_layers)
-                .await?;
+                .await
+                .map_err(|_| DeleteTimelineError::Cancelled)?;
         }
 
         fail::fail_point!("timeline-delete-before-index-delete", |_| {
-            Err(anyhow::anyhow!(
+            Err(DeleteTimelineError::Other(anyhow::anyhow!(
                 "failpoint: timeline-delete-before-index-delete"
-            ))?
+            )))?
         });
 
         debug!("enqueuing index part deletion");
         self.deletion_queue_client
             .push_immediate([latest_index].to_vec())
-            .await?;
+            .await
+            .map_err(|_| DeleteTimelineError::Cancelled)?;
 
         // Timeline deletion is rare and we have probably emitted a reasonably number of objects: wait
         // for a flush to a persistent deletion list so that we may be sure deletion will occur.
-        self.flush_deletion_queue().await?;
+        self.flush_deletion_queue()
+            .await
+            .map_err(|_| DeleteTimelineError::Cancelled)?;
 
         fail::fail_point!("timeline-delete-after-index-delete", |_| {
-            Err(anyhow::anyhow!(
+            Err(DeleteTimelineError::Other(anyhow::anyhow!(
                 "failpoint: timeline-delete-after-index-delete"
-            ))?
+            )))?
         });
 
         info!(prefix=%timeline_storage_path, referenced=layer_deletion_count, not_referenced=%not_referenced_count, "done deleting in timeline prefix, including index_part.json");

pageserver/src/tenant/timeline.rs

@@ -5043,7 +5043,7 @@ impl Timeline {
 
             // 1. Is it newer than GC horizon cutoff point?
             if l.get_lsn_range().end > space_cutoff {
-                debug!(
+                info!(
                     "keeping {} because it's newer than space_cutoff {}",
                     l.layer_name(),
                     space_cutoff,
@@ -5054,7 +5054,7 @@ impl Timeline {
 
             // 2. It is newer than PiTR cutoff point?
             if l.get_lsn_range().end > time_cutoff {
-                debug!(
+                info!(
                     "keeping {} because it's newer than time_cutoff {}",
                     l.layer_name(),
                     time_cutoff,
@@ -5073,7 +5073,7 @@ impl Timeline {
             for retain_lsn in &retain_lsns {
                 // start_lsn is inclusive
                 if &l.get_lsn_range().start <= retain_lsn {
-                    debug!(
+                    info!(
                         "keeping {} because it's still might be referenced by child branch forked at {} is_dropped: xx is_incremental: {}",
                         l.layer_name(),
                         retain_lsn,
@@ -5088,7 +5088,7 @@ impl Timeline {
             if let Some(lsn) = &max_lsn_with_valid_lease {
                 // keep if layer start <= any of the lease
                 if &l.get_lsn_range().start <= lsn {
-                    debug!(
+                    info!(
                         "keeping {} because there is a valid lease preventing GC at {}",
                         l.layer_name(),
                         lsn,
@@ -5120,13 +5120,13 @@ impl Timeline {
             if !layers
                 .image_layer_exists(&l.get_key_range(), &(l.get_lsn_range().end..new_gc_cutoff))
             {
-                debug!("keeping {} because it is the latest layer", l.layer_name());
+                info!("keeping {} because it is the latest layer", l.layer_name());
                 result.layers_not_updated += 1;
                 continue 'outer;
             }
 
             // We didn't find any reason to keep this file, so remove it.
-            debug!(
+            info!(
                 "garbage collecting {} is_dropped: xx is_incremental: {}",
                 l.layer_name(),
                 l.is_incremental(),

pageserver/src/tenant/timeline/delete.rs

@@ -5,6 +5,7 @@ use std::{
 
 use anyhow::Context;
 use pageserver_api::{models::TimelineState, shard::TenantShardId};
+use remote_storage::DownloadError;
 use tokio::sync::OwnedMutexGuard;
 use tracing::{error, info, info_span, instrument, Instrument};
 use utils::{crashsafe, fs_ext, id::TimelineId, pausable_failpoint};
@@ -16,7 +17,7 @@ use crate::{
         metadata::TimelineMetadata,
         remote_timeline_client::{PersistIndexPartWithDeletedFlagError, RemoteTimelineClient},
         CreateTimelineCause, DeleteTimelineError, MaybeDeletedIndexPart, Tenant,
-        TimelineOrOffloaded,
+        TenantManifestError, TimelineOrOffloaded,
     },
     virtual_file::MaybeFatalIo,
 };
@@ -110,13 +111,6 @@ pub(super) async fn delete_local_timeline_directory(
     info!("finished deleting layer files, releasing locks");
 }
 
-/// Removes remote layers and an index file after them.
-async fn delete_remote_layers_and_index(
-    remote_client: &Arc<RemoteTimelineClient>,
-) -> anyhow::Result<()> {
-    remote_client.delete_all().await.context("delete_all")
-}
-
 /// It is important that this gets called when DeletionGuard is being held.
 /// For more context see comments in [`DeleteTimelineFlow::prepare`]
 async fn remove_maybe_offloaded_timeline_from_tenant(
@@ -222,11 +216,24 @@ impl DeleteTimelineFlow {
             None => {
                 let remote_client = tenant
                     .build_timeline_client(timeline.timeline_id(), tenant.remote_storage.clone());
-                let result = remote_client
+                let result = match remote_client
                     .download_index_file(&tenant.cancel)
                     .instrument(info_span!("download_index_file"))
                     .await
-                    .map_err(|e| DeleteTimelineError::Other(anyhow::anyhow!("error: {:?}", e)))?;
+                {
+                    Ok(r) => r,
+                    Err(DownloadError::NotFound) => {
+                        // Deletion is already complete
+                        tracing::info!("Timeline already deleted in remote storage");
+                        return Ok(());
+                    }
+                    Err(e) => {
+                        return Err(DeleteTimelineError::Other(anyhow::anyhow!(
+                            "error: {:?}",
+                            e
+                        )));
+                    }
+                };
                 let index_part = match result {
                     MaybeDeletedIndexPart::Deleted(p) => {
                         tracing::info!("Timeline already set as deleted in remote index");
@@ -407,7 +414,12 @@ impl DeleteTimelineFlow {
             "timeline_delete",
             async move {
                 if let Err(err) = Self::background(guard, conf, &tenant, &timeline, remote_client).await {
-                    error!("Error: {err:#}");
+                    // Only log as an error if it's not a cancellation.
+                    if matches!(err, DeleteTimelineError::Cancelled) {
+                        info!("Shutdown during timeline deletion");
+                    }else {
+                        error!("Error: {err:#}");
+                    }
                     if let TimelineOrOffloaded::Timeline(timeline) = timeline {
                         timeline.set_broken(format!("{err:#}"))
                     }
@@ -439,7 +451,7 @@ impl DeleteTimelineFlow {
             Err(anyhow::anyhow!("failpoint: timeline-delete-after-rm"))?
         });
 
-        delete_remote_layers_and_index(&remote_client).await?;
+        remote_client.delete_all().await?;
 
         pausable_failpoint!("in_progress_delete");
 
@@ -450,10 +462,10 @@ impl DeleteTimelineFlow {
         // So indeed, the tenant manifest might refer to an offloaded timeline which has already been deleted.
         // However, we handle this case in tenant loading code so the next time we attach, the issue is
         // resolved.
-        tenant
-            .store_tenant_manifest()
-            .await
-            .map_err(|e| DeleteTimelineError::Other(anyhow::anyhow!(e)))?;
+        tenant.store_tenant_manifest().await.map_err(|e| match e {
+            TenantManifestError::Cancelled => DeleteTimelineError::Cancelled,
+            _ => DeleteTimelineError::Other(e.into()),
+        })?;
 
         *guard = Self::Finished;