walingest createdb record both for CREATE_DATABASE_FROM_WAL and CREATE_DATABASE_FILE_COPY

.github/workflows/build_and_test.yml

@@ -834,7 +834,7 @@ jobs:
       run:
         shell: sh -eu {0}
     env:
-      VM_BUILDER_VERSION: v0.17.12
+      VM_BUILDER_VERSION: v0.17.10
 
     steps:
       - name: Checkout
@@ -1091,9 +1091,8 @@ jobs:
           GH_TOKEN: ${{ secrets.CI_ACCESS_TOKEN }}
         run: |
           if [[ "$GITHUB_REF_NAME" == "main" ]]; then
-            gh workflow --repo neondatabase/aws run deploy-dev.yml --ref main -f branch=main -f dockerTag=${{needs.tag.outputs.build-tag}} -f deployPreprodRegion=false
+            gh workflow --repo neondatabase/aws run deploy-dev.yml --ref main -f branch=main -f dockerTag=${{needs.tag.outputs.build-tag}}
           elif [[ "$GITHUB_REF_NAME" == "release" ]]; then
-            gh workflow --repo neondatabase/aws run deploy-dev.yml --ref main -f branch=main -f dockerTag=${{needs.tag.outputs.build-tag}} -f deployPreprodRegion=true
             gh workflow --repo neondatabase/aws run deploy-prod.yml --ref main -f branch=main -f dockerTag=${{needs.tag.outputs.build-tag}} -f disclamerAcknowledged=true
           else
             echo "GITHUB_REF_NAME (value '$GITHUB_REF_NAME') is not set to either 'main' or 'release'"

Cargo.lock

@@ -636,7 +636,7 @@ dependencies = [
  "sha1",
  "sync_wrapper",
  "tokio",
- "tokio-tungstenite",
+ "tokio-tungstenite 0.20.0",
  "tower",
  "tower-layer",
  "tower-service",
@@ -1941,15 +1941,15 @@ dependencies = [
 
 [[package]]
 name = "hyper-tungstenite"
-version = "0.11.1"
+version = "0.9.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7cc7dcb1ab67cd336f468a12491765672e61a3b6b148634dbfe2fe8acd3fe7d9"
+checksum = "880b8b1c98a5ec2a505c7c90db6d3f6f1f480af5655d9c5b55facc9382a5a5b5"
 dependencies = [
  "hyper",
- "pin-project-lite",
+ "pin-project",
  "tokio",
- "tokio-tungstenite",
- "tungstenite",
+ "tokio-tungstenite 0.18.0",
+ "tungstenite 0.18.0",
 ]
 
 [[package]]
@@ -2908,9 +2908,9 @@ dependencies = [
 
 [[package]]
 name = "pin-project-lite"
-version = "0.2.13"
+version = "0.2.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8afb450f006bf6385ca15ef45d71d2288452bc3683ce2e2cacc0d18e4be60b58"
+checksum = "e0a7ae3ac2f1173085d398531c705756c94a4c56843785df85a60c1a0afac116"
 
 [[package]]
 name = "pin-utils"
@@ -3246,7 +3246,6 @@ dependencies = [
  "reqwest-tracing",
  "routerify",
  "rstest",
- "rustc-hash",
  "rustls",
  "rustls-pemfile",
  "scopeguard",
@@ -3418,7 +3417,6 @@ dependencies = [
  "metrics",
  "once_cell",
  "pin-project-lite",
- "rand",
  "scopeguard",
  "serde",
  "serde_json",
@@ -4643,6 +4641,18 @@ dependencies = [
  "xattr",
 ]
 
+[[package]]
+name = "tokio-tungstenite"
+version = "0.18.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "54319c93411147bced34cb5609a80e0a8e44c5999c93903a81cd866630ec0bfd"
+dependencies = [
+ "futures-util",
+ "log",
+ "tokio",
+ "tungstenite 0.18.0",
+]
+
 [[package]]
 name = "tokio-tungstenite"
 version = "0.20.0"
@@ -4652,7 +4662,7 @@ dependencies = [
  "futures-util",
  "log",
  "tokio",
- "tungstenite",
+ "tungstenite 0.20.0",
 ]
 
 [[package]]
@@ -4967,9 +4977,28 @@ checksum = "3528ecfd12c466c6f163363caf2d02a71161dd5e1cc6ae7b34207ea2d42d81ed"
 
 [[package]]
 name = "tungstenite"
-version = "0.20.1"
+version = "0.18.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9e3dac10fd62eaf6617d3a904ae222845979aec67c615d1c842b4002c7666fb9"
+checksum = "30ee6ab729cd4cf0fd55218530c4522ed30b7b6081752839b68fcec8d0960788"
+dependencies = [
+ "base64 0.13.1",
+ "byteorder",
+ "bytes",
+ "http",
+ "httparse",
+ "log",
+ "rand",
+ "sha1",
+ "thiserror",
+ "url",
+ "utf-8",
+]
+
+[[package]]
+name = "tungstenite"
+version = "0.20.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e862a1c4128df0112ab625f55cd5c934bcb4312ba80b39ae4b4835a3fd58e649"
 dependencies = [
  "byteorder",
  "bytes",
@@ -5619,7 +5648,6 @@ dependencies = [
  "tower",
  "tracing",
  "tracing-core",
- "tungstenite",
  "url",
  "uuid",
 ]

Cargo.toml

@@ -78,7 +78,7 @@ hostname = "0.3.1"
 humantime = "2.1"
 humantime-serde = "1.1.1"
 hyper = "0.14"
-hyper-tungstenite = "0.11"
+hyper-tungstenite = "0.9"
 inotify = "0.10.2"
 itertools = "0.10"
 jsonwebtoken = "8"
@@ -107,7 +107,6 @@ reqwest-middleware = "0.2.0"
 reqwest-retry = "0.2.2"
 routerify = "3"
 rpds = "0.13"
-rustc-hash = "1.1.0"
 rustls = "0.21"
 rustls-pemfile = "1"
 rustls-split = "0.3"

Dockerfile.compute-node

@@ -589,7 +589,8 @@ RUN case "${PG_VERSION}" in \
     echo "${PG_EMBEDDING_CHECKSUM} 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
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/embedding.control
 
 #########################################################################################
 #

Makefile

@@ -153,6 +153,18 @@ neon-pg-ext-%: postgres-%
 		-C $(POSTGRES_INSTALL_DIR)/build/neon-utils-$* \
 		-f $(ROOT_PROJECT_DIR)/pgxn/neon_utils/Makefile install
 
+# pg_embedding was temporarily released as hnsw from this repo, when we only
+# supported PostgreSQL 14 and 15
+neon-pg-ext-v14: neon-pg-ext-hnsw-v14
+neon-pg-ext-v15: neon-pg-ext-hnsw-v15
+
+neon-pg-ext-hnsw-%: postgres-headers-% postgres-%
+	+@echo "Compiling hnsw $*"
+	mkdir -p $(POSTGRES_INSTALL_DIR)/build/hnsw-$*
+	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/$*/bin/pg_config CFLAGS='$(PG_CFLAGS) $(COPT)' \
+		-C $(POSTGRES_INSTALL_DIR)/build/hnsw-$* \
+		-f $(ROOT_PROJECT_DIR)/pgxn/hnsw/Makefile install
+
 .PHONY: neon-pg-ext-clean-%
 neon-pg-ext-clean-%:
 	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/$*/bin/pg_config \
@@ -167,6 +179,9 @@ neon-pg-ext-clean-%:
 	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/$*/bin/pg_config \
 	-C $(POSTGRES_INSTALL_DIR)/build/neon-utils-$* \
 	-f $(ROOT_PROJECT_DIR)/pgxn/neon_utils/Makefile clean
+	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/$*/bin/pg_config \
+	-C $(POSTGRES_INSTALL_DIR)/build/hnsw-$* \
+	-f $(ROOT_PROJECT_DIR)/pgxn/hnsw/Makefile clean
 
 .PHONY: neon-pg-ext
 neon-pg-ext: \

README.md

@@ -29,13 +29,13 @@ See developer documentation in [SUMMARY.md](/docs/SUMMARY.md) for more informati
 ```bash
 apt install build-essential libtool libreadline-dev zlib1g-dev flex bison libseccomp-dev \
 libssl-dev clang pkg-config libpq-dev cmake postgresql-client protobuf-compiler \
-libcurl4-openssl-dev openssl python-poetry lsof libicu-dev
+libcurl4-openssl-dev openssl python-poetry lsof
 ```
 * On Fedora, these packages are needed:
 ```bash
 dnf install flex bison readline-devel zlib-devel openssl-devel \
   libseccomp-devel perl clang cmake postgresql postgresql-contrib protobuf-compiler \
-  protobuf-devel libcurl-devel openssl poetry lsof libicu-devel
+  protobuf-devel libcurl-devel openssl poetry lsof
 ```
 * On Arch based systems, these packages are needed:
 ```bash

control_plane/src/bin/attachment_service.rs

@@ -223,7 +223,6 @@ async fn handle_attach_hook(mut req: Request<Body>) -> Result<Response<Body>, Ap
     if attach_req.pageserver_id.is_some() {
         tenant_state.generation += 1;
     }
-    tenant_state.pageserver = attach_req.pageserver_id;
     let generation = tenant_state.generation;
 
     locked.save().await.map_err(ApiError::InternalServerError)?;

docs/rfcs/028-pageserver-migration.md

@@ -1,599 +0,0 @@
-# Seamless tenant migration
-
-- Author: john@neon.tech
-- Created on 2023-08-11
-- Implemented on ..
-
-## Summary
-
-The preceding [generation numbers RFC](025-generation-numbers.md) may be thought of as "making tenant
-migration safe". Following that,
-this RFC is about how those migrations are to be done:
-
-1. Seamlessly (without interruption to client availability)
-2. Quickly (enabling faster operations)
-3. Efficiently (minimizing I/O and $ cost)
-
-These points are in priority order: if we have to sacrifice
-efficiency to make a migration seamless for clients, we will
-do so, etc.
-
-This is accomplished by introducing two high level changes:
-
-- A dual-attached state for tenants, used in a control-plane-orchestrated
-  migration procedure that preserves availability during a migration.
-- Warm secondary locations for tenants, where on-disk content is primed
-  for a fast migration of the tenant from its current attachment to this
-  secondary location.
-
-## Motivation
-
-Migrating tenants between pageservers is essential to operating a service
-at scale, in several contexts:
-
-1. Responding to a pageserver node failure by migrating tenants to other pageservers
-2. Balancing load and capacity across pageservers, for example when a user expands their
-   database and they need to migrate to a pageserver with more capacity.
-3. Restarting pageservers for upgrades and maintenance
-
-The current situation steps for migration are:
-
-- detach from old node; skip if old node is dead; (the [skip part is still WIP](https://github.com/neondatabase/cloud/issues/5426)).
-- attach to new node
-- re-configure endpoints to use the new node
-
-Once [generation numbers](025-generation-numbers.md) are implemented,
-the detach step is no longer critical for correctness. So, we can
-
-- attach to a new node,
-- re-configure endpoints to use the new node, and then
-- detach from the old node.
-
-However, this still does not meet our seamless/fast/efficient goals:
-
-- Not fast: The new node will have to download potentially large amounts
-  of data from S3, which may take many minutes.
-- Not seamless: If we attach to a new pageserver before detaching an old one,
-  the new one might delete some objects that interrupt availability of reads on the old one.
-- Not efficient: the old pageserver will continue uploading
-  S3 content during the migration that will never be read.
-
-The user expectations for availability are:
-
-- For planned maintenance, there should be zero availability
-  gap. This expectation is fulfilled by this RFC.
-- For unplanned changes (e.g. node failures), there should be
-  minimal availability gap. This RFC provides the _mechanism_
-  to fail over quickly, but does not provide the failure _detection_
-  nor failover _policy_.
-
-## Non Goals
-
-- Defining service tiers with different storage strategies: the same
-  level of HA & overhead will apply to all tenants. This doesn't rule out
-  adding such tiers in future.
-- Enabling pageserver failover in the absence of a control plane: the control
-  plane will remain the source of truth for what should be attached where.
-- Totally avoiding availability gaps on unplanned migrations during
-  a failure (we expect a small, bounded window of
-  read unavailability of very recent LSNs)
-- Workload balancing: this RFC defines the mechanism for moving tenants
-  around, not the higher level logic for deciding who goes where.
-- Defining all possible configuration flows for tenants: the migration process
-  defined in this RFC demonstrates the sufficiency of the pageserver API, but
-  is not the only kind of configuration change the control plane will ever do.
-  The APIs defined here should let the control plane move tenants around in
-  whatever way is needed while preserving data safety and read availability.
-
-## Impacted components
-
-Pageserver, control plane
-
-## Terminology
-
-- **Attachment**: a tenant is _attached_ to a pageserver if it has
-  been issued a generation number, and is running an instance of
-  the `Tenant` type, ingesting the WAL, and available to serve
-  page reads.
-- **Location**: locations are a superset of attachments. A location
-  is a combination of a tenant and a pageserver. We may _attach_ at a _location_.
-
-- **Secondary location**: a location which is not currently attached.
-- **Warm secondary location**: a location which is not currently attached, but is endeavoring to maintain a warm local cache of layers. We avoid calling this a _warm standby_ to avoid confusion with similar postgres features.
-
-## Implementation (high level)
-
-### Warm secondary locations
-
-To enable faster migrations, we will identify at least one _secondary location_
-for each tenant. This secondary location will keep a warm cache of layers
-for the tenant, so that if it is later attached, it can catch up with the
-latest LSN quickly: rather than downloading everything, it only has to replay
-the recent part of the WAL to advance from the remote_consistent_offset to the
-most recent LSN in the WAL.
-
-The control plane is responsible for selecting secondary locations, and
-calling into pageservers to configure tenants into a secondary mode at this
-new location, as well as attaching the tenant in its existing primary location.
-
-The attached pageserver for a tenant will publish a [layer heatmap](#layer-heatmap)
-to advise secondaries of which layers should be downloaded.
-
-### Location modes
-
-Currently, we consider a tenant to be in one of two states on a pageserver:
-
-- Attached: active `Tenant` object, and layers on local disk
-- Detached: no layers on local disk, no runtime state.
-
-We will extend this with finer-grained modes, whose purpose will become
-clear in later sections:
-
-- **AttachedSingle**: equivalent the existing attached state.
-- **AttachedMulti**: like AttachedSingle, holds an up to date generation, but
-  does not do deletions.
-- **AttachedStale**: like AttachedSingle, holds a stale generation,
-  do not do any remote storage operations.
-- **Secondary**: keep local state on disk, periodically update from S3.
-- **Detached**: equivalent to existing detached state.
-
-To control these finer grained states, a new pageserver API endpoint will be added.
-
-### Cutover procedure
-
-Define old location and new location as "Node A" and "Node B". Consider
-the case where both nodes are available, and Node B was previously configured
-as a secondary location for the tenant we are migrating.
-
-The cutover procedure is orchestrated by the control plane, calling into
-the pageservers' APIs:
-
-1. Call to Node A requesting it to flush to S3 and enter AttachedStale state
-2. Increment generation, and call to Node B requesting it to enter AttachedMulti
-   state with the new generation.
-3. Call to Node B, requesting it to download the latest hot layers from remote storage,
-   according to the latest heatmap flushed by Node A.
-4. Wait for Node B's WAL ingestion to catch up with node A's
-5. Update endpoints to use node B instead of node A
-6. Call to node B requesting it to enter state AttachedSingle.
-7. Call to node A requesting it to enter state Secondary
-
-The following table summarizes how the state of the system advances:
-
-|     Step      |     Node A     |     Node B     | Node used by endpoints |
-| :-----------: | :------------: | :------------: | :--------------------: |
-| 1 (_initial_) | AttachedSingle |   Secondary    |           A            |
-|       2       | AttachedStale  | AttachedMulti  |           A            |
-|       3       | AttachedStale  | AttachedMulti  |           A            |
-|       4       | AttachedStale  | AttachedMulti  |           A            |
-| 5 (_cutover_) | AttachedStale  | AttachedMulti  |           B            |
-|       6       | AttachedStale  | AttachedSingle |           B            |
-|  7 (_final_)  |   Secondary    | AttachedSingle |           B            |
-
-The procedure described for a clean handover from a live node to a secondary
-is also used for failure cases and for migrations to a location that is not
-configured as a secondary, by simply skipping irrelevant steps, as described in
-the following sections.
-
-#### Migration from an unresponsive node
-
-If node A is unavailable, then all calls into
-node A are skipped and we don't wait for B to catch up before
-switching updating the endpoints to use B.
-
-#### Migration to a location that is not a secondary
-
-If node B is initially in Detached state, the procedure is identical. Since Node B
-is coming from a Detached state rather than Secondary, the download of layers and
-catch up with WAL will take much longer.
-
-We might do this if:
-
-- Attached and secondary locations are both critically low on disk, and we need
-  to migrate to a third node with more resources available.
-- We are migrating a tenant which does not use secondary locations to save on cost.
-
-#### Permanent migration away from a node
-
-In the final step of the migration, we generally request the original node to enter a Secondary
-state. This is typical if we are doing a planned migration during maintenance, or to
-balance CPU/network load away from a node.
-
-One might also want to permanently migrate away: this can be done by simply removing the secondary
-location after the migration is complete, or as an optimization by substituting the Detached state
-for the Secondary state in the final step.
-
-#### Cutover diagram
-
-```mermaid
-sequenceDiagram
-participant CP as Control plane
-participant A as Node A
-participant B as Node B
-participant E as Endpoint
-
-CP->>A: PUT Flush & go to AttachedStale
-note right of A: A continues to ingest WAL
-CP->>B: PUT AttachedMulti
-CP->>B: PUT Download layers from latest heatmap
-note right of B: B downloads from S3
-loop Poll until download complete
-CP->>B: GET download status
-end
-activate B
-note right of B: B ingests WAL
-loop Poll until catch up
-CP->>B: GET visible WAL
-CP->>A: GET visible WAL
-end
-deactivate B
-CP->>E: Configure to use Node B
-E->>B: Connect for reads
-CP->>B: PUT AttachedSingle
-CP->>A: PUT Secondary
-```
-
-#### Cutover from an unavailable pageserver
-
-This case is far simpler: we may skip straight to our intended
-end state.
-
-```mermaid
-sequenceDiagram
-participant A as Node A
-participant CP as Control plane
-participant B as Node B
-participant E as Endpoint
-
-note right of A: Node A offline
-activate A
-CP->>B: PUT AttachedSingle
-CP->>E: Configure to use Node B
-E->>B: Connect for reads
-deactivate A
-```
-
-## Implementation (detail)
-
-### Purpose of AttachedMulti, AttachedStale
-
-#### AttachedMulti
-
-Ordinarily, an attached pageserver whose generation is the latest may delete
-layers at will (e.g. during compaction). If a previous generation pageserver
-is also still attached, and in use by endpoints, then this layer deletion could
-lead to a loss of availability for the endpoint when reading from the previous
-generation pageserver.
-
-The _AttachedMulti_ state simply disables deletions. These will be enqueued
-in `RemoteTimelineClient` until the control plane transitions the
-node into AttachedSingle, which unblocks deletions.  Other remote storage operations
-such as uploads are not blocked.
-
-AttachedMulti is not required for data safety, only to preserve availability
-on pageservers running with stale generations.
-
-A node enters AttachedMulti only when explicitly asked to by the control plane. It should
-only remain in this state for the duration of a migration.
-
-If a control plane bug leaves
-the node in AttachedMulti for a long time, then we must avoid unbounded memory use from enqueued
-deletions. This may be accomplished simply, by dropping enqueued deletions when some modest
-threshold of delayed deletions (e.g. 10k layers per tenant) is reached. As with all deletions,
-it is safe to skip them, and the leaked objects will be eventually cleaned up by scrub or
-by timeline deletion.
-
-During AttachedMulti, the Tenant is free to drop layers from local disk in response to
-disk pressure: only the deletion of remote layers is blocked.
-
-#### AttachedStale
-
-Currently, a pageserver with a stale generation number will continue to
-upload layers, but be prevented from completing deletions. This is safe, but inefficient: layers uploaded by this stale generation
-will not be read back by future generations of pageservers.
-
-The _AttachedStale_ state disables S3 uploads. The stale pageserver
-will continue to ingest the WAL and write layers to local disk, but not to
-do any uploads to S3.
-
-A node may enter AttachedStale in two ways:
-
-- Explicitly, when control plane calls into the node at the start of a migration.
-- Implicitly, when the node tries to validate some deletions and discovers
-  that its generation is stale.
-
-The AttachedStale state also disables sending consumption metrics from
-that location: it is interpreted as an indication that some other pageserver
-is already attached or is about to be attached, and that new pageserver will
-be responsible for sending consumption metrics.
-
-#### Disk Pressure & AttachedStale
-
-Over long periods of time, a tenant location in AttachedStale will accumulate data
-on local disk, as it cannot evict any layers written since it entered the
-AttachStale state. We rely on the control plane to revert the location to
-Secondary or Detached at the end of a migration.
-
-This scenario is particularly noteworthy when evacuating all tenants on a pageserver:
-since _all_ the attached tenants will go into AttachedStale, we will be doing no
-uploads at all, therefore ingested data will cause disk usage to increase continuously.
-Under nominal conditions, the available disk space on pageservers should be sufficient
-to complete the evacuation before this becomes a problem, but we must also handle
-the case where we hit a low disk situation while in this state.
-
-The concept of disk pressure already exists in the pageserver: the `disk_usage_eviction_task`
-touches each Tenant when it determines that a low-disk condition requires
-some layer eviction. Having selected layers for eviction, the eviction
-task calls `Timeline::evict_layers`.
-
-**Safety**: If evict_layers is called while in AttachedStale state, and some of the to-be-evicted
-layers are not yet uploaded to S3, then the block on uploads will be lifted. This
-will result in leaking some objects once a migration is complete, but will enable
-the node to manage its disk space properly: if a node is left with some tenants
-in AttachedStale indefinitely due to a network partition or control plane bug,
-these tenants will not cause a full disk condition.
-
-### Warm secondary updates
-
-#### Layer heatmap
-
-The secondary location's job is to serve reads **with the same quality of service as the original location
-was serving them around the time of a migration**. This does not mean the secondary
-location needs the whole set of layers: inactive layers that might soon
-be evicted on the attached pageserver need not be downloaded by the
-secondary. A totally idle tenant only needs to maintain enough on-disk
-state to enable a fast cold start (i.e. the most recent image layers are
-typically sufficient).
-
-To enable this, we introduce the concept of a _layer heatmap_, which
-acts as an advisory input to secondary locations to decide which
-layers to download from S3.
-
-#### Attached pageserver
-
-The attached pageserver, if in state AttachedSingle, periodically
-uploads a serialized heat map to S3. It may skip this if there
-is no change since the last time it uploaded (e.g. if the tenant
-is totally idle).
-
-Additionally, when the tenant is flushed to remote storage prior to a migration
-(the first step in [cutover procedure](#cutover-procedure)), 
-the heatmap is written out. This enables a future attached pageserver
-to get an up to date view when deciding which layers to download.
-
-#### Secondary location behavior
-
-Secondary warm locations run a simple loop, implemented separately from
-the main `Tenant` type, which represents attached tenants:
-
-- Download the layer heatmap
-- Select any "hot enough" layers to download, if there is sufficient
-  free disk space.
-- Download layers, if they were not previously evicted (see below)
-- Download the latest index_part.json
-- Check if any layers currently on disk are no longer referenced by
-  IndexPart & delete them
-
-Note that the heatmap is only advisory: if a secondary location has plenty
-of disk space, it may choose to retain layers that aren't referenced
-by the heatmap, as long as they are still referenced by the IndexPart. Conversely,
-if a node is very low on disk space, it might opt to raise the heat threshold required
-to both downloading a layer, until more disk space is available.
-
-#### Secondary locations & disk pressure
-
-Secondary locations are subject to eviction on disk pressure, just as
-attached locations are.  For eviction purposes, the access time of a
-layer in a secondary location will be the access time given in the heatmap,
-rather than the literal time at which the local layer file was accessed.
-
-The heatmap will indicate which layers are in local storage on the attached
-location.  The secondary will always attempt to get back to having that
-set of layers on disk, but to avoid flapping, it will remember the access
-time of the layer it was most recently asked to evict, and layers whose
-access time is below that will not be re-downloaded.
-
-The resulting behavior is that after a layer is evicted from a secondary
-location, it is only re-downloaded once the attached pageserver accesses
-the layer and uploads a heatmap reflecting that access time.  On a pageserver
-restart, the secondary location will attempt to download all layers in
-the heatmap again, if they are not on local disk.
-
-This behavior will be slightly different when secondary locations are
-used for "low energy tenants", but that is beyond the scope of this RFC.
-
-### Location configuration API
-
-Currently, the `/tenant/<tenant_id>/config` API defines various
-tunables like compaction settings, which apply to the tenant irrespective
-of which pageserver it is running on.
-
-A new "location config" structure will be introduced, which defines
-configuration which is per-tenant, but local to a particular pageserver,
-such as the attachment mode and whether it is a secondary.
-
-The pageserver will expose a new per-tenant API for setting
-the state: `/tenant/<tenant_id>/location/config`.
-
-Body content:
-
-```
-{
-  state: 'enum{Detached, Secondary, AttachedSingle, AttachedMulti, AttachedStale}',
-  generation: Option<u32>,
-  configuration: `Option<TenantConfig>`
-  flush: bool
-}
-```
-
-Existing `/attach` and `/detach` endpoint will have the same
-behavior as calling `/location/config` with `AttachedSingle` and `Detached`
-states respectively. These endpoints will be deprecated and later
-removed.
-
-The generation attribute is mandatory for entering `AttachedSingle` or
-`AttachedMulti`.
-
-The configuration attribute is mandatory when entering any state other
-than `Detached`. This configuration is the same as the body for
-the existing `/tenant/<tenant_id>/config` endpoint.
-
-The `flush` argument indicates whether the pageservers should flush
-to S3 before proceeding: this only has any effect if the node is
-currently in AttachedSingle or AttachedMulti. This is used
-during the first phase of migration, when transitioning the
-old pageserver to AttachedSingle.
-
-The `/re-attach` API response will be extended to include a `state` as
-well as a `generation`, enabling the pageserver to enter the
-correct state for each tenant on startup.
-
-### Database schema for locations
-
-A new table `ProjectLocation`:
-
-- pageserver_id: int
-- tenant_id: TenantId
-- generation: Option<int>
-- state: `enum(Secondary, AttachedSingle, AttachedMulti)`
-
-Notes:
-
-- It is legacy for a Project to have zero `ProjectLocation`s
-- The `pageserver` column in `Project` now means "to which pageserver should
-  endpoints connect", rather than simply which pageserver is attached.
-- The `generation` column in `Project` remains, and is incremented and used
-  to set the generation of `ProjectLocation` rows when they are set into
-  an attached state.
-- The `Detached` state is implicitly represented as the absence of
-  a `ProjectLocation`.
-
-### Executing migrations
-
-Migrations will be implemented as Go functions, within the
-existing `Operation` framework in the control plane. These
-operations are persistent, such that they will always keep
-trying until completion: this property is important to avoid
-leaving garbage behind on pageservers, such as AttachedStale
-locations.
-
-### Recovery from failures during migration
-
-During migration, the control plane may encounter failures of either
-the original or new pageserver, or both:
-
-- If the original fails, skip past waiting for the new pageserver
-  to catch up, and put it into AttachedSingle immediately.
-- If the new node fails, put the old pageserver into Secondary
-  and then back into AttachedSingle (this has the effect of
-  retaining on-disk state and granting it a fresh generation number).
-- If both nodes fail, keep trying until one of them is available
-  again.
-
-### Control plane -> Pageserver reconciliation
-
-A migration may be done while the old node is unavailable,
-in which case the old node may still be running in an AttachedStale
-state.
-
-In this case, it is undesirable to have the migration `Operation`
-stay alive until the old node eventually comes back online
-and can be cleaned up. To handle this, the control plane
-should run a background reconciliation process to compare
-a pageserver's attachments with the database, and clean up
-any that shouldn't be there any more.
-
-Note that there will be no work to do if the old node was really
-offline, as during startup it will call into `/re-attach` and
-be updated that way. The reconciliation will only be needed
-if the node was unavailable but still running.
-
-## Alternatives considered
-
-### Only enabling secondary locations for tenants on a higher service tier
-
-This will make sense in future, especially for tiny databases that may be
-downloaded from S3 in milliseconds when needed.
-
-However, it is not wise to do it immediately, because pageservers contain
-a mixture of higher and lower tier workloads. If we had 1 tenant with
-a secondary location and 9 without, then those other 9 tenants will do
-a lot of I/O as they try to recover from S3, which may degrade the
-service of the tenant which had a secondary location.
-
-Until we segregate tenant on different service tiers on different pageserver
-nodes, or implement & test QoS to ensure that tenants with secondaries are
-not harmed by tenants without, we should use the same failover approach
-for all the tenants.
-
-### Hot secondary locations (continuous WAL replay)
-
-Instead of secondary locations populating their caches from S3, we could
-have them consume the WAL from safekeepers. The downsides of this would be:
-
-- Double load on safekeepers, which are a less scalable service than S3
-- Secondary locations' on-disk state would end up subtly different to
-  the remote state, which would make synchronizing with S3 more complex/expensive
-  when going into attached state.
-
-The downside of only updating secondary locations from S3 is that we will
-have a delay during migration from replaying the LSN range between what's
-in S3 and what's in the pageserver. This range will be very small on
-planned migrations, as we have the old pageserver flush to S3 immediately
-before attaching the new pageserver. On unplanned migrations (old pageserver
-is unavailable), the range of LSNs to replay is bounded by the flush frequency
-on the old pageserver. However, the migration doesn't have to wait for the
-replay: it's just that not-yet-replayed LSNs will be unavailable for read
-until the new pageserver catches up.
-
-We expect that pageserver reads of the most recent LSNs will be relatively
-rare, as for an active endpoint those pages will usually still be in the postgres
-page cache: this leads us to prefer synchronizing from S3 on secondary
-locations, rather than consuming the WAL from safekeepers.
-
-### Cold secondary locations
-
-It is not functionally necessary to keep warm caches on secondary locations at all. However, if we do not, then
-we would experience a de-facto availability loss in unplanned migrations, as reads to the new node would take an extremely long time (many seconds, perhaps minutes).
-
-Warm caches on secondary locations are necessary to meet
-our availability goals.
-
-### Pageserver-granularity failover
-
-Instead of migrating tenants individually, we could have entire spare nodes,
-and on a node death, move all its work to one of these spares.
-
-This approach is avoided for several reasons:
-
-- we would still need fine-grained tenant migration for other
-  purposes such as balancing load
-- by sharing the spare capacity over many peers rather than one spare node,
-  these peers may use the capacity for other purposes, until it is needed
-  to handle migrated tenants. e.g. for keeping a deeper cache of their
-  attached tenants.
-
-### Readonly during migration
-
-We could simplify migrations by making both previous and new nodes go into a
-readonly state, then flush remote content from the previous node, then activate
-attachment on the secondary node.
-
-The downside to this approach is a potentially large gap in readability of
-recent LSNs while loading data onto the new node. To avoid this, it is worthwhile
-to incur the extra cost of double-replaying the WAL onto old and new nodes' local
-storage during a migration.
-
-### Peer-to-peer pageserver communication
-
-Rather than uploading the heatmap to S3, attached pageservers could make it
-available to peers.
-
-Currently, pageservers have no peer to peer communication, so adding this
-for heatmaps would incur significant overhead in deployment and configuration
-of the service, and ensuring that when a new pageserver is deployed, other
-pageservers are updated to be aware of it.
-
-As well as simplifying implementation, putting heatmaps in S3 will be useful
-for future analytics purposes -- gathering aggregated statistics on activity
-pattersn across many tenants may be done directly from data in S3.

libs/consumption_metrics/src/lib.rs

@@ -107,7 +107,7 @@ pub const CHUNK_SIZE: usize = 1000;
 
 // Just a wrapper around a slice of events
 // to serialize it as `{"events" : [ ] }
-#[derive(serde::Serialize, serde::Deserialize)]
+#[derive(serde::Serialize)]
 pub struct EventChunk<'a, T: Clone> {
     pub events: std::borrow::Cow<'a, [T]>,
 }

libs/pageserver_api/src/models.rs

@@ -363,15 +363,8 @@ pub struct TimelineInfo {
     pub latest_gc_cutoff_lsn: Lsn,
     #[serde_as(as = "DisplayFromStr")]
     pub disk_consistent_lsn: Lsn,
-
-    /// The LSN that we have succesfully uploaded to remote storage
     #[serde_as(as = "DisplayFromStr")]
     pub remote_consistent_lsn: Lsn,
-
-    /// The LSN that we are advertizing to safekeepers
-    #[serde_as(as = "DisplayFromStr")]
-    pub remote_consistent_lsn_visible: Lsn,
-
     pub current_logical_size: Option<u64>, // is None when timeline is Unloaded
     /// Sum of the size of all layer files.
     /// If a layer is present in both local FS and S3, it counts only once.

libs/remote_storage/Cargo.toml

@@ -29,4 +29,3 @@ workspace_hack.workspace = true
 [dev-dependencies]
 tempfile.workspace = true
 test-context.workspace = true
-rand.workspace = true

libs/remote_storage/src/lib.rs

@@ -20,7 +20,6 @@ use std::{
 
 use anyhow::{bail, Context};
 
-use serde::{Deserialize, Serialize};
 use tokio::io;
 use toml_edit::Item;
 use tracing::info;
@@ -43,9 +42,6 @@ pub const DEFAULT_REMOTE_STORAGE_S3_CONCURRENCY_LIMIT: usize = 100;
 /// <https://docs.aws.amazon.com/AmazonS3/latest/API/API_ListObjectsV2.html#API_ListObjectsV2_RequestSyntax>
 pub const DEFAULT_MAX_KEYS_PER_LIST_RESPONSE: Option<i32> = None;
 
-/// As defined in S3 docs
-pub const MAX_KEYS_PER_DELETE: usize = 1000;
-
 const REMOTE_STORAGE_PREFIX_SEPARATOR: char = '/';
 
 /// Path on the remote storage, relative to some inner prefix.
@@ -54,25 +50,6 @@ const REMOTE_STORAGE_PREFIX_SEPARATOR: char = '/';
 #[derive(Debug, Clone, PartialEq, Eq, PartialOrd, Ord, Hash)]
 pub struct RemotePath(PathBuf);
 
-impl Serialize for RemotePath {
-    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
-    where
-        S: serde::Serializer,
-    {
-        serializer.collect_str(self)
-    }
-}
-
-impl<'de> Deserialize<'de> for RemotePath {
-    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
-    where
-        D: serde::Deserializer<'de>,
-    {
-        let str = String::deserialize(deserializer)?;
-        Ok(Self(PathBuf::from(&str)))
-    }
-}
-
 impl std::fmt::Display for RemotePath {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         write!(f, "{}", self.0.display())
@@ -111,10 +88,6 @@ impl RemotePath {
     pub fn extension(&self) -> Option<&str> {
         self.0.extension()?.to_str()
     }
-
-    pub fn strip_prefix(&self, p: &RemotePath) -> Result<&Path, std::path::StripPrefixError> {
-        self.0.strip_prefix(&p.0)
-    }
 }
 
 /// Storage (potentially remote) API to manage its state.

libs/remote_storage/src/s3_bucket.rs

@@ -33,10 +33,11 @@ use tracing::debug;
 
 use super::StorageMetadata;
 use crate::{
-    Download, DownloadError, RemotePath, RemoteStorage, S3Config, MAX_KEYS_PER_DELETE,
-    REMOTE_STORAGE_PREFIX_SEPARATOR,
+    Download, DownloadError, RemotePath, RemoteStorage, S3Config, REMOTE_STORAGE_PREFIX_SEPARATOR,
 };
 
+const MAX_DELETE_OBJECTS_REQUEST_SIZE: usize = 1000;
+
 pub(super) mod metrics;
 
 use self::metrics::{AttemptOutcome, RequestKind};
@@ -499,7 +500,7 @@ impl RemoteStorage for S3Bucket {
             delete_objects.push(obj_id);
         }
 
-        for chunk in delete_objects.chunks(MAX_KEYS_PER_DELETE) {
+        for chunk in delete_objects.chunks(MAX_DELETE_OBJECTS_REQUEST_SIZE) {
             let started_at = start_measuring_requests(kind);
 
             let resp = self

libs/remote_storage/tests/test_real_s3.rs

@@ -378,30 +378,21 @@ impl AsyncTestContext for MaybeEnabledS3WithSimpleTestBlobs {
 fn create_s3_client(
     max_keys_per_list_response: Option<i32>,
 ) -> anyhow::Result<Arc<GenericRemoteStorage>> {
-    use rand::Rng;
-
     let remote_storage_s3_bucket = env::var("REMOTE_STORAGE_S3_BUCKET")
         .context("`REMOTE_STORAGE_S3_BUCKET` env var is not set, but real S3 tests are enabled")?;
     let remote_storage_s3_region = env::var("REMOTE_STORAGE_S3_REGION")
         .context("`REMOTE_STORAGE_S3_REGION` env var is not set, but real S3 tests are enabled")?;
-
-    // due to how time works, we've had test runners use the same nanos as bucket prefixes.
-    // millis is just a debugging aid for easier finding the prefix later.
-    let millis = std::time::SystemTime::now()
+    let random_prefix_part = std::time::SystemTime::now()
         .duration_since(UNIX_EPOCH)
         .context("random s3 test prefix part calculation")?
-        .as_millis();
-
-    // because nanos can be the same for two threads so can millis, add randomness
-    let random = rand::thread_rng().gen::<u32>();
-
+        .as_nanos();
     let remote_storage_config = RemoteStorageConfig {
         max_concurrent_syncs: NonZeroUsize::new(100).unwrap(),
         max_sync_errors: NonZeroU32::new(5).unwrap(),
         storage: RemoteStorageKind::AwsS3(S3Config {
             bucket_name: remote_storage_s3_bucket,
             bucket_region: remote_storage_s3_region,
-            prefix_in_bucket: Some(format!("test_{millis}_{random:08x}/")),
+            prefix_in_bucket: Some(format!("pagination_should_work_test_{random_prefix_part}/")),
             endpoint: None,
             concurrency_limit: NonZeroUsize::new(100).unwrap(),
             max_keys_per_list_response,

libs/utils/src/generation.rs

@@ -89,22 +89,6 @@ impl Generation {
             Self::Broken => panic!("Attempted to use a broken generation"),
         }
     }
-
-    pub fn next(&self) -> Generation {
-        match self {
-            Self::Valid(n) => Self::Valid(*n + 1),
-            Self::None => Self::Valid(1),
-            Self::Broken => panic!("Attempted to use a broken generation"),
-        }
-    }
-
-    pub fn into(self) -> Option<u32> {
-        if let Self::Valid(v) = self {
-            Some(v)
-        } else {
-            None
-        }
-    }
 }
 
 impl Serialize for Generation {

libs/utils/src/http/error.rs

@@ -24,9 +24,6 @@ pub enum ApiError {
     #[error("Precondition failed: {0}")]
     PreconditionFailed(Box<str>),
 
-    #[error("Shutting down")]
-    ShuttingDown,
-
     #[error(transparent)]
     InternalServerError(anyhow::Error),
 }
@@ -55,10 +52,6 @@ impl ApiError {
                 self.to_string(),
                 StatusCode::PRECONDITION_FAILED,
             ),
-            ApiError::ShuttingDown => HttpErrorBody::response_from_msg_and_status(
-                "Shutting down".to_string(),
-                StatusCode::SERVICE_UNAVAILABLE,
-            ),
             ApiError::InternalServerError(err) => HttpErrorBody::response_from_msg_and_status(
                 err.to_string(),
                 StatusCode::INTERNAL_SERVER_ERROR,

libs/utils/src/logging.rs

@@ -216,24 +216,6 @@ impl std::fmt::Debug for PrettyLocation<'_, '_> {
     }
 }
 
-/// When you will store a secret but want to make sure it won't
-/// be accidentally logged, wrap it in a SecretString, whose Debug
-/// implementation does not expose the contents.
-#[derive(Clone, Eq, PartialEq)]
-pub struct SecretString(String);
-
-impl SecretString {
-    pub fn get_contents(&self) -> &str {
-        self.0.as_str()
-    }
-}
-
-impl std::fmt::Debug for SecretString {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(f, "[SECRET]")
-    }
-}
-
 #[cfg(test)]
 mod tests {
     use metrics::{core::Opts, IntCounterVec};

libs/vm_monitor/src/cgroup.rs

@@ -431,14 +431,14 @@ impl CgroupWatcher {
                             .context("failed to request upscale")?;
 
                         let memory_high =
-                            self.get_memory_high_bytes().context("failed to get memory.high")?;
+                            self.get_high_bytes().context("failed to get memory.high")?;
                         let new_high = memory_high + self.config.memory_high_increase_by_bytes;
                         info!(
                             current_high_bytes = memory_high,
                             new_high_bytes = new_high,
                             "updating memory.high"
                         );
-                        self.set_memory_high_bytes(new_high)
+                        self.set_high_bytes(new_high)
                             .context("failed to set memory.high")?;
                         last_memory_high_increase_at = Some(Instant::now());
                         continue;
@@ -556,6 +556,14 @@ impl CgroupWatcher {
     }
 }
 
+/// Represents a set of limits we apply to a cgroup to control memory usage.
+///
+/// Setting these values also affects the thresholds for receiving usage alerts.
+#[derive(Debug)]
+pub struct MemoryLimits {
+    pub high: u64,
+}
+
 // Methods for manipulating the actual cgroup
 impl CgroupWatcher {
     /// Get a handle on the freezer subsystem.
@@ -616,29 +624,50 @@ impl CgroupWatcher {
     }
 
     /// Set cgroup memory.high threshold.
-    pub fn set_memory_high_bytes(&self, bytes: u64) -> anyhow::Result<()> {
-        self.set_memory_high_internal(MaxValue::Value(u64::min(bytes, i64::MAX as u64) as i64))
-    }
-
-    /// Set the cgroup's memory.high to 'max', disabling it.
-    pub fn unset_memory_high(&self) -> anyhow::Result<()> {
-        self.set_memory_high_internal(MaxValue::Max)
-    }
-
-    fn set_memory_high_internal(&self, value: MaxValue) -> anyhow::Result<()> {
+    pub fn set_high_bytes(&self, bytes: u64) -> anyhow::Result<()> {
         self.memory()
             .context("failed to get memory subsystem")?
             .set_mem(cgroups_rs::memory::SetMemory {
                 low: None,
-                high: Some(value),
+                high: Some(MaxValue::Value(u64::min(bytes, i64::MAX as u64) as i64)),
                 min: None,
                 max: None,
             })
-            .map_err(anyhow::Error::from)
+            .context("failed to set memory.high")
+    }
+
+    /// Set cgroup memory.high and memory.max.
+    pub fn set_limits(&self, limits: &MemoryLimits) -> anyhow::Result<()> {
+        info!(limits.high, path = self.path(), "writing new memory limits",);
+        self.memory()
+            .context("failed to get memory subsystem while setting memory limits")?
+            .set_mem(cgroups_rs::memory::SetMemory {
+                min: None,
+                low: None,
+                high: Some(MaxValue::Value(
+                    u64::min(limits.high, i64::MAX as u64) as i64
+                )),
+                max: None,
+            })
+            .context("failed to set memory limits")
+    }
+
+    /// Given some amount of available memory, set the desired cgroup memory limits
+    pub fn set_memory_limits(&mut self, available_memory: u64) -> anyhow::Result<()> {
+        let new_high = self.config.calculate_memory_high_value(available_memory);
+        let limits = MemoryLimits { high: new_high };
+        info!(
+            path = self.path(),
+            memory = ?limits,
+            "setting cgroup memory",
+        );
+        self.set_limits(&limits)
+            .context("failed to set cgroup memory limits")?;
+        Ok(())
     }
 
     /// Get memory.high threshold.
-    pub fn get_memory_high_bytes(&self) -> anyhow::Result<u64> {
+    pub fn get_high_bytes(&self) -> anyhow::Result<u64> {
         let high = self
             .memory()
             .context("failed to get memory subsystem while getting memory statistics")?

libs/vm_monitor/src/runner.rs

@@ -16,7 +16,7 @@ use tokio::sync::mpsc;
 use tokio_util::sync::CancellationToken;
 use tracing::{error, info, warn};
 
-use crate::cgroup::{CgroupWatcher, Sequenced};
+use crate::cgroup::{CgroupWatcher, MemoryLimits, Sequenced};
 use crate::dispatcher::Dispatcher;
 use crate::filecache::{FileCacheConfig, FileCacheState};
 use crate::protocol::{InboundMsg, InboundMsgKind, OutboundMsg, OutboundMsgKind, Resources};
@@ -106,51 +106,6 @@ impl Runner {
             kill,
         };
 
-        // If we have both the cgroup and file cache integrations enabled, it's possible for
-        // temporary failures to result in cgroup throttling (from memory.high), that in turn makes
-        // it near-impossible to connect to the file cache (because it times out). Unfortunately,
-        // we *do* still want to determine the file cache size before setting the cgroup's
-        // memory.high, so it's not as simple as just swapping the order.
-        //
-        // Instead, the resolution here is that on vm-monitor startup (note: happens on each
-        // connection from autoscaler-agent, possibly multiple times per compute_ctl lifecycle), we
-        // temporarily unset memory.high, to allow any existing throttling to dissipate. It's a bit
-        // of a hacky solution, but helps with reliability.
-        if let Some(name) = &args.cgroup {
-            // Best not to set up cgroup stuff more than once, so we'll initialize cgroup state
-            // now, and then set limits later.
-            info!("initializing cgroup");
-
-            let (cgroup, cgroup_event_stream) = CgroupWatcher::new(name.clone(), requesting_send)
-                .context("failed to create cgroup manager")?;
-
-            info!("temporarily unsetting memory.high");
-
-            // Temporarily un-set cgroup memory.high; see above.
-            cgroup
-                .unset_memory_high()
-                .context("failed to unset memory.high")?;
-
-            let cgroup = Arc::new(cgroup);
-
-            let cgroup_clone = Arc::clone(&cgroup);
-            spawn_with_cancel(
-                token.clone(),
-                |_| error!("cgroup watcher terminated"),
-                async move { cgroup_clone.watch(notified_recv, cgroup_event_stream).await },
-            );
-
-            state.cgroup = Some(cgroup);
-        } else {
-            // *NOTE*: We need to forget the sender so that its drop impl does not get ran.
-            // This allows us to poll it in `Monitor::run` regardless of whether we
-            // are managing a cgroup or not. If we don't forget it, all receives will
-            // immediately return an error because the sender is droped and it will
-            // claim all select! statements, effectively turning `Monitor::run` into
-            // `loop { fail to receive }`.
-            mem::forget(requesting_send);
-        }
-
         let mut file_cache_reserved_bytes = 0;
         let mem = get_total_system_memory();
 
@@ -164,7 +119,7 @@ impl Runner {
                 false => FileCacheConfig::default_in_memory(),
             };
 
-            let mut file_cache = FileCacheState::new(connstr, config, token)
+            let mut file_cache = FileCacheState::new(connstr, config, token.clone())
                 .await
                 .context("failed to create file cache")?;
 
@@ -197,15 +152,35 @@ impl Runner {
             state.filecache = Some(file_cache);
         }
 
-        if let Some(cgroup) = &state.cgroup {
-            let available = mem - file_cache_reserved_bytes;
-            let value = cgroup.config.calculate_memory_high_value(available);
+        if let Some(name) = &args.cgroup {
+            let (mut cgroup, cgroup_event_stream) =
+                CgroupWatcher::new(name.clone(), requesting_send)
+                    .context("failed to create cgroup manager")?;
 
-            info!(value, "setting memory.high");
+            let available = mem - file_cache_reserved_bytes;
 
             cgroup
-                .set_memory_high_bytes(value)
-                .context("failed to set cgroup memory.high")?;
+                .set_memory_limits(available)
+                .context("failed to set cgroup memory limits")?;
+
+            let cgroup = Arc::new(cgroup);
+
+            // Some might call this . . . cgroup v2
+            let cgroup_clone = Arc::clone(&cgroup);
+
+            spawn_with_cancel(token, |_| error!("cgroup watcher terminated"), async move {
+                cgroup_clone.watch(notified_recv, cgroup_event_stream).await
+            });
+
+            state.cgroup = Some(cgroup);
+        } else {
+            // *NOTE*: We need to forget the sender so that its drop impl does not get ran.
+            // This allows us to poll it in `Monitor::run` regardless of whether we
+            // are managing a cgroup or not. If we don't forget it, all receives will
+            // immediately return an error because the sender is droped and it will
+            // claim all select! statements, effectively turning `Monitor::run` into
+            // `loop { fail to receive }`.
+            mem::forget(requesting_send);
         }
 
         Ok(state)
@@ -282,11 +257,14 @@ impl Runner {
                 new_cgroup_mem_high = cgroup.config.calculate_memory_high_value(available_memory);
             }
 
-            // new_cgroup_mem_high is initialized to 0 but it is guaranteed to not be here
-            // since it is properly initialized in the previous cgroup if let block
+            let limits = MemoryLimits {
+                // new_cgroup_mem_high is initialized to 0 but it is guarancontextd to not be here
+                // since it is properly initialized in the previous cgroup if let block
+                high: new_cgroup_mem_high,
+            };
             cgroup
-                .set_memory_high_bytes(new_cgroup_mem_high)
-                .context("failed to set cgroup memory.high")?;
+                .set_limits(&limits)
+                .context("failed to set cgroup memory limits")?;
 
             let message = format!(
                 "set cgroup memory.high to {} MiB, of new max {} MiB",
@@ -349,9 +327,12 @@ impl Runner {
                 name = cgroup.path(),
                 "updating cgroup memory.high",
             );
+            let limits = MemoryLimits {
+                high: new_cgroup_mem_high,
+            };
             cgroup
-                .set_memory_high_bytes(new_cgroup_mem_high)
-                .context("failed to set cgroup memory.high")?;
+                .set_limits(&limits)
+                .context("failed to set file cache size")?;
         }
 
         Ok(())

pageserver/ctl/src/layer_map_analyzer.rs

@@ -3,8 +3,6 @@
 //! Currently it only analyzes holes, which are regions within the layer range that the layer contains no updates for. In the future it might do more analysis (maybe key quantiles?) but it should never return sensitive data.
 
 use anyhow::Result;
-use pageserver::context::{DownloadBehavior, RequestContext};
-use pageserver::task_mgr::TaskKind;
 use pageserver::tenant::{TENANTS_SEGMENT_NAME, TIMELINES_SEGMENT_NAME};
 use std::cmp::Ordering;
 use std::collections::BinaryHeap;
@@ -98,9 +96,9 @@ pub(crate) fn parse_filename(name: &str) -> Option<LayerFile> {
 }
 
 // Finds the max_holes largest holes, ignoring any that are smaller than MIN_HOLE_LENGTH"
-async fn get_holes(path: &Path, max_holes: usize, ctx: &RequestContext) -> Result<Vec<Hole>> {
+async fn get_holes(path: &Path, max_holes: usize) -> Result<Vec<Hole>> {
     let file = FileBlockReader::new(VirtualFile::open(path).await?);
-    let summary_blk = file.read_blk(0, ctx).await?;
+    let summary_blk = file.read_blk(0).await?;
     let actual_summary = Summary::des_prefix(summary_blk.as_ref())?;
     let tree_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
         actual_summary.index_start_blk,
@@ -127,7 +125,6 @@ async fn get_holes(path: &Path, max_holes: usize, ctx: &RequestContext) -> Resul
                 prev_key = Some(curr.next());
                 true
             },
-            ctx,
         )
         .await?;
     let mut holes = heap.into_vec();
@@ -138,7 +135,6 @@ async fn get_holes(path: &Path, max_holes: usize, ctx: &RequestContext) -> Resul
 pub(crate) async fn main(cmd: &AnalyzeLayerMapCmd) -> Result<()> {
     let storage_path = &cmd.path;
     let max_holes = cmd.max_holes.unwrap_or(DEFAULT_MAX_HOLES);
-    let ctx = RequestContext::new(TaskKind::DebugTool, DownloadBehavior::Error);
 
     // Initialize virtual_file (file desriptor cache) and page cache which are needed to access layer persistent B-Tree.
     pageserver::virtual_file::init(10);
@@ -167,7 +163,7 @@ pub(crate) async fn main(cmd: &AnalyzeLayerMapCmd) -> Result<()> {
                     parse_filename(&layer.file_name().into_string().unwrap())
                 {
                     if layer_file.is_delta {
-                        layer_file.holes = get_holes(&layer.path(), max_holes, &ctx).await?;
+                        layer_file.holes = get_holes(&layer.path(), max_holes).await?;
                         n_deltas += 1;
                     }
                     layers.push(layer_file);

pageserver/ctl/src/layers.rs

@@ -2,8 +2,6 @@ use std::path::{Path, PathBuf};
 
 use anyhow::Result;
 use clap::Subcommand;
-use pageserver::context::{DownloadBehavior, RequestContext};
-use pageserver::task_mgr::TaskKind;
 use pageserver::tenant::block_io::BlockCursor;
 use pageserver::tenant::disk_btree::DiskBtreeReader;
 use pageserver::tenant::storage_layer::delta_layer::{BlobRef, Summary};
@@ -46,12 +44,12 @@ pub(crate) enum LayerCmd {
     },
 }
 
-async fn read_delta_file(path: impl AsRef<Path>, ctx: &RequestContext) -> Result<()> {
+async fn read_delta_file(path: impl AsRef<Path>) -> Result<()> {
     let path = path.as_ref();
     virtual_file::init(10);
     page_cache::init(100);
     let file = FileBlockReader::new(VirtualFile::open(path).await?);
-    let summary_blk = file.read_blk(0, ctx).await?;
+    let summary_blk = file.read_blk(0).await?;
     let actual_summary = Summary::des_prefix(summary_blk.as_ref())?;
     let tree_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
         actual_summary.index_start_blk,
@@ -69,12 +67,11 @@ async fn read_delta_file(path: impl AsRef<Path>, ctx: &RequestContext) -> Result
                 all.push((curr, BlobRef(value_offset)));
                 true
             },
-            ctx,
         )
         .await?;
     let cursor = BlockCursor::new_fileblockreader(&file);
     for (k, v) in all {
-        let value = cursor.read_blob(v.pos(), ctx).await?;
+        let value = cursor.read_blob(v.pos()).await?;
         println!("key:{} value_len:{}", k, value.len());
     }
     // TODO(chi): special handling for last key?
@@ -82,7 +79,6 @@ async fn read_delta_file(path: impl AsRef<Path>, ctx: &RequestContext) -> Result
 }
 
 pub(crate) async fn main(cmd: &LayerCmd) -> Result<()> {
-    let ctx = RequestContext::new(TaskKind::DebugTool, DownloadBehavior::Error);
     match cmd {
         LayerCmd::List { path } => {
             for tenant in fs::read_dir(path.join(TENANTS_SEGMENT_NAME))? {
@@ -157,7 +153,7 @@ pub(crate) async fn main(cmd: &LayerCmd) -> Result<()> {
                         );
 
                         if layer_file.is_delta {
-                            read_delta_file(layer.path(), &ctx).await?;
+                            read_delta_file(layer.path()).await?;
                         } else {
                             anyhow::bail!("not supported yet :(");
                         }

pageserver/src/bin/pageserver.rs

@@ -8,7 +8,6 @@ use anyhow::{anyhow, Context};
 use clap::{Arg, ArgAction, Command};
 
 use metrics::launch_timestamp::{set_launch_timestamp_metric, LaunchTimestamp};
-use pageserver::control_plane_client::ControlPlaneClient;
 use pageserver::disk_usage_eviction_task::{self, launch_disk_usage_global_eviction_task};
 use pageserver::metrics::{STARTUP_DURATION, STARTUP_IS_LOADING};
 use pageserver::task_mgr::WALRECEIVER_RUNTIME;
@@ -21,7 +20,6 @@ use metrics::set_build_info_metric;
 use pageserver::{
     config::{defaults::*, PageServerConf},
     context::{DownloadBehavior, RequestContext},
-    deletion_queue::DeletionQueue,
     http, page_cache, page_service, task_mgr,
     task_mgr::TaskKind,
     task_mgr::{BACKGROUND_RUNTIME, COMPUTE_REQUEST_RUNTIME, MGMT_REQUEST_RUNTIME},
@@ -348,22 +346,9 @@ fn start_pageserver(
         }
     };
 
-    // Top-level cancellation token for the process
-    let shutdown_pageserver = tokio_util::sync::CancellationToken::new();
-
     // Set up remote storage client
     let remote_storage = create_remote_storage_client(conf)?;
 
-    // Set up deletion queue
-    let (deletion_queue, deletion_workers) = DeletionQueue::new(
-        remote_storage.clone(),
-        ControlPlaneClient::new(conf, &shutdown_pageserver),
-        conf,
-    );
-    if let Some(deletion_workers) = deletion_workers {
-        deletion_workers.spawn_with(BACKGROUND_RUNTIME.handle());
-    }
-
     // Up to this point no significant I/O has been done: this should have been fast.  Record
     // duration prior to starting I/O intensive phase of startup.
     startup_checkpoint("initial", "Starting loading tenants");
@@ -394,13 +379,13 @@ fn start_pageserver(
     };
 
     // Scan the local 'tenants/' directory and start loading the tenants
-    let deletion_queue_client = deletion_queue.new_client();
+    let shutdown_pageserver = tokio_util::sync::CancellationToken::new();
+
     BACKGROUND_RUNTIME.block_on(mgr::init_tenant_mgr(
         conf,
         TenantSharedResources {
             broker_client: broker_client.clone(),
             remote_storage: remote_storage.clone(),
-            deletion_queue_client,
         },
         order,
         shutdown_pageserver.clone(),
@@ -496,10 +481,9 @@ fn start_pageserver(
             http::routes::State::new(
                 conf,
                 http_auth.clone(),
-                remote_storage.clone(),
+                remote_storage,
                 broker_client.clone(),
                 disk_usage_eviction_state,
-                deletion_queue.new_client(),
             )
             .context("Failed to initialize router state")?,
         );
@@ -627,12 +611,7 @@ fn start_pageserver(
             // Right now that tree doesn't reach very far, and `task_mgr` is used instead.
             // The plan is to change that over time.
             shutdown_pageserver.take();
-            let bg_remote_storage = remote_storage.clone();
-            let bg_deletion_queue = deletion_queue.clone();
-            BACKGROUND_RUNTIME.block_on(pageserver::shutdown_pageserver(
-                bg_remote_storage.map(|_| bg_deletion_queue),
-                0,
-            ));
+            BACKGROUND_RUNTIME.block_on(pageserver::shutdown_pageserver(0));
             unreachable!()
         }
     })
@@ -644,7 +623,7 @@ fn create_remote_storage_client(
     let config = if let Some(config) = &conf.remote_storage_config {
         config
     } else {
-        tracing::warn!("no remote storage configured, this is a deprecated configuration");
+        // No remote storage configured.
         return Ok(None);
     };
 

pageserver/src/config.rs

@@ -11,7 +11,6 @@ use std::env;
 use storage_broker::Uri;
 use utils::crashsafe::path_with_suffix_extension;
 use utils::id::ConnectionId;
-use utils::logging::SecretString;
 
 use once_cell::sync::OnceCell;
 use reqwest::Url;
@@ -208,9 +207,6 @@ pub struct PageServerConf {
     pub background_task_maximum_delay: Duration,
 
     pub control_plane_api: Option<Url>,
-
-    /// JWT token for use with the control plane API.
-    pub control_plane_api_token: Option<SecretString>,
 }
 
 /// We do not want to store this in a PageServerConf because the latter may be logged
@@ -287,7 +283,6 @@ struct PageServerConfigBuilder {
     background_task_maximum_delay: BuilderValue<Duration>,
 
     control_plane_api: BuilderValue<Option<Url>>,
-    control_plane_api_token: BuilderValue<Option<SecretString>>,
 }
 
 impl Default for PageServerConfigBuilder {
@@ -352,7 +347,6 @@ impl Default for PageServerConfigBuilder {
             .unwrap()),
 
             control_plane_api: Set(None),
-            control_plane_api_token: Set(None),
         }
     }
 }
@@ -481,8 +475,8 @@ impl PageServerConfigBuilder {
         self.background_task_maximum_delay = BuilderValue::Set(delay);
     }
 
-    pub fn control_plane_api(&mut self, api: Option<Url>) {
-        self.control_plane_api = BuilderValue::Set(api)
+    pub fn control_plane_api(&mut self, api: Url) {
+        self.control_plane_api = BuilderValue::Set(Some(api))
     }
 
     pub fn build(self) -> anyhow::Result<PageServerConf> {
@@ -573,9 +567,6 @@ impl PageServerConfigBuilder {
             control_plane_api: self
                 .control_plane_api
                 .ok_or(anyhow!("missing control_plane_api"))?,
-            control_plane_api_token: self
-                .control_plane_api_token
-                .ok_or(anyhow!("missing control_plane_api_token"))?,
         })
     }
 }
@@ -589,27 +580,6 @@ impl PageServerConf {
         self.workdir.join(TENANTS_SEGMENT_NAME)
     }
 
-    pub fn deletion_prefix(&self) -> PathBuf {
-        self.workdir.join("deletion")
-    }
-
-    pub fn deletion_list_path(&self, sequence: u64) -> PathBuf {
-        // Encode a version in the filename, so that if we ever switch away from JSON we can
-        // increment this.
-        const VERSION: u8 = 1;
-
-        self.deletion_prefix()
-            .join(format!("{sequence:016x}-{VERSION:02x}.list"))
-    }
-
-    pub fn deletion_header_path(&self) -> PathBuf {
-        // Encode a version in the filename, so that if we ever switch away from JSON we can
-        // increment this.
-        const VERSION: u8 = 1;
-
-        self.deletion_prefix().join(format!("header-{VERSION:02x}"))
-    }
-
     pub fn tenant_path(&self, tenant_id: &TenantId) -> PathBuf {
         self.tenants_path().join(tenant_id.to_string())
     }
@@ -777,14 +747,7 @@ impl PageServerConf {
                 },
                 "ondemand_download_behavior_treat_error_as_warn" => builder.ondemand_download_behavior_treat_error_as_warn(parse_toml_bool(key, item)?),
                 "background_task_maximum_delay" => builder.background_task_maximum_delay(parse_toml_duration(key, item)?),
-                "control_plane_api" => {
-                    let parsed = parse_toml_string(key, item)?;
-                    if parsed.is_empty() {
-                        builder.control_plane_api(None)
-                    } else {
-                        builder.control_plane_api(Some(parsed.parse().context("failed to parse control plane URL")?))
-                    }
-                },
+                "control_plane_api" => builder.control_plane_api(parse_toml_string(key, item)?.parse().context("failed to parse control plane URL")?),
                 _ => bail!("unrecognized pageserver option '{key}'"),
             }
         }
@@ -954,7 +917,6 @@ impl PageServerConf {
             ondemand_download_behavior_treat_error_as_warn: false,
             background_task_maximum_delay: Duration::ZERO,
             control_plane_api: None,
-            control_plane_api_token: None,
         }
     }
 }
@@ -1178,8 +1140,7 @@ background_task_maximum_delay = '334 s'
                 background_task_maximum_delay: humantime::parse_duration(
                     defaults::DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY
                 )?,
-                control_plane_api: None,
-                control_plane_api_token: None
+                control_plane_api: None
             },
             "Correct defaults should be used when no config values are provided"
         );
@@ -1235,8 +1196,7 @@ background_task_maximum_delay = '334 s'
                 test_remote_failures: 0,
                 ondemand_download_behavior_treat_error_as_warn: false,
                 background_task_maximum_delay: Duration::from_secs(334),
-                control_plane_api: None,
-                control_plane_api_token: None
+                control_plane_api: None
             },
             "Should be able to parse all basic config values correctly"
         );

pageserver/src/context.rs

@@ -94,18 +94,6 @@ pub struct RequestContext {
     task_kind: TaskKind,
     download_behavior: DownloadBehavior,
     access_stats_behavior: AccessStatsBehavior,
-    page_content_kind: PageContentKind,
-}
-
-/// The kind of access to the page cache.
-#[derive(Clone, Copy, PartialEq, Eq, Debug, enum_map::Enum, strum_macros::IntoStaticStr)]
-pub enum PageContentKind {
-    Unknown,
-    DeltaLayerBtreeNode,
-    DeltaLayerValue,
-    ImageLayerBtreeNode,
-    ImageLayerValue,
-    InMemoryLayer,
 }
 
 /// Desired behavior if the operation requires an on-demand download
@@ -149,7 +137,6 @@ impl RequestContextBuilder {
                 task_kind,
                 download_behavior: DownloadBehavior::Download,
                 access_stats_behavior: AccessStatsBehavior::Update,
-                page_content_kind: PageContentKind::Unknown,
             },
         }
     }
@@ -162,7 +149,6 @@ impl RequestContextBuilder {
                 task_kind: original.task_kind,
                 download_behavior: original.download_behavior,
                 access_stats_behavior: original.access_stats_behavior,
-                page_content_kind: original.page_content_kind,
             },
         }
     }
@@ -181,11 +167,6 @@ impl RequestContextBuilder {
         self
     }
 
-    pub(crate) fn page_content_kind(mut self, k: PageContentKind) -> Self {
-        self.inner.page_content_kind = k;
-        self
-    }
-
     pub fn build(self) -> RequestContext {
         self.inner
     }
@@ -282,8 +263,4 @@ impl RequestContext {
     pub(crate) fn access_stats_behavior(&self) -> AccessStatsBehavior {
         self.access_stats_behavior
     }
-
-    pub(crate) fn page_content_kind(&self) -> PageContentKind {
-        self.page_content_kind
-    }
 }

pageserver/src/control_plane_client.rs

@@ -1,9 +1,7 @@
 use std::collections::HashMap;
 
-use pageserver_api::control_api::{
-    ReAttachRequest, ReAttachResponse, ValidateRequest, ValidateRequestTenant, ValidateResponse,
-};
-use serde::{de::DeserializeOwned, Serialize};
+use hyper::StatusCode;
+use pageserver_api::control_api::{ReAttachRequest, ReAttachResponse};
 use tokio_util::sync::CancellationToken;
 use url::Url;
 use utils::{
@@ -14,34 +12,25 @@ use utils::{
 
 use crate::config::PageServerConf;
 
+// Backoffs when control plane requests do not succeed: compromise between reducing load
+// on control plane, and retrying frequently when we are blocked on a control plane
+// response to make progress.
+const BACKOFF_INCREMENT: f64 = 0.1;
+const BACKOFF_MAX: f64 = 10.0;
+
 /// The Pageserver's client for using the control plane API: this is a small subset
 /// of the overall control plane API, for dealing with generations (see docs/rfcs/025-generation-numbers.md)
-pub struct ControlPlaneClient {
+pub(crate) struct ControlPlaneClient {
     http_client: reqwest::Client,
     base_url: Url,
     node_id: NodeId,
     cancel: CancellationToken,
 }
 
-/// Represent operations which internally retry on all errors other than
-/// cancellation token firing: the only way they can fail is ShuttingDown.
-pub enum RetryForeverError {
-    ShuttingDown,
-}
-
-#[async_trait::async_trait]
-pub trait ControlPlaneGenerationsApi {
-    async fn re_attach(&self) -> Result<HashMap<TenantId, Generation>, RetryForeverError>;
-    async fn validate(
-        &self,
-        tenants: Vec<(TenantId, Generation)>,
-    ) -> Result<HashMap<TenantId, bool>, RetryForeverError>;
-}
-
 impl ControlPlaneClient {
     /// A None return value indicates that the input `conf` object does not have control
     /// plane API enabled.
-    pub fn new(conf: &'static PageServerConf, cancel: &CancellationToken) -> Option<Self> {
+    pub(crate) fn new(conf: &'static PageServerConf, cancel: &CancellationToken) -> Option<Self> {
         let mut url = match conf.control_plane_api.as_ref() {
             Some(u) => u.clone(),
             None => return None,
@@ -53,78 +42,39 @@ impl ControlPlaneClient {
             segs.pop_if_empty().push("");
         }
 
-        let mut client = reqwest::ClientBuilder::new();
-
-        if let Some(jwt) = &conf.control_plane_api_token {
-            let mut headers = hyper::HeaderMap::new();
-            headers.insert("Authorization", jwt.get_contents().parse().unwrap());
-            client = client.default_headers(headers);
-        }
+        let client = reqwest::ClientBuilder::new()
+            .build()
+            .expect("Failed to construct http client");
 
         Some(Self {
-            http_client: client.build().expect("Failed to construct HTTP client"),
+            http_client: client,
             base_url: url,
             node_id: conf.id,
             cancel: cancel.clone(),
         })
     }
 
-    async fn retry_http_forever<R, T>(
+    async fn try_re_attach(
         &self,
-        url: &url::Url,
-        request: R,
-    ) -> Result<T, RetryForeverError>
-    where
-        R: Serialize,
-        T: DeserializeOwned,
-    {
-        #[derive(thiserror::Error, Debug)]
-        enum RemoteAttemptError {
-            #[error("shutdown")]
-            Shutdown,
-            #[error("remote: {0}")]
-            Remote(reqwest::Error),
-        }
-
-        match backoff::retry(
-            || async {
-                let response = self
-                    .http_client
-                    .post(url.clone())
-                    .json(&request)
-                    .send()
-                    .await
-                    .map_err(RemoteAttemptError::Remote)?;
-
-                response
-                    .error_for_status_ref()
-                    .map_err(RemoteAttemptError::Remote)?;
-                response
-                    .json::<T>()
-                    .await
-                    .map_err(RemoteAttemptError::Remote)
-            },
-            |_| false,
-            3,
-            u32::MAX,
-            "calling control plane generation validation API",
-            backoff::Cancel::new(self.cancel.clone(), || RemoteAttemptError::Shutdown),
-        )
-        .await
-        {
-            Err(RemoteAttemptError::Shutdown) => Err(RetryForeverError::ShuttingDown),
-            Err(RemoteAttemptError::Remote(_)) => {
-                panic!("We retry forever, this should never be reached");
+        url: Url,
+        request: &ReAttachRequest,
+    ) -> anyhow::Result<ReAttachResponse> {
+        match self.http_client.post(url).json(request).send().await {
+            Err(e) => Err(anyhow::Error::from(e)),
+            Ok(r) => {
+                if r.status() == StatusCode::OK {
+                    r.json::<ReAttachResponse>()
+                        .await
+                        .map_err(anyhow::Error::from)
+                } else {
+                    Err(anyhow::anyhow!("Unexpected status {}", r.status()))
+                }
             }
-            Ok(r) => Ok(r),
         }
     }
-}
 
-#[async_trait::async_trait]
-impl ControlPlaneGenerationsApi for ControlPlaneClient {
-    /// Block until we get a successful response, or error out if we are shut down
-    async fn re_attach(&self) -> Result<HashMap<TenantId, Generation>, RetryForeverError> {
+    /// Block until we get a successful response
+    pub(crate) async fn re_attach(&self) -> anyhow::Result<HashMap<TenantId, Generation>> {
         let re_attach_path = self
             .base_url
             .join("re-attach")
@@ -133,47 +83,37 @@ impl ControlPlaneGenerationsApi for ControlPlaneClient {
             node_id: self.node_id,
         };
 
-        let response: ReAttachResponse = self.retry_http_forever(&re_attach_path, request).await?;
-        tracing::info!(
-            "Received re-attach response with {} tenants",
-            response.tenants.len()
-        );
+        let mut attempt = 0;
+        loop {
+            let result = self.try_re_attach(re_attach_path.clone(), &request).await;
+            match result {
+                Ok(res) => {
+                    tracing::info!(
+                        "Received re-attach response with {} tenants",
+                        res.tenants.len()
+                    );
 
-        Ok(response
-            .tenants
-            .into_iter()
-            .map(|t| (t.id, Generation::new(t.generation)))
-            .collect::<HashMap<_, _>>())
-    }
-
-    /// Block until we get a successful response, or error out if we are shut down
-    async fn validate(
-        &self,
-        tenants: Vec<(TenantId, Generation)>,
-    ) -> Result<HashMap<TenantId, bool>, RetryForeverError> {
-        let re_attach_path = self
-            .base_url
-            .join("validate")
-            .expect("Failed to build validate path");
-
-        let request = ValidateRequest {
-            tenants: tenants
-                .into_iter()
-                .map(|(id, gen)| ValidateRequestTenant {
-                    id,
-                    gen: gen
-                        .into()
-                        .expect("Generation should always be valid for a Tenant doing deletions"),
-                })
-                .collect(),
-        };
-
-        let response: ValidateResponse = self.retry_http_forever(&re_attach_path, request).await?;
-
-        Ok(response
-            .tenants
-            .into_iter()
-            .map(|rt| (rt.id, rt.valid))
-            .collect())
+                    return Ok(res
+                        .tenants
+                        .into_iter()
+                        .map(|t| (t.id, Generation::new(t.generation)))
+                        .collect::<HashMap<_, _>>());
+                }
+                Err(e) => {
+                    tracing::error!("Error re-attaching tenants, retrying: {e:#}");
+                    backoff::exponential_backoff(
+                        attempt,
+                        BACKOFF_INCREMENT,
+                        BACKOFF_MAX,
+                        &self.cancel,
+                    )
+                    .await;
+                    if self.cancel.is_cancelled() {
+                        return Err(anyhow::anyhow!("Shutting down"));
+                    }
+                    attempt += 1;
+                }
+            }
+        }
     }
 }

pageserver/src/deletion_queue/deleter.rs

@@ -1,156 +0,0 @@
-//! The deleter is the final stage in the deletion queue.  It accumulates remote
-//! paths to delete, and periodically executes them in batches of up to 1000
-//! using the DeleteObjects request.
-//!
-//! Its purpose is to increase efficiency of remote storage I/O by issuing a smaller
-//! number of full-sized DeleteObjects requests, rather than a larger number of
-//! smaller requests.
-
-use remote_storage::GenericRemoteStorage;
-use remote_storage::RemotePath;
-use remote_storage::MAX_KEYS_PER_DELETE;
-use std::time::Duration;
-use tokio_util::sync::CancellationToken;
-use tracing::info;
-use tracing::warn;
-
-use crate::metrics;
-
-use super::DeletionQueueError;
-use super::FlushOp;
-
-const AUTOFLUSH_INTERVAL: Duration = Duration::from_secs(10);
-
-pub(super) enum DeleterMessage {
-    Delete(Vec<RemotePath>),
-    Flush(FlushOp),
-}
-
-/// Non-persistent deletion queue, for coalescing multiple object deletes into
-/// larger DeleteObjects requests.
-pub(super) struct Deleter {
-    // Accumulate up to 1000 keys for the next deletion operation
-    accumulator: Vec<RemotePath>,
-
-    rx: tokio::sync::mpsc::Receiver<DeleterMessage>,
-
-    cancel: CancellationToken,
-    remote_storage: GenericRemoteStorage,
-}
-
-impl Deleter {
-    pub(super) fn new(
-        remote_storage: GenericRemoteStorage,
-        rx: tokio::sync::mpsc::Receiver<DeleterMessage>,
-        cancel: CancellationToken,
-    ) -> Self {
-        Self {
-            remote_storage,
-            rx,
-            cancel,
-            accumulator: Vec::new(),
-        }
-    }
-
-    /// Wrap the remote `delete_objects` with a failpoint
-    async fn remote_delete(&self) -> Result<(), anyhow::Error> {
-        fail::fail_point!("deletion-queue-before-execute", |_| {
-            info!("Skipping execution, failpoint set");
-            metrics::DELETION_QUEUE
-                .remote_errors
-                .with_label_values(&["failpoint"])
-                .inc();
-            Err(anyhow::anyhow!("failpoint hit"))
-        });
-
-        self.remote_storage.delete_objects(&self.accumulator).await
-    }
-
-    /// Block until everything in accumulator has been executed
-    async fn flush(&mut self) -> Result<(), DeletionQueueError> {
-        while !self.accumulator.is_empty() && !self.cancel.is_cancelled() {
-            match self.remote_delete().await {
-                Ok(()) => {
-                    // Note: we assume that the remote storage layer returns Ok(()) if some
-                    // or all of the deleted objects were already gone.
-                    metrics::DELETION_QUEUE
-                        .keys_executed
-                        .inc_by(self.accumulator.len() as u64);
-                    info!(
-                        "Executed deletion batch {}..{}",
-                        self.accumulator
-                            .first()
-                            .expect("accumulator should be non-empty"),
-                        self.accumulator
-                            .last()
-                            .expect("accumulator should be non-empty"),
-                    );
-                    self.accumulator.clear();
-                }
-                Err(e) => {
-                    warn!("DeleteObjects request failed: {e:#}, will retry");
-                    metrics::DELETION_QUEUE
-                        .remote_errors
-                        .with_label_values(&["execute"])
-                        .inc();
-                }
-            };
-        }
-        if self.cancel.is_cancelled() {
-            // Expose an error because we may not have actually flushed everything
-            Err(DeletionQueueError::ShuttingDown)
-        } else {
-            Ok(())
-        }
-    }
-
-    pub(super) async fn background(&mut self) -> Result<(), DeletionQueueError> {
-        self.accumulator.reserve(MAX_KEYS_PER_DELETE);
-
-        loop {
-            if self.cancel.is_cancelled() {
-                return Err(DeletionQueueError::ShuttingDown);
-            }
-
-            let msg = match tokio::time::timeout(AUTOFLUSH_INTERVAL, self.rx.recv()).await {
-                Ok(Some(m)) => m,
-                Ok(None) => {
-                    // All queue senders closed
-                    info!("Shutting down");
-                    return Err(DeletionQueueError::ShuttingDown);
-                }
-                Err(_) => {
-                    // Timeout, we hit deadline to execute whatever we have in hand.  These functions will
-                    // return immediately if no work is pending
-                    self.flush().await?;
-
-                    continue;
-                }
-            };
-
-            match msg {
-                DeleterMessage::Delete(mut list) => {
-                    while !list.is_empty() || self.accumulator.len() == MAX_KEYS_PER_DELETE {
-                        if self.accumulator.len() == MAX_KEYS_PER_DELETE {
-                            self.flush().await?;
-                            // If we have received this number of keys, proceed with attempting to execute
-                            assert_eq!(self.accumulator.len(), 0);
-                        }
-
-                        let available_slots = MAX_KEYS_PER_DELETE - self.accumulator.len();
-                        let take_count = std::cmp::min(available_slots, list.len());
-                        for path in list.drain(list.len() - take_count..) {
-                            self.accumulator.push(path);
-                        }
-                    }
-                }
-                DeleterMessage::Flush(flush_op) => {
-                    // If flush() errors, we drop the flush_op and the caller will get
-                    // an error recv()'ing their oneshot channel.
-                    self.flush().await?;
-                    flush_op.notify();
-                }
-            }
-        }
-    }
-}

pageserver/src/deletion_queue/list_writer.rs

@@ -1,487 +0,0 @@
-//! The list writer is the first stage in the deletion queue.  It accumulates
-//! layers to delete, and periodically writes out these layers into a persistent
-//! DeletionList.
-//!
-//! The purpose of writing DeletionLists is to decouple the decision to
-//! delete an object from the validation required to execute it: even if
-//! validation is not possible, e.g. due to a control plane outage, we can
-//! still persist our intent to delete an object, in a way that would
-//! survive a restart.
-//!
-//! DeletionLists are passed onwards to the Validator.
-
-use super::DeletionHeader;
-use super::DeletionList;
-use super::FlushOp;
-use super::ValidatorQueueMessage;
-
-use std::collections::HashMap;
-use std::fs::create_dir_all;
-use std::time::Duration;
-
-use regex::Regex;
-use remote_storage::RemotePath;
-use tokio_util::sync::CancellationToken;
-use tracing::debug;
-use tracing::info;
-use tracing::warn;
-use utils::generation::Generation;
-use utils::id::TenantId;
-use utils::id::TimelineId;
-
-use crate::config::PageServerConf;
-use crate::deletion_queue::TEMP_SUFFIX;
-use crate::metrics;
-use crate::tenant::remote_timeline_client::remote_layer_path;
-use crate::tenant::storage_layer::LayerFileName;
-
-// The number of keys in a DeletionList before we will proactively persist it
-// (without reaching a flush deadline).  This aims to deliver objects of the order
-// of magnitude 1MB when we are under heavy delete load.
-const DELETION_LIST_TARGET_SIZE: usize = 16384;
-
-// Ordinarily, we only flush to DeletionList periodically, to bound the window during
-// which we might leak objects from not flushing a DeletionList after
-// the objects are already unlinked from timeline metadata.
-const FRONTEND_DEFAULT_TIMEOUT: Duration = Duration::from_millis(10000);
-
-// If someone is waiting for a flush to DeletionList, only delay a little to accumulate
-// more objects before doing the flush.
-const FRONTEND_FLUSHING_TIMEOUT: Duration = Duration::from_millis(100);
-
-#[derive(Debug)]
-pub(super) struct DeletionOp {
-    pub(super) tenant_id: TenantId,
-    pub(super) timeline_id: TimelineId,
-    // `layers` and `objects` are both just lists of objects.  `layers` is used if you do not
-    // have a config object handy to project it to a remote key, and need the consuming worker
-    // to do it for you.
-    pub(super) layers: Vec<(LayerFileName, Generation)>,
-    pub(super) objects: Vec<RemotePath>,
-
-    /// The _current_ generation of the Tenant attachment in which we are enqueuing
-    /// this deletion.
-    pub(super) generation: Generation,
-}
-
-#[derive(Debug)]
-pub(super) struct RecoverOp {
-    pub(super) attached_tenants: HashMap<TenantId, Generation>,
-}
-
-#[derive(Debug)]
-pub(super) enum ListWriterQueueMessage {
-    Delete(DeletionOp),
-    // Wait until all prior deletions make it into a persistent DeletionList
-    Flush(FlushOp),
-    // Wait until all prior deletions have been executed (i.e. objects are actually deleted)
-    FlushExecute(FlushOp),
-    // Call once after re-attaching to control plane, to notify the deletion queue about
-    // latest attached generations & load any saved deletion lists from disk.
-    Recover(RecoverOp),
-}
-
-pub(super) struct ListWriter {
-    conf: &'static PageServerConf,
-
-    // Incoming frontend requests to delete some keys
-    rx: tokio::sync::mpsc::Receiver<ListWriterQueueMessage>,
-
-    // Outbound requests to the backend to execute deletion lists we have composed.
-    tx: tokio::sync::mpsc::Sender<ValidatorQueueMessage>,
-
-    // The list we are currently building, contains a buffer of keys to delete
-    // and our next sequence number
-    pending: DeletionList,
-
-    // These FlushOps should notify the next time we flush
-    pending_flushes: Vec<FlushOp>,
-
-    // Worker loop is torn down when this fires.
-    cancel: CancellationToken,
-
-    // Safety guard to do recovery exactly once
-    recovered: bool,
-}
-
-impl ListWriter {
-    // Initially DeletionHeader.validated_sequence is zero.  The place we start our
-    // sequence numbers must be higher than that.
-    const BASE_SEQUENCE: u64 = 1;
-
-    pub(super) fn new(
-        conf: &'static PageServerConf,
-        rx: tokio::sync::mpsc::Receiver<ListWriterQueueMessage>,
-        tx: tokio::sync::mpsc::Sender<ValidatorQueueMessage>,
-        cancel: CancellationToken,
-    ) -> Self {
-        Self {
-            pending: DeletionList::new(Self::BASE_SEQUENCE),
-            conf,
-            rx,
-            tx,
-            pending_flushes: Vec::new(),
-            cancel,
-            recovered: false,
-        }
-    }
-
-    /// Try to flush `list` to persistent storage
-    ///
-    /// This does not return errors, because on failure to flush we do not lose
-    /// any state: flushing will be retried implicitly on the next deadline
-    async fn flush(&mut self) {
-        if self.pending.is_empty() {
-            for f in self.pending_flushes.drain(..) {
-                f.notify();
-            }
-            return;
-        }
-
-        match self.pending.save(self.conf).await {
-            Ok(_) => {
-                info!(sequence = self.pending.sequence, "Stored deletion list");
-
-                for f in self.pending_flushes.drain(..) {
-                    f.notify();
-                }
-
-                // Take the list we've accumulated, replace it with a fresh list for the next sequence
-                let next_list = DeletionList::new(self.pending.sequence + 1);
-                let list = std::mem::replace(&mut self.pending, next_list);
-
-                if let Err(e) = self.tx.send(ValidatorQueueMessage::Delete(list)).await {
-                    // This is allowed to fail: it will only happen if the backend worker is shut down,
-                    // so we can just drop this on the floor.
-                    info!("Deletion list dropped, this is normal during shutdown ({e:#})");
-                }
-            }
-            Err(e) => {
-                metrics::DELETION_QUEUE.unexpected_errors.inc();
-                warn!(
-                    sequence = self.pending.sequence,
-                    "Failed to write deletion list, will retry later ({e:#})"
-                );
-            }
-        }
-    }
-
-    /// Load the header, to learn the sequence number up to which deletions
-    /// have been validated.  We will apply validated=true to DeletionLists
-    /// <= this sequence when loading them.
-    ///
-    /// It is not an error for the header to not exist: we return None, and
-    /// the caller should act as if validated_sequence is 0
-    async fn load_validated_sequence(&self) -> Result<Option<u64>, anyhow::Error> {
-        let header_path = self.conf.deletion_header_path();
-        match tokio::fs::read(&header_path).await {
-            Ok(header_bytes) => {
-                match serde_json::from_slice::<DeletionHeader>(&header_bytes) {
-                    Ok(h) => Ok(Some(h.validated_sequence)),
-                    Err(e) => {
-                        warn!(
-                            "Failed to deserialize deletion header, ignoring {}: {e:#}",
-                            header_path.display()
-                        );
-                        // This should never happen unless we make a mistake with our serialization.
-                        // Ignoring a deletion header is not consequential for correctnes because all deletions
-                        // are ultimately allowed to fail: worst case we leak some objects for the scrubber to clean up.
-                        metrics::DELETION_QUEUE.unexpected_errors.inc();
-                        Ok(None)
-                    }
-                }
-            }
-            Err(e) => {
-                if e.kind() == std::io::ErrorKind::NotFound {
-                    debug!(
-                        "Deletion header {} not found, first start?",
-                        header_path.display()
-                    );
-                    Ok(None)
-                } else {
-                    Err(anyhow::anyhow!(e))
-                }
-            }
-        }
-    }
-
-    async fn recover(
-        &mut self,
-        attached_tenants: HashMap<TenantId, Generation>,
-    ) -> Result<(), anyhow::Error> {
-        debug!(
-            "recovering with {} attached tenants",
-            attached_tenants.len()
-        );
-
-        // Load the header
-        let validated_sequence = self.load_validated_sequence().await?.unwrap_or(0);
-
-        self.pending.sequence = validated_sequence + 1;
-
-        let deletion_directory = self.conf.deletion_prefix();
-        let mut dir = match tokio::fs::read_dir(&deletion_directory).await {
-            Ok(d) => d,
-            Err(e) => {
-                warn!(
-                    "Failed to open deletion list directory {}: {e:#}",
-                    deletion_directory.display(),
-                );
-
-                // Give up: if we can't read the deletion list directory, we probably can't
-                // write lists into it later, so the queue won't work.
-                return Err(e.into());
-            }
-        };
-
-        let list_name_pattern =
-            Regex::new("(?<sequence>[a-zA-Z0-9]{16})-(?<version>[a-zA-Z0-9]{2}).list").unwrap();
-
-        let header_path = self.conf.deletion_header_path();
-        let mut seqs: Vec<u64> = Vec::new();
-        while let Some(dentry) = dir.next_entry().await? {
-            let file_name = dentry.file_name();
-            let dentry_str = file_name.to_string_lossy();
-
-            if Some(file_name.as_os_str()) == header_path.file_name() {
-                // Don't try and parse the header's name like a list
-                continue;
-            }
-
-            if dentry_str.ends_with(TEMP_SUFFIX) {
-                info!("Cleaning up temporary file {dentry_str}");
-                let absolute_path = deletion_directory.join(dentry.file_name());
-                if let Err(e) = tokio::fs::remove_file(&absolute_path).await {
-                    // Non-fatal error: we will just leave the file behind but not
-                    // try and load it.
-                    warn!(
-                        "Failed to clean up temporary file {}: {e:#}",
-                        absolute_path.display()
-                    );
-                }
-
-                continue;
-            }
-
-            let file_name = dentry.file_name().to_owned();
-            let basename = file_name.to_string_lossy();
-            let seq_part = if let Some(m) = list_name_pattern.captures(&basename) {
-                m.name("sequence")
-                    .expect("Non optional group should be present")
-                    .as_str()
-            } else {
-                warn!("Unexpected key in deletion queue: {basename}");
-                metrics::DELETION_QUEUE.unexpected_errors.inc();
-                continue;
-            };
-
-            let seq: u64 = match u64::from_str_radix(seq_part, 16) {
-                Ok(s) => s,
-                Err(e) => {
-                    warn!("Malformed key '{basename}': {e}");
-                    metrics::DELETION_QUEUE.unexpected_errors.inc();
-                    continue;
-                }
-            };
-            seqs.push(seq);
-        }
-        seqs.sort();
-
-        // Start our next deletion list from after the last location validated by
-        // previous process lifetime, or after the last location found (it is updated
-        // below after enumerating the deletion lists)
-        self.pending.sequence = validated_sequence + 1;
-        if let Some(max_list_seq) = seqs.last() {
-            self.pending.sequence = std::cmp::max(self.pending.sequence, max_list_seq + 1);
-        }
-
-        for s in seqs {
-            let list_path = self.conf.deletion_list_path(s);
-
-            let list_bytes = tokio::fs::read(&list_path).await?;
-
-            let mut deletion_list = match serde_json::from_slice::<DeletionList>(&list_bytes) {
-                Ok(l) => l,
-                Err(e) => {
-                    // Drop the list on the floor: any objects it referenced will be left behind
-                    // for scrubbing to clean up.  This should never happen unless we have a serialization bug.
-                    warn!(sequence = s, "Failed to deserialize deletion list: {e}");
-                    metrics::DELETION_QUEUE.unexpected_errors.inc();
-                    continue;
-                }
-            };
-
-            if deletion_list.sequence <= validated_sequence {
-                // If the deletion list falls below valid_seq, we may assume that it was
-                // already validated the last time this pageserver ran.  Otherwise, we still
-                // load it, as it may still contain content valid in this generation.
-                deletion_list.validated = true;
-            } else {
-                // Special case optimization: if a tenant is still attached, and no other
-                // generation was issued to another node in the interval while we restarted,
-                // then we may treat deletion lists from the previous generation as if they
-                // belong to our currently attached generation, and proceed to validate & execute.
-                for (tenant_id, tenant_list) in &mut deletion_list.tenants {
-                    if let Some(attached_gen) = attached_tenants.get(tenant_id) {
-                        if attached_gen.previous() == tenant_list.generation {
-                            tenant_list.generation = *attached_gen;
-                        }
-                    }
-                }
-            }
-
-            info!(
-                validated = deletion_list.validated,
-                sequence = deletion_list.sequence,
-                "Recovered deletion list"
-            );
-
-            // We will drop out of recovery if this fails: it indicates that we are shutting down
-            // or the backend has panicked
-            metrics::DELETION_QUEUE
-                .keys_submitted
-                .inc_by(deletion_list.len() as u64);
-            self.tx
-                .send(ValidatorQueueMessage::Delete(deletion_list))
-                .await?;
-        }
-
-        info!(next_sequence = self.pending.sequence, "Replay complete");
-
-        Ok(())
-    }
-
-    /// This is the front-end ingest, where we bundle up deletion requests into DeletionList
-    /// and write them out, for later validation by the backend and execution by the executor.
-    pub(super) async fn background(&mut self) {
-        info!("Started deletion frontend worker");
-
-        // Synchronous, but we only do it once per process lifetime so it's tolerable
-        if let Err(e) = create_dir_all(&self.conf.deletion_prefix()) {
-            tracing::error!(
-                "Failed to create deletion list directory {}, deletions will not be executed ({e})",
-                self.conf.deletion_prefix().display()
-            );
-            metrics::DELETION_QUEUE.unexpected_errors.inc();
-            return;
-        }
-
-        while !self.cancel.is_cancelled() {
-            let timeout = if self.pending_flushes.is_empty() {
-                FRONTEND_DEFAULT_TIMEOUT
-            } else {
-                FRONTEND_FLUSHING_TIMEOUT
-            };
-
-            let msg = match tokio::time::timeout(timeout, self.rx.recv()).await {
-                Ok(Some(msg)) => msg,
-                Ok(None) => {
-                    // Queue sender destroyed, shutting down
-                    break;
-                }
-                Err(_) => {
-                    // Hit deadline, flush.
-                    self.flush().await;
-                    continue;
-                }
-            };
-
-            match msg {
-                ListWriterQueueMessage::Delete(op) => {
-                    assert!(
-                        self.recovered,
-                        "Cannot process deletions before recovery.  This is a bug."
-                    );
-
-                    debug!(
-                        "Delete: ingesting {} layers, {} other objects",
-                        op.layers.len(),
-                        op.objects.len()
-                    );
-
-                    let mut layer_paths = Vec::new();
-                    for (layer, generation) in op.layers {
-                        layer_paths.push(remote_layer_path(
-                            &op.tenant_id,
-                            &op.timeline_id,
-                            &layer,
-                            generation,
-                        ));
-                    }
-                    layer_paths.extend(op.objects);
-
-                    if !self.pending.push(
-                        &op.tenant_id,
-                        &op.timeline_id,
-                        op.generation,
-                        &mut layer_paths,
-                    ) {
-                        self.flush().await;
-                        let retry_succeeded = self.pending.push(
-                            &op.tenant_id,
-                            &op.timeline_id,
-                            op.generation,
-                            &mut layer_paths,
-                        );
-                        if !retry_succeeded {
-                            // Unexpected: after we flush, we should have
-                            // drained self.pending, so a conflict on
-                            // generation numbers should be impossible.
-                            tracing::error!(
-                                "Failed to enqueue deletions, leaking objects.  This is a bug."
-                            );
-                            metrics::DELETION_QUEUE.unexpected_errors.inc();
-                        }
-                    }
-                }
-                ListWriterQueueMessage::Flush(op) => {
-                    if self.pending.is_empty() {
-                        // Execute immediately
-                        debug!("Flush: No pending objects, flushing immediately");
-                        op.notify()
-                    } else {
-                        // Execute next time we flush
-                        debug!("Flush: adding to pending flush list for next deadline flush");
-                        self.pending_flushes.push(op);
-                    }
-                }
-                ListWriterQueueMessage::FlushExecute(op) => {
-                    debug!("FlushExecute: passing through to backend");
-                    // We do not flush to a deletion list here: the client sends a Flush before the FlushExecute
-                    if let Err(e) = self.tx.send(ValidatorQueueMessage::Flush(op)).await {
-                        info!("Can't flush, shutting down ({e})");
-                        // Caller will get error when their oneshot sender was dropped.
-                    }
-                }
-                ListWriterQueueMessage::Recover(op) => {
-                    if self.recovered {
-                        tracing::error!(
-                            "Deletion queue recovery called more than once.  This is a bug."
-                        );
-                        metrics::DELETION_QUEUE.unexpected_errors.inc();
-                        // Non-fatal: although this is a bug, since we did recovery at least once we may proceed.
-                        continue;
-                    }
-
-                    if let Err(e) = self.recover(op.attached_tenants).await {
-                        // This should only happen in truly unrecoverable cases, like the recovery finding that the backend
-                        // queue receiver has been dropped, or something is critically broken with
-                        // the local filesystem holding deletion lists.
-                        info!(
-                            "Deletion queue recover aborted, deletion queue will not proceed ({e})"
-                        );
-                        metrics::DELETION_QUEUE.unexpected_errors.inc();
-                        return;
-                    } else {
-                        self.recovered = true;
-                    }
-                }
-            }
-
-            if self.pending.len() > DELETION_LIST_TARGET_SIZE || !self.pending_flushes.is_empty() {
-                self.flush().await;
-            }
-        }
-        info!("Deletion queue shut down.");
-    }
-}

pageserver/src/deletion_queue/validator.rs

@@ -1,414 +0,0 @@
-//! The validator is responsible for validating DeletionLists for execution,
-//! based on whethe the generation in the DeletionList is still the latest
-//! generation for a tenant.
-//!
-//! The purpose of validation is to ensure split-brain safety in the cluster
-//! of pageservers: a deletion may only be executed if the tenant generation
-//! that originated it is still current.  See docs/rfcs/025-generation-numbers.md
-//! The purpose of accumulating lists before validating them is to reduce load
-//! on the control plane API by issuing fewer, larger requests.
-//!
-//! In addition to validating DeletionLists, the validator validates updates to remote_consistent_lsn
-//! for timelines: these are logically deletions because the safekeepers use remote_consistent_lsn
-//! to decide when old
-//!
-//! Deletions are passed onward to the Deleter.
-
-use std::collections::HashMap;
-use std::path::PathBuf;
-use std::sync::Arc;
-use std::time::Duration;
-
-use tokio_util::sync::CancellationToken;
-use tracing::debug;
-use tracing::info;
-use tracing::warn;
-
-use crate::config::PageServerConf;
-use crate::control_plane_client::ControlPlaneGenerationsApi;
-use crate::control_plane_client::RetryForeverError;
-use crate::metrics;
-
-use super::deleter::DeleterMessage;
-use super::DeletionHeader;
-use super::DeletionList;
-use super::DeletionQueueError;
-use super::FlushOp;
-use super::VisibleLsnUpdates;
-
-// After this length of time, do any validation work that is pending,
-// even if we haven't accumulated many keys to delete.
-//
-// This also causes updates to remote_consistent_lsn to be validated, even
-// if there were no deletions enqueued.
-const AUTOFLUSH_INTERVAL: Duration = Duration::from_secs(10);
-
-// If we have received this number of keys, proceed with attempting to execute
-const AUTOFLUSH_KEY_COUNT: usize = 16384;
-
-#[derive(Debug)]
-pub(super) enum ValidatorQueueMessage {
-    Delete(DeletionList),
-    Flush(FlushOp),
-}
-pub(super) struct Validator<C>
-where
-    C: ControlPlaneGenerationsApi,
-{
-    conf: &'static PageServerConf,
-    rx: tokio::sync::mpsc::Receiver<ValidatorQueueMessage>,
-    tx: tokio::sync::mpsc::Sender<DeleterMessage>,
-
-    // Client for calling into control plane API for validation of deletes
-    control_plane_client: Option<C>,
-
-    // DeletionLists which are waiting generation validation.  Not safe to
-    // execute until [`validate`] has processed them.
-    pending_lists: Vec<DeletionList>,
-
-    // DeletionLists which have passed validation and are ready to execute.
-    validated_lists: Vec<DeletionList>,
-
-    // Sum of all the lengths of lists in pending_lists
-    pending_key_count: usize,
-
-    // Lsn validation state: we read projected LSNs and write back visible LSNs
-    // after validation.  This is the LSN equivalent of `pending_validation_lists`:
-    // it is drained in [`validate`]
-    lsn_table: Arc<std::sync::RwLock<VisibleLsnUpdates>>,
-
-    // If we failed to rewrite a deletion list due to local filesystem I/O failure,
-    // we must remember that and refuse to advance our persistent validated sequence
-    // number past the failure.
-    list_write_failed: Option<u64>,
-
-    cancel: CancellationToken,
-}
-
-impl<C> Validator<C>
-where
-    C: ControlPlaneGenerationsApi,
-{
-    pub(super) fn new(
-        conf: &'static PageServerConf,
-        rx: tokio::sync::mpsc::Receiver<ValidatorQueueMessage>,
-        tx: tokio::sync::mpsc::Sender<DeleterMessage>,
-        control_plane_client: Option<C>,
-        lsn_table: Arc<std::sync::RwLock<VisibleLsnUpdates>>,
-        cancel: CancellationToken,
-    ) -> Self {
-        Self {
-            conf,
-            rx,
-            tx,
-            control_plane_client,
-            lsn_table,
-            pending_lists: Vec::new(),
-            validated_lists: Vec::new(),
-            pending_key_count: 0,
-            list_write_failed: None,
-            cancel,
-        }
-    }
-    /// Process any outstanding validations of generations of pending LSN updates or pending
-    /// DeletionLists.
-    ///
-    /// Valid LSN updates propagate back to Timelines immediately, valid DeletionLists
-    /// go into the queue of ready-to-execute lists.
-    async fn validate(&mut self) -> Result<(), DeletionQueueError> {
-        let mut tenant_generations = HashMap::new();
-        for list in &self.pending_lists {
-            for (tenant_id, tenant_list) in &list.tenants {
-                // Note: DeletionLists are in logical time order, so generation always
-                // goes up.  By doing a simple insert() we will always end up with
-                // the latest generation seen for a tenant.
-                tenant_generations.insert(*tenant_id, tenant_list.generation);
-            }
-        }
-
-        let pending_lsn_updates = {
-            let mut lsn_table = self.lsn_table.write().expect("Lock should not be poisoned");
-            std::mem::take(&mut *lsn_table)
-        };
-        for (tenant_id, update) in &pending_lsn_updates.tenants {
-            let entry = tenant_generations
-                .entry(*tenant_id)
-                .or_insert(update.generation);
-            if update.generation > *entry {
-                *entry = update.generation;
-            }
-        }
-
-        if tenant_generations.is_empty() {
-            // No work to do
-            return Ok(());
-        }
-
-        let tenants_valid = if let Some(control_plane_client) = &self.control_plane_client {
-            match control_plane_client
-                .validate(tenant_generations.iter().map(|(k, v)| (*k, *v)).collect())
-                .await
-            {
-                Ok(tenants) => tenants,
-                Err(RetryForeverError::ShuttingDown) => {
-                    // The only way a validation call returns an error is when the cancellation token fires
-                    return Err(DeletionQueueError::ShuttingDown);
-                }
-            }
-        } else {
-            // Control plane API disabled.  In legacy mode we consider everything valid.
-            tenant_generations.keys().map(|k| (*k, true)).collect()
-        };
-
-        let mut validated_sequence: Option<u64> = None;
-
-        // Apply the validation results to the pending LSN updates
-        for (tenant_id, tenant_lsn_state) in pending_lsn_updates.tenants {
-            let validated_generation = tenant_generations
-                .get(&tenant_id)
-                .expect("Map was built from the same keys we're reading");
-
-            let valid = tenants_valid
-                .get(&tenant_id)
-                .copied()
-                // If the tenant was missing from the validation response, it has been deleted.
-                // The Timeline that requested the LSN update is probably already torn down,
-                // or will be torn down soon.  In this case, drop the update by setting valid=false.
-                .unwrap_or(false);
-
-            if valid && *validated_generation == tenant_lsn_state.generation {
-                for (_timeline_id, pending_lsn) in tenant_lsn_state.timelines {
-                    pending_lsn.result_slot.store(pending_lsn.projected);
-                }
-            } else {
-                // If we failed validation, then do not apply any of the projected updates
-                warn!("Dropped remote consistent LSN updates for tenant {tenant_id} in stale generation {:?}", tenant_lsn_state.generation);
-                metrics::DELETION_QUEUE.dropped_lsn_updates.inc();
-            }
-        }
-
-        // Apply the validation results to the pending deletion lists
-        for list in &mut self.pending_lists {
-            // Filter the list based on whether the server responded valid: true.
-            // If a tenant is omitted in the response, it has been deleted, and we should
-            // proceed with deletion.
-            let mut mutated = false;
-            list.tenants.retain(|tenant_id, tenant| {
-                let validated_generation = tenant_generations
-                    .get(tenant_id)
-                    .expect("Map was built from the same keys we're reading");
-
-                // If the tenant was missing from the validation response, it has been deleted.
-                // This means that a deletion is valid, but also redundant since the tenant's
-                // objects should have already been deleted.  Treat it as invalid to drop the
-                // redundant deletion.
-                let valid = tenants_valid.get(tenant_id).copied().unwrap_or(false);
-
-                // A list is valid if it comes from the current _or previous_ generation.
-                // - The previous generation case is permitted due to how we store deletion lists locally:
-                // if we see the immediately previous generation in a locally stored deletion list,
-                // it proves that this node's disk was used for both current & previous generations,
-                // and therefore no other node was involved in between: the two generations may be
-                // logically treated as the same.
-                // - In that previous generation case, we rewrote it to the current generation
-                // in recover(), so the comparison here is simply an equality.
-
-                let this_list_valid = valid
-                    && (tenant.generation == *validated_generation);
-
-                if !this_list_valid {
-                    warn!("Dropping stale deletions for tenant {tenant_id} in generation {:?}, objects may be leaked", tenant.generation);
-                    metrics::DELETION_QUEUE.keys_dropped.inc_by(tenant.len() as u64);
-                    mutated = true;
-                }
-                this_list_valid
-            });
-            list.validated = true;
-
-            if mutated {
-                // Save the deletion list if we had to make changes due to stale generations.  The
-                // saved list is valid for execution.
-                if let Err(e) = list.save(self.conf).await {
-                    // Highly unexpected.  Could happen if e.g. disk full.
-                    // If we didn't save the trimmed list, it is _not_ valid to execute.
-                    warn!("Failed to save modified deletion list {list}: {e:#}");
-                    metrics::DELETION_QUEUE.unexpected_errors.inc();
-
-                    // Rather than have a complex retry process, just drop it and leak the objects,
-                    // scrubber will clean up eventually.
-                    list.tenants.clear(); // Result is a valid-but-empty list, which is a no-op for execution.
-
-                    // We must remember this failure, to prevent later writing out a header that
-                    // would imply the unwritable list was valid on disk.
-                    if self.list_write_failed.is_none() {
-                        self.list_write_failed = Some(list.sequence);
-                    }
-                }
-            }
-
-            validated_sequence = Some(list.sequence);
-        }
-
-        if let Some(validated_sequence) = validated_sequence {
-            if let Some(list_write_failed) = self.list_write_failed {
-                // Rare error case: we failed to write out a deletion list to excise invalid
-                // entries, so we cannot advance the header's valid sequence number past that point.
-                //
-                // In this state we will continue to validate, execute and delete deletion lists,
-                // we just cannot update the header.  It should be noticed and fixed by a human due to
-                // the nonzero value of our unexpected_errors metric.
-                warn!(
-                    sequence_number = list_write_failed,
-                    "Cannot write header because writing a deletion list failed earlier",
-                );
-            } else {
-                // Write the queue header to record how far validation progressed.  This avoids having
-                // to rewrite each DeletionList to set validated=true in it.
-                let header = DeletionHeader::new(validated_sequence);
-
-                // Drop result because the validated_sequence is an optimization.  If we fail to save it,
-                // then restart, we will drop some deletion lists, creating work for scrubber.
-                // The save() function logs a warning on error.
-                if let Err(e) = header.save(self.conf).await {
-                    warn!("Failed to write deletion queue header: {e:#}");
-                    metrics::DELETION_QUEUE.unexpected_errors.inc();
-                }
-            }
-        }
-
-        // Transfer the validated lists to the validated queue, for eventual execution
-        self.validated_lists.append(&mut self.pending_lists);
-
-        Ok(())
-    }
-
-    async fn cleanup_lists(&mut self, list_paths: Vec<PathBuf>) {
-        for list_path in list_paths {
-            debug!("Removing deletion list {}", list_path.display());
-
-            if let Err(e) = tokio::fs::remove_file(&list_path).await {
-                // Unexpected: we should have permissions and nothing else should
-                // be touching these files.  We will leave the file behind.  Subsequent
-                // pageservers will try and load it again: hopefully whatever storage
-                // issue (probably permissions) has been fixed by then.
-                tracing::error!("Failed to delete {}: {e:#}", list_path.display());
-                metrics::DELETION_QUEUE.unexpected_errors.inc();
-                break;
-            }
-        }
-    }
-
-    async fn flush(&mut self) -> Result<(), DeletionQueueError> {
-        tracing::debug!("Flushing with {} pending lists", self.pending_lists.len());
-
-        // Issue any required generation validation calls to the control plane
-        self.validate().await?;
-
-        // After successful validation, nothing is pending: any lists that
-        // made it through validation will be in validated_lists.
-        assert!(self.pending_lists.is_empty());
-        self.pending_key_count = 0;
-
-        tracing::debug!(
-            "Validation complete, have {} validated lists",
-            self.validated_lists.len()
-        );
-
-        // Return quickly if we have no validated lists to execute.  This avoids flushing the
-        // executor when an idle backend hits its autoflush interval
-        if self.validated_lists.is_empty() {
-            return Ok(());
-        }
-
-        // Drain `validated_lists` into the executor
-        let mut executing_lists = Vec::new();
-        for list in self.validated_lists.drain(..) {
-            let list_path = self.conf.deletion_list_path(list.sequence);
-            let objects = list.into_remote_paths();
-            self.tx
-                .send(DeleterMessage::Delete(objects))
-                .await
-                .map_err(|_| DeletionQueueError::ShuttingDown)?;
-            executing_lists.push(list_path);
-        }
-
-        self.flush_executor().await?;
-
-        // Erase the deletion lists whose keys have all be deleted from remote storage
-        self.cleanup_lists(executing_lists).await;
-
-        Ok(())
-    }
-
-    async fn flush_executor(&mut self) -> Result<(), DeletionQueueError> {
-        // Flush the executor, so that all the keys referenced by these deletion lists
-        // are actually removed from remote storage.  This is a precondition to deleting
-        // the deletion lists themselves.
-        let (flush_op, rx) = FlushOp::new();
-        self.tx
-            .send(DeleterMessage::Flush(flush_op))
-            .await
-            .map_err(|_| DeletionQueueError::ShuttingDown)?;
-
-        rx.await.map_err(|_| DeletionQueueError::ShuttingDown)
-    }
-
-    pub(super) async fn background(&mut self) {
-        tracing::info!("Started deletion backend worker");
-
-        while !self.cancel.is_cancelled() {
-            let msg = match tokio::time::timeout(AUTOFLUSH_INTERVAL, self.rx.recv()).await {
-                Ok(Some(m)) => m,
-                Ok(None) => {
-                    // All queue senders closed
-                    info!("Shutting down");
-                    break;
-                }
-                Err(_) => {
-                    // Timeout, we hit deadline to execute whatever we have in hand.  These functions will
-                    // return immediately if no work is pending.
-                    match self.flush().await {
-                        Ok(()) => {}
-                        Err(DeletionQueueError::ShuttingDown) => {
-                            // If we are shutting down, then auto-flush can safely be skipped
-                        }
-                    }
-
-                    continue;
-                }
-            };
-
-            match msg {
-                ValidatorQueueMessage::Delete(list) => {
-                    if list.validated {
-                        // A pre-validated list may only be seen during recovery, if we are recovering
-                        // a DeletionList whose on-disk state has validated=true
-                        self.validated_lists.push(list)
-                    } else {
-                        self.pending_key_count += list.len();
-                        self.pending_lists.push(list);
-                    }
-
-                    if self.pending_key_count > AUTOFLUSH_KEY_COUNT {
-                        match self.flush().await {
-                            Ok(()) => {}
-                            Err(DeletionQueueError::ShuttingDown) => {
-                                // If we are shutting down, then auto-flush can safely be skipped
-                            }
-                        }
-                    }
-                }
-                ValidatorQueueMessage::Flush(op) => {
-                    match self.flush().await {
-                        Ok(()) => {
-                            op.notify();
-                        }
-                        Err(DeletionQueueError::ShuttingDown) => {
-                            // If we fail due to shutting down, we will just drop `op` to propagate that status.
-                        }
-                    }
-                }
-            }
-        }
-    }
-}

pageserver/src/http/openapi_spec.yml

@@ -1093,9 +1093,6 @@ components:
         remote_consistent_lsn:
           type: string
           format: hex
-        remote_consistent_lsn_visible:
-          type: string
-          format: hex
         ancestor_timeline_id:
           type: string
           format: hex

pageserver/src/http/routes.rs

@@ -5,7 +5,6 @@ use std::collections::HashMap;
 use std::sync::Arc;
 
 use anyhow::{anyhow, Context, Result};
-use futures::TryFutureExt;
 use hyper::StatusCode;
 use hyper::{Body, Request, Response, Uri};
 use metrics::launch_timestamp::LaunchTimestamp;
@@ -25,7 +24,6 @@ use super::models::{
     TimelineCreateRequest, TimelineGcRequest, TimelineInfo,
 };
 use crate::context::{DownloadBehavior, RequestContext};
-use crate::deletion_queue::DeletionQueueClient;
 use crate::metrics::{StorageTimeOperation, STORAGE_TIME_GLOBAL};
 use crate::pgdatadir_mapping::LsnForTimestamp;
 use crate::task_mgr::TaskKind;
@@ -36,7 +34,7 @@ use crate::tenant::mgr::{
 use crate::tenant::size::ModelInputs;
 use crate::tenant::storage_layer::LayerAccessStatsReset;
 use crate::tenant::timeline::Timeline;
-use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError, TenantSharedResources};
+use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError};
 use crate::{config::PageServerConf, tenant::mgr};
 use crate::{disk_usage_eviction_task, tenant};
 use utils::{
@@ -63,7 +61,6 @@ pub struct State {
     remote_storage: Option<GenericRemoteStorage>,
     broker_client: storage_broker::BrokerClientChannel,
     disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
-    deletion_queue_client: DeletionQueueClient,
 }
 
 impl State {
@@ -73,7 +70,6 @@ impl State {
         remote_storage: Option<GenericRemoteStorage>,
         broker_client: storage_broker::BrokerClientChannel,
         disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
-        deletion_queue_client: DeletionQueueClient,
     ) -> anyhow::Result<Self> {
         let allowlist_routes = ["/v1/status", "/v1/doc", "/swagger.yml"]
             .iter()
@@ -86,17 +82,8 @@ impl State {
             remote_storage,
             broker_client,
             disk_usage_eviction_state,
-            deletion_queue_client,
         })
     }
-
-    fn tenant_resources(&self) -> TenantSharedResources {
-        TenantSharedResources {
-            broker_client: self.broker_client.clone(),
-            remote_storage: self.remote_storage.clone(),
-            deletion_queue_client: self.deletion_queue_client.clone(),
-        }
-    }
 }
 
 #[inline(always)]
@@ -296,12 +283,7 @@ async fn build_timeline_info_common(
     };
     let current_physical_size = Some(timeline.layer_size_sum().await);
     let state = timeline.current_state();
-    let remote_consistent_lsn_projected = timeline
-        .get_remote_consistent_lsn_projected()
-        .unwrap_or(Lsn(0));
-    let remote_consistent_lsn_visible = timeline
-        .get_remote_consistent_lsn_visible()
-        .unwrap_or(Lsn(0));
+    let remote_consistent_lsn = timeline.get_remote_consistent_lsn().unwrap_or(Lsn(0));
 
     let walreceiver_status = timeline.walreceiver_status();
 
@@ -311,8 +293,7 @@ async fn build_timeline_info_common(
         ancestor_timeline_id,
         ancestor_lsn,
         disk_consistent_lsn: timeline.get_disk_consistent_lsn(),
-        remote_consistent_lsn: remote_consistent_lsn_projected,
-        remote_consistent_lsn_visible,
+        remote_consistent_lsn,
         last_record_lsn,
         prev_record_lsn: Some(timeline.get_prev_record_lsn()),
         latest_gc_cutoff_lsn: *timeline.get_latest_gc_cutoff_lsn(),
@@ -511,23 +492,24 @@ async fn tenant_attach_handler(
 
     let generation = get_request_generation(state, maybe_body.as_ref().and_then(|r| r.generation))?;
 
-    if state.remote_storage.is_none() {
+    if let Some(remote_storage) = &state.remote_storage {
+        mgr::attach_tenant(
+            state.conf,
+            tenant_id,
+            generation,
+            tenant_conf,
+            state.broker_client.clone(),
+            remote_storage.clone(),
+            &ctx,
+        )
+        .instrument(info_span!("tenant_attach", %tenant_id))
+        .await?;
+    } else {
         return Err(ApiError::BadRequest(anyhow!(
             "attach_tenant is not possible because pageserver was configured without remote storage"
         )));
     }
 
-    mgr::attach_tenant(
-        state.conf,
-        tenant_id,
-        generation,
-        tenant_conf,
-        state.tenant_resources(),
-        &ctx,
-    )
-    .instrument(info_span!("tenant_attach", %tenant_id))
-    .await?;
-
     json_response(StatusCode::ACCEPTED, ())
 }
 
@@ -588,7 +570,6 @@ async fn tenant_load_handler(
         generation,
         state.broker_client.clone(),
         state.remote_storage.clone(),
-        state.deletion_queue_client.clone(),
         &ctx,
     )
     .instrument(info_span!("load", %tenant_id))
@@ -930,7 +911,8 @@ async fn tenant_create_handler(
         tenant_conf,
         target_tenant_id,
         generation,
-        state.tenant_resources(),
+        state.broker_client.clone(),
+        state.remote_storage.clone(),
         &ctx,
     )
     .instrument(info_span!("tenant_create", tenant_id = %target_tenant_id))
@@ -1147,39 +1129,6 @@ async fn timeline_download_remote_layers_handler_get(
     json_response(StatusCode::OK, info)
 }
 
-async fn deletion_queue_flush(
-    r: Request<Body>,
-    cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    let state = get_state(&r);
-
-    if state.remote_storage.is_none() {
-        // Nothing to do if remote storage is disabled.
-        return json_response(StatusCode::OK, ());
-    }
-
-    let execute = parse_query_param(&r, "execute")?.unwrap_or(false);
-
-    let flush = async {
-        if execute {
-            state.deletion_queue_client.flush_execute().await
-        } else {
-            state.deletion_queue_client.flush().await
-        }
-    }
-    // DeletionQueueError's only case is shutting down.
-    .map_err(|_| ApiError::ShuttingDown);
-
-    tokio::select! {
-        res = flush => {
-            res.map(|()| json_response(StatusCode::OK, ()))?
-        }
-        _ = cancel.cancelled() => {
-            Err(ApiError::ShuttingDown)
-        }
-    }
-}
-
 async fn active_timeline_of_active_tenant(
     tenant_id: TenantId,
     timeline_id: TimelineId,
@@ -1514,9 +1463,6 @@ pub fn make_router(
         .put("/v1/disk_usage_eviction/run", |r| {
             api_handler(r, disk_usage_eviction_run)
         })
-        .put("/v1/deletion_queue/flush", |r| {
-            api_handler(r, deletion_queue_flush)
-        })
         .put("/v1/tenant/:tenant_id/break", |r| {
             testing_api_handler("set tenant state to broken", r, handle_tenant_break)
         })

pageserver/src/import_datadir.rs

@@ -75,12 +75,12 @@ pub async fn import_timeline_from_postgres_datadir(
             {
                 pg_control = Some(control_file);
             }
-            modification.flush(ctx).await?;
+            modification.flush().await?;
         }
     }
 
     // We're done importing all the data files.
-    modification.commit(ctx).await?;
+    modification.commit().await?;
 
     // We expect the Postgres server to be shut down cleanly.
     let pg_control = pg_control.context("pg_control file not found")?;
@@ -359,7 +359,7 @@ pub async fn import_basebackup_from_tar(
                     // We found the pg_control file.
                     pg_control = Some(res);
                 }
-                modification.flush(ctx).await?;
+                modification.flush().await?;
             }
             tokio_tar::EntryType::Directory => {
                 debug!("directory {:?}", file_path);
@@ -377,7 +377,7 @@ pub async fn import_basebackup_from_tar(
     // sanity check: ensure that pg_control is loaded
     let _pg_control = pg_control.context("pg_control file not found")?;
 
-    modification.commit(ctx).await?;
+    modification.commit().await?;
     Ok(())
 }
 

pageserver/src/lib.rs

@@ -3,8 +3,7 @@ pub mod basebackup;
 pub mod config;
 pub mod consumption_metrics;
 pub mod context;
-pub mod control_plane_client;
-pub mod deletion_queue;
+mod control_plane_client;
 pub mod disk_usage_eviction_task;
 pub mod http;
 pub mod import_datadir;
@@ -28,7 +27,6 @@ pub mod failpoint_support;
 use std::path::Path;
 
 use crate::task_mgr::TaskKind;
-use deletion_queue::DeletionQueue;
 use tracing::info;
 
 /// Current storage format version
@@ -50,8 +48,8 @@ static ZERO_PAGE: bytes::Bytes = bytes::Bytes::from_static(&[0u8; 8192]);
 
 pub use crate::metrics::preinitialize_metrics;
 
-#[tracing::instrument(skip_all, fields(%exit_code))]
-pub async fn shutdown_pageserver(deletion_queue: Option<DeletionQueue>, exit_code: i32) {
+#[tracing::instrument]
+pub async fn shutdown_pageserver(exit_code: i32) {
     use std::time::Duration;
     // Shut down the libpq endpoint task. This prevents new connections from
     // being accepted.
@@ -79,11 +77,6 @@ pub async fn shutdown_pageserver(deletion_queue: Option<DeletionQueue>, exit_cod
     )
     .await;
 
-    // Best effort to persist any outstanding deletions, to avoid leaking objects
-    if let Some(mut deletion_queue) = deletion_queue {
-        deletion_queue.shutdown(Duration::from_secs(5)).await;
-    }
-
     // Shut down the HTTP endpoint last, so that you can still check the server's
     // status while it's shutting down.
     // FIXME: We should probably stop accepting commands like attach/detach earlier.

pageserver/src/metrics.rs

@@ -1,4 +1,3 @@
-use enum_map::EnumMap;
 use metrics::metric_vec_duration::DurationResultObserver;
 use metrics::{
     register_counter_vec, register_gauge_vec, register_histogram, register_histogram_vec,
@@ -128,24 +127,22 @@ pub(crate) static MATERIALIZED_PAGE_CACHE_HIT: Lazy<IntCounter> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
-pub struct PageCacheMetricsForTaskKind {
+pub struct PageCacheMetrics {
     pub read_accesses_materialized_page: IntCounter,
+    pub read_accesses_ephemeral: IntCounter,
     pub read_accesses_immutable: IntCounter,
 
+    pub read_hits_ephemeral: IntCounter,
     pub read_hits_immutable: IntCounter,
     pub read_hits_materialized_page_exact: IntCounter,
     pub read_hits_materialized_page_older_lsn: IntCounter,
 }
 
-pub struct PageCacheMetrics {
-    map: EnumMap<TaskKind, EnumMap<PageContentKind, PageCacheMetricsForTaskKind>>,
-}
-
 static PAGE_CACHE_READ_HITS: Lazy<IntCounterVec> = Lazy::new(|| {
     register_int_counter_vec!(
         "pageserver_page_cache_read_hits_total",
         "Number of read accesses to the page cache that hit",
-        &["task_kind", "key_kind", "content_kind", "hit_kind"]
+        &["key_kind", "hit_kind"]
     )
     .expect("failed to define a metric")
 });
@@ -154,73 +151,55 @@ static PAGE_CACHE_READ_ACCESSES: Lazy<IntCounterVec> = Lazy::new(|| {
     register_int_counter_vec!(
         "pageserver_page_cache_read_accesses_total",
         "Number of read accesses to the page cache",
-        &["task_kind", "key_kind", "content_kind"]
+        &["key_kind"]
     )
     .expect("failed to define a metric")
 });
 
 pub static PAGE_CACHE: Lazy<PageCacheMetrics> = Lazy::new(|| PageCacheMetrics {
-    map: EnumMap::from_array(std::array::from_fn(|task_kind| {
-        let task_kind = <TaskKind as enum_map::Enum>::from_usize(task_kind);
-        let task_kind: &'static str = task_kind.into();
-        EnumMap::from_array(std::array::from_fn(|content_kind| {
-            let content_kind = <PageContentKind as enum_map::Enum>::from_usize(content_kind);
-            let content_kind: &'static str = content_kind.into();
-            PageCacheMetricsForTaskKind {
-                read_accesses_materialized_page: {
-                    PAGE_CACHE_READ_ACCESSES
-                        .get_metric_with_label_values(&[
-                            task_kind,
-                            "materialized_page",
-                            content_kind,
-                        ])
-                        .unwrap()
-                },
+    read_accesses_materialized_page: {
+        PAGE_CACHE_READ_ACCESSES
+            .get_metric_with_label_values(&["materialized_page"])
+            .unwrap()
+    },
 
-                read_accesses_immutable: {
-                    PAGE_CACHE_READ_ACCESSES
-                        .get_metric_with_label_values(&[task_kind, "immutable", content_kind])
-                        .unwrap()
-                },
+    read_accesses_ephemeral: {
+        PAGE_CACHE_READ_ACCESSES
+            .get_metric_with_label_values(&["ephemeral"])
+            .unwrap()
+    },
 
-                read_hits_immutable: {
-                    PAGE_CACHE_READ_HITS
-                        .get_metric_with_label_values(&[task_kind, "immutable", content_kind, "-"])
-                        .unwrap()
-                },
+    read_accesses_immutable: {
+        PAGE_CACHE_READ_ACCESSES
+            .get_metric_with_label_values(&["immutable"])
+            .unwrap()
+    },
 
-                read_hits_materialized_page_exact: {
-                    PAGE_CACHE_READ_HITS
-                        .get_metric_with_label_values(&[
-                            task_kind,
-                            "materialized_page",
-                            content_kind,
-                            "exact",
-                        ])
-                        .unwrap()
-                },
+    read_hits_ephemeral: {
+        PAGE_CACHE_READ_HITS
+            .get_metric_with_label_values(&["ephemeral", "-"])
+            .unwrap()
+    },
 
-                read_hits_materialized_page_older_lsn: {
-                    PAGE_CACHE_READ_HITS
-                        .get_metric_with_label_values(&[
-                            task_kind,
-                            "materialized_page",
-                            content_kind,
-                            "older_lsn",
-                        ])
-                        .unwrap()
-                },
-            }
-        }))
-    })),
+    read_hits_immutable: {
+        PAGE_CACHE_READ_HITS
+            .get_metric_with_label_values(&["immutable", "-"])
+            .unwrap()
+    },
+
+    read_hits_materialized_page_exact: {
+        PAGE_CACHE_READ_HITS
+            .get_metric_with_label_values(&["materialized_page", "exact"])
+            .unwrap()
+    },
+
+    read_hits_materialized_page_older_lsn: {
+        PAGE_CACHE_READ_HITS
+            .get_metric_with_label_values(&["materialized_page", "older_lsn"])
+            .unwrap()
+    },
 });
 
-impl PageCacheMetrics {
-    pub(crate) fn for_ctx(&self, ctx: &RequestContext) -> &PageCacheMetricsForTaskKind {
-        &self.map[ctx.task_kind()][ctx.page_content_kind()]
-    }
-}
-
 pub struct PageCacheSizeMetrics {
     pub max_bytes: UIntGauge,
 
@@ -291,14 +270,6 @@ static RESIDENT_PHYSICAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
-pub(crate) static RESIDENT_PHYSICAL_SIZE_GLOBAL: Lazy<UIntGauge> = Lazy::new(|| {
-    register_uint_gauge!(
-        "pageserver_resident_physical_size_global",
-        "Like `pageserver_resident_physical_size`, but without tenant/timeline dimensions."
-    )
-    .expect("failed to define a metric")
-});
-
 static REMOTE_PHYSICAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
     register_uint_gauge_vec!(
         "pageserver_remote_physical_size",
@@ -309,14 +280,6 @@ static REMOTE_PHYSICAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
-static REMOTE_PHYSICAL_SIZE_GLOBAL: Lazy<UIntGauge> = Lazy::new(|| {
-    register_uint_gauge!(
-        "pageserver_remote_physical_size_global",
-        "Like `pageserver_remote_physical_size`, but without tenant/timeline dimensions."
-    )
-    .expect("failed to define a metric")
-});
-
 pub(crate) static REMOTE_ONDEMAND_DOWNLOADED_LAYERS: Lazy<IntCounter> = Lazy::new(|| {
     register_int_counter!(
         "pageserver_remote_ondemand_downloaded_layers_total",
@@ -903,54 +866,6 @@ static REMOTE_TIMELINE_CLIENT_BYTES_FINISHED_COUNTER: Lazy<IntCounterVec> = Lazy
     .expect("failed to define a metric")
 });
 
-pub(crate) struct DeletionQueueMetrics {
-    pub(crate) keys_submitted: IntCounter,
-    pub(crate) keys_dropped: IntCounter,
-    pub(crate) keys_executed: IntCounter,
-    pub(crate) dropped_lsn_updates: IntCounter,
-    pub(crate) unexpected_errors: IntCounter,
-    pub(crate) remote_errors: IntCounterVec,
-}
-pub(crate) static DELETION_QUEUE: Lazy<DeletionQueueMetrics> = Lazy::new(|| {
-    DeletionQueueMetrics{
-
-    keys_submitted: register_int_counter!(
-        "pageserver_deletion_queue_submitted_total",
-        "Number of objects submitted for deletion"
-    )
-    .expect("failed to define a metric"),
-
-    keys_dropped: register_int_counter!(
-        "pageserver_deletion_queue_dropped_total",
-        "Number of object deletions dropped due to stale generation."
-    )
-    .expect("failed to define a metric"),
-
-    keys_executed: register_int_counter!(
-        "pageserver_deletion_queue_executed_total",
-        "Number of objects deleted. Only includes objects that we actually deleted, sum with pageserver_deletion_queue_dropped_total for the total number of keys processed."
-    )
-    .expect("failed to define a metric"),
-
-    dropped_lsn_updates: register_int_counter!(
-        "pageserver_deletion_queue_dropped_lsn_updates_total",
-        "Updates to remote_consistent_lsn dropped due to stale generation number."
-    )
-    .expect("failed to define a metric"),
-    unexpected_errors: register_int_counter!(
-        "pageserver_deletion_queue_unexpected_errors_total",
-        "Number of unexpected condiions that may stall the queue: any value above zero is unexpected."
-    )
-    .expect("failed to define a metric"),
-    remote_errors: register_int_counter_vec!(
-        "pageserver_deletion_queue_remote_errors_total",
-        "Retryable remote I/O errors while executing deletions, for example 503 responses to DeleteObjects",
-        &["op_kind"],
-    )
-    .expect("failed to define a metric")
-}
-});
-
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
 pub enum RemoteOpKind {
     Upload,
@@ -1225,7 +1140,7 @@ pub struct TimelineMetrics {
     pub load_layer_map_histo: StorageTimeMetrics,
     pub garbage_collect_histo: StorageTimeMetrics,
     pub last_record_gauge: IntGauge,
-    resident_physical_size_gauge: UIntGauge,
+    pub resident_physical_size_gauge: UIntGauge,
     /// copy of LayeredTimeline.current_logical_size
     pub current_logical_size_gauge: UIntGauge,
     pub num_persistent_files_created: IntCounter,
@@ -1303,29 +1218,10 @@ impl TimelineMetrics {
     }
 
     pub fn record_new_file_metrics(&self, sz: u64) {
-        self.resident_physical_size_add(sz);
+        self.resident_physical_size_gauge.add(sz);
         self.num_persistent_files_created.inc_by(1);
         self.persistent_bytes_written.inc_by(sz);
     }
-
-    pub fn resident_physical_size_sub(&self, sz: u64) {
-        self.resident_physical_size_gauge.sub(sz);
-        crate::metrics::RESIDENT_PHYSICAL_SIZE_GLOBAL.sub(sz);
-    }
-
-    pub fn resident_physical_size_add(&self, sz: u64) {
-        self.resident_physical_size_gauge.add(sz);
-        crate::metrics::RESIDENT_PHYSICAL_SIZE_GLOBAL.add(sz);
-    }
-
-    pub fn resident_physical_size_set(&self, sz: u64) {
-        self.resident_physical_size_gauge.set(sz);
-        crate::metrics::RESIDENT_PHYSICAL_SIZE_GLOBAL.set(sz);
-    }
-
-    pub fn resident_physical_size_get(&self) -> u64 {
-        self.resident_physical_size_gauge.get()
-    }
 }
 
 impl Drop for TimelineMetrics {
@@ -1333,10 +1229,7 @@ impl Drop for TimelineMetrics {
         let tenant_id = &self.tenant_id;
         let timeline_id = &self.timeline_id;
         let _ = LAST_RECORD_LSN.remove_label_values(&[tenant_id, timeline_id]);
-        {
-            RESIDENT_PHYSICAL_SIZE_GLOBAL.sub(self.resident_physical_size_get());
-            let _ = RESIDENT_PHYSICAL_SIZE.remove_label_values(&[tenant_id, timeline_id]);
-        }
+        let _ = RESIDENT_PHYSICAL_SIZE.remove_label_values(&[tenant_id, timeline_id]);
         let _ = CURRENT_LOGICAL_SIZE.remove_label_values(&[tenant_id, timeline_id]);
         let _ = NUM_PERSISTENT_FILES_CREATED.remove_label_values(&[tenant_id, timeline_id]);
         let _ = PERSISTENT_BYTES_WRITTEN.remove_label_values(&[tenant_id, timeline_id]);
@@ -1387,46 +1280,10 @@ use std::sync::{Arc, Mutex};
 use std::task::{Context, Poll};
 use std::time::{Duration, Instant};
 
-use crate::context::{PageContentKind, RequestContext};
-use crate::task_mgr::TaskKind;
-
-/// Maintain a per timeline gauge in addition to the global gauge.
-struct PerTimelineRemotePhysicalSizeGauge {
-    last_set: u64,
-    gauge: UIntGauge,
-}
-
-impl PerTimelineRemotePhysicalSizeGauge {
-    fn new(per_timeline_gauge: UIntGauge) -> Self {
-        Self {
-            last_set: per_timeline_gauge.get(),
-            gauge: per_timeline_gauge,
-        }
-    }
-    fn set(&mut self, sz: u64) {
-        self.gauge.set(sz);
-        if sz < self.last_set {
-            REMOTE_PHYSICAL_SIZE_GLOBAL.sub(self.last_set - sz);
-        } else {
-            REMOTE_PHYSICAL_SIZE_GLOBAL.add(sz - self.last_set);
-        };
-        self.last_set = sz;
-    }
-    fn get(&self) -> u64 {
-        self.gauge.get()
-    }
-}
-
-impl Drop for PerTimelineRemotePhysicalSizeGauge {
-    fn drop(&mut self) {
-        REMOTE_PHYSICAL_SIZE_GLOBAL.sub(self.last_set);
-    }
-}
-
 pub struct RemoteTimelineClientMetrics {
     tenant_id: String,
     timeline_id: String,
-    remote_physical_size_gauge: Mutex<Option<PerTimelineRemotePhysicalSizeGauge>>,
+    remote_physical_size_gauge: Mutex<Option<UIntGauge>>,
     calls_unfinished_gauge: Mutex<HashMap<(&'static str, &'static str), IntGauge>>,
     bytes_started_counter: Mutex<HashMap<(&'static str, &'static str), IntCounter>>,
     bytes_finished_counter: Mutex<HashMap<(&'static str, &'static str), IntCounter>>,
@@ -1444,24 +1301,18 @@ impl RemoteTimelineClientMetrics {
         }
     }
 
-    pub(crate) fn remote_physical_size_set(&self, sz: u64) {
+    pub fn remote_physical_size_gauge(&self) -> UIntGauge {
         let mut guard = self.remote_physical_size_gauge.lock().unwrap();
-        let gauge = guard.get_or_insert_with(|| {
-            PerTimelineRemotePhysicalSizeGauge::new(
+        guard
+            .get_or_insert_with(|| {
                 REMOTE_PHYSICAL_SIZE
                     .get_metric_with_label_values(&[
                         &self.tenant_id.to_string(),
                         &self.timeline_id.to_string(),
                     ])
-                    .unwrap(),
-            )
-        });
-        gauge.set(sz);
-    }
-
-    pub(crate) fn remote_physical_size_get(&self) -> u64 {
-        let guard = self.remote_physical_size_gauge.lock().unwrap();
-        guard.as_ref().map(|gauge| gauge.get()).unwrap_or(0)
+                    .unwrap()
+            })
+            .clone()
     }
 
     pub fn remote_operation_time(
@@ -1800,9 +1651,6 @@ pub fn preinitialize_metrics() {
         Lazy::force(c);
     });
 
-    // Deletion queue stats
-    Lazy::force(&DELETION_QUEUE);
-
     // countervecs
     [&BACKGROUND_LOOP_PERIOD_OVERRUN_COUNT]
         .into_iter()

pageserver/src/page_cache.rs

@@ -85,7 +85,7 @@ use utils::{
     lsn::Lsn,
 };
 
-use crate::{context::RequestContext, metrics::PageCacheSizeMetrics, repository::Key};
+use crate::{metrics::PageCacheSizeMetrics, repository::Key};
 
 static PAGE_CACHE: OnceCell<PageCache> = OnceCell::new();
 const TEST_PAGE_CACHE_SIZE: usize = 50;
@@ -346,10 +346,8 @@ impl PageCache {
         timeline_id: TimelineId,
         key: &Key,
         lsn: Lsn,
-        ctx: &RequestContext,
     ) -> Option<(Lsn, PageReadGuard)> {
         crate::metrics::PAGE_CACHE
-            .for_ctx(ctx)
             .read_accesses_materialized_page
             .inc();
 
@@ -370,12 +368,10 @@ impl PageCache {
             {
                 if available_lsn == lsn {
                     crate::metrics::PAGE_CACHE
-                        .for_ctx(ctx)
                         .read_hits_materialized_page_exact
                         .inc();
                 } else {
                     crate::metrics::PAGE_CACHE
-                        .for_ctx(ctx)
                         .read_hits_materialized_page_older_lsn
                         .inc();
                 }
@@ -430,11 +426,10 @@ impl PageCache {
         &self,
         file_id: FileId,
         blkno: u32,
-        ctx: &RequestContext,
     ) -> anyhow::Result<ReadBufResult> {
         let mut cache_key = CacheKey::ImmutableFilePage { file_id, blkno };
 
-        self.lock_for_read(&mut cache_key, ctx).await
+        self.lock_for_read(&mut cache_key).await
     }
 
     //
@@ -502,20 +497,14 @@ impl PageCache {
     /// }
     /// ```
     ///
-    async fn lock_for_read(
-        &self,
-        cache_key: &mut CacheKey,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<ReadBufResult> {
+    async fn lock_for_read(&self, cache_key: &mut CacheKey) -> anyhow::Result<ReadBufResult> {
         let (read_access, hit) = match cache_key {
             CacheKey::MaterializedPage { .. } => {
                 unreachable!("Materialized pages use lookup_materialized_page")
             }
             CacheKey::ImmutableFilePage { .. } => (
-                &crate::metrics::PAGE_CACHE
-                    .for_ctx(ctx)
-                    .read_accesses_immutable,
-                &crate::metrics::PAGE_CACHE.for_ctx(ctx).read_hits_immutable,
+                &crate::metrics::PAGE_CACHE.read_accesses_immutable,
+                &crate::metrics::PAGE_CACHE.read_hits_immutable,
             ),
         };
         read_access.inc();

pageserver/src/pgdatadir_mapping.rs

@@ -1138,7 +1138,7 @@ impl<'a> DatadirModification<'a> {
     /// retains all the metadata, but data pages are flushed. That's again OK
     /// for bulk import, where you are just loading data pages and won't try to
     /// modify the same pages twice.
-    pub async fn flush(&mut self, ctx: &RequestContext) -> anyhow::Result<()> {
+    pub async fn flush(&mut self) -> anyhow::Result<()> {
         // Unless we have accumulated a decent amount of changes, it's not worth it
         // to scan through the pending_updates list.
         let pending_nblocks = self.pending_nblocks;
@@ -1154,7 +1154,7 @@ impl<'a> DatadirModification<'a> {
             if is_rel_block_key(key) || is_slru_block_key(key) {
                 // This bails out on first error without modifying pending_updates.
                 // That's Ok, cf this function's doc comment.
-                writer.put(key, self.lsn, &value, ctx).await?;
+                writer.put(key, self.lsn, &value).await?;
             } else {
                 retained_pending_updates.insert(key, value);
             }
@@ -1174,14 +1174,14 @@ impl<'a> DatadirModification<'a> {
     /// underlying timeline.
     /// All the modifications in this atomic update are stamped by the specified LSN.
     ///
-    pub async fn commit(&mut self, ctx: &RequestContext) -> anyhow::Result<()> {
+    pub async fn commit(&mut self) -> anyhow::Result<()> {
         let writer = self.tline.writer().await;
         let lsn = self.lsn;
         let pending_nblocks = self.pending_nblocks;
         self.pending_nblocks = 0;
 
         for (key, value) in self.pending_updates.drain() {
-            writer.put(key, lsn, &value, ctx).await?;
+            writer.put(key, lsn, &value).await?;
         }
         for key_range in self.pending_deletions.drain(..) {
             writer.delete(key_range, lsn).await?;

pageserver/src/repository.rs

@@ -37,7 +37,7 @@ impl Key {
             | self.field6 as i128
     }
 
-    pub const fn from_i128(x: i128) -> Self {
+    pub fn from_i128(x: i128) -> Self {
         Key {
             field1: ((x >> 120) & 0xf) as u8,
             field2: ((x >> 104) & 0xFFFF) as u32,

pageserver/src/task_mgr.rs

@@ -187,7 +187,6 @@ task_local! {
     Debug,
     // NB: enumset::EnumSetType derives PartialEq, Eq, Clone, Copy
     enumset::EnumSetType,
-    enum_map::Enum,
     serde::Serialize,
     serde::Deserialize,
     strum_macros::IntoStaticStr,
@@ -456,7 +455,7 @@ async fn task_finish(
     }
 
     if shutdown_process {
-        shutdown_pageserver(None, 1).await;
+        shutdown_pageserver(1).await;
     }
 }
 

pageserver/src/tenant.rs

@@ -57,7 +57,6 @@ use self::timeline::EvictionTaskTenantState;
 use self::timeline::TimelineResources;
 use crate::config::PageServerConf;
 use crate::context::{DownloadBehavior, RequestContext};
-use crate::deletion_queue::DeletionQueueClient;
 use crate::import_datadir;
 use crate::is_uninit_mark;
 use crate::metrics::TENANT_ACTIVATION;
@@ -118,7 +117,7 @@ mod span;
 
 pub mod metadata;
 mod par_fsync;
-pub mod remote_timeline_client;
+mod remote_timeline_client;
 pub mod storage_layer;
 
 pub mod config;
@@ -158,7 +157,6 @@ pub const TENANT_DELETED_MARKER_FILE_NAME: &str = "deleted";
 pub struct TenantSharedResources {
     pub broker_client: storage_broker::BrokerClientChannel,
     pub remote_storage: Option<GenericRemoteStorage>,
-    pub deletion_queue_client: DeletionQueueClient,
 }
 
 ///
@@ -199,9 +197,6 @@ pub struct Tenant {
     // provides access to timeline data sitting in the remote storage
     pub(crate) remote_storage: Option<GenericRemoteStorage>,
 
-    // Access to global deletion queue for when this tenant wants to schedule a deletion
-    deletion_queue_client: DeletionQueueClient,
-
     /// Cached logical sizes updated updated on each [`Tenant::gather_size_inputs`].
     cached_logical_sizes: tokio::sync::Mutex<HashMap<(TimelineId, Lsn), u64>>,
     cached_synthetic_tenant_size: Arc<AtomicU64>,
@@ -528,20 +523,15 @@ impl Tenant {
         conf: &'static PageServerConf,
         tenant_id: TenantId,
         generation: Generation,
-        resources: TenantSharedResources,
+        broker_client: storage_broker::BrokerClientChannel,
         tenants: &'static tokio::sync::RwLock<TenantsMap>,
+        remote_storage: GenericRemoteStorage,
         ctx: &RequestContext,
     ) -> anyhow::Result<Arc<Tenant>> {
         // TODO dedup with spawn_load
         let tenant_conf =
             Self::load_tenant_config(conf, &tenant_id).context("load tenant config")?;
 
-        let TenantSharedResources {
-            broker_client,
-            remote_storage,
-            deletion_queue_client,
-        } = resources;
-
         let wal_redo_manager = Arc::new(PostgresRedoManager::new(conf, tenant_id));
         let tenant = Arc::new(Tenant::new(
             TenantState::Attaching,
@@ -550,8 +540,7 @@ impl Tenant {
             wal_redo_manager,
             tenant_id,
             generation,
-            remote_storage.clone(),
-            deletion_queue_client,
+            Some(remote_storage.clone()),
         ));
 
         // Do all the hard work in the background
@@ -582,7 +571,7 @@ impl Tenant {
                 let pending_deletion = {
                     match DeleteTenantFlow::should_resume_deletion(
                         conf,
-                        remote_storage.as_ref(),
+                        Some(&remote_storage),
                         &tenant_clone,
                     )
                     .await
@@ -671,7 +660,6 @@ impl Tenant {
         for timeline_id in remote_timeline_ids {
             let client = RemoteTimelineClient::new(
                 remote_storage.clone(),
-                self.deletion_queue_client.clone(),
                 self.conf,
                 self.tenant_id,
                 timeline_id,
@@ -738,7 +726,6 @@ impl Tenant {
                 remote_metadata,
                 TimelineResources {
                     remote_client: Some(remote_client),
-                    deletion_queue_client: self.deletion_queue_client.clone(),
                 },
                 ctx,
             )
@@ -763,7 +750,6 @@ impl Tenant {
                 timeline_id,
                 &index_part.metadata,
                 Some(remote_timeline_client),
-                self.deletion_queue_client.clone(),
                 None,
             )
             .await
@@ -865,7 +851,6 @@ impl Tenant {
             tenant_id,
             Generation::broken(),
             None,
-            DeletionQueueClient::broken(),
         ))
     }
 
@@ -910,7 +895,6 @@ impl Tenant {
             tenant_id,
             generation,
             remote_storage.clone(),
-            resources.deletion_queue_client.clone(),
         );
         let tenant = Arc::new(tenant);
 
@@ -1318,7 +1302,6 @@ impl Tenant {
                                 timeline_id,
                                 &local_metadata,
                                 Some(remote_client),
-                                self.deletion_queue_client.clone(),
                                 init_order,
                             )
                             .await
@@ -1368,7 +1351,6 @@ impl Tenant {
                         timeline_id,
                         &local_metadata,
                         None,
-                        self.deletion_queue_client.clone(),
                         init_order,
                     )
                     .await
@@ -1522,7 +1504,7 @@ impl Tenant {
             .init_empty_test_timeline()
             .context("init_empty_test_timeline")?;
         modification
-            .commit(ctx)
+            .commit()
             .await
             .context("commit init_empty_test_timeline modification")?;
 
@@ -2260,9 +2242,6 @@ impl Tenant {
         Ok(timeline)
     }
 
-    // Allow too_many_arguments because a constructor's argument list naturally grows with the
-    // number of attributes in the struct: breaking these out into a builder wouldn't be helpful.
-    #[allow(clippy::too_many_arguments)]
     fn new(
         state: TenantState,
         conf: &'static PageServerConf,
@@ -2271,7 +2250,6 @@ impl Tenant {
         tenant_id: TenantId,
         generation: Generation,
         remote_storage: Option<GenericRemoteStorage>,
-        deletion_queue_client: DeletionQueueClient,
     ) -> Tenant {
         let (state, mut rx) = watch::channel(state);
 
@@ -2339,7 +2317,6 @@ impl Tenant {
             gc_cs: tokio::sync::Mutex::new(()),
             walredo_mgr,
             remote_storage,
-            deletion_queue_client,
             state,
             cached_logical_sizes: tokio::sync::Mutex::new(HashMap::new()),
             cached_synthetic_tenant_size: Arc::new(AtomicU64::new(0)),
@@ -2879,7 +2856,6 @@ impl Tenant {
         let remote_client = if let Some(remote_storage) = self.remote_storage.as_ref() {
             let remote_client = RemoteTimelineClient::new(
                 remote_storage.clone(),
-                self.deletion_queue_client.clone(),
                 self.conf,
                 self.tenant_id,
                 timeline_id,
@@ -2890,10 +2866,7 @@ impl Tenant {
             None
         };
 
-        TimelineResources {
-            remote_client,
-            deletion_queue_client: self.deletion_queue_client.clone(),
-        }
+        TimelineResources { remote_client }
     }
 
     /// Creates intermediate timeline structure and its files.
@@ -3349,7 +3322,6 @@ pub mod harness {
     use utils::logging;
     use utils::lsn::Lsn;
 
-    use crate::deletion_queue::mock::MockDeletionQueue;
     use crate::{
         config::PageServerConf,
         repository::Key,
@@ -3411,7 +3383,6 @@ pub mod harness {
         pub generation: Generation,
         pub remote_storage: GenericRemoteStorage,
         pub remote_fs_dir: PathBuf,
-        pub deletion_queue: MockDeletionQueue,
     }
 
     static LOG_HANDLE: OnceCell<()> = OnceCell::new();
@@ -3460,7 +3431,6 @@ pub mod harness {
                 storage: RemoteStorageKind::LocalFs(remote_fs_dir.clone()),
             };
             let remote_storage = GenericRemoteStorage::from_config(&config).unwrap();
-            let deletion_queue = MockDeletionQueue::new(Some(remote_storage.clone()));
 
             Ok(Self {
                 conf,
@@ -3469,7 +3439,6 @@ pub mod harness {
                 generation: Generation::new(0xdeadbeef),
                 remote_storage,
                 remote_fs_dir,
-                deletion_queue,
             })
         }
 
@@ -3494,7 +3463,6 @@ pub mod harness {
                 self.tenant_id,
                 self.generation,
                 Some(self.remote_storage.clone()),
-                self.deletion_queue.new_client(),
             ));
             tenant
                 .load(None, ctx)
@@ -3570,24 +3538,14 @@ mod tests {
 
         let writer = tline.writer().await;
         writer
-            .put(
-                *TEST_KEY,
-                Lsn(0x10),
-                &Value::Image(TEST_IMG("foo at 0x10")),
-                &ctx,
-            )
+            .put(*TEST_KEY, Lsn(0x10), &Value::Image(TEST_IMG("foo at 0x10")))
             .await?;
         writer.finish_write(Lsn(0x10));
         drop(writer);
 
         let writer = tline.writer().await;
         writer
-            .put(
-                *TEST_KEY,
-                Lsn(0x20),
-                &Value::Image(TEST_IMG("foo at 0x20")),
-                &ctx,
-            )
+            .put(*TEST_KEY, Lsn(0x20), &Value::Image(TEST_IMG("foo at 0x20")))
             .await?;
         writer.finish_write(Lsn(0x20));
         drop(writer);
@@ -3661,19 +3619,19 @@ mod tests {
 
         // Insert a value on the timeline
         writer
-            .put(TEST_KEY_A, Lsn(0x20), &test_value("foo at 0x20"), &ctx)
+            .put(TEST_KEY_A, Lsn(0x20), &test_value("foo at 0x20"))
             .await?;
         writer
-            .put(TEST_KEY_B, Lsn(0x20), &test_value("foobar at 0x20"), &ctx)
+            .put(TEST_KEY_B, Lsn(0x20), &test_value("foobar at 0x20"))
             .await?;
         writer.finish_write(Lsn(0x20));
 
         writer
-            .put(TEST_KEY_A, Lsn(0x30), &test_value("foo at 0x30"), &ctx)
+            .put(TEST_KEY_A, Lsn(0x30), &test_value("foo at 0x30"))
             .await?;
         writer.finish_write(Lsn(0x30));
         writer
-            .put(TEST_KEY_A, Lsn(0x40), &test_value("foo at 0x40"), &ctx)
+            .put(TEST_KEY_A, Lsn(0x40), &test_value("foo at 0x40"))
             .await?;
         writer.finish_write(Lsn(0x40));
 
@@ -3688,7 +3646,7 @@ mod tests {
             .expect("Should have a local timeline");
         let new_writer = newtline.writer().await;
         new_writer
-            .put(TEST_KEY_A, Lsn(0x40), &test_value("bar at 0x40"), &ctx)
+            .put(TEST_KEY_A, Lsn(0x40), &test_value("bar at 0x40"))
             .await?;
         new_writer.finish_write(Lsn(0x40));
 
@@ -3711,11 +3669,7 @@ mod tests {
         Ok(())
     }
 
-    async fn make_some_layers(
-        tline: &Timeline,
-        start_lsn: Lsn,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<()> {
+    async fn make_some_layers(tline: &Timeline, start_lsn: Lsn) -> anyhow::Result<()> {
         let mut lsn = start_lsn;
         #[allow(non_snake_case)]
         {
@@ -3726,7 +3680,6 @@ mod tests {
                     *TEST_KEY,
                     lsn,
                     &Value::Image(TEST_IMG(&format!("foo at {}", lsn))),
-                    ctx,
                 )
                 .await?;
             writer.finish_write(lsn);
@@ -3736,7 +3689,6 @@ mod tests {
                     *TEST_KEY,
                     lsn,
                     &Value::Image(TEST_IMG(&format!("foo at {}", lsn))),
-                    ctx,
                 )
                 .await?;
             writer.finish_write(lsn);
@@ -3750,7 +3702,6 @@ mod tests {
                     *TEST_KEY,
                     lsn,
                     &Value::Image(TEST_IMG(&format!("foo at {}", lsn))),
-                    ctx,
                 )
                 .await?;
             writer.finish_write(lsn);
@@ -3760,7 +3711,6 @@ mod tests {
                     *TEST_KEY,
                     lsn,
                     &Value::Image(TEST_IMG(&format!("foo at {}", lsn))),
-                    ctx,
                 )
                 .await?;
             writer.finish_write(lsn);
@@ -3777,7 +3727,7 @@ mod tests {
         let tline = tenant
             .create_test_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
             .await?;
-        make_some_layers(tline.as_ref(), Lsn(0x20), &ctx).await?;
+        make_some_layers(tline.as_ref(), Lsn(0x20)).await?;
 
         // this removes layers before lsn 40 (50 minus 10), so there are two remaining layers, image and delta for 31-50
         // FIXME: this doesn't actually remove any layer currently, given how the flushing
@@ -3851,7 +3801,7 @@ mod tests {
             .load();
 
         let tline = repo.create_empty_timeline(TIMELINE_ID, Lsn(0), DEFAULT_PG_VERSION)?;
-        make_some_layers(tline.as_ref(), Lsn(0x20), &ctx).await?;
+        make_some_layers(tline.as_ref(), Lsn(0x20)).await?;
 
         repo.gc_iteration(Some(TIMELINE_ID), 0x10, Duration::ZERO)?;
         let latest_gc_cutoff_lsn = tline.get_latest_gc_cutoff_lsn();
@@ -3873,7 +3823,7 @@ mod tests {
         let tline = tenant
             .create_test_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
             .await?;
-        make_some_layers(tline.as_ref(), Lsn(0x20), &ctx).await?;
+        make_some_layers(tline.as_ref(), Lsn(0x20)).await?;
 
         tenant
             .branch_timeline_test(&tline, NEW_TIMELINE_ID, Some(Lsn(0x40)), &ctx)
@@ -3882,7 +3832,7 @@ mod tests {
             .get_timeline(NEW_TIMELINE_ID, true)
             .expect("Should have a local timeline");
 
-        make_some_layers(newtline.as_ref(), Lsn(0x60), &ctx).await?;
+        make_some_layers(newtline.as_ref(), Lsn(0x60)).await?;
 
         tline.set_broken("test".to_owned());
 
@@ -3923,7 +3873,7 @@ mod tests {
         let tline = tenant
             .create_test_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
             .await?;
-        make_some_layers(tline.as_ref(), Lsn(0x20), &ctx).await?;
+        make_some_layers(tline.as_ref(), Lsn(0x20)).await?;
 
         tenant
             .branch_timeline_test(&tline, NEW_TIMELINE_ID, Some(Lsn(0x40)), &ctx)
@@ -3948,7 +3898,7 @@ mod tests {
         let tline = tenant
             .create_test_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
             .await?;
-        make_some_layers(tline.as_ref(), Lsn(0x20), &ctx).await?;
+        make_some_layers(tline.as_ref(), Lsn(0x20)).await?;
 
         tenant
             .branch_timeline_test(&tline, NEW_TIMELINE_ID, Some(Lsn(0x40)), &ctx)
@@ -3957,7 +3907,7 @@ mod tests {
             .get_timeline(NEW_TIMELINE_ID, true)
             .expect("Should have a local timeline");
 
-        make_some_layers(newtline.as_ref(), Lsn(0x60), &ctx).await?;
+        make_some_layers(newtline.as_ref(), Lsn(0x60)).await?;
 
         // run gc on parent
         tenant
@@ -3982,7 +3932,7 @@ mod tests {
             let tline = tenant
                 .create_test_timeline(TIMELINE_ID, Lsn(0x7000), DEFAULT_PG_VERSION, &ctx)
                 .await?;
-            make_some_layers(tline.as_ref(), Lsn(0x8000), &ctx).await?;
+            make_some_layers(tline.as_ref(), Lsn(0x8000)).await?;
             // so that all uploads finish & we can call harness.load() below again
             tenant
                 .shutdown(Default::default(), true)
@@ -4011,7 +3961,7 @@ mod tests {
                 .create_test_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
                 .await?;
 
-            make_some_layers(tline.as_ref(), Lsn(0x20), &ctx).await?;
+            make_some_layers(tline.as_ref(), Lsn(0x20)).await?;
 
             let child_tline = tenant
                 .branch_timeline_test(&tline, NEW_TIMELINE_ID, Some(Lsn(0x40)), &ctx)
@@ -4022,7 +3972,7 @@ mod tests {
                 .get_timeline(NEW_TIMELINE_ID, true)
                 .expect("Should have a local timeline");
 
-            make_some_layers(newtline.as_ref(), Lsn(0x60), &ctx).await?;
+            make_some_layers(newtline.as_ref(), Lsn(0x60)).await?;
 
             // so that all uploads finish & we can call harness.load() below again
             tenant
@@ -4054,7 +4004,7 @@ mod tests {
         let tline = tenant
             .create_test_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
             .await?;
-        make_some_layers(tline.as_ref(), Lsn(0x20), &ctx).await?;
+        make_some_layers(tline.as_ref(), Lsn(0x20)).await?;
 
         let layer_map = tline.layers.read().await;
         let level0_deltas = layer_map.layer_map().get_level0_deltas()?;
@@ -4137,12 +4087,7 @@ mod tests {
 
         let writer = tline.writer().await;
         writer
-            .put(
-                *TEST_KEY,
-                Lsn(0x10),
-                &Value::Image(TEST_IMG("foo at 0x10")),
-                &ctx,
-            )
+            .put(*TEST_KEY, Lsn(0x10), &Value::Image(TEST_IMG("foo at 0x10")))
             .await?;
         writer.finish_write(Lsn(0x10));
         drop(writer);
@@ -4152,12 +4097,7 @@ mod tests {
 
         let writer = tline.writer().await;
         writer
-            .put(
-                *TEST_KEY,
-                Lsn(0x20),
-                &Value::Image(TEST_IMG("foo at 0x20")),
-                &ctx,
-            )
+            .put(*TEST_KEY, Lsn(0x20), &Value::Image(TEST_IMG("foo at 0x20")))
             .await?;
         writer.finish_write(Lsn(0x20));
         drop(writer);
@@ -4167,12 +4107,7 @@ mod tests {
 
         let writer = tline.writer().await;
         writer
-            .put(
-                *TEST_KEY,
-                Lsn(0x30),
-                &Value::Image(TEST_IMG("foo at 0x30")),
-                &ctx,
-            )
+            .put(*TEST_KEY, Lsn(0x30), &Value::Image(TEST_IMG("foo at 0x30")))
             .await?;
         writer.finish_write(Lsn(0x30));
         drop(writer);
@@ -4182,12 +4117,7 @@ mod tests {
 
         let writer = tline.writer().await;
         writer
-            .put(
-                *TEST_KEY,
-                Lsn(0x40),
-                &Value::Image(TEST_IMG("foo at 0x40")),
-                &ctx,
-            )
+            .put(*TEST_KEY, Lsn(0x40), &Value::Image(TEST_IMG("foo at 0x40")))
             .await?;
         writer.finish_write(Lsn(0x40));
         drop(writer);
@@ -4225,8 +4155,7 @@ mod tests {
     //
     #[tokio::test]
     async fn test_bulk_insert() -> anyhow::Result<()> {
-        let harness = TenantHarness::create("test_bulk_insert")?;
-        let (tenant, ctx) = harness.load().await;
+        let (tenant, ctx) = TenantHarness::create("test_bulk_insert")?.load().await;
         let tline = tenant
             .create_test_timeline(TIMELINE_ID, Lsn(0x08), DEFAULT_PG_VERSION, &ctx)
             .await?;
@@ -4246,7 +4175,6 @@ mod tests {
                         test_key,
                         lsn,
                         &Value::Image(TEST_IMG(&format!("{} at {}", blknum, lsn))),
-                        &ctx,
                     )
                     .await?;
                 writer.finish_write(lsn);
@@ -4273,8 +4201,7 @@ mod tests {
 
     #[tokio::test]
     async fn test_random_updates() -> anyhow::Result<()> {
-        let harness = TenantHarness::create("test_random_updates")?;
-        let (tenant, ctx) = harness.load().await;
+        let (tenant, ctx) = TenantHarness::create("test_random_updates")?.load().await;
         let tline = tenant
             .create_test_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
             .await?;
@@ -4300,7 +4227,6 @@ mod tests {
                     test_key,
                     lsn,
                     &Value::Image(TEST_IMG(&format!("{} at {}", blknum, lsn))),
-                    &ctx,
                 )
                 .await?;
             writer.finish_write(lsn);
@@ -4321,7 +4247,6 @@ mod tests {
                         test_key,
                         lsn,
                         &Value::Image(TEST_IMG(&format!("{} at {}", blknum, lsn))),
-                        &ctx,
                     )
                     .await?;
                 writer.finish_write(lsn);
@@ -4381,7 +4306,6 @@ mod tests {
                     test_key,
                     lsn,
                     &Value::Image(TEST_IMG(&format!("{} at {}", blknum, lsn))),
-                    &ctx,
                 )
                 .await?;
             writer.finish_write(lsn);
@@ -4410,7 +4334,6 @@ mod tests {
                         test_key,
                         lsn,
                         &Value::Image(TEST_IMG(&format!("{} at {}", blknum, lsn))),
-                        &ctx,
                     )
                     .await?;
                 println!("updating {} at {}", blknum, lsn);
@@ -4479,7 +4402,6 @@ mod tests {
                         test_key,
                         lsn,
                         &Value::Image(TEST_IMG(&format!("{} {} at {}", idx, blknum, lsn))),
-                        &ctx,
                     )
                     .await?;
                 println!("updating [{}][{}] at {}", idx, blknum, lsn);
@@ -4552,7 +4474,7 @@ mod tests {
             .init_empty_test_timeline()
             .context("init_empty_test_timeline")?;
         modification
-            .commit(&ctx)
+            .commit()
             .await
             .context("commit init_empty_test_timeline modification")?;
 

pageserver/src/tenant/blob_io.rs

@@ -11,7 +11,6 @@
 //! len <  128: 0XXXXXXX
 //! len >= 128: 1XXXXXXX XXXXXXXX XXXXXXXX XXXXXXXX
 //!
-use crate::context::RequestContext;
 use crate::page_cache::PAGE_SZ;
 use crate::tenant::block_io::BlockCursor;
 use crate::virtual_file::VirtualFile;
@@ -20,13 +19,9 @@ use std::io::{Error, ErrorKind};
 
 impl<'a> BlockCursor<'a> {
     /// Read a blob into a new buffer.
-    pub async fn read_blob(
-        &self,
-        offset: u64,
-        ctx: &RequestContext,
-    ) -> Result<Vec<u8>, std::io::Error> {
+    pub async fn read_blob(&self, offset: u64) -> Result<Vec<u8>, std::io::Error> {
         let mut buf = Vec::new();
-        self.read_blob_into_buf(offset, &mut buf, ctx).await?;
+        self.read_blob_into_buf(offset, &mut buf).await?;
         Ok(buf)
     }
     /// Read blob into the given buffer. Any previous contents in the buffer
@@ -35,12 +30,11 @@ impl<'a> BlockCursor<'a> {
         &self,
         offset: u64,
         dstbuf: &mut Vec<u8>,
-        ctx: &RequestContext,
     ) -> Result<(), std::io::Error> {
         let mut blknum = (offset / PAGE_SZ as u64) as u32;
         let mut off = (offset % PAGE_SZ as u64) as usize;
 
-        let mut buf = self.read_blk(blknum, ctx).await?;
+        let mut buf = self.read_blk(blknum).await?;
 
         // peek at the first byte, to determine if it's a 1- or 4-byte length
         let first_len_byte = buf[off];
@@ -56,7 +50,7 @@ impl<'a> BlockCursor<'a> {
                 // it is split across two pages
                 len_buf[..thislen].copy_from_slice(&buf[off..PAGE_SZ]);
                 blknum += 1;
-                buf = self.read_blk(blknum, ctx).await?;
+                buf = self.read_blk(blknum).await?;
                 len_buf[thislen..].copy_from_slice(&buf[0..4 - thislen]);
                 off = 4 - thislen;
             } else {
@@ -77,7 +71,7 @@ impl<'a> BlockCursor<'a> {
             if page_remain == 0 {
                 // continue on next page
                 blknum += 1;
-                buf = self.read_blk(blknum, ctx).await?;
+                buf = self.read_blk(blknum).await?;
                 off = 0;
                 page_remain = PAGE_SZ;
             }
@@ -234,13 +228,12 @@ impl BlobWriter<false> {
 #[cfg(test)]
 mod tests {
     use super::*;
-    use crate::{context::DownloadBehavior, task_mgr::TaskKind, tenant::block_io::BlockReaderRef};
+    use crate::tenant::block_io::BlockReaderRef;
     use rand::{Rng, SeedableRng};
 
     async fn round_trip_test<const BUFFERED: bool>(blobs: &[Vec<u8>]) -> Result<(), Error> {
         let temp_dir = tempfile::tempdir()?;
         let path = temp_dir.path().join("file");
-        let ctx = RequestContext::new(TaskKind::UnitTest, DownloadBehavior::Error);
 
         // Write part (in block to drop the file)
         let mut offsets = Vec::new();
@@ -262,7 +255,7 @@ mod tests {
         let rdr = BlockReaderRef::VirtualFile(&file);
         let rdr = BlockCursor::new(rdr);
         for (idx, (blob, offset)) in blobs.iter().zip(offsets.iter()).enumerate() {
-            let blob_read = rdr.read_blob(*offset, &ctx).await?;
+            let blob_read = rdr.read_blob(*offset).await?;
             assert_eq!(
                 blob, &blob_read,
                 "mismatch for idx={idx} at offset={offset}"

pageserver/src/tenant/block_io.rs

@@ -4,7 +4,6 @@
 
 use super::ephemeral_file::EphemeralFile;
 use super::storage_layer::delta_layer::{Adapter, DeltaLayerInner};
-use crate::context::RequestContext;
 use crate::page_cache::{self, PageReadGuard, ReadBufResult, PAGE_SZ};
 use crate::virtual_file::VirtualFile;
 use bytes::Bytes;
@@ -83,16 +82,12 @@ pub(crate) enum BlockReaderRef<'a> {
 
 impl<'a> BlockReaderRef<'a> {
     #[inline(always)]
-    async fn read_blk(
-        &self,
-        blknum: u32,
-        ctx: &RequestContext,
-    ) -> Result<BlockLease, std::io::Error> {
+    async fn read_blk(&self, blknum: u32) -> Result<BlockLease, std::io::Error> {
         use BlockReaderRef::*;
         match self {
-            FileBlockReader(r) => r.read_blk(blknum, ctx).await,
-            EphemeralFile(r) => r.read_blk(blknum, ctx).await,
-            Adapter(r) => r.read_blk(blknum, ctx).await,
+            FileBlockReader(r) => r.read_blk(blknum).await,
+            EphemeralFile(r) => r.read_blk(blknum).await,
+            Adapter(r) => r.read_blk(blknum).await,
             #[cfg(test)]
             TestDisk(r) => r.read_blk(blknum),
             #[cfg(test)]
@@ -110,13 +105,11 @@ impl<'a> BlockReaderRef<'a> {
 ///
 /// ```no_run
 /// # use pageserver::tenant::block_io::{BlockReader, FileBlockReader};
-/// # use pageserver::context::RequestContext;
 /// # let reader: FileBlockReader = unimplemented!("stub");
-/// # let ctx: RequestContext = unimplemented!("stub");
 /// let cursor = reader.block_cursor();
-/// let buf = cursor.read_blk(1, &ctx);
+/// let buf = cursor.read_blk(1);
 /// // do stuff with 'buf'
-/// let buf = cursor.read_blk(2, &ctx);
+/// let buf = cursor.read_blk(2);
 /// // do stuff with 'buf'
 /// ```
 ///
@@ -141,12 +134,8 @@ impl<'a> BlockCursor<'a> {
     /// access to the contents of the page. (For the page cache, the
     /// lease object represents a lock on the buffer.)
     #[inline(always)]
-    pub async fn read_blk(
-        &self,
-        blknum: u32,
-        ctx: &RequestContext,
-    ) -> Result<BlockLease, std::io::Error> {
-        self.reader.read_blk(blknum, ctx).await
+    pub async fn read_blk(&self, blknum: u32) -> Result<BlockLease, std::io::Error> {
+        self.reader.read_blk(blknum).await
     }
 }
 
@@ -180,15 +169,11 @@ impl FileBlockReader {
     /// Returns a "lease" object that can be used to
     /// access to the contents of the page. (For the page cache, the
     /// lease object represents a lock on the buffer.)
-    pub async fn read_blk(
-        &self,
-        blknum: u32,
-        ctx: &RequestContext,
-    ) -> Result<BlockLease, std::io::Error> {
+    pub async fn read_blk(&self, blknum: u32) -> Result<BlockLease, std::io::Error> {
         let cache = page_cache::get();
         loop {
             match cache
-                .read_immutable_buf(self.file_id, blknum, ctx)
+                .read_immutable_buf(self.file_id, blknum)
                 .await
                 .map_err(|e| {
                     std::io::Error::new(

pageserver/src/tenant/disk_btree.rs

@@ -26,11 +26,7 @@ use std::{cmp::Ordering, io, result};
 use thiserror::Error;
 use tracing::error;
 
-use crate::{
-    context::{DownloadBehavior, RequestContext},
-    task_mgr::TaskKind,
-    tenant::block_io::{BlockReader, BlockWriter},
-};
+use crate::tenant::block_io::{BlockReader, BlockWriter};
 
 // The maximum size of a value stored in the B-tree. 5 bytes is enough currently.
 pub const VALUE_SZ: usize = 5;
@@ -235,19 +231,14 @@ where
     ///
     /// Read the value for given key. Returns the value, or None if it doesn't exist.
     ///
-    pub async fn get(&self, search_key: &[u8; L], ctx: &RequestContext) -> Result<Option<u64>> {
+    pub async fn get(&self, search_key: &[u8; L]) -> Result<Option<u64>> {
         let mut result: Option<u64> = None;
-        self.visit(
-            search_key,
-            VisitDirection::Forwards,
-            |key, value| {
-                if key == search_key {
-                    result = Some(value);
-                }
-                false
-            },
-            ctx,
-        )
+        self.visit(search_key, VisitDirection::Forwards, |key, value| {
+            if key == search_key {
+                result = Some(value);
+            }
+            false
+        })
         .await?;
         Ok(result)
     }
@@ -262,7 +253,6 @@ where
         search_key: &[u8; L],
         dir: VisitDirection,
         mut visitor: V,
-        ctx: &RequestContext,
     ) -> Result<bool>
     where
         V: FnMut(&[u8], u64) -> bool,
@@ -272,9 +262,7 @@ where
         let block_cursor = self.reader.block_cursor();
         while let Some((node_blknum, opt_iter)) = stack.pop() {
             // Locate the node.
-            let node_buf = block_cursor
-                .read_blk(self.start_blk + node_blknum, ctx)
-                .await?;
+            let node_buf = block_cursor.read_blk(self.start_blk + node_blknum).await?;
 
             let node = OnDiskNode::deparse(node_buf.as_ref())?;
             let prefix_len = node.prefix_len as usize;
@@ -363,14 +351,13 @@ where
     #[allow(dead_code)]
     pub async fn dump(&self) -> Result<()> {
         let mut stack = Vec::new();
-        let ctx = RequestContext::new(TaskKind::DebugTool, DownloadBehavior::Error);
 
         stack.push((self.root_blk, String::new(), 0, 0, 0));
 
         let block_cursor = self.reader.block_cursor();
 
         while let Some((blknum, path, depth, child_idx, key_off)) = stack.pop() {
-            let blk = block_cursor.read_blk(self.start_blk + blknum, &ctx).await?;
+            let blk = block_cursor.read_blk(self.start_blk + blknum).await?;
             let buf: &[u8] = blk.as_ref();
             let node = OnDiskNode::<L>::deparse(buf)?;
 
@@ -701,8 +688,6 @@ impl<const L: usize> BuildNode<L> {
 #[cfg(test)]
 pub(crate) mod tests {
     use super::*;
-    use crate::context::DownloadBehavior;
-    use crate::task_mgr::TaskKind;
     use crate::tenant::block_io::{BlockCursor, BlockLease, BlockReaderRef};
     use rand::Rng;
     use std::collections::BTreeMap;
@@ -740,8 +725,6 @@ pub(crate) mod tests {
         let mut disk = TestDisk::new();
         let mut writer = DiskBtreeBuilder::<_, 6>::new(&mut disk);
 
-        let ctx = RequestContext::new(TaskKind::UnitTest, DownloadBehavior::Error);
-
         let all_keys: Vec<&[u8; 6]> = vec![
             b"xaaaaa", b"xaaaba", b"xaaaca", b"xabaaa", b"xababa", b"xabaca", b"xabada", b"xabadb",
         ];
@@ -762,12 +745,12 @@ pub(crate) mod tests {
 
         // Test the `get` function on all the keys.
         for (key, val) in all_data.iter() {
-            assert_eq!(reader.get(key, &ctx).await?, Some(*val));
+            assert_eq!(reader.get(key).await?, Some(*val));
         }
         // And on some keys that don't exist
-        assert_eq!(reader.get(b"aaaaaa", &ctx).await?, None);
-        assert_eq!(reader.get(b"zzzzzz", &ctx).await?, None);
-        assert_eq!(reader.get(b"xaaabx", &ctx).await?, None);
+        assert_eq!(reader.get(b"aaaaaa").await?, None);
+        assert_eq!(reader.get(b"zzzzzz").await?, None);
+        assert_eq!(reader.get(b"xaaabx").await?, None);
 
         // Test search with `visit` function
         let search_key = b"xabaaa";
@@ -779,15 +762,10 @@ pub(crate) mod tests {
 
         let mut data = Vec::new();
         reader
-            .visit(
-                search_key,
-                VisitDirection::Forwards,
-                |key, value| {
-                    data.push((key.to_vec(), value));
-                    true
-                },
-                &ctx,
-            )
+            .visit(search_key, VisitDirection::Forwards, |key, value| {
+                data.push((key.to_vec(), value));
+                true
+            })
             .await?;
         assert_eq!(data, expected);
 
@@ -800,28 +778,18 @@ pub(crate) mod tests {
         expected.reverse();
         let mut data = Vec::new();
         reader
-            .visit(
-                search_key,
-                VisitDirection::Backwards,
-                |key, value| {
-                    data.push((key.to_vec(), value));
-                    true
-                },
-                &ctx,
-            )
+            .visit(search_key, VisitDirection::Backwards, |key, value| {
+                data.push((key.to_vec(), value));
+                true
+            })
             .await?;
         assert_eq!(data, expected);
 
         // Backward scan where nothing matches
         reader
-            .visit(
-                b"aaaaaa",
-                VisitDirection::Backwards,
-                |key, value| {
-                    panic!("found unexpected key {}: {}", hex::encode(key), value);
-                },
-                &ctx,
-            )
+            .visit(b"aaaaaa", VisitDirection::Backwards, |key, value| {
+                panic!("found unexpected key {}: {}", hex::encode(key), value);
+            })
             .await?;
 
         // Full scan
@@ -831,15 +799,10 @@ pub(crate) mod tests {
             .collect();
         let mut data = Vec::new();
         reader
-            .visit(
-                &[0u8; 6],
-                VisitDirection::Forwards,
-                |key, value| {
-                    data.push((key.to_vec(), value));
-                    true
-                },
-                &ctx,
-            )
+            .visit(&[0u8; 6], VisitDirection::Forwards, |key, value| {
+                data.push((key.to_vec(), value));
+                true
+            })
             .await?;
         assert_eq!(data, expected);
 
@@ -850,7 +813,6 @@ pub(crate) mod tests {
     async fn lots_of_keys() -> Result<()> {
         let mut disk = TestDisk::new();
         let mut writer = DiskBtreeBuilder::<_, 8>::new(&mut disk);
-        let ctx = RequestContext::new(TaskKind::UnitTest, DownloadBehavior::Error);
 
         const NUM_KEYS: u64 = 1000;
 
@@ -889,14 +851,14 @@ pub(crate) mod tests {
         for search_key_int in 0..(NUM_KEYS * 2 + 10) {
             let search_key = u64::to_be_bytes(search_key_int);
             assert_eq!(
-                reader.get(&search_key, &ctx).await?,
+                reader.get(&search_key).await?,
                 all_data.get(&search_key_int).cloned()
             );
 
             // Test a forward scan starting with this key
             result.lock().unwrap().clear();
             reader
-                .visit(&search_key, VisitDirection::Forwards, take_ten, &ctx)
+                .visit(&search_key, VisitDirection::Forwards, take_ten)
                 .await?;
             let expected = all_data
                 .range(search_key_int..)
@@ -908,7 +870,7 @@ pub(crate) mod tests {
             // And a backwards scan
             result.lock().unwrap().clear();
             reader
-                .visit(&search_key, VisitDirection::Backwards, take_ten, &ctx)
+                .visit(&search_key, VisitDirection::Backwards, take_ten)
                 .await?;
             let expected = all_data
                 .range(..=search_key_int)
@@ -924,7 +886,7 @@ pub(crate) mod tests {
         limit.store(usize::MAX, Ordering::Relaxed);
         result.lock().unwrap().clear();
         reader
-            .visit(&search_key, VisitDirection::Forwards, take_ten, &ctx)
+            .visit(&search_key, VisitDirection::Forwards, take_ten)
             .await?;
         let expected = all_data
             .iter()
@@ -937,7 +899,7 @@ pub(crate) mod tests {
         limit.store(usize::MAX, Ordering::Relaxed);
         result.lock().unwrap().clear();
         reader
-            .visit(&search_key, VisitDirection::Backwards, take_ten, &ctx)
+            .visit(&search_key, VisitDirection::Backwards, take_ten)
             .await?;
         let expected = all_data
             .iter()
@@ -951,8 +913,6 @@ pub(crate) mod tests {
 
     #[tokio::test]
     async fn random_data() -> Result<()> {
-        let ctx = RequestContext::new(TaskKind::UnitTest, DownloadBehavior::Error);
-
         // Generate random keys with exponential distribution, to
         // exercise the prefix compression
         const NUM_KEYS: usize = 100000;
@@ -979,24 +939,22 @@ pub(crate) mod tests {
         // Test get() operation on all the keys
         for (&key, &val) in all_data.iter() {
             let search_key = u128::to_be_bytes(key);
-            assert_eq!(reader.get(&search_key, &ctx).await?, Some(val));
+            assert_eq!(reader.get(&search_key).await?, Some(val));
         }
 
         // Test get() operations on random keys, most of which will not exist
         for _ in 0..100000 {
             let key_int = rand::thread_rng().gen::<u128>();
             let search_key = u128::to_be_bytes(key_int);
-            assert!(reader.get(&search_key, &ctx).await? == all_data.get(&key_int).cloned());
+            assert!(reader.get(&search_key).await? == all_data.get(&key_int).cloned());
         }
 
         // Test boundary cases
         assert!(
-            reader.get(&u128::to_be_bytes(u128::MIN), &ctx).await?
-                == all_data.get(&u128::MIN).cloned()
+            reader.get(&u128::to_be_bytes(u128::MIN)).await? == all_data.get(&u128::MIN).cloned()
         );
         assert!(
-            reader.get(&u128::to_be_bytes(u128::MAX), &ctx).await?
-                == all_data.get(&u128::MAX).cloned()
+            reader.get(&u128::to_be_bytes(u128::MAX)).await? == all_data.get(&u128::MAX).cloned()
         );
 
         Ok(())
@@ -1027,7 +985,6 @@ pub(crate) mod tests {
         // Build a tree from it
         let mut disk = TestDisk::new();
         let mut writer = DiskBtreeBuilder::<_, 26>::new(&mut disk);
-        let ctx = RequestContext::new(TaskKind::UnitTest, DownloadBehavior::Error);
 
         for (key, val) in disk_btree_test_data::TEST_DATA {
             writer.append(&key, val)?;
@@ -1040,21 +997,16 @@ pub(crate) mod tests {
 
         // Test get() operation on all the keys
         for (key, val) in disk_btree_test_data::TEST_DATA {
-            assert_eq!(reader.get(&key, &ctx).await?, Some(val));
+            assert_eq!(reader.get(&key).await?, Some(val));
         }
 
         // Test full scan
         let mut count = 0;
         reader
-            .visit(
-                &[0u8; 26],
-                VisitDirection::Forwards,
-                |_key, _value| {
-                    count += 1;
-                    true
-                },
-                &ctx,
-            )
+            .visit(&[0u8; 26], VisitDirection::Forwards, |_key, _value| {
+                count += 1;
+                true
+            })
             .await?;
         assert_eq!(count, disk_btree_test_data::TEST_DATA.len());
 

pageserver/src/tenant/ephemeral_file.rs

@@ -2,7 +2,6 @@
 //! used to keep in-memory layers spilled on disk.
 
 use crate::config::PageServerConf;
-use crate::context::RequestContext;
 use crate::page_cache::{self, PAGE_SZ};
 use crate::tenant::block_io::{BlockCursor, BlockLease, BlockReader};
 use crate::virtual_file::VirtualFile;
@@ -62,17 +61,13 @@ impl EphemeralFile {
         self.len
     }
 
-    pub(crate) async fn read_blk(
-        &self,
-        blknum: u32,
-        ctx: &RequestContext,
-    ) -> Result<BlockLease, io::Error> {
+    pub(crate) async fn read_blk(&self, blknum: u32) -> Result<BlockLease, io::Error> {
         let flushed_blknums = 0..self.len / PAGE_SZ as u64;
         if flushed_blknums.contains(&(blknum as u64)) {
             let cache = page_cache::get();
             loop {
                 match cache
-                    .read_immutable_buf(self.page_cache_file_id, blknum, ctx)
+                    .read_immutable_buf(self.page_cache_file_id, blknum)
                     .await
                     .map_err(|e| {
                         std::io::Error::new(
@@ -108,11 +103,7 @@ impl EphemeralFile {
         }
     }
 
-    pub(crate) async fn write_blob(
-        &mut self,
-        srcbuf: &[u8],
-        ctx: &RequestContext,
-    ) -> Result<u64, io::Error> {
+    pub(crate) async fn write_blob(&mut self, srcbuf: &[u8]) -> Result<u64, io::Error> {
         struct Writer<'a> {
             ephemeral_file: &'a mut EphemeralFile,
             /// The block to which the next [`push_bytes`] will write.
@@ -129,11 +120,7 @@ impl EphemeralFile {
                 })
             }
             #[inline(always)]
-            async fn push_bytes(
-                &mut self,
-                src: &[u8],
-                ctx: &RequestContext,
-            ) -> Result<(), io::Error> {
+            async fn push_bytes(&mut self, src: &[u8]) -> Result<(), io::Error> {
                 let mut src_remaining = src;
                 while !src_remaining.is_empty() {
                     let dst_remaining = &mut self.ephemeral_file.mutable_tail[self.off..];
@@ -159,7 +146,6 @@ impl EphemeralFile {
                                     .read_immutable_buf(
                                         self.ephemeral_file.page_cache_file_id,
                                         self.blknum,
-                                        ctx,
                                     )
                                     .await
                                 {
@@ -213,15 +199,15 @@ impl EphemeralFile {
         if srcbuf.len() < 0x80 {
             // short one-byte length header
             let len_buf = [srcbuf.len() as u8];
-            writer.push_bytes(&len_buf, ctx).await?;
+            writer.push_bytes(&len_buf).await?;
         } else {
             let mut len_buf = u32::to_be_bytes(srcbuf.len() as u32);
             len_buf[0] |= 0x80;
-            writer.push_bytes(&len_buf, ctx).await?;
+            writer.push_bytes(&len_buf).await?;
         }
 
         // Write the payload
-        writer.push_bytes(srcbuf, ctx).await?;
+        writer.push_bytes(srcbuf).await?;
 
         if srcbuf.len() < 0x80 {
             self.len += 1;
@@ -275,8 +261,6 @@ impl BlockReader for EphemeralFile {
 #[cfg(test)]
 mod tests {
     use super::*;
-    use crate::context::DownloadBehavior;
-    use crate::task_mgr::TaskKind;
     use crate::tenant::block_io::{BlockCursor, BlockReaderRef};
     use rand::{thread_rng, RngCore};
     use std::fs;
@@ -284,15 +268,7 @@ mod tests {
 
     fn harness(
         test_name: &str,
-    ) -> Result<
-        (
-            &'static PageServerConf,
-            TenantId,
-            TimelineId,
-            RequestContext,
-        ),
-        io::Error,
-    > {
+    ) -> Result<(&'static PageServerConf, TenantId, TimelineId), io::Error> {
         let repo_dir = PageServerConf::test_repo_dir(test_name);
         let _ = fs::remove_dir_all(&repo_dir);
         let conf = PageServerConf::dummy_conf(repo_dir);
@@ -304,57 +280,46 @@ mod tests {
         let timeline_id = TimelineId::from_str("22000000000000000000000000000000").unwrap();
         fs::create_dir_all(conf.timeline_path(&tenant_id, &timeline_id))?;
 
-        let ctx = RequestContext::new(TaskKind::UnitTest, DownloadBehavior::Error);
-
-        Ok((conf, tenant_id, timeline_id, ctx))
+        Ok((conf, tenant_id, timeline_id))
     }
 
     #[tokio::test]
     async fn test_ephemeral_blobs() -> Result<(), io::Error> {
-        let (conf, tenant_id, timeline_id, ctx) = harness("ephemeral_blobs")?;
+        let (conf, tenant_id, timeline_id) = harness("ephemeral_blobs")?;
 
         let mut file = EphemeralFile::create(conf, tenant_id, timeline_id).await?;
 
-        let pos_foo = file.write_blob(b"foo", &ctx).await?;
+        let pos_foo = file.write_blob(b"foo").await?;
         assert_eq!(
             b"foo",
-            file.block_cursor()
-                .read_blob(pos_foo, &ctx)
-                .await?
-                .as_slice()
+            file.block_cursor().read_blob(pos_foo).await?.as_slice()
         );
-        let pos_bar = file.write_blob(b"bar", &ctx).await?;
+        let pos_bar = file.write_blob(b"bar").await?;
         assert_eq!(
             b"foo",
-            file.block_cursor()
-                .read_blob(pos_foo, &ctx)
-                .await?
-                .as_slice()
+            file.block_cursor().read_blob(pos_foo).await?.as_slice()
         );
         assert_eq!(
             b"bar",
-            file.block_cursor()
-                .read_blob(pos_bar, &ctx)
-                .await?
-                .as_slice()
+            file.block_cursor().read_blob(pos_bar).await?.as_slice()
         );
 
         let mut blobs = Vec::new();
         for i in 0..10000 {
             let data = Vec::from(format!("blob{}", i).as_bytes());
-            let pos = file.write_blob(&data, &ctx).await?;
+            let pos = file.write_blob(&data).await?;
             blobs.push((pos, data));
         }
         // also test with a large blobs
         for i in 0..100 {
             let data = format!("blob{}", i).as_bytes().repeat(100);
-            let pos = file.write_blob(&data, &ctx).await?;
+            let pos = file.write_blob(&data).await?;
             blobs.push((pos, data));
         }
 
         let cursor = BlockCursor::new(BlockReaderRef::EphemeralFile(&file));
         for (pos, expected) in blobs {
-            let actual = cursor.read_blob(pos, &ctx).await?;
+            let actual = cursor.read_blob(pos).await?;
             assert_eq!(actual, expected);
         }
 
@@ -362,8 +327,8 @@ mod tests {
         let mut large_data = Vec::new();
         large_data.resize(20000, 0);
         thread_rng().fill_bytes(&mut large_data);
-        let pos_large = file.write_blob(&large_data, &ctx).await?;
-        let result = file.block_cursor().read_blob(pos_large, &ctx).await?;
+        let pos_large = file.write_blob(&large_data).await?;
+        let result = file.block_cursor().read_blob(pos_large).await?;
         assert_eq!(result, large_data);
 
         Ok(())

pageserver/src/tenant/mgr.rs

@@ -20,10 +20,7 @@ use utils::crashsafe;
 
 use crate::config::PageServerConf;
 use crate::context::{DownloadBehavior, RequestContext};
-use crate::control_plane_client::{
-    ControlPlaneClient, ControlPlaneGenerationsApi, RetryForeverError,
-};
-use crate::deletion_queue::DeletionQueueClient;
+use crate::control_plane_client::ControlPlaneClient;
 use crate::task_mgr::{self, TaskKind};
 use crate::tenant::config::TenantConfOpt;
 use crate::tenant::delete::DeleteTenantFlow;
@@ -119,23 +116,7 @@ pub async fn init_tenant_mgr(
 
     // If we are configured to use the control plane API, then it is the source of truth for what tenants to load.
     let tenant_generations = if let Some(client) = ControlPlaneClient::new(conf, &cancel) {
-        let result = match client.re_attach().await {
-            Ok(tenants) => tenants,
-            Err(RetryForeverError::ShuttingDown) => {
-                anyhow::bail!("Shut down while waiting for control plane re-attach response")
-            }
-        };
-
-        // The deletion queue needs to know about the startup attachment state to decide which (if any) stored
-        // deletion list entries may still be valid.  We provide that by pushing a recovery operation into
-        // the queue. Sequential processing of te queue ensures that recovery is done before any new tenant deletions
-        // are processed, even though we don't block on recovery completing here.
-        resources
-            .deletion_queue_client
-            .recover(result.clone())
-            .await?;
-
-        Some(result)
+        Some(client.re_attach().await?)
     } else {
         info!("Control plane API not configured, tenant generations are disabled");
         None
@@ -304,21 +285,29 @@ pub(crate) fn schedule_local_tenant_processing(
 
     let tenant = if conf.tenant_attaching_mark_file_path(&tenant_id).exists() {
         info!("tenant {tenant_id} has attaching mark file, resuming its attach operation");
-        if resources.remote_storage.is_none() {
-            warn!("tenant {tenant_id} has attaching mark file, but pageserver has no remote storage configured");
-            Tenant::create_broken_tenant(
+        if let Some(remote_storage) = resources.remote_storage {
+            match Tenant::spawn_attach(
                 conf,
                 tenant_id,
-                "attaching mark file present but no remote storage configured".to_string(),
-            )
-        } else {
-            match Tenant::spawn_attach(conf, tenant_id, generation, resources, tenants, ctx) {
+                generation,
+                resources.broker_client,
+                tenants,
+                remote_storage,
+                ctx,
+            ) {
                 Ok(tenant) => tenant,
                 Err(e) => {
                     error!("Failed to spawn_attach tenant {tenant_id}, reason: {e:#}");
                     Tenant::create_broken_tenant(conf, tenant_id, format!("{e:#}"))
                 }
             }
+        } else {
+            warn!("tenant {tenant_id} has attaching mark file, but pageserver has no remote storage configured");
+            Tenant::create_broken_tenant(
+                conf,
+                tenant_id,
+                "attaching mark file present but no remote storage configured".to_string(),
+            )
         }
     } else {
         info!("tenant {tenant_id} is assumed to be loadable, starting load operation");
@@ -449,7 +438,8 @@ pub async fn create_tenant(
     tenant_conf: TenantConfOpt,
     tenant_id: TenantId,
     generation: Generation,
-    resources: TenantSharedResources,
+    broker_client: storage_broker::BrokerClientChannel,
+    remote_storage: Option<GenericRemoteStorage>,
     ctx: &RequestContext,
 ) -> Result<Arc<Tenant>, TenantMapInsertError> {
     tenant_map_insert(tenant_id, || async {
@@ -460,9 +450,13 @@ pub async fn create_tenant(
         // TODO: tenant directory remains on disk if we bail out from here on.
         //       See https://github.com/neondatabase/neon/issues/4233
 
+        let tenant_resources = TenantSharedResources {
+            broker_client,
+            remote_storage,
+        };
         let created_tenant =
             schedule_local_tenant_processing(conf, tenant_id, &tenant_directory,
-                generation, resources, None, &TENANTS, ctx)?;
+                generation, tenant_resources, None, &TENANTS, ctx)?;
         // TODO: tenant object & its background loops remain, untracked in tenant map, if we fail here.
         //      See https://github.com/neondatabase/neon/issues/4233
 
@@ -628,7 +622,6 @@ pub async fn load_tenant(
     generation: Generation,
     broker_client: storage_broker::BrokerClientChannel,
     remote_storage: Option<GenericRemoteStorage>,
-    deletion_queue_client: DeletionQueueClient,
     ctx: &RequestContext,
 ) -> Result<(), TenantMapInsertError> {
     tenant_map_insert(tenant_id, || async {
@@ -642,7 +635,6 @@ pub async fn load_tenant(
         let resources = TenantSharedResources {
             broker_client,
             remote_storage,
-            deletion_queue_client
         };
         let new_tenant = schedule_local_tenant_processing(conf, tenant_id, &tenant_path, generation, resources, None,  &TENANTS, ctx)
             .with_context(|| {
@@ -710,7 +702,8 @@ pub async fn attach_tenant(
     tenant_id: TenantId,
     generation: Generation,
     tenant_conf: TenantConfOpt,
-    resources: TenantSharedResources,
+    broker_client: storage_broker::BrokerClientChannel,
+    remote_storage: GenericRemoteStorage,
     ctx: &RequestContext,
 ) -> Result<(), TenantMapInsertError> {
     tenant_map_insert(tenant_id, || async {
@@ -725,7 +718,10 @@ pub async fn attach_tenant(
             .context("check for attach marker file existence")?;
         anyhow::ensure!(marker_file_exists, "create_tenant_files should have created the attach marker file");
 
-
+        let resources = TenantSharedResources {
+            broker_client,
+            remote_storage: Some(remote_storage),
+        };
         let attached_tenant = schedule_local_tenant_processing(conf, tenant_id, &tenant_dir, generation, resources, None, &TENANTS, ctx)?;
         // TODO: tenant object & its background loops remain, untracked in tenant map, if we fail here.
         //      See https://github.com/neondatabase/neon/issues/4233

pageserver/src/tenant/remote_timeline_client.rs

@@ -116,12 +116,8 @@
 //! # Completion
 //!
 //! Once an operation has completed, we update
-//! [`UploadQueueInitialized::projected_remote_consistent_lsn`] immediately,
-//! and submit a request through the DeletionQueue to update
-//! [`UploadQueueInitialized::visible_remote_consistent_lsn`] after it has
-//! validated that our generation is not stale.  It is this visible value
-//! that is advertized to safekeepers as a signal that that they can
-//! delete the WAL up to that LSN.
+//! [`UploadQueueInitialized::last_uploaded_consistent_lsn`] which indicates
+//! to safekeepers that they can delete the WAL up to that LSN.
 //!
 //! The [`RemoteTimelineClient::wait_completion`] method can be used to wait
 //! for all pending operations to complete. It does not prevent more
@@ -204,6 +200,7 @@
 //! [`Tenant::timeline_init_and_sync`]: super::Tenant::timeline_init_and_sync
 //! [`Timeline::load_layer_map`]: super::Timeline::load_layer_map
 
+mod delete;
 mod download;
 pub mod index;
 mod upload;
@@ -229,7 +226,6 @@ use tracing::{debug, error, info, instrument, warn};
 use tracing::{info_span, Instrument};
 use utils::lsn::Lsn;
 
-use crate::deletion_queue::DeletionQueueClient;
 use crate::metrics::{
     MeasureRemoteOp, RemoteOpFileKind, RemoteOpKind, RemoteTimelineClientMetrics,
     RemoteTimelineClientMetricsCallTrackSize, REMOTE_ONDEMAND_DOWNLOADED_BYTES,
@@ -328,8 +324,6 @@ pub struct RemoteTimelineClient {
     metrics: Arc<RemoteTimelineClientMetrics>,
 
     storage_impl: GenericRemoteStorage,
-
-    deletion_queue_client: DeletionQueueClient,
 }
 
 impl RemoteTimelineClient {
@@ -341,7 +335,6 @@ impl RemoteTimelineClient {
     ///
     pub fn new(
         remote_storage: GenericRemoteStorage,
-        deletion_queue_client: DeletionQueueClient,
         conf: &'static PageServerConf,
         tenant_id: TenantId,
         timeline_id: TimelineId,
@@ -359,7 +352,6 @@ impl RemoteTimelineClient {
             timeline_id,
             generation,
             storage_impl: remote_storage,
-            deletion_queue_client,
             upload_queue: Mutex::new(UploadQueue::Uninitialized),
             metrics: Arc::new(RemoteTimelineClientMetrics::new(&tenant_id, &timeline_id)),
         }
@@ -421,24 +413,13 @@ impl RemoteTimelineClient {
         Ok(())
     }
 
-    pub fn remote_consistent_lsn_projected(&self) -> Option<Lsn> {
-        match &mut *self.upload_queue.lock().unwrap() {
+    pub fn last_uploaded_consistent_lsn(&self) -> Option<Lsn> {
+        match &*self.upload_queue.lock().unwrap() {
             UploadQueue::Uninitialized => None,
-            UploadQueue::Initialized(q) => q.get_last_remote_consistent_lsn_projected(),
-            UploadQueue::Stopped(q) => q
-                .upload_queue_for_deletion
-                .get_last_remote_consistent_lsn_projected(),
-        }
-    }
-
-    pub fn remote_consistent_lsn_visible(&self) -> Option<Lsn> {
-        match &mut *self.upload_queue.lock().unwrap() {
-            UploadQueue::Uninitialized => None,
-            UploadQueue::Initialized(q) => Some(q.get_last_remote_consistent_lsn_visible()),
-            UploadQueue::Stopped(q) => Some(
-                q.upload_queue_for_deletion
-                    .get_last_remote_consistent_lsn_visible(),
-            ),
+            UploadQueue::Initialized(q) => Some(q.last_uploaded_consistent_lsn),
+            UploadQueue::Stopped(q) => {
+                Some(q.upload_queue_for_deletion.last_uploaded_consistent_lsn)
+            }
         }
     }
 
@@ -453,11 +434,11 @@ impl RemoteTimelineClient {
         } else {
             0
         };
-        self.metrics.remote_physical_size_set(size);
+        self.metrics.remote_physical_size_gauge().set(size);
     }
 
     pub fn get_remote_physical_size(&self) -> u64 {
-        self.metrics.remote_physical_size_get()
+        self.metrics.remote_physical_size_gauge().get()
     }
 
     //
@@ -662,7 +643,7 @@ impl RemoteTimelineClient {
     /// successfully.
     pub fn schedule_layer_file_deletion(
         self: &Arc<Self>,
-        names: Vec<LayerFileName>,
+        names: &[LayerFileName],
     ) -> anyhow::Result<()> {
         let mut guard = self.upload_queue.lock().unwrap();
         let upload_queue = guard.initialized_mut()?;
@@ -682,10 +663,10 @@ impl RemoteTimelineClient {
             // Decorate our list of names with each name's generation, dropping
             // makes that are unexpectedly missing from our metadata.
             let with_generations: Vec<_> = names
-                .into_iter()
+                .iter()
                 .filter_map(|name| {
                     // Remove from latest_files, learning the file's remote generation in the process
-                    let meta = upload_queue.latest_files.remove(&name);
+                    let meta = upload_queue.latest_files.remove(name);
 
                     if let Some(meta) = meta {
                         upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
@@ -707,16 +688,18 @@ impl RemoteTimelineClient {
                 self.schedule_index_upload(upload_queue, metadata);
             }
 
-            for (name, gen) in &with_generations {
-                info!("scheduling deletion of layer {}{}", name, gen.get_suffix());
-            }
-
             // schedule the actual deletions
-            let op = UploadOp::Delete(Delete {
-                layers: with_generations,
-            });
-            self.calls_unfinished_metric_begin(&op);
-            upload_queue.queued_operations.push_back(op);
+            for (name, generation) in with_generations {
+                let op = UploadOp::Delete(Delete {
+                    file_kind: RemoteOpFileKind::Layer,
+                    layer_file_name: name.clone(),
+                    scheduled_from_timeline_delete: false,
+                    generation,
+                });
+                self.calls_unfinished_metric_begin(&op);
+                upload_queue.queued_operations.push_back(op);
+                info!("scheduled layer file deletion {name}");
+            }
 
             // Launch the tasks immediately, if possible
             self.launch_queued_tasks(upload_queue);
@@ -850,7 +833,9 @@ impl RemoteTimelineClient {
     pub(crate) async fn delete_all(self: &Arc<Self>) -> anyhow::Result<()> {
         debug_assert_current_span_has_tenant_and_timeline_id();
 
-        let layers: Vec<RemotePath> = {
+        let (mut receiver, deletions_queued) = {
+            let mut deletions_queued = 0;
+
             let mut locked = self.upload_queue.lock().unwrap();
             let stopped = locked.stopped_mut()?;
 
@@ -862,30 +847,42 @@ impl RemoteTimelineClient {
 
             stopped
                 .upload_queue_for_deletion
-                .latest_files
-                .drain()
-                .map(|(file_name, meta)| {
-                    remote_layer_path(
-                        &self.tenant_id,
-                        &self.timeline_id,
-                        &file_name,
-                        meta.generation,
-                    )
-                })
-                .collect()
+                .queued_operations
+                .reserve(stopped.upload_queue_for_deletion.latest_files.len());
+
+            // schedule the actual deletions
+            for (name, meta) in &stopped.upload_queue_for_deletion.latest_files {
+                let op = UploadOp::Delete(Delete {
+                    file_kind: RemoteOpFileKind::Layer,
+                    layer_file_name: name.clone(),
+                    scheduled_from_timeline_delete: true,
+                    generation: meta.generation,
+                });
+
+                self.calls_unfinished_metric_begin(&op);
+                stopped
+                    .upload_queue_for_deletion
+                    .queued_operations
+                    .push_back(op);
+
+                info!("scheduled layer file deletion {name}");
+                deletions_queued += 1;
+            }
+
+            self.launch_queued_tasks(&mut stopped.upload_queue_for_deletion);
+
+            (
+                self.schedule_barrier(&mut stopped.upload_queue_for_deletion),
+                deletions_queued,
+            )
         };
 
-        let layer_deletion_count = layers.len();
-        self.deletion_queue_client.push_immediate(layers).await?;
+        receiver.changed().await.context("upload queue shut down")?;
 
         // 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_id, &self.timeline_id);
 
-        // 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.deletion_queue_client.flush_immediate().await?;
-
         let remaining = backoff::retry(
             || async {
                 self.storage_impl
@@ -913,9 +910,17 @@ impl RemoteTimelineClient {
             })
             .collect();
 
-        let not_referenced_count = remaining.len();
         if !remaining.is_empty() {
-            self.deletion_queue_client.push_immediate(remaining).await?;
+            backoff::retry(
+                || async { self.storage_impl.delete_objects(&remaining).await },
+                |_e| false,
+                FAILED_UPLOAD_WARN_THRESHOLD,
+                FAILED_REMOTE_OP_RETRIES,
+                "delete_objects",
+                backoff::Cancel::new(shutdown_token(), || anyhow::anyhow!("Cancelled!")),
+            )
+            .await
+            .context("delete_objects")?;
         }
 
         fail::fail_point!("timeline-delete-before-index-delete", |_| {
@@ -926,14 +931,18 @@ impl RemoteTimelineClient {
 
         let index_file_path = timeline_storage_path.join(Path::new(IndexPart::FILE_NAME));
 
-        debug!("enqueuing index part deletion");
-        self.deletion_queue_client
-            .push_immediate([index_file_path].to_vec())
-            .await?;
+        debug!("deleting index part");
 
-        // 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.deletion_queue_client.flush_immediate().await?;
+        backoff::retry(
+            || async { self.storage_impl.delete(&index_file_path).await },
+            |_e| false,
+            FAILED_UPLOAD_WARN_THRESHOLD,
+            FAILED_REMOTE_OP_RETRIES,
+            "delete_index",
+            backoff::Cancel::new(shutdown_token(), || anyhow::anyhow!("Cancelled")),
+        )
+        .await
+        .context("delete_index")?;
 
         fail::fail_point!("timeline-delete-after-index-delete", |_| {
             Err(anyhow::anyhow!(
@@ -941,7 +950,7 @@ impl RemoteTimelineClient {
             ))?
         });
 
-        info!(prefix=%timeline_storage_path, referenced=layer_deletion_count, not_referenced=%not_referenced_count, "done deleting in timeline prefix, including index_part.json");
+        info!(prefix=%timeline_storage_path, referenced=deletions_queued, not_referenced=%remaining.len(), "done deleting in timeline prefix, including index_part.json");
 
         Ok(())
     }
@@ -1131,16 +1140,21 @@ impl RemoteTimelineClient {
                     }
                     res
                 }
-                UploadOp::Delete(delete) => self
-                    .deletion_queue_client
-                    .push_layers(
-                        self.tenant_id,
-                        self.timeline_id,
-                        self.generation,
-                        delete.layers.clone(),
-                    )
-                    .await
-                    .map_err(|e| anyhow::anyhow!(e)),
+                UploadOp::Delete(delete) => {
+                    let path = &self
+                        .conf
+                        .timeline_path(&self.tenant_id, &self.timeline_id)
+                        .join(delete.layer_file_name.file_name());
+                    delete::delete_layer(self.conf, &self.storage_impl, path, delete.generation)
+                        .measure_remote_op(
+                            self.tenant_id,
+                            self.timeline_id,
+                            delete.file_kind,
+                            RemoteOpKind::Delete,
+                            Arc::clone(&self.metrics),
+                        )
+                        .await
+                }
                 UploadOp::Barrier(_) => {
                     // unreachable. Barrier operations are handled synchronously in
                     // launch_queued_tasks
@@ -1196,12 +1210,18 @@ impl RemoteTimelineClient {
         }
 
         // The task has completed successfully. Remove it from the in-progress list.
-        let lsn_update = {
+        {
             let mut upload_queue_guard = self.upload_queue.lock().unwrap();
             let upload_queue = match upload_queue_guard.deref_mut() {
                 UploadQueue::Uninitialized => panic!("callers are responsible for ensuring this is only called on an initialized queue"),
-                UploadQueue::Stopped(_stopped) => {
-                    None
+                UploadQueue::Stopped(stopped) => {
+                    // Special care is needed for deletions, if it was an earlier deletion (not scheduled from deletion)
+                    // then stop() took care of it so we just return.
+                    // For deletions that come from delete_all we still want to maintain metrics, launch following tasks, etc.
+                    match &task.op {
+                        UploadOp::Delete(delete) if delete.scheduled_from_timeline_delete => Some(&mut stopped.upload_queue_for_deletion),
+                        _ => None
+                    }
                 },
                 UploadQueue::Initialized(qi) => { Some(qi) }
             };
@@ -1216,51 +1236,23 @@ impl RemoteTimelineClient {
 
             upload_queue.inprogress_tasks.remove(&task.task_id);
 
-            let lsn_update = match task.op {
+            match task.op {
                 UploadOp::UploadLayer(_, _) => {
                     upload_queue.num_inprogress_layer_uploads -= 1;
-                    None
                 }
                 UploadOp::UploadMetadata(_, lsn) => {
                     upload_queue.num_inprogress_metadata_uploads -= 1;
-                    // XXX monotonicity check?
-
-                    upload_queue.projected_remote_consistent_lsn = Some(lsn);
-                    if self.generation.is_none() {
-                        // Legacy mode: skip validating generation
-                        upload_queue.visible_remote_consistent_lsn.store(lsn);
-                        None
-                    } else {
-                        Some((lsn, upload_queue.visible_remote_consistent_lsn.clone()))
-                    }
+                    upload_queue.last_uploaded_consistent_lsn = lsn; // XXX monotonicity check?
                 }
                 UploadOp::Delete(_) => {
                     upload_queue.num_inprogress_deletions -= 1;
-                    None
                 }
                 UploadOp::Barrier(_) => unreachable!(),
             };
 
             // Launch any queued tasks that were unblocked by this one.
             self.launch_queued_tasks(upload_queue);
-            lsn_update
-        };
-
-        if let Some((lsn, slot)) = lsn_update {
-            // Updates to the remote_consistent_lsn we advertise to pageservers
-            // are all routed through the DeletionQueue, to enforce important
-            // data safety guarantees (see docs/rfcs/025-generation-numbers.md)
-            self.deletion_queue_client
-                .update_remote_consistent_lsn(
-                    self.tenant_id,
-                    self.timeline_id,
-                    self.generation,
-                    lsn,
-                    slot,
-                )
-                .await;
         }
-
         self.calls_unfinished_metric_end(&task.op);
     }
 
@@ -1286,8 +1278,8 @@ impl RemoteTimelineClient {
                     reason: "metadata uploads are tiny",
                 },
             ),
-            UploadOp::Delete(_delete) => (
-                RemoteOpFileKind::Layer,
+            UploadOp::Delete(delete) => (
+                delete.file_kind,
                 RemoteOpKind::Delete,
                 DontTrackSize {
                     reason: "should we track deletes? positive or negative sign?",
@@ -1349,10 +1341,7 @@ impl RemoteTimelineClient {
                         latest_files: initialized.latest_files.clone(),
                         latest_files_changes_since_metadata_upload_scheduled: 0,
                         latest_metadata: initialized.latest_metadata.clone(),
-                        projected_remote_consistent_lsn: None,
-                        visible_remote_consistent_lsn: initialized
-                            .visible_remote_consistent_lsn
-                            .clone(),
+                        last_uploaded_consistent_lsn: initialized.last_uploaded_consistent_lsn,
                         num_inprogress_layer_uploads: 0,
                         num_inprogress_metadata_uploads: 0,
                         num_inprogress_deletions: 0,
@@ -1416,13 +1405,13 @@ pub fn remote_layer_path(
     tenant_id: &TenantId,
     timeline_id: &TimelineId,
     layer_file_name: &LayerFileName,
-    generation: Generation,
+    layer_meta: &LayerFileMetadata,
 ) -> RemotePath {
     // Generation-aware key format
     let path = format!(
         "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{0}{1}",
         layer_file_name.file_name(),
-        generation.get_suffix()
+        layer_meta.generation.get_suffix()
     );
 
     RemotePath::from_string(&path).expect("Failed to construct path")
@@ -1565,6 +1554,7 @@ mod tests {
 
     impl TestSetup {
         async fn new(test_name: &str) -> anyhow::Result<Self> {
+            // Use a current-thread runtime in the test
             let test_name = Box::leak(Box::new(format!("remote_timeline_client__{test_name}")));
             let harness = TenantHarness::create(test_name)?;
             let (tenant, ctx) = harness.load().await;
@@ -1590,7 +1580,6 @@ mod tests {
                 timeline_id: TIMELINE_ID,
                 generation,
                 storage_impl: self.harness.remote_storage.clone(),
-                deletion_queue_client: self.harness.deletion_queue.new_client(),
                 upload_queue: Mutex::new(UploadQueue::Uninitialized),
                 metrics: Arc::new(RemoteTimelineClientMetrics::new(
                     &self.harness.tenant_id,
@@ -1760,7 +1749,7 @@ mod tests {
             )
             .unwrap();
         client
-            .schedule_layer_file_deletion([layer_file_name_1.clone()].to_vec())
+            .schedule_layer_file_deletion(&[layer_file_name_1.clone()])
             .unwrap();
         {
             let mut guard = client.upload_queue.lock().unwrap();
@@ -1786,7 +1775,6 @@ mod tests {
 
         // Finish them
         client.wait_completion().await.unwrap();
-        harness.deletion_queue.pump().await;
 
         assert_remote_files(
             &[

pageserver/src/tenant/remote_timeline_client/delete.rs

@@ -0,0 +1,34 @@
+//! Helper functions to delete files from remote storage with a RemoteStorage
+use anyhow::Context;
+use std::path::Path;
+use tracing::debug;
+
+use remote_storage::GenericRemoteStorage;
+
+use crate::{
+    config::PageServerConf,
+    tenant::{remote_timeline_client::remote_path, Generation},
+};
+
+pub(super) async fn delete_layer<'a>(
+    conf: &'static PageServerConf,
+    storage: &'a GenericRemoteStorage,
+    local_layer_path: &'a Path,
+    generation: Generation,
+) -> anyhow::Result<()> {
+    fail::fail_point!("before-delete-layer", |_| {
+        anyhow::bail!("failpoint before-delete-layer")
+    });
+    debug!("Deleting layer from remote storage: {local_layer_path:?}",);
+
+    let path_to_delete = remote_path(conf, local_layer_path, generation)?;
+
+    // We don't want to print an error if the delete failed if the file has
+    // already been deleted. Thankfully, in this situation S3 already
+    // does not yield an error. While OS-provided local file system APIs do yield
+    // errors, we avoid them in the `LocalFs` wrapper.
+    storage
+        .delete(&path_to_delete)
+        .await
+        .with_context(|| format!("delete remote layer from storage at {path_to_delete:?}"))
+}

pageserver/src/tenant/remote_timeline_client/download.rs

@@ -50,12 +50,7 @@ pub async fn download_layer_file<'a>(
         .timeline_path(&tenant_id, &timeline_id)
         .join(layer_file_name.file_name());
 
-    let remote_path = remote_layer_path(
-        &tenant_id,
-        &timeline_id,
-        layer_file_name,
-        layer_metadata.generation,
-    );
+    let remote_path = remote_layer_path(&tenant_id, &timeline_id, layer_file_name, layer_metadata);
 
     // Perform a rename inspired by durable_rename from file_utils.c.
     // The sequence:

pageserver/src/tenant/storage_layer/delta_layer.rs

@@ -28,7 +28,7 @@
 //! "values" part.
 //!
 use crate::config::PageServerConf;
-use crate::context::{PageContentKind, RequestContext, RequestContextBuilder};
+use crate::context::RequestContext;
 use crate::page_cache::PAGE_SZ;
 use crate::repository::{Key, Value, KEY_SIZE};
 use crate::tenant::blob_io::BlobWriter;
@@ -317,11 +317,11 @@ impl DeltaLayer {
 
         tree_reader.dump().await?;
 
-        let keys = DeltaLayerInner::load_keys(&inner, ctx).await?;
+        let keys = DeltaLayerInner::load_keys(&inner).await?;
 
         // A subroutine to dump a single blob
-        async fn dump_blob(val: ValueRef<'_>, ctx: &RequestContext) -> Result<String> {
-            let buf = val.reader.read_blob(val.blob_ref.pos(), ctx).await?;
+        async fn dump_blob(val: ValueRef<'_>) -> Result<String> {
+            let buf = val.reader.read_blob(val.blob_ref.pos()).await?;
             let val = Value::des(&buf)?;
             let desc = match val {
                 Value::Image(img) => {
@@ -342,7 +342,7 @@ impl DeltaLayer {
 
         for entry in keys {
             let DeltaEntry { key, lsn, val, .. } = entry;
-            let desc = match dump_blob(val, ctx).await {
+            let desc = match dump_blob(val).await {
                 Ok(desc) => desc,
                 Err(err) => {
                     let err: anyhow::Error = err;
@@ -370,7 +370,7 @@ impl DeltaLayer {
             .load(LayerAccessKind::GetValueReconstructData, ctx)
             .await?;
         inner
-            .get_value_reconstruct_data(key, lsn_range, reconstruct_state, ctx)
+            .get_value_reconstruct_data(key, lsn_range, reconstruct_state)
             .await
     }
 
@@ -453,12 +453,12 @@ impl DeltaLayer {
         self.access_stats.record_access(access_kind, ctx);
         // Quick exit if already loaded
         self.inner
-            .get_or_try_init(|| self.load_inner(ctx))
+            .get_or_try_init(|| self.load_inner())
             .await
             .with_context(|| format!("Failed to load delta layer {}", self.path().display()))
     }
 
-    async fn load_inner(&self, ctx: &RequestContext) -> Result<Arc<DeltaLayerInner>> {
+    async fn load_inner(&self) -> Result<Arc<DeltaLayerInner>> {
         let path = self.path();
 
         let summary = match &self.path_or_conf {
@@ -466,7 +466,7 @@ impl DeltaLayer {
             PathOrConf::Path(_) => None,
         };
 
-        let loaded = DeltaLayerInner::load(&path, summary, ctx).await?;
+        let loaded = DeltaLayerInner::load(&path, summary).await?;
 
         if let PathOrConf::Path(ref path) = self.path_or_conf {
             // not production code
@@ -554,7 +554,7 @@ impl DeltaLayer {
             .load(LayerAccessKind::KeyIter, ctx)
             .await
             .context("load delta layer keys")?;
-        DeltaLayerInner::load_keys(inner, ctx)
+        DeltaLayerInner::load_keys(inner)
             .await
             .context("Layer index is corrupted")
     }
@@ -849,14 +849,13 @@ impl DeltaLayerInner {
     pub(super) async fn load(
         path: &std::path::Path,
         summary: Option<Summary>,
-        ctx: &RequestContext,
     ) -> anyhow::Result<Self> {
         let file = VirtualFile::open(path)
             .await
             .with_context(|| format!("Failed to open file '{}'", path.display()))?;
         let file = FileBlockReader::new(file);
 
-        let summary_blk = file.read_blk(0, ctx).await?;
+        let summary_blk = file.read_blk(0).await?;
         let actual_summary = Summary::des_prefix(summary_blk.as_ref())?;
 
         if let Some(mut expected_summary) = summary {
@@ -884,7 +883,6 @@ impl DeltaLayerInner {
         key: Key,
         lsn_range: Range<Lsn>,
         reconstruct_state: &mut ValueReconstructState,
-        ctx: &RequestContext,
     ) -> anyhow::Result<ValueReconstructResult> {
         let mut need_image = true;
         // Scan the page versions backwards, starting from `lsn`.
@@ -899,38 +897,27 @@ impl DeltaLayerInner {
         let mut offsets: Vec<(Lsn, u64)> = Vec::new();
 
         tree_reader
-            .visit(
-                &search_key.0,
-                VisitDirection::Backwards,
-                |key, value| {
-                    let blob_ref = BlobRef(value);
-                    if key[..KEY_SIZE] != search_key.0[..KEY_SIZE] {
-                        return false;
-                    }
-                    let entry_lsn = DeltaKey::extract_lsn_from_buf(key);
-                    if entry_lsn < lsn_range.start {
-                        return false;
-                    }
-                    offsets.push((entry_lsn, blob_ref.pos()));
+            .visit(&search_key.0, VisitDirection::Backwards, |key, value| {
+                let blob_ref = BlobRef(value);
+                if key[..KEY_SIZE] != search_key.0[..KEY_SIZE] {
+                    return false;
+                }
+                let entry_lsn = DeltaKey::extract_lsn_from_buf(key);
+                if entry_lsn < lsn_range.start {
+                    return false;
+                }
+                offsets.push((entry_lsn, blob_ref.pos()));
 
-                    !blob_ref.will_init()
-                },
-                &RequestContextBuilder::extend(ctx)
-                    .page_content_kind(PageContentKind::DeltaLayerBtreeNode)
-                    .build(),
-            )
+                !blob_ref.will_init()
+            })
             .await?;
 
-        let ctx = &RequestContextBuilder::extend(ctx)
-            .page_content_kind(PageContentKind::DeltaLayerValue)
-            .build();
-
         // Ok, 'offsets' now contains the offsets of all the entries we need to read
         let cursor = file.block_cursor();
         let mut buf = Vec::new();
         for (entry_lsn, pos) in offsets {
             cursor
-                .read_blob_into_buf(pos, &mut buf, ctx)
+                .read_blob_into_buf(pos, &mut buf)
                 .await
                 .with_context(|| {
                     format!(
@@ -971,10 +958,9 @@ impl DeltaLayerInner {
         }
     }
 
-    pub(super) async fn load_keys<'a, 'b, T: AsRef<DeltaLayerInner> + Clone>(
-        this: &'a T,
-        ctx: &'b RequestContext,
-    ) -> Result<Vec<DeltaEntry<'a>>> {
+    pub(super) async fn load_keys<T: AsRef<DeltaLayerInner> + Clone>(
+        this: &T,
+    ) -> Result<Vec<DeltaEntry<'_>>> {
         let dl = this.as_ref();
         let file = &dl.file;
 
@@ -1011,9 +997,6 @@ impl DeltaLayerInner {
                     all_keys.push(entry);
                     true
                 },
-                &RequestContextBuilder::extend(ctx)
-                    .page_content_kind(PageContentKind::DeltaLayerBtreeNode)
-                    .build(),
             )
             .await?;
         if let Some(last) = all_keys.last_mut() {
@@ -1043,9 +1026,9 @@ pub struct ValueRef<'a> {
 
 impl<'a> ValueRef<'a> {
     /// Loads the value from disk
-    pub async fn load(&self, ctx: &RequestContext) -> Result<Value> {
+    pub async fn load(&self) -> Result<Value> {
         // theoretically we *could* record an access time for each, but it does not really matter
-        let buf = self.reader.read_blob(self.blob_ref.pos(), ctx).await?;
+        let buf = self.reader.read_blob(self.blob_ref.pos()).await?;
         let val = Value::des(&buf)?;
         Ok(val)
     }
@@ -1054,11 +1037,7 @@ impl<'a> ValueRef<'a> {
 pub(crate) struct Adapter<T>(T);
 
 impl<T: AsRef<DeltaLayerInner>> Adapter<T> {
-    pub(crate) async fn read_blk(
-        &self,
-        blknum: u32,
-        ctx: &RequestContext,
-    ) -> Result<BlockLease, std::io::Error> {
-        self.0.as_ref().file.read_blk(blknum, ctx).await
+    pub(crate) async fn read_blk(&self, blknum: u32) -> Result<BlockLease, std::io::Error> {
+        self.0.as_ref().file.read_blk(blknum).await
     }
 }

pageserver/src/tenant/storage_layer/image_layer.rs

@@ -24,7 +24,7 @@
 //! mapping from Key to an offset in the "values" part.  The
 //! actual page images are stored in the "values" part.
 use crate::config::PageServerConf;
-use crate::context::{PageContentKind, RequestContext, RequestContextBuilder};
+use crate::context::RequestContext;
 use crate::page_cache::PAGE_SZ;
 use crate::repository::{Key, KEY_SIZE};
 use crate::tenant::blob_io::BlobWriter;
@@ -237,15 +237,10 @@ impl ImageLayer {
         tree_reader.dump().await?;
 
         tree_reader
-            .visit(
-                &[0u8; KEY_SIZE],
-                VisitDirection::Forwards,
-                |key, value| {
-                    println!("key: {} offset {}", hex::encode(key), value);
-                    true
-                },
-                ctx,
-            )
+            .visit(&[0u8; KEY_SIZE], VisitDirection::Forwards, |key, value| {
+                println!("key: {} offset {}", hex::encode(key), value);
+                true
+            })
             .await?;
 
         Ok(())
@@ -266,7 +261,7 @@ impl ImageLayer {
             .load(LayerAccessKind::GetValueReconstructData, ctx)
             .await?;
         inner
-            .get_value_reconstruct_data(key, reconstruct_state, ctx)
+            .get_value_reconstruct_data(key, reconstruct_state)
             .await
             // FIXME: makes no sense to dump paths
             .with_context(|| format!("read {}", self.path().display()))
@@ -340,12 +335,12 @@ impl ImageLayer {
     ) -> Result<&ImageLayerInner> {
         self.access_stats.record_access(access_kind, ctx);
         self.inner
-            .get_or_try_init(|| self.load_inner(ctx))
+            .get_or_try_init(|| self.load_inner())
             .await
             .with_context(|| format!("Failed to load image layer {}", self.path().display()))
     }
 
-    async fn load_inner(&self, ctx: &RequestContext) -> Result<ImageLayerInner> {
+    async fn load_inner(&self) -> Result<ImageLayerInner> {
         let path = self.path();
 
         let expected_summary = match &self.path_or_conf {
@@ -354,8 +349,7 @@ impl ImageLayer {
         };
 
         let loaded =
-            ImageLayerInner::load(&path, self.desc.image_layer_lsn(), expected_summary, ctx)
-                .await?;
+            ImageLayerInner::load(&path, self.desc.image_layer_lsn(), expected_summary).await?;
 
         if let PathOrConf::Path(ref path) = self.path_or_conf {
             // not production code
@@ -442,13 +436,12 @@ impl ImageLayerInner {
         path: &std::path::Path,
         lsn: Lsn,
         summary: Option<Summary>,
-        ctx: &RequestContext,
     ) -> anyhow::Result<Self> {
         let file = VirtualFile::open(path)
             .await
             .with_context(|| format!("Failed to open file '{}'", path.display()))?;
         let file = FileBlockReader::new(file);
-        let summary_blk = file.read_blk(0, ctx).await?;
+        let summary_blk = file.read_blk(0).await?;
         let actual_summary = Summary::des_prefix(summary_blk.as_ref())?;
 
         if let Some(mut expected_summary) = summary {
@@ -477,30 +470,16 @@ impl ImageLayerInner {
         &self,
         key: Key,
         reconstruct_state: &mut ValueReconstructState,
-        ctx: &RequestContext,
     ) -> anyhow::Result<ValueReconstructResult> {
         let file = &self.file;
         let tree_reader = DiskBtreeReader::new(self.index_start_blk, self.index_root_blk, file);
 
         let mut keybuf: [u8; KEY_SIZE] = [0u8; KEY_SIZE];
         key.write_to_byte_slice(&mut keybuf);
-        if let Some(offset) = tree_reader
-            .get(
-                &keybuf,
-                &RequestContextBuilder::extend(ctx)
-                    .page_content_kind(PageContentKind::ImageLayerBtreeNode)
-                    .build(),
-            )
-            .await?
-        {
+        if let Some(offset) = tree_reader.get(&keybuf).await? {
             let blob = file
                 .block_cursor()
-                .read_blob(
-                    offset,
-                    &RequestContextBuilder::extend(ctx)
-                        .page_content_kind(PageContentKind::ImageLayerValue)
-                        .build(),
-                )
+                .read_blob(offset)
                 .await
                 .with_context(|| format!("failed to read value from offset {}", offset))?;
             let value = Bytes::from(blob);

pageserver/src/tenant/storage_layer/inmemory_layer.rs

@@ -5,7 +5,7 @@
 //! its position in the file, is kept in memory, though.
 //!
 use crate::config::PageServerConf;
-use crate::context::{PageContentKind, RequestContext, RequestContextBuilder};
+use crate::context::RequestContext;
 use crate::repository::{Key, Value};
 use crate::tenant::block_io::BlockReader;
 use crate::tenant::ephemeral_file::EphemeralFile;
@@ -106,7 +106,7 @@ impl InMemoryLayer {
     /// debugging function to print out the contents of the layer
     ///
     /// this is likely completly unused
-    pub async fn dump(&self, verbose: bool, ctx: &RequestContext) -> Result<()> {
+    pub async fn dump(&self, verbose: bool, _ctx: &RequestContext) -> Result<()> {
         let inner = self.inner.read().await;
 
         let end_str = self.end_lsn_or_max();
@@ -125,7 +125,7 @@ impl InMemoryLayer {
         for (key, vec_map) in inner.index.iter() {
             for (lsn, pos) in vec_map.as_slice() {
                 let mut desc = String::new();
-                cursor.read_blob_into_buf(*pos, &mut buf, ctx).await?;
+                cursor.read_blob_into_buf(*pos, &mut buf).await?;
                 let val = Value::des(&buf);
                 match val {
                     Ok(Value::Image(img)) => {
@@ -158,15 +158,11 @@ impl InMemoryLayer {
         key: Key,
         lsn_range: Range<Lsn>,
         reconstruct_state: &mut ValueReconstructState,
-        ctx: &RequestContext,
+        _ctx: &RequestContext,
     ) -> anyhow::Result<ValueReconstructResult> {
         ensure!(lsn_range.start >= self.start_lsn);
         let mut need_image = true;
 
-        let ctx = RequestContextBuilder::extend(ctx)
-            .page_content_kind(PageContentKind::InMemoryLayer)
-            .build();
-
         let inner = self.inner.read().await;
 
         let reader = inner.file.block_cursor();
@@ -175,7 +171,7 @@ impl InMemoryLayer {
         if let Some(vec_map) = inner.index.get(&key) {
             let slice = vec_map.slice_range(lsn_range);
             for (entry_lsn, pos) in slice.iter().rev() {
-                let buf = reader.read_blob(*pos, &ctx).await?;
+                let buf = reader.read_blob(*pos).await?;
                 let value = Value::des(&buf)?;
                 match value {
                     Value::Image(img) => {
@@ -267,13 +263,7 @@ impl InMemoryLayer {
 
     /// Common subroutine of the public put_wal_record() and put_page_image() functions.
     /// Adds the page version to the in-memory tree
-    pub async fn put_value(
-        &self,
-        key: Key,
-        lsn: Lsn,
-        val: &Value,
-        ctx: &RequestContext,
-    ) -> Result<()> {
+    pub async fn put_value(&self, key: Key, lsn: Lsn, val: &Value) -> Result<()> {
         trace!("put_value key {} at {}/{}", key, self.timeline_id, lsn);
         let inner: &mut _ = &mut *self.inner.write().await;
         self.assert_writable();
@@ -285,15 +275,7 @@ impl InMemoryLayer {
             let mut buf = smallvec::SmallVec::<[u8; 256]>::new();
             buf.clear();
             val.ser_into(&mut buf)?;
-            inner
-                .file
-                .write_blob(
-                    &buf,
-                    &RequestContextBuilder::extend(ctx)
-                        .page_content_kind(PageContentKind::InMemoryLayer)
-                        .build(),
-                )
-                .await?
+            inner.file.write_blob(&buf).await?
         };
 
         let vec_map = inner.index.entry(key).or_default();
@@ -331,7 +313,7 @@ impl InMemoryLayer {
     /// Write this frozen in-memory layer to disk.
     ///
     /// Returns a new delta layer with all the same data as this in-memory layer
-    pub(crate) async fn write_to_disk(&self, ctx: &RequestContext) -> Result<DeltaLayer> {
+    pub(crate) async fn write_to_disk(&self) -> Result<DeltaLayer> {
         // Grab the lock in read-mode. We hold it over the I/O, but because this
         // layer is not writeable anymore, no one should be trying to acquire the
         // write lock on it, so we shouldn't block anyone. There's one exception
@@ -361,14 +343,11 @@ impl InMemoryLayer {
         let mut keys: Vec<(&Key, &VecMap<Lsn, u64>)> = inner.index.iter().collect();
         keys.sort_by_key(|k| k.0);
 
-        let ctx = RequestContextBuilder::extend(ctx)
-            .page_content_kind(PageContentKind::InMemoryLayer)
-            .build();
         for (key, vec_map) in keys.iter() {
             let key = **key;
             // Write all page versions
             for (lsn, pos) in vec_map.as_slice() {
-                cursor.read_blob_into_buf(*pos, &mut buf, &ctx).await?;
+                cursor.read_blob_into_buf(*pos, &mut buf).await?;
                 let will_init = Value::des(&buf)?.will_init();
                 delta_layer_writer
                     .put_value_bytes(key, *lsn, &buf, will_init)

pageserver/src/tenant/timeline.rs

@@ -38,7 +38,6 @@ use std::time::{Duration, Instant, SystemTime};
 use crate::context::{
     AccessStatsBehavior, DownloadBehavior, RequestContext, RequestContextBuilder,
 };
-use crate::deletion_queue::DeletionQueueClient;
 use crate::tenant::remote_timeline_client::index::LayerFileMetadata;
 use crate::tenant::storage_layer::delta_layer::DeltaEntry;
 use crate::tenant::storage_layer::{
@@ -144,7 +143,6 @@ fn drop_wlock<T>(rlock: tokio::sync::RwLockWriteGuard<'_, T>) {
 /// The outward-facing resources required to build a Timeline
 pub struct TimelineResources {
     pub remote_client: Option<RemoteTimelineClient>,
-    pub deletion_queue_client: DeletionQueueClient,
 }
 
 pub struct Timeline {
@@ -473,7 +471,7 @@ impl Timeline {
         // The cached image can be returned directly if there is no WAL between the cached image
         // and requested LSN. The cached image can also be used to reduce the amount of WAL needed
         // for redo.
-        let cached_page_img = match self.lookup_cached_page(&key, lsn, ctx).await {
+        let cached_page_img = match self.lookup_cached_page(&key, lsn).await {
             Some((cached_lsn, cached_img)) => {
                 match cached_lsn.cmp(&lsn) {
                     Ordering::Less => {} // there might be WAL between cached_lsn and lsn, we need to check
@@ -523,23 +521,9 @@ impl Timeline {
         self.disk_consistent_lsn.load()
     }
 
-    /// remote_consistent_lsn from the perspective of the tenant's current generation,
-    /// not validated with control plane yet.
-    /// See [`Self::get_remote_consistent_lsn_visible`].
-    pub fn get_remote_consistent_lsn_projected(&self) -> Option<Lsn> {
+    pub fn get_remote_consistent_lsn(&self) -> Option<Lsn> {
         if let Some(remote_client) = &self.remote_client {
-            remote_client.remote_consistent_lsn_projected()
-        } else {
-            None
-        }
-    }
-
-    /// remote_consistent_lsn which the tenant is guaranteed not to go backward from,
-    /// i.e. a value of remote_consistent_lsn_projected which has undergone
-    /// generation validation in the deletion queue.
-    pub fn get_remote_consistent_lsn_visible(&self) -> Option<Lsn> {
-        if let Some(remote_client) = &self.remote_client {
-            remote_client.remote_consistent_lsn_visible()
+            remote_client.last_uploaded_consistent_lsn()
         } else {
             None
         }
@@ -559,7 +543,7 @@ impl Timeline {
     }
 
     pub fn resident_physical_size(&self) -> u64 {
-        self.metrics.resident_physical_size_get()
+        self.metrics.resident_physical_size_gauge.get()
     }
 
     ///
@@ -1309,7 +1293,10 @@ impl Timeline {
         // will treat the file as a local layer again, count it towards resident size,
         // and it'll be like the layer removal never happened.
         // The bump in resident size is perhaps unexpected but overall a robust behavior.
-        self.metrics.resident_physical_size_sub(layer_file_size);
+        self.metrics
+            .resident_physical_size_gauge
+            .sub(layer_file_size);
+
         self.metrics.evictions.inc();
 
         if let Some(delta) = local_layer_residence_duration {
@@ -1833,7 +1820,7 @@ impl Timeline {
             for (layer, m) in needs_upload {
                 rtc.schedule_layer_file_upload(&layer.layer_desc().filename(), &m)?;
             }
-            rtc.schedule_layer_file_deletion(needs_cleanup)?;
+            rtc.schedule_layer_file_deletion(&needs_cleanup)?;
             rtc.schedule_index_upload_for_file_changes()?;
             // Tenant::create_timeline will wait for these uploads to happen before returning, or
             // on retry.
@@ -1843,7 +1830,9 @@ impl Timeline {
             "loaded layer map with {} layers at {}, total physical size: {}",
             num_layers, disk_consistent_lsn, total_physical_size
         );
-        self.metrics.resident_physical_size_set(total_physical_size);
+        self.metrics
+            .resident_physical_size_gauge
+            .set(total_physical_size);
 
         timer.stop_and_record();
         Ok(())
@@ -2529,18 +2518,13 @@ impl Timeline {
         }
     }
 
-    async fn lookup_cached_page(
-        &self,
-        key: &Key,
-        lsn: Lsn,
-        ctx: &RequestContext,
-    ) -> Option<(Lsn, Bytes)> {
+    async fn lookup_cached_page(&self, key: &Key, lsn: Lsn) -> Option<(Lsn, Bytes)> {
         let cache = page_cache::get();
 
         // FIXME: It's pointless to check the cache for things that are not 8kB pages.
         // We should look at the key to determine if it's a cacheable object
         let (lsn, read_guard) = cache
-            .lookup_materialized_page(self.tenant_id, self.timeline_id, key, lsn, ctx)
+            .lookup_materialized_page(self.tenant_id, self.timeline_id, key, lsn)
             .await?;
         let img = Bytes::from(read_guard.to_vec());
         Some((lsn, img))
@@ -2574,16 +2558,10 @@ impl Timeline {
         Ok(layer)
     }
 
-    async fn put_value(
-        &self,
-        key: Key,
-        lsn: Lsn,
-        val: &Value,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<()> {
+    async fn put_value(&self, key: Key, lsn: Lsn, val: &Value) -> anyhow::Result<()> {
         //info!("PUT: key {} at {}", key, lsn);
         let layer = self.get_layer_for_write(lsn).await?;
-        layer.put_value(key, lsn, val, ctx).await?;
+        layer.put_value(key, lsn, val).await?;
         Ok(())
     }
 
@@ -2755,7 +2733,7 @@ impl Timeline {
                 // Normal case, write out a L0 delta layer file.
                 // `create_delta_layer` will not modify the layer map.
                 // We will remove frozen layer and add delta layer in one atomic operation later.
-                let layer = self.create_delta_layer(&frozen_layer, ctx).await?;
+                let layer = self.create_delta_layer(&frozen_layer).await?;
                 (
                     HashMap::from([(
                         layer.filename(),
@@ -2878,21 +2856,19 @@ impl Timeline {
     async fn create_delta_layer(
         self: &Arc<Self>,
         frozen_layer: &Arc<InMemoryLayer>,
-        ctx: &RequestContext,
     ) -> anyhow::Result<DeltaLayer> {
         let span = tracing::info_span!("blocking");
         let new_delta: DeltaLayer = tokio::task::spawn_blocking({
             let _g = span.entered();
             let self_clone = Arc::clone(self);
             let frozen_layer = Arc::clone(frozen_layer);
-            let ctx = ctx.attached_child();
             move || {
                 // Write it out
                 // Keep this inside `spawn_blocking` and `Handle::current`
                 // as long as the write path is still sync and the read impl
                 // is still not fully async. Otherwise executor threads would
                 // be blocked.
-                let new_delta = Handle::current().block_on(frozen_layer.write_to_disk(&ctx))?;
+                let new_delta = Handle::current().block_on(frozen_layer.write_to_disk())?;
                 let new_delta_path = new_delta.path();
 
                 // Sync it to disk.
@@ -3598,7 +3574,7 @@ impl Timeline {
             key, lsn, ref val, ..
         } in all_values_iter
         {
-            let value = val.load(ctx).await?;
+            let value = val.load().await?;
             let same_key = prev_key.map_or(false, |prev_key| prev_key == key);
             // We need to check key boundaries once we reach next key or end of layer with the same key
             if !same_key || lsn == dup_end_lsn {
@@ -3886,7 +3862,7 @@ impl Timeline {
 
         // Also schedule the deletions in remote storage
         if let Some(remote_client) = &self.remote_client {
-            remote_client.schedule_layer_file_deletion(layer_names_to_delete)?;
+            remote_client.schedule_layer_file_deletion(&layer_names_to_delete)?;
         }
 
         Ok(())
@@ -4221,7 +4197,7 @@ impl Timeline {
             }
 
             if let Some(remote_client) = &self.remote_client {
-                remote_client.schedule_layer_file_deletion(layer_names_to_delete)?;
+                remote_client.schedule_layer_file_deletion(&layer_names_to_delete)?;
             }
 
             apply.flush();
@@ -4393,7 +4369,7 @@ impl Timeline {
 
                     // XXX the temp file is still around in Err() case
                     // and consumes space until we clean up upon pageserver restart.
-                    self_clone.metrics.resident_physical_size_add(*size);
+                    self_clone.metrics.resident_physical_size_gauge.add(*size);
 
                     // Download complete. Replace the RemoteLayer with the corresponding
                     // Delta- or ImageLayer in the layer map.
@@ -4723,14 +4699,8 @@ impl<'a> TimelineWriter<'a> {
     ///
     /// This will implicitly extend the relation, if the page is beyond the
     /// current end-of-file.
-    pub async fn put(
-        &self,
-        key: Key,
-        lsn: Lsn,
-        value: &Value,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<()> {
-        self.tl.put_value(key, lsn, value, ctx).await
+    pub async fn put(&self, key: Key, lsn: Lsn, value: &Value) -> anyhow::Result<()> {
+        self.tl.put_value(key, lsn, value).await
     }
 
     pub async fn delete(&self, key_range: Range<Key>, lsn: Lsn) -> anyhow::Result<()> {

pageserver/src/tenant/timeline/delete.rs

@@ -14,7 +14,6 @@ use utils::{
 
 use crate::{
     config::PageServerConf,
-    deletion_queue::DeletionQueueClient,
     task_mgr::{self, TaskKind},
     tenant::{
         metadata::TimelineMetadata,
@@ -408,7 +407,6 @@ impl DeleteTimelineFlow {
         timeline_id: TimelineId,
         local_metadata: &TimelineMetadata,
         remote_client: Option<RemoteTimelineClient>,
-        deletion_queue_client: DeletionQueueClient,
         init_order: Option<&InitializationOrder>,
     ) -> anyhow::Result<()> {
         // Note: here we even skip populating layer map. Timeline is essentially uninitialized.
@@ -418,10 +416,7 @@ impl DeleteTimelineFlow {
                 timeline_id,
                 local_metadata,
                 None, // Ancestor is not needed for deletion.
-                TimelineResources {
-                    remote_client,
-                    deletion_queue_client,
-                },
+                TimelineResources { remote_client },
                 init_order,
                 // Important. We dont pass ancestor above because it can be missing.
                 // Thus we need to skip the validation here.

pageserver/src/tenant/timeline/layer_manager.rs

@@ -263,7 +263,7 @@ impl LayerManager {
         let desc = layer.layer_desc();
         if !layer.is_remote_layer() {
             layer.delete_resident_layer_file()?;
-            metrics.resident_physical_size_sub(desc.file_size);
+            metrics.resident_physical_size_gauge.sub(desc.file_size);
         }
 
         // TODO Removing from the bottom of the layer map is expensive.

pageserver/src/tenant/timeline/walreceiver/walreceiver_connection.rs

@@ -370,9 +370,8 @@ pub(super) async fn handle_walreceiver_connection(
             })?;
 
         if let Some(last_lsn) = status_update {
-            let timeline_remote_consistent_lsn = timeline
-                .get_remote_consistent_lsn_visible()
-                .unwrap_or(Lsn(0));
+            let timeline_remote_consistent_lsn =
+                timeline.get_remote_consistent_lsn().unwrap_or(Lsn(0));
 
             // The last LSN we processed. It is not guaranteed to survive pageserver crash.
             let last_received_lsn = last_lsn;

pageserver/src/tenant/upload_queue.rs

@@ -1,3 +1,5 @@
+use crate::metrics::RemoteOpFileKind;
+
 use super::storage_layer::LayerFileName;
 use super::Generation;
 use crate::tenant::metadata::TimelineMetadata;
@@ -9,7 +11,6 @@ use std::fmt::Debug;
 use chrono::NaiveDateTime;
 use std::sync::Arc;
 use tracing::info;
-use utils::lsn::AtomicLsn;
 
 use std::sync::atomic::AtomicU32;
 use utils::lsn::Lsn;
@@ -57,12 +58,7 @@ pub(crate) struct UploadQueueInitialized {
     /// uploaded. `Lsn(0)` if nothing was uploaded yet.
     /// Unlike `latest_files` or `latest_metadata`, this value is never ahead.
     /// Safekeeper can rely on it to make decisions for WAL storage.
-    ///
-    /// visible_remote_consistent_lsn is only updated after our generation has been validated with
-    /// the control plane (unlesss a timeline's generation is None, in which case
-    /// we skip validation)
-    pub(crate) projected_remote_consistent_lsn: Option<Lsn>,
-    pub(crate) visible_remote_consistent_lsn: Arc<AtomicLsn>,
+    pub(crate) last_uploaded_consistent_lsn: Lsn,
 
     // Breakdown of different kinds of tasks currently in-progress
     pub(crate) num_inprogress_layer_uploads: usize,
@@ -85,14 +81,6 @@ impl UploadQueueInitialized {
     pub(super) fn no_pending_work(&self) -> bool {
         self.inprogress_tasks.is_empty() && self.queued_operations.is_empty()
     }
-
-    pub(super) fn get_last_remote_consistent_lsn_visible(&self) -> Lsn {
-        self.visible_remote_consistent_lsn.load()
-    }
-
-    pub(super) fn get_last_remote_consistent_lsn_projected(&self) -> Option<Lsn> {
-        self.projected_remote_consistent_lsn
-    }
 }
 
 #[derive(Clone, Copy)]
@@ -126,8 +114,9 @@ impl UploadQueue {
             latest_files: HashMap::new(),
             latest_files_changes_since_metadata_upload_scheduled: 0,
             latest_metadata: metadata.clone(),
-            projected_remote_consistent_lsn: None,
-            visible_remote_consistent_lsn: Arc::new(AtomicLsn::new(0)),
+            // We haven't uploaded anything yet, so, `last_uploaded_consistent_lsn` must be 0 to prevent
+            // safekeepers from garbage-collecting anything.
+            last_uploaded_consistent_lsn: Lsn(0),
             // what follows are boring default initializations
             task_counter: 0,
             num_inprogress_layer_uploads: 0,
@@ -169,10 +158,7 @@ impl UploadQueue {
             latest_files: files,
             latest_files_changes_since_metadata_upload_scheduled: 0,
             latest_metadata: index_part.metadata.clone(),
-            projected_remote_consistent_lsn: Some(index_part.metadata.disk_consistent_lsn()),
-            visible_remote_consistent_lsn: Arc::new(
-                index_part.metadata.disk_consistent_lsn().into(),
-            ),
+            last_uploaded_consistent_lsn: index_part.metadata.disk_consistent_lsn(),
             // what follows are boring default initializations
             task_counter: 0,
             num_inprogress_layer_uploads: 0,
@@ -215,11 +201,12 @@ pub(crate) struct UploadTask {
     pub(crate) op: UploadOp,
 }
 
-/// A deletion of some layers within the lifetime of a timeline.  This is not used
-/// for timeline deletion, which skips this queue and goes directly to DeletionQueue.
 #[derive(Debug)]
 pub(crate) struct Delete {
-    pub(crate) layers: Vec<(LayerFileName, Generation)>,
+    pub(crate) file_kind: RemoteOpFileKind,
+    pub(crate) layer_file_name: LayerFileName,
+    pub(crate) scheduled_from_timeline_delete: bool,
+    pub(crate) generation: Generation,
 }
 
 #[derive(Debug)]
@@ -230,7 +217,7 @@ pub(crate) enum UploadOp {
     /// Upload the metadata file
     UploadMetadata(IndexPart, Lsn),
 
-    /// Delete layer files
+    /// Delete a layer file
     Delete(Delete),
 
     /// Barrier. When the barrier operation is reached,
@@ -252,9 +239,13 @@ impl std::fmt::Display for UploadOp {
             UploadOp::UploadMetadata(_, lsn) => {
                 write!(f, "UploadMetadata(lsn: {})", lsn)
             }
-            UploadOp::Delete(delete) => {
-                write!(f, "Delete({} layers)", delete.layers.len(),)
-            }
+            UploadOp::Delete(delete) => write!(
+                f,
+                "Delete(path: {}, scheduled_from_timeline_delete: {}, gen: {:?})",
+                delete.layer_file_name.file_name(),
+                delete.scheduled_from_timeline_delete,
+                delete.generation
+            ),
             UploadOp::Barrier(_) => write!(f, "Barrier"),
         }
     }

pageserver/src/virtual_file.rs

@@ -650,12 +650,6 @@ mod tests {
         File(File),
     }
 
-    impl From<VirtualFile> for MaybeVirtualFile {
-        fn from(vf: VirtualFile) -> Self {
-            MaybeVirtualFile::VirtualFile(vf)
-        }
-    }
-
     impl MaybeVirtualFile {
         async fn read_exact_at(&self, buf: &mut [u8], offset: u64) -> Result<(), Error> {
             match self {
@@ -893,54 +887,4 @@ mod tests {
 
         Ok(())
     }
-
-    #[tokio::test]
-    async fn test_atomic_overwrite_basic() {
-        let testdir = crate::config::PageServerConf::test_repo_dir("test_atomic_overwrite_basic");
-        std::fs::create_dir_all(&testdir).unwrap();
-
-        let path = testdir.join("myfile");
-        let tmp_path = testdir.join("myfile.tmp");
-
-        VirtualFile::crashsafe_overwrite(&path, &tmp_path, b"foo")
-            .await
-            .unwrap();
-        let mut file = MaybeVirtualFile::from(VirtualFile::open(&path).await.unwrap());
-        let post = file.read_string().await.unwrap();
-        assert_eq!(post, "foo");
-        assert!(!tmp_path.exists());
-        drop(file);
-
-        VirtualFile::crashsafe_overwrite(&path, &tmp_path, b"bar")
-            .await
-            .unwrap();
-        let mut file = MaybeVirtualFile::from(VirtualFile::open(&path).await.unwrap());
-        let post = file.read_string().await.unwrap();
-        assert_eq!(post, "bar");
-        assert!(!tmp_path.exists());
-        drop(file);
-    }
-
-    #[tokio::test]
-    async fn test_atomic_overwrite_preexisting_tmp() {
-        let testdir =
-            crate::config::PageServerConf::test_repo_dir("test_atomic_overwrite_preexisting_tmp");
-        std::fs::create_dir_all(&testdir).unwrap();
-
-        let path = testdir.join("myfile");
-        let tmp_path = testdir.join("myfile.tmp");
-
-        std::fs::write(&tmp_path, "some preexisting junk that should be removed").unwrap();
-        assert!(tmp_path.exists());
-
-        VirtualFile::crashsafe_overwrite(&path, &tmp_path, b"foo")
-            .await
-            .unwrap();
-
-        let mut file = MaybeVirtualFile::from(VirtualFile::open(&path).await.unwrap());
-        let post = file.read_string().await.unwrap();
-        assert_eq!(post, "foo");
-        assert!(!tmp_path.exists());
-        drop(file);
-    }
 }

pageserver/src/walingest.rs

@@ -151,56 +151,50 @@ impl<'a> WalIngest<'a> {
                     }
                 }
             } else if self.timeline.pg_version == 15 {
-                if (decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK)
-                    == postgres_ffi::v15::bindings::XLOG_DBASE_CREATE_WAL_LOG
-                {
-                    debug!("XLOG_DBASE_CREATE_WAL_LOG: noop");
-                } else if (decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK)
-                    == postgres_ffi::v15::bindings::XLOG_DBASE_CREATE_FILE_COPY
-                {
-                    // The XLOG record was renamed between v14 and v15,
-                    // but the record format is the same.
-                    // So we can reuse XlCreateDatabase here.
-                    debug!("XLOG_DBASE_CREATE_FILE_COPY");
-                    let createdb = XlCreateDatabase::decode(&mut buf);
-                    self.ingest_xlog_dbase_create(modification, &createdb, ctx)
-                        .await?;
-                } else if (decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK)
-                    == postgres_ffi::v15::bindings::XLOG_DBASE_DROP
-                {
-                    let dropdb = XlDropDatabase::decode(&mut buf);
-                    for tablespace_id in dropdb.tablespace_ids {
-                        trace!("Drop db {}, {}", tablespace_id, dropdb.db_id);
-                        modification
-                            .drop_dbdir(tablespace_id, dropdb.db_id, ctx)
+                match decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK {
+                    postgres_ffi::v15::bindings::XLOG_DBASE_CREATE_WAL_LOG
+                    | postgres_ffi::v15::bindings::XLOG_DBASE_CREATE_FILE_COPY => {
+                        // The XLOG record was renamed between v14 and v15,
+                        // but the record format is the same.
+                        // So we can reuse XlCreateDatabase here.
+                        debug!("XLOG_DBASE_CREATE_WAL_LOG");
+                        let createdb = XlCreateDatabase::decode(&mut buf);
+                        self.ingest_xlog_dbase_create(modification, &createdb, ctx)
                             .await?;
                     }
+                    postgres_ffi::v15::bindings::XLOG_DBASE_DROP => {
+                        let dropdb = XlDropDatabase::decode(&mut buf);
+                        for tablespace_id in dropdb.tablespace_ids {
+                            trace!("Drop db {}, {}", tablespace_id, dropdb.db_id);
+                            modification
+                                .drop_dbdir(tablespace_id, dropdb.db_id, ctx)
+                                .await?;
+                        }
+                    }
+                    _ => {}
                 }
             } else if self.timeline.pg_version == 16 {
-                if (decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK)
-                    == postgres_ffi::v16::bindings::XLOG_DBASE_CREATE_WAL_LOG
-                {
-                    debug!("XLOG_DBASE_CREATE_WAL_LOG: noop");
-                } else if (decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK)
-                    == postgres_ffi::v16::bindings::XLOG_DBASE_CREATE_FILE_COPY
-                {
-                    // The XLOG record was renamed between v14 and v15,
-                    // but the record format is the same.
-                    // So we can reuse XlCreateDatabase here.
-                    debug!("XLOG_DBASE_CREATE_FILE_COPY");
-                    let createdb = XlCreateDatabase::decode(&mut buf);
-                    self.ingest_xlog_dbase_create(modification, &createdb, ctx)
-                        .await?;
-                } else if (decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK)
-                    == postgres_ffi::v16::bindings::XLOG_DBASE_DROP
-                {
-                    let dropdb = XlDropDatabase::decode(&mut buf);
-                    for tablespace_id in dropdb.tablespace_ids {
-                        trace!("Drop db {}, {}", tablespace_id, dropdb.db_id);
-                        modification
-                            .drop_dbdir(tablespace_id, dropdb.db_id, ctx)
+                match decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK {
+                    postgres_ffi::v16::bindings::XLOG_DBASE_CREATE_WAL_LOG
+                    | postgres_ffi::v16::bindings::XLOG_DBASE_CREATE_FILE_COPY => {
+                        // The XLOG record was renamed between v14 and v15,
+                        // but the record format is the same.
+                        // So we can reuse XlCreateDatabase here.
+                        debug!("XLOG_DBASE_CREATE_WAL_LOG");
+                        let createdb = XlCreateDatabase::decode(&mut buf);
+                        self.ingest_xlog_dbase_create(modification, &createdb, ctx)
                             .await?;
                     }
+                    postgres_ffi::v16::bindings::XLOG_DBASE_DROP => {
+                        let dropdb = XlDropDatabase::decode(&mut buf);
+                        for tablespace_id in dropdb.tablespace_ids {
+                            trace!("Drop db {}, {}", tablespace_id, dropdb.db_id);
+                            modification
+                                .drop_dbdir(tablespace_id, dropdb.db_id, ctx)
+                                .await?;
+                        }
+                    }
+                    _ => {}
                 }
             }
         } else if decoded.xl_rmid == pg_constants::RM_TBLSPC_ID {
@@ -363,7 +357,7 @@ impl<'a> WalIngest<'a> {
 
         // Now that this record has been fully handled, including updating the
         // checkpoint data, let the repository know that it is up-to-date to this LSN
-        modification.commit(ctx).await?;
+        modification.commit().await?;
 
         Ok(())
     }
@@ -1561,7 +1555,7 @@ mod tests {
         let mut m = tline.begin_modification(Lsn(0x10));
         m.put_checkpoint(ZERO_CHECKPOINT.clone())?;
         m.put_relmap_file(0, 111, Bytes::from(""), ctx).await?; // dummy relmapper file
-        m.commit(ctx).await?;
+        m.commit().await?;
         let walingest = WalIngest::new(tline, Lsn(0x10), ctx).await?;
 
         Ok(walingest)
@@ -1580,22 +1574,22 @@ mod tests {
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 0, TEST_IMG("foo blk 0 at 2"), &ctx)
             .await?;
-        m.commit(&ctx).await?;
+        m.commit().await?;
         let mut m = tline.begin_modification(Lsn(0x30));
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 0, TEST_IMG("foo blk 0 at 3"), &ctx)
             .await?;
-        m.commit(&ctx).await?;
+        m.commit().await?;
         let mut m = tline.begin_modification(Lsn(0x40));
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 1, TEST_IMG("foo blk 1 at 4"), &ctx)
             .await?;
-        m.commit(&ctx).await?;
+        m.commit().await?;
         let mut m = tline.begin_modification(Lsn(0x50));
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 2, TEST_IMG("foo blk 2 at 5"), &ctx)
             .await?;
-        m.commit(&ctx).await?;
+        m.commit().await?;
 
         assert_current_logical_size(&tline, Lsn(0x50));
 
@@ -1681,7 +1675,7 @@ mod tests {
         walingest
             .put_rel_truncation(&mut m, TESTREL_A, 2, &ctx)
             .await?;
-        m.commit(&ctx).await?;
+        m.commit().await?;
         assert_current_logical_size(&tline, Lsn(0x60));
 
         // Check reported size and contents after truncation
@@ -1723,7 +1717,7 @@ mod tests {
         walingest
             .put_rel_truncation(&mut m, TESTREL_A, 0, &ctx)
             .await?;
-        m.commit(&ctx).await?;
+        m.commit().await?;
         assert_eq!(
             tline
                 .get_rel_size(TESTREL_A, Lsn(0x68), false, &ctx)
@@ -1736,7 +1730,7 @@ mod tests {
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 1, TEST_IMG("foo blk 1"), &ctx)
             .await?;
-        m.commit(&ctx).await?;
+        m.commit().await?;
         assert_eq!(
             tline
                 .get_rel_size(TESTREL_A, Lsn(0x70), false, &ctx)
@@ -1761,7 +1755,7 @@ mod tests {
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 1500, TEST_IMG("foo blk 1500"), &ctx)
             .await?;
-        m.commit(&ctx).await?;
+        m.commit().await?;
         assert_eq!(
             tline
                 .get_rel_size(TESTREL_A, Lsn(0x80), false, &ctx)
@@ -1800,7 +1794,7 @@ mod tests {
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 0, TEST_IMG("foo blk 0 at 2"), &ctx)
             .await?;
-        m.commit(&ctx).await?;
+        m.commit().await?;
 
         // Check that rel exists and size is correct
         assert_eq!(
@@ -1819,7 +1813,7 @@ mod tests {
         // Drop rel
         let mut m = tline.begin_modification(Lsn(0x30));
         walingest.put_rel_drop(&mut m, TESTREL_A, &ctx).await?;
-        m.commit(&ctx).await?;
+        m.commit().await?;
 
         // Check that rel is not visible anymore
         assert_eq!(
@@ -1837,7 +1831,7 @@ mod tests {
         walingest
             .put_rel_page_image(&mut m, TESTREL_A, 0, TEST_IMG("foo blk 0 at 4"), &ctx)
             .await?;
-        m.commit(&ctx).await?;
+        m.commit().await?;
 
         // Check that rel exists and size is correct
         assert_eq!(
@@ -1876,7 +1870,7 @@ mod tests {
                 .put_rel_page_image(&mut m, TESTREL_A, blkno, TEST_IMG(&data), &ctx)
                 .await?;
         }
-        m.commit(&ctx).await?;
+        m.commit().await?;
 
         // The relation was created at LSN 20, not visible at LSN 1 yet.
         assert_eq!(
@@ -1921,7 +1915,7 @@ mod tests {
         walingest
             .put_rel_truncation(&mut m, TESTREL_A, 1, &ctx)
             .await?;
-        m.commit(&ctx).await?;
+        m.commit().await?;
 
         // Check reported size and contents after truncation
         assert_eq!(
@@ -1970,7 +1964,7 @@ mod tests {
                 .put_rel_page_image(&mut m, TESTREL_A, blkno, TEST_IMG(&data), &ctx)
                 .await?;
         }
-        m.commit(&ctx).await?;
+        m.commit().await?;
 
         assert_eq!(
             tline
@@ -2017,7 +2011,7 @@ mod tests {
             walingest
                 .put_rel_page_image(&mut m, TESTREL_A, blknum as BlockNumber, img, &ctx)
                 .await?;
-            m.commit(&ctx).await?;
+            m.commit().await?;
         }
 
         assert_current_logical_size(&tline, Lsn(lsn));
@@ -2033,7 +2027,7 @@ mod tests {
         walingest
             .put_rel_truncation(&mut m, TESTREL_A, RELSEG_SIZE, &ctx)
             .await?;
-        m.commit(&ctx).await?;
+        m.commit().await?;
         assert_eq!(
             tline.get_rel_size(TESTREL_A, Lsn(lsn), false, &ctx).await?,
             RELSEG_SIZE
@@ -2046,7 +2040,7 @@ mod tests {
         walingest
             .put_rel_truncation(&mut m, TESTREL_A, RELSEG_SIZE - 1, &ctx)
             .await?;
-        m.commit(&ctx).await?;
+        m.commit().await?;
         assert_eq!(
             tline.get_rel_size(TESTREL_A, Lsn(lsn), false, &ctx).await?,
             RELSEG_SIZE - 1
@@ -2062,7 +2056,7 @@ mod tests {
             walingest
                 .put_rel_truncation(&mut m, TESTREL_A, size as BlockNumber, &ctx)
                 .await?;
-            m.commit(&ctx).await?;
+            m.commit().await?;
             assert_eq!(
                 tline.get_rel_size(TESTREL_A, Lsn(lsn), false, &ctx).await?,
                 size as BlockNumber

pgxn/hnsw/hnsw.control

@@ -2,3 +2,4 @@ comment = '** Deprecated ** Please use pg_embedding instead'
 default_version = '0.1.0'
 module_pathname = '$libdir/hnsw'
 relocatable = true
+trusted = true

pgxn/neon/file_cache.c

@@ -222,9 +222,8 @@ lfc_change_limit_hook(int newval, void *extra)
 	/*
 	 * Stats collector detach shared memory, so we should not try to access shared memory here.
 	 * Parallel workers first assign default value (0), so not perform truncation in parallel workers.
-	 * The Postmaster can handle SIGHUP and it has access to shared memory (UsedShmemSegAddr != NULL), but has no PGPROC.
 	 */
-	if (!lfc_ctl || !MyProc || !UsedShmemSegAddr || IsParallelWorker())
+	if (!lfc_ctl || !UsedShmemSegAddr || IsParallelWorker())
 		return;
 
 	/* Open cache file if not done yet */

pgxn/neon/pagestore_smgr.c

@@ -1790,14 +1790,6 @@ neon_zeroextend(SMgrRelation reln, ForkNumber forkNum, BlockNumber blocknum,
 	if (!XLogInsertAllowed())
 		return;
 
-	/* ensure we have enough xlog buffers to log max-sized records */
-	XLogEnsureRecordSpace(Min(remblocks, (XLR_MAX_BLOCK_ID - 1)), 0);
-
-	/*
-	 * Iterate over all the pages. They are collected into batches of
-	 * XLR_MAX_BLOCK_ID pages, and a single WAL-record is written for each
-	 * batch.
-	 */
 	while (remblocks > 0)
 	{
 		int			count = Min(remblocks, XLR_MAX_BLOCK_ID);

poetry.lock

@@ -1,4 +1,4 @@
-# This file is automatically @generated by Poetry 1.6.1 and should not be changed by hand.
+# This file is automatically @generated by Poetry 1.5.1 and should not be changed by hand.
 
 [[package]]
 name = "aiohttp"
@@ -887,34 +887,34 @@ files = [
 
 [[package]]
 name = "cryptography"
-version = "41.0.4"
+version = "41.0.3"
 description = "cryptography is a package which provides cryptographic recipes and primitives to Python developers."
 optional = false
 python-versions = ">=3.7"
 files = [
-    {file = "cryptography-41.0.4-cp37-abi3-macosx_10_12_universal2.whl", hash = "sha256:80907d3faa55dc5434a16579952ac6da800935cd98d14dbd62f6f042c7f5e839"},
-    {file = "cryptography-41.0.4-cp37-abi3-macosx_10_12_x86_64.whl", hash = "sha256:35c00f637cd0b9d5b6c6bd11b6c3359194a8eba9c46d4e875a3660e3b400005f"},
-    {file = "cryptography-41.0.4-cp37-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:cecfefa17042941f94ab54f769c8ce0fe14beff2694e9ac684176a2535bf9714"},
-    {file = "cryptography-41.0.4-cp37-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:e40211b4923ba5a6dc9769eab704bdb3fbb58d56c5b336d30996c24fcf12aadb"},
-    {file = "cryptography-41.0.4-cp37-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:23a25c09dfd0d9f28da2352503b23e086f8e78096b9fd585d1d14eca01613e13"},
-    {file = "cryptography-41.0.4-cp37-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:2ed09183922d66c4ec5fdaa59b4d14e105c084dd0febd27452de8f6f74704143"},
-    {file = "cryptography-41.0.4-cp37-abi3-musllinux_1_1_aarch64.whl", hash = "sha256:5a0f09cefded00e648a127048119f77bc2b2ec61e736660b5789e638f43cc397"},
-    {file = "cryptography-41.0.4-cp37-abi3-musllinux_1_1_x86_64.whl", hash = "sha256:9eeb77214afae972a00dee47382d2591abe77bdae166bda672fb1e24702a3860"},
-    {file = "cryptography-41.0.4-cp37-abi3-win32.whl", hash = "sha256:3b224890962a2d7b57cf5eeb16ccaafba6083f7b811829f00476309bce2fe0fd"},
-    {file = "cryptography-41.0.4-cp37-abi3-win_amd64.whl", hash = "sha256:c880eba5175f4307129784eca96f4e70b88e57aa3f680aeba3bab0e980b0f37d"},
-    {file = "cryptography-41.0.4-pp310-pypy310_pp73-macosx_10_12_x86_64.whl", hash = "sha256:004b6ccc95943f6a9ad3142cfabcc769d7ee38a3f60fb0dddbfb431f818c3a67"},
-    {file = "cryptography-41.0.4-pp310-pypy310_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:86defa8d248c3fa029da68ce61fe735432b047e32179883bdb1e79ed9bb8195e"},
-    {file = "cryptography-41.0.4-pp310-pypy310_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:37480760ae08065437e6573d14be973112c9e6dcaf5f11d00147ee74f37a3829"},
-    {file = "cryptography-41.0.4-pp310-pypy310_pp73-win_amd64.whl", hash = "sha256:b5f4dfe950ff0479f1f00eda09c18798d4f49b98f4e2006d644b3301682ebdca"},
-    {file = "cryptography-41.0.4-pp38-pypy38_pp73-macosx_10_12_x86_64.whl", hash = "sha256:7e53db173370dea832190870e975a1e09c86a879b613948f09eb49324218c14d"},
-    {file = "cryptography-41.0.4-pp38-pypy38_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:5b72205a360f3b6176485a333256b9bcd48700fc755fef51c8e7e67c4b63e3ac"},
-    {file = "cryptography-41.0.4-pp38-pypy38_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:93530900d14c37a46ce3d6c9e6fd35dbe5f5601bf6b3a5c325c7bffc030344d9"},
-    {file = "cryptography-41.0.4-pp38-pypy38_pp73-win_amd64.whl", hash = "sha256:efc8ad4e6fc4f1752ebfb58aefece8b4e3c4cae940b0994d43649bdfce8d0d4f"},
-    {file = "cryptography-41.0.4-pp39-pypy39_pp73-macosx_10_12_x86_64.whl", hash = "sha256:c3391bd8e6de35f6f1140e50aaeb3e2b3d6a9012536ca23ab0d9c35ec18c8a91"},
-    {file = "cryptography-41.0.4-pp39-pypy39_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:0d9409894f495d465fe6fda92cb70e8323e9648af912d5b9141d616df40a87b8"},
-    {file = "cryptography-41.0.4-pp39-pypy39_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:8ac4f9ead4bbd0bc8ab2d318f97d85147167a488be0e08814a37eb2f439d5cf6"},
-    {file = "cryptography-41.0.4-pp39-pypy39_pp73-win_amd64.whl", hash = "sha256:047c4603aeb4bbd8db2756e38f5b8bd7e94318c047cfe4efeb5d715e08b49311"},
-    {file = "cryptography-41.0.4.tar.gz", hash = "sha256:7febc3094125fc126a7f6fb1f420d0da639f3f32cb15c8ff0dc3997c4549f51a"},
+    {file = "cryptography-41.0.3-cp37-abi3-macosx_10_12_universal2.whl", hash = "sha256:652627a055cb52a84f8c448185922241dd5217443ca194d5739b44612c5e6507"},
+    {file = "cryptography-41.0.3-cp37-abi3-macosx_10_12_x86_64.whl", hash = "sha256:8f09daa483aedea50d249ef98ed500569841d6498aa9c9f4b0531b9964658922"},
+    {file = "cryptography-41.0.3-cp37-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:4fd871184321100fb400d759ad0cddddf284c4b696568204d281c902fc7b0d81"},
+    {file = "cryptography-41.0.3-cp37-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:84537453d57f55a50a5b6835622ee405816999a7113267739a1b4581f83535bd"},
+    {file = "cryptography-41.0.3-cp37-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:3fb248989b6363906827284cd20cca63bb1a757e0a2864d4c1682a985e3dca47"},
+    {file = "cryptography-41.0.3-cp37-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:42cb413e01a5d36da9929baa9d70ca90d90b969269e5a12d39c1e0d475010116"},
+    {file = "cryptography-41.0.3-cp37-abi3-musllinux_1_1_aarch64.whl", hash = "sha256:aeb57c421b34af8f9fe830e1955bf493a86a7996cc1338fe41b30047d16e962c"},
+    {file = "cryptography-41.0.3-cp37-abi3-musllinux_1_1_x86_64.whl", hash = "sha256:6af1c6387c531cd364b72c28daa29232162010d952ceb7e5ca8e2827526aceae"},
+    {file = "cryptography-41.0.3-cp37-abi3-win32.whl", hash = "sha256:0d09fb5356f975974dbcb595ad2d178305e5050656affb7890a1583f5e02a306"},
+    {file = "cryptography-41.0.3-cp37-abi3-win_amd64.whl", hash = "sha256:a983e441a00a9d57a4d7c91b3116a37ae602907a7618b882c8013b5762e80574"},
+    {file = "cryptography-41.0.3-pp310-pypy310_pp73-macosx_10_12_x86_64.whl", hash = "sha256:5259cb659aa43005eb55a0e4ff2c825ca111a0da1814202c64d28a985d33b087"},
+    {file = "cryptography-41.0.3-pp310-pypy310_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:67e120e9a577c64fe1f611e53b30b3e69744e5910ff3b6e97e935aeb96005858"},
+    {file = "cryptography-41.0.3-pp310-pypy310_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:7efe8041897fe7a50863e51b77789b657a133c75c3b094e51b5e4b5cec7bf906"},
+    {file = "cryptography-41.0.3-pp310-pypy310_pp73-win_amd64.whl", hash = "sha256:ce785cf81a7bdade534297ef9e490ddff800d956625020ab2ec2780a556c313e"},
+    {file = "cryptography-41.0.3-pp38-pypy38_pp73-macosx_10_12_x86_64.whl", hash = "sha256:57a51b89f954f216a81c9d057bf1a24e2f36e764a1ca9a501a6964eb4a6800dd"},
+    {file = "cryptography-41.0.3-pp38-pypy38_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:4c2f0d35703d61002a2bbdcf15548ebb701cfdd83cdc12471d2bae80878a4207"},
+    {file = "cryptography-41.0.3-pp38-pypy38_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:23c2d778cf829f7d0ae180600b17e9fceea3c2ef8b31a99e3c694cbbf3a24b84"},
+    {file = "cryptography-41.0.3-pp38-pypy38_pp73-win_amd64.whl", hash = "sha256:95dd7f261bb76948b52a5330ba5202b91a26fbac13ad0e9fc8a3ac04752058c7"},
+    {file = "cryptography-41.0.3-pp39-pypy39_pp73-macosx_10_12_x86_64.whl", hash = "sha256:41d7aa7cdfded09b3d73a47f429c298e80796c8e825ddfadc84c8a7f12df212d"},
+    {file = "cryptography-41.0.3-pp39-pypy39_pp73-manylinux_2_28_aarch64.whl", hash = "sha256:d0d651aa754ef58d75cec6edfbd21259d93810b73f6ec246436a21b7841908de"},
+    {file = "cryptography-41.0.3-pp39-pypy39_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:ab8de0d091acbf778f74286f4989cf3d1528336af1b59f3e5d2ebca8b5fe49e1"},
+    {file = "cryptography-41.0.3-pp39-pypy39_pp73-win_amd64.whl", hash = "sha256:a74fbcdb2a0d46fe00504f571a2a540532f4c188e6ccf26f1f178480117b33c4"},
+    {file = "cryptography-41.0.3.tar.gz", hash = "sha256:6d192741113ef5e30d89dcb5b956ef4e1578f304708701b8b73d38e3e1461f34"},
 ]
 
 [package.dependencies]

proxy/Cargo.toml

@@ -42,7 +42,6 @@ reqwest-middleware.workspace = true
 reqwest-retry.workspace = true
 reqwest-tracing.workspace = true
 routerify.workspace = true
-rustc-hash.workspace = true
 rustls-pemfile.workspace = true
 rustls.workspace = true
 scopeguard.workspace = true

proxy/src/http/conn_pool.rs

@@ -17,12 +17,11 @@ use std::{
 use tokio::time;
 use tokio_postgres::AsyncMessage;
 
-use crate::{
-    auth, console,
-    metrics::{Ids, MetricCounter, USAGE_METRICS},
-};
+use crate::{auth, console};
 use crate::{compute, config};
 
+use super::sql_over_http::MAX_RESPONSE_SIZE;
+
 use crate::proxy::ConnectMechanism;
 
 use tracing::{error, warn};
@@ -401,6 +400,7 @@ async fn connect_to_compute_once(
         .user(&conn_info.username)
         .password(&conn_info.password)
         .dbname(&conn_info.dbname)
+        .max_backend_message_size(MAX_RESPONSE_SIZE)
         .connect_timeout(timeout)
         .connect(tokio_postgres::NoTls)
         .await?;
@@ -412,10 +412,6 @@ async fn connect_to_compute_once(
     span.in_scope(|| {
         info!(%conn_info, %session, "new connection");
     });
-    let ids = Ids {
-        endpoint_id: node_info.aux.endpoint_id.to_string(),
-        branch_id: node_info.aux.branch_id.to_string(),
-    };
 
     tokio::spawn(
         poll_fn(move |cx| {
@@ -454,18 +450,10 @@ async fn connect_to_compute_once(
     Ok(Client {
         inner: client,
         session: tx,
-        ids,
     })
 }
 
 pub struct Client {
     pub inner: tokio_postgres::Client,
     session: tokio::sync::watch::Sender<uuid::Uuid>,
-    ids: Ids,
-}
-
-impl Client {
-    pub fn metrics(&self) -> Arc<MetricCounter> {
-        USAGE_METRICS.register(self.ids.clone())
-    }
 }

proxy/src/http/sql_over_http.rs

@@ -3,12 +3,10 @@ use std::sync::Arc;
 use anyhow::bail;
 use futures::pin_mut;
 use futures::StreamExt;
+use hashbrown::HashMap;
 use hyper::body::HttpBody;
-use hyper::header;
 use hyper::http::HeaderName;
 use hyper::http::HeaderValue;
-use hyper::Response;
-use hyper::StatusCode;
 use hyper::{Body, HeaderMap, Request};
 use serde_json::json;
 use serde_json::Map;
@@ -18,11 +16,7 @@ use tokio_postgres::types::Type;
 use tokio_postgres::GenericClient;
 use tokio_postgres::IsolationLevel;
 use tokio_postgres::Row;
-use tracing::error;
-use tracing::instrument;
 use url::Url;
-use utils::http::error::ApiError;
-use utils::http::json::json_response;
 
 use super::conn_pool::ConnInfo;
 use super::conn_pool::GlobalConnPool;
@@ -45,6 +39,7 @@ enum Payload {
     Batch(BatchQueryData),
 }
 
+pub const MAX_RESPONSE_SIZE: usize = 10 * 1024 * 1024; // 10 MB
 const MAX_REQUEST_SIZE: u64 = 1024 * 1024; // 1 MB
 
 static RAW_TEXT_OUTPUT: HeaderName = HeaderName::from_static("neon-raw-text-output");
@@ -187,45 +182,7 @@ pub async fn handle(
     sni_hostname: Option<String>,
     conn_pool: Arc<GlobalConnPool>,
     session_id: uuid::Uuid,
-) -> Result<Response<Body>, ApiError> {
-    let result = handle_inner(request, sni_hostname, conn_pool, session_id).await;
-
-    let mut response = match result {
-        Ok(r) => r,
-        Err(e) => {
-            let message = format!("{:?}", e);
-            let code = match e.downcast_ref::<tokio_postgres::Error>() {
-                Some(e) => match e.code() {
-                    Some(e) => serde_json::to_value(e.code()).unwrap(),
-                    None => Value::Null,
-                },
-                None => Value::Null,
-            };
-            error!(
-                ?code,
-                "sql-over-http per-client task finished with an error: {e:#}"
-            );
-            // TODO: this shouldn't always be bad request.
-            json_response(
-                StatusCode::BAD_REQUEST,
-                json!({ "message": message, "code": code }),
-            )?
-        }
-    };
-    response.headers_mut().insert(
-        "Access-Control-Allow-Origin",
-        hyper::http::HeaderValue::from_static("*"),
-    );
-    Ok(response)
-}
-
-#[instrument(name = "sql-over-http", skip_all)]
-async fn handle_inner(
-    request: Request<Body>,
-    sni_hostname: Option<String>,
-    conn_pool: Arc<GlobalConnPool>,
-    session_id: uuid::Uuid,
-) -> anyhow::Result<Response<Body>> {
+) -> anyhow::Result<(Value, HashMap<HeaderName, HeaderValue>)> {
     //
     // Determine the destination and connection params
     //
@@ -276,18 +233,13 @@ async fn handle_inner(
 
     let mut client = conn_pool.get(&conn_info, !allow_pool, session_id).await?;
 
-    let mut response = Response::builder()
-        .status(StatusCode::OK)
-        .header(header::CONTENT_TYPE, "application/json");
-
     //
     // Now execute the query and return the result
     //
-    let mut size = 0;
     let result = match payload {
-        Payload::Single(query) => {
-            query_to_json(&client.inner, query, &mut size, raw_output, array_mode).await
-        }
+        Payload::Single(query) => query_to_json(&client.inner, query, raw_output, array_mode)
+            .await
+            .map(|x| (x, HashMap::default())),
         Payload::Batch(batch_query) => {
             let mut results = Vec::new();
             let mut builder = client.inner.build_transaction();
@@ -302,8 +254,7 @@ async fn handle_inner(
             }
             let transaction = builder.start().await?;
             for query in batch_query.queries {
-                let result =
-                    query_to_json(&transaction, query, &mut size, raw_output, array_mode).await;
+                let result = query_to_json(&transaction, query, raw_output, array_mode).await;
                 match result {
                     Ok(r) => results.push(r),
                     Err(e) => {
@@ -313,27 +264,26 @@ async fn handle_inner(
                 }
             }
             transaction.commit().await?;
+            let mut headers = HashMap::default();
             if txn_read_only {
-                response = response.header(
+                headers.insert(
                     TXN_READ_ONLY.clone(),
                     HeaderValue::try_from(txn_read_only.to_string())?,
                 );
             }
             if txn_deferrable {
-                response = response.header(
+                headers.insert(
                     TXN_DEFERRABLE.clone(),
                     HeaderValue::try_from(txn_deferrable.to_string())?,
                 );
             }
             if let Some(txn_isolation_level) = txn_isolation_level_raw {
-                response = response.header(TXN_ISOLATION_LEVEL.clone(), txn_isolation_level);
+                headers.insert(TXN_ISOLATION_LEVEL.clone(), txn_isolation_level);
             }
-            Ok(json!({ "results": results }))
+            Ok((json!({ "results": results }), headers))
         }
     };
 
-    let metrics = client.metrics();
-
     if allow_pool {
         let current_span = tracing::Span::current();
         // return connection to the pool
@@ -343,30 +293,12 @@ async fn handle_inner(
         });
     }
 
-    match result {
-        Ok(value) => {
-            // how could this possibly fail
-            let body = serde_json::to_string(&value).expect("json serialization should not fail");
-            let len = body.len();
-            let response = response
-                .body(Body::from(body))
-                // only fails if invalid status code or invalid header/values are given.
-                // these are not user configurable so it cannot fail dynamically
-                .expect("building response payload should not fail");
-
-            // count the egress bytes - we miss the TLS and header overhead but oh well...
-            // moving this later in the stack is going to be a lot of effort and ehhhh
-            metrics.record_egress(len as u64);
-            Ok(response)
-        }
-        Err(e) => Err(e),
-    }
+    result
 }
 
 async fn query_to_json<T: GenericClient>(
     client: &T,
     data: QueryData,
-    current_size: &mut usize,
     raw_output: bool,
     array_mode: bool,
 ) -> anyhow::Result<Value> {
@@ -380,10 +312,16 @@ async fn query_to_json<T: GenericClient>(
     // big.
     pin_mut!(row_stream);
     let mut rows: Vec<tokio_postgres::Row> = Vec::new();
+    let mut current_size = 0;
     while let Some(row) = row_stream.next().await {
         let row = row?;
-        *current_size += row.body_len();
+        current_size += row.body_len();
         rows.push(row);
+        if current_size > MAX_RESPONSE_SIZE {
+            return Err(anyhow::anyhow!(
+                "response is too large (max is {MAX_RESPONSE_SIZE} bytes)"
+            ));
+        }
     }
 
     // grab the command tag and number of rows affected

proxy/src/http/websocket.rs

@@ -7,6 +7,7 @@ use crate::{
 };
 use bytes::{Buf, Bytes};
 use futures::{Sink, Stream, StreamExt};
+use hashbrown::HashMap;
 use hyper::{
     server::{
         accept,
@@ -17,6 +18,7 @@ use hyper::{
 };
 use hyper_tungstenite::{tungstenite::Message, HyperWebsocket, WebSocketStream};
 use pin_project_lite::pin_project;
+use serde_json::{json, Value};
 
 use std::{
     convert::Infallible,
@@ -202,7 +204,44 @@ async fn ws_handler(
     // TODO: that deserves a refactor as now this function also handles http json client besides websockets.
     // Right now I don't want to blow up sql-over-http patch with file renames and do that as a follow up instead.
     } else if request.uri().path() == "/sql" && request.method() == Method::POST {
-        sql_over_http::handle(request, sni_hostname, conn_pool, session_id).await
+        let result = sql_over_http::handle(request, sni_hostname, conn_pool, session_id)
+            .instrument(info_span!("sql-over-http"))
+            .await;
+        let status_code = match result {
+            Ok(_) => StatusCode::OK,
+            Err(_) => StatusCode::BAD_REQUEST,
+        };
+        let (json, headers) = match result {
+            Ok(r) => r,
+            Err(e) => {
+                let message = format!("{:?}", e);
+                let code = match e.downcast_ref::<tokio_postgres::Error>() {
+                    Some(e) => match e.code() {
+                        Some(e) => serde_json::to_value(e.code()).unwrap(),
+                        None => Value::Null,
+                    },
+                    None => Value::Null,
+                };
+                error!(
+                    ?code,
+                    "sql-over-http per-client task finished with an error: {e:#}"
+                );
+                (
+                    json!({ "message": message, "code": code }),
+                    HashMap::default(),
+                )
+            }
+        };
+        json_response(status_code, json).map(|mut r| {
+            r.headers_mut().insert(
+                "Access-Control-Allow-Origin",
+                hyper::http::HeaderValue::from_static("*"),
+            );
+            for (k, v) in headers {
+                r.headers_mut().insert(k, v);
+            }
+            r
+        })
     } else if request.uri().path() == "/sql" && request.method() == Method::OPTIONS {
         Response::builder()
             .header("Allow", "OPTIONS, POST")
@@ -214,7 +253,7 @@ async fn ws_handler(
             .header("Access-Control-Max-Age", "86400" /* 24 hours */)
             .status(StatusCode::OK) // 204 is also valid, but see: https://developer.mozilla.org/en-US/docs/Web/HTTP/Methods/OPTIONS#status_code
             .body(Body::empty())
-            .map_err(|e| ApiError::InternalServerError(e.into()))
+            .map_err(|e| ApiError::BadRequest(e.into()))
     } else {
         json_response(StatusCode::BAD_REQUEST, "query is not supported")
     }

proxy/src/metrics.rs

@@ -3,18 +3,9 @@
 use crate::{config::MetricCollectionConfig, http};
 use chrono::{DateTime, Utc};
 use consumption_metrics::{idempotency_key, Event, EventChunk, EventType, CHUNK_SIZE};
-use dashmap::{mapref::entry::Entry, DashMap};
-use once_cell::sync::Lazy;
-use serde::{Deserialize, Serialize};
-use std::{
-    convert::Infallible,
-    sync::{
-        atomic::{AtomicU64, AtomicUsize, Ordering},
-        Arc,
-    },
-    time::Duration,
-};
-use tracing::{error, info, instrument, trace};
+use serde::Serialize;
+use std::{collections::HashMap, convert::Infallible, time::Duration};
+use tracing::{error, info, instrument, trace, warn};
 
 const PROXY_IO_BYTES_PER_CLIENT: &str = "proxy_io_bytes_per_client";
 
@@ -27,95 +18,12 @@ const DEFAULT_HTTP_REPORTING_TIMEOUT: Duration = Duration::from_secs(60);
 /// Both the proxy and the ingestion endpoint will live in the same region (or cell)
 /// so while the project-id is unique across regions the whole pipeline will work correctly
 /// because we enrich the event with project_id in the control-plane endpoint.
-#[derive(Eq, Hash, PartialEq, Serialize, Deserialize, Debug, Clone)]
+#[derive(Eq, Hash, PartialEq, Serialize, Debug, Clone)]
 pub struct Ids {
     pub endpoint_id: String,
     pub branch_id: String,
 }
 
-#[derive(Debug)]
-pub struct MetricCounter {
-    transmitted: AtomicU64,
-    opened_connections: AtomicUsize,
-}
-
-impl MetricCounter {
-    /// Record that some bytes were sent from the proxy to the client
-    pub fn record_egress(&self, bytes: u64) {
-        self.transmitted.fetch_add(bytes, Ordering::AcqRel);
-    }
-
-    /// extract the value that should be reported
-    fn should_report(self: &Arc<Self>) -> Option<u64> {
-        // heuristic to see if the branch is still open
-        // if a clone happens while we are observing, the heuristic will be incorrect.
-        //
-        // Worst case is that we won't report an event for this endpoint.
-        // However, for the strong count to be 1 it must have occured that at one instant
-        // all the endpoints were closed, so missing a report because the endpoints are closed is valid.
-        let is_open = Arc::strong_count(self) > 1;
-        let opened = self.opened_connections.swap(0, Ordering::AcqRel);
-
-        // update cached metrics eagerly, even if they can't get sent
-        // (to avoid sending the same metrics twice)
-        // see the relevant discussion on why to do so even if the status is not success:
-        // https://github.com/neondatabase/neon/pull/4563#discussion_r1246710956
-        let value = self.transmitted.swap(0, Ordering::AcqRel);
-
-        // Our only requirement is that we report in every interval if there was an open connection
-        // if there were no opened connections since, then we don't need to report
-        if value == 0 && !is_open && opened == 0 {
-            None
-        } else {
-            Some(value)
-        }
-    }
-
-    /// Determine whether the counter should be cleared from the global map.
-    fn should_clear(self: &mut Arc<Self>) -> bool {
-        // we can't clear this entry if it's acquired elsewhere
-        let Some(counter) = Arc::get_mut(self) else {
-            return false;
-        };
-        let opened = *counter.opened_connections.get_mut();
-        let value = *counter.transmitted.get_mut();
-        // clear if there's no data to report
-        value == 0 && opened == 0
-    }
-}
-
-// endpoint and branch IDs are not user generated so we don't run the risk of hash-dos
-type FastHasher = std::hash::BuildHasherDefault<rustc_hash::FxHasher>;
-
-#[derive(Default)]
-pub struct Metrics {
-    endpoints: DashMap<Ids, Arc<MetricCounter>, FastHasher>,
-}
-
-impl Metrics {
-    /// Register a new byte metrics counter for this endpoint
-    pub fn register(&self, ids: Ids) -> Arc<MetricCounter> {
-        let entry = if let Some(entry) = self.endpoints.get(&ids) {
-            entry.clone()
-        } else {
-            self.endpoints
-                .entry(ids)
-                .or_insert_with(|| {
-                    Arc::new(MetricCounter {
-                        transmitted: AtomicU64::new(0),
-                        opened_connections: AtomicUsize::new(0),
-                    })
-                })
-                .clone()
-        };
-
-        entry.opened_connections.fetch_add(1, Ordering::AcqRel);
-        entry
-    }
-}
-
-pub static USAGE_METRICS: Lazy<Metrics> = Lazy::new(Metrics::default);
-
 pub async fn task_main(config: &MetricCollectionConfig) -> anyhow::Result<Infallible> {
     info!("metrics collector config: {config:?}");
     scopeguard::defer! {
@@ -123,83 +31,145 @@ pub async fn task_main(config: &MetricCollectionConfig) -> anyhow::Result<Infall
     }
 
     let http_client = http::new_client_with_timeout(DEFAULT_HTTP_REPORTING_TIMEOUT);
+    let mut cached_metrics: HashMap<Ids, (u64, DateTime<Utc>)> = HashMap::new();
     let hostname = hostname::get()?.as_os_str().to_string_lossy().into_owned();
 
-    let mut prev = Utc::now();
     let mut ticker = tokio::time::interval(config.interval);
     loop {
         ticker.tick().await;
 
-        let now = Utc::now();
-        collect_metrics_iteration(
-            &USAGE_METRICS,
+        let res = collect_metrics_iteration(
             &http_client,
+            &mut cached_metrics,
             &config.endpoint,
             &hostname,
-            prev,
-            now,
         )
         .await;
-        prev = now;
+
+        match res {
+            Err(e) => error!("failed to send consumption metrics: {e} "),
+            Ok(_) => trace!("periodic metrics collection completed successfully"),
+        }
     }
 }
 
+fn gather_proxy_io_bytes_per_client() -> Vec<(Ids, (u64, DateTime<Utc>))> {
+    let mut current_metrics: Vec<(Ids, (u64, DateTime<Utc>))> = Vec::new();
+    let metrics = prometheus::default_registry().gather();
+
+    for m in metrics {
+        if m.get_name() == "proxy_io_bytes_per_client" {
+            for ms in m.get_metric() {
+                let direction = ms
+                    .get_label()
+                    .iter()
+                    .find(|l| l.get_name() == "direction")
+                    .unwrap()
+                    .get_value();
+
+                // Only collect metric for outbound traffic
+                if direction == "tx" {
+                    let endpoint_id = ms
+                        .get_label()
+                        .iter()
+                        .find(|l| l.get_name() == "endpoint_id")
+                        .unwrap()
+                        .get_value();
+                    let branch_id = ms
+                        .get_label()
+                        .iter()
+                        .find(|l| l.get_name() == "branch_id")
+                        .unwrap()
+                        .get_value();
+
+                    let value = ms.get_counter().get_value() as u64;
+
+                    // Report if the metric value is suspiciously large
+                    if value > (1u64 << 40) {
+                        warn!(
+                            "potentially abnormal counter value: branch_id {} endpoint_id {} val: {}",
+                            branch_id, endpoint_id, value
+                        );
+                    }
+
+                    current_metrics.push((
+                        Ids {
+                            endpoint_id: endpoint_id.to_string(),
+                            branch_id: branch_id.to_string(),
+                        },
+                        (value, Utc::now()),
+                    ));
+                }
+            }
+        }
+    }
+
+    current_metrics
+}
+
 #[instrument(skip_all)]
 async fn collect_metrics_iteration(
-    metrics: &Metrics,
     client: &http::ClientWithMiddleware,
+    cached_metrics: &mut HashMap<Ids, (u64, DateTime<Utc>)>,
     metric_collection_endpoint: &reqwest::Url,
     hostname: &str,
-    prev: DateTime<Utc>,
-    now: DateTime<Utc>,
-) {
+) -> anyhow::Result<()> {
     info!(
         "starting collect_metrics_iteration. metric_collection_endpoint: {}",
         metric_collection_endpoint
     );
 
-    let mut metrics_to_clear = Vec::new();
+    let current_metrics = gather_proxy_io_bytes_per_client();
 
-    let metrics_to_send: Vec<(Ids, u64)> = metrics
-        .endpoints
+    let metrics_to_send: Vec<Event<Ids, &'static str>> = current_metrics
         .iter()
-        .filter_map(|counter| {
-            let key = counter.key().clone();
-            let Some(value) = counter.should_report() else {
-                metrics_to_clear.push(key);
-                return None;
+        .filter_map(|(curr_key, (curr_val, curr_time))| {
+            let mut start_time = *curr_time;
+            let mut value = *curr_val;
+
+            if let Some((prev_val, prev_time)) = cached_metrics.get(curr_key) {
+                // Only send metrics updates if the metric has increased
+                if curr_val > prev_val {
+                    value = curr_val - prev_val;
+                    start_time = *prev_time;
+                } else {
+                    if curr_val < prev_val {
+                        error!("proxy_io_bytes_per_client metric value decreased from {} to {} for key {:?}",
+                        prev_val, curr_val, curr_key);
+                    }
+                    return None;
+                }
             };
-            Some((key, value))
+
+            Some(Event {
+                kind: EventType::Incremental {
+                    start_time,
+                    stop_time: *curr_time,
+                },
+                metric: PROXY_IO_BYTES_PER_CLIENT,
+                idempotency_key: idempotency_key(hostname),
+                value,
+                extra: Ids {
+                    endpoint_id: curr_key.endpoint_id.clone(),
+                    branch_id: curr_key.branch_id.clone(),
+                },
+            })
         })
         .collect();
 
     if metrics_to_send.is_empty() {
         trace!("no new metrics to send");
+        return Ok(());
     }
 
     // Send metrics.
     // Split into chunks of 1000 metrics to avoid exceeding the max request size
     for chunk in metrics_to_send.chunks(CHUNK_SIZE) {
-        let events = chunk
-            .iter()
-            .map(|(ids, value)| Event {
-                kind: EventType::Incremental {
-                    start_time: prev,
-                    stop_time: now,
-                },
-                metric: PROXY_IO_BYTES_PER_CLIENT,
-                idempotency_key: idempotency_key(hostname),
-                value: *value,
-                extra: Ids {
-                    endpoint_id: ids.endpoint_id.clone(),
-                    branch_id: ids.branch_id.clone(),
-                },
-            })
-            .collect();
-
         let res = client
             .post(metric_collection_endpoint.clone())
-            .json(&EventChunk { events })
+            .json(&EventChunk {
+                events: chunk.into(),
+            })
             .send()
             .await;
 
@@ -213,113 +183,34 @@ async fn collect_metrics_iteration(
 
         if !res.status().is_success() {
             error!("metrics endpoint refused the sent metrics: {:?}", res);
-            for metric in chunk.iter().filter(|(_, value)| *value > (1u64 << 40)) {
+            for metric in chunk.iter().filter(|metric| metric.value > (1u64 << 40)) {
                 // Report if the metric value is suspiciously large
                 error!("potentially abnormal metric value: {:?}", metric);
             }
         }
-    }
+        // update cached metrics after they were sent
+        // (to avoid sending the same metrics twice)
+        // see the relevant discussion on why to do so even if the status is not success:
+        // https://github.com/neondatabase/neon/pull/4563#discussion_r1246710956
+        for send_metric in chunk {
+            let stop_time = match send_metric.kind {
+                EventType::Incremental { stop_time, .. } => stop_time,
+                _ => unreachable!(),
+            };
 
-    for metric in metrics_to_clear {
-        match metrics.endpoints.entry(metric) {
-            Entry::Occupied(mut counter) => {
-                if counter.get_mut().should_clear() {
-                    counter.remove_entry();
-                }
-            }
-            Entry::Vacant(_) => {}
+            cached_metrics
+                .entry(Ids {
+                    endpoint_id: send_metric.extra.endpoint_id.clone(),
+                    branch_id: send_metric.extra.branch_id.clone(),
+                })
+                // update cached value (add delta) and time
+                .and_modify(|e| {
+                    e.0 = e.0.saturating_add(send_metric.value);
+                    e.1 = stop_time
+                })
+                // cache new metric
+                .or_insert((send_metric.value, stop_time));
         }
     }
-}
-
-#[cfg(test)]
-mod tests {
-    use std::{
-        net::TcpListener,
-        sync::{Arc, Mutex},
-    };
-
-    use anyhow::Error;
-    use chrono::Utc;
-    use consumption_metrics::{Event, EventChunk};
-    use hyper::{
-        service::{make_service_fn, service_fn},
-        Body, Response,
-    };
-    use url::Url;
-
-    use super::{collect_metrics_iteration, Ids, Metrics};
-    use crate::http;
-
-    #[tokio::test]
-    async fn metrics() {
-        let listener = TcpListener::bind("0.0.0.0:0").unwrap();
-
-        let reports = Arc::new(Mutex::new(vec![]));
-        let reports2 = reports.clone();
-
-        let server = hyper::server::Server::from_tcp(listener)
-            .unwrap()
-            .serve(make_service_fn(move |_| {
-                let reports = reports.clone();
-                async move {
-                    Ok::<_, Error>(service_fn(move |req| {
-                        let reports = reports.clone();
-                        async move {
-                            let bytes = hyper::body::to_bytes(req.into_body()).await?;
-                            let events: EventChunk<'static, Event<Ids, String>> =
-                                serde_json::from_slice(&bytes)?;
-                            reports.lock().unwrap().push(events);
-                            Ok::<_, Error>(Response::new(Body::from(vec![])))
-                        }
-                    }))
-                }
-            }));
-        let addr = server.local_addr();
-        tokio::spawn(server);
-
-        let metrics = Metrics::default();
-        let client = http::new_client();
-        let endpoint = Url::parse(&format!("http://{addr}")).unwrap();
-        let now = Utc::now();
-
-        // no counters have been registered
-        collect_metrics_iteration(&metrics, &client, &endpoint, "foo", now, now).await;
-        let r = std::mem::take(&mut *reports2.lock().unwrap());
-        assert!(r.is_empty());
-
-        // register a new counter
-        let counter = metrics.register(Ids {
-            endpoint_id: "e1".to_string(),
-            branch_id: "b1".to_string(),
-        });
-
-        // the counter should be observed despite 0 egress
-        collect_metrics_iteration(&metrics, &client, &endpoint, "foo", now, now).await;
-        let r = std::mem::take(&mut *reports2.lock().unwrap());
-        assert_eq!(r.len(), 1);
-        assert_eq!(r[0].events.len(), 1);
-        assert_eq!(r[0].events[0].value, 0);
-
-        // record egress
-        counter.record_egress(1);
-
-        // egress should be observered
-        collect_metrics_iteration(&metrics, &client, &endpoint, "foo", now, now).await;
-        let r = std::mem::take(&mut *reports2.lock().unwrap());
-        assert_eq!(r.len(), 1);
-        assert_eq!(r[0].events.len(), 1);
-        assert_eq!(r[0].events[0].value, 1);
-
-        // release counter
-        drop(counter);
-
-        // we do not observe the counter
-        collect_metrics_iteration(&metrics, &client, &endpoint, "foo", now, now).await;
-        let r = std::mem::take(&mut *reports2.lock().unwrap());
-        assert!(r.is_empty());
-
-        // counter is unregistered
-        assert!(metrics.endpoints.is_empty());
-    }
+    Ok(())
 }

proxy/src/protocol2.rs

@@ -129,18 +129,12 @@ impl<T: AsyncRead> WithClientIp<T> {
             // exit for bad header
             let len = usize::min(self.buf.len(), HEADER.len());
             if self.buf[..len] != HEADER[..len] {
-                return Poll::Ready(Err(io::Error::new(
-                    io::ErrorKind::InvalidInput,
-                    "invalid proxy protocol v2 header",
-                )));
+                return Poll::Ready(Ok(None));
             }
 
             // if no more bytes available then exit
             if ready!(bytes_read) == 0 {
-                return Poll::Ready(Err(io::Error::new(
-                    io::ErrorKind::UnexpectedEof,
-                    "missing proxy protocol v2 header",
-                )));
+                return Poll::Ready(Ok(None));
             };
         }
 
@@ -152,27 +146,27 @@ impl<T: AsyncRead> WithClientIp<T> {
         let command = vc & 0b1111;
         if version != 2 {
             return Poll::Ready(Err(io::Error::new(
-                io::ErrorKind::InvalidInput,
+                io::ErrorKind::Other,
                 "invalid proxy protocol version. expected version 2",
             )));
         }
-        let local = match command {
+        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 => true,
+            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 => false,
+            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::InvalidInput,
+                    io::ErrorKind::Other,
                     "invalid proxy protocol command. expected local (0) or proxy (1)",
                 )))
             }
@@ -192,29 +186,8 @@ impl<T: AsyncRead> WithClientIp<T> {
             // - \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,
-
-            // - \x31 : UNIX stream : the forwarded connection uses SOCK_STREAM over the
-            //   AF_UNIX protocol family. Address length is 2*108 = 216 bytes.
-            // - \x32 : UNIX datagram : the forwarded connection uses SOCK_DGRAM over the
-            //   AF_UNIX protocol family. Address length is 2*108 = 216 bytes.
-            0x31 | 0x32 => 216,
-
-            // UNSPEC : the connection is forwarded for an unknown, unspecified
-            // or unsupported protocol. The sender should use this family when sending
-            // LOCAL commands or when dealing with unsupported protocol families. When
-            // used with a LOCAL command, the receiver must accept the connection and
-            // ignore any address information. For other commands, the receiver is free
-            // to accept the connection anyway and use the real endpoints addresses or to
-            // reject the connection. The receiver should ignore address information.
-            0x00 | 0x01 | 0x02 | 0x10 | 0x20 | 0x30 if local => 0,
-
-            // unspecified or invalid. ignore the addresses
-            _ => {
-                return Poll::Ready(Err(io::Error::new(
-                    io::ErrorKind::InvalidInput,
-                    "invalid proxy protocol address family/transport protocol",
-                )))
-            }
+            // unspecified or unix stream. ignore the addresses
+            _ => 0,
         };
 
         // The 15th and 16th bytes is the address length in bytes in network endian order.
@@ -446,7 +419,6 @@ mod tests {
     }
 
     #[tokio::test]
-    #[should_panic(expected = "invalid proxy protocol v2 header")]
     async fn test_invalid() {
         let data = [0x55; 256];
 
@@ -454,15 +426,20 @@ mod tests {
 
         let mut bytes = vec![];
         read.read_to_end(&mut bytes).await.unwrap();
+        assert_eq!(bytes, data);
+        assert_eq!(read.state, ProxyParse::None);
     }
 
     #[tokio::test]
-    #[should_panic(expected = "missing proxy protocol v2 header")]
     async fn test_short() {
-        let mut read = pin!(WithClientIp::new(&super::HEADER.as_slice()[..10]));
+        let data = [0x55; 10];
+
+        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!(read.state, ProxyParse::None);
     }
 
     #[tokio::test]

proxy/src/proxy.rs

@@ -7,7 +7,6 @@ use crate::{
     compute::{self, PostgresConnection},
     config::{ProxyConfig, TlsConfig},
     console::{self, errors::WakeComputeError, messages::MetricsAuxInfo, Api},
-    metrics::{Ids, USAGE_METRICS},
     protocol2::WithClientIp,
     stream::{PqStream, Stream},
 };
@@ -29,11 +28,10 @@ use tracing::{error, info, info_span, warn, Instrument};
 use utils::measured_stream::MeasuredStream;
 
 /// Number of times we should retry the `/proxy_wake_compute` http request.
-/// Retry duration is BASE_RETRY_WAIT_DURATION * RETRY_WAIT_EXPONENT_BASE ^ n, where n starts at 0
-pub const NUM_RETRIES_CONNECT: u32 = 16;
+/// Retry duration is BASE_RETRY_WAIT_DURATION * 1.5^n
+pub const NUM_RETRIES_CONNECT: u32 = 10;
 const CONNECT_TIMEOUT: time::Duration = time::Duration::from_secs(2);
-const BASE_RETRY_WAIT_DURATION: time::Duration = time::Duration::from_millis(25);
-const RETRY_WAIT_EXPONENT_BASE: f64 = std::f64::consts::SQRT_2;
+const BASE_RETRY_WAIT_DURATION: time::Duration = time::Duration::from_millis(100);
 
 const ERR_INSECURE_CONNECTION: &str = "connection is insecure (try using `sslmode=require`)";
 const ERR_PROTO_VIOLATION: &str = "protocol violation";
@@ -555,7 +553,8 @@ impl ShouldRetry for compute::ConnectionError {
 }
 
 pub fn retry_after(num_retries: u32) -> time::Duration {
-    BASE_RETRY_WAIT_DURATION.mul_f64(RETRY_WAIT_EXPONENT_BASE.powi((num_retries as i32) - 1))
+    // 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.
@@ -603,11 +602,6 @@ pub async fn proxy_pass(
     compute: impl AsyncRead + AsyncWrite + Unpin,
     aux: &MetricsAuxInfo,
 ) -> anyhow::Result<()> {
-    let usage = USAGE_METRICS.register(Ids {
-        endpoint_id: aux.endpoint_id.to_string(),
-        branch_id: aux.branch_id.to_string(),
-    });
-
     let m_sent = NUM_BYTES_PROXIED_COUNTER.with_label_values(&aux.traffic_labels("tx"));
     let mut client = MeasuredStream::new(
         client,
@@ -615,7 +609,6 @@ pub async fn proxy_pass(
         |cnt| {
             // Number of bytes we sent to the client (outbound).
             m_sent.inc_by(cnt as u64);
-            usage.record_egress(cnt as u64);
         },
     );
 

proxy/src/proxy/tests.rs

@@ -303,7 +303,7 @@ 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 1..NUM_RETRIES_CONNECT {
+    for num_retries in 1..10 {
         total_wait += retry_after(num_retries);
     }
     assert!(total_wait < tokio::time::Duration::from_secs(12));
@@ -494,11 +494,11 @@ async fn connect_to_compute_non_retry_2() {
 /// Retry for at most `NUM_RETRIES_CONNECT` times.
 #[tokio::test]
 async fn connect_to_compute_non_retry_3() {
-    assert_eq!(NUM_RETRIES_CONNECT, 16);
+    assert_eq!(NUM_RETRIES_CONNECT, 10);
     use ConnectAction::*;
     let mechanism = TestConnectMechanism::new(vec![
-        Retry, Wake, Retry, Retry, Retry, Retry, Retry, Retry, Retry, Retry, Retry, Retry, Retry,
-        Retry, Retry, Retry, Retry, /* the 17th time */ Retry,
+        Retry, Wake, Retry, Retry, Retry, Retry, Retry, Retry, Retry, Retry, Retry,
+        /* the 11th time */ Retry,
     ]);
     let (cache, extra, creds) = helper_create_connect_info(&mechanism);
     connect_to_compute(&mechanism, cache, &extra, &creds)

rust-toolchain.toml

@@ -1,5 +1,5 @@
 [toolchain]
-channel = "1.72.1"
+channel = "1.72.0"
 profile = "default"
 # The default profile includes rustc, rust-std, cargo, rust-docs, rustfmt and clippy.
 # https://rust-lang.github.io/rustup/concepts/profiles.html

scripts/download_basebackup.py

@@ -1,62 +0,0 @@
-#!/usr/bin/env python3
-#
-# Script to download the basebackup from a pageserver to a tar file.
-#
-# This can be useful in disaster recovery.
-#
-import argparse
-
-import psycopg2
-from psycopg2.extensions import connection as PgConnection
-
-
-def main(args: argparse.Namespace):
-    pageserver_connstr = args.pageserver_connstr
-    tenant_id = args.tenant
-    timeline_id = args.timeline
-    lsn = args.lsn
-    output_path = args.output_path
-
-    psconn: PgConnection = psycopg2.connect(pageserver_connstr)
-    psconn.autocommit = True
-
-    output = open(output_path, "wb")
-
-    with psconn.cursor() as pscur:
-        pscur.copy_expert(f"basebackup {tenant_id} {timeline_id} {lsn}", output)
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        "--tenant-id",
-        dest="tenant",
-        required=True,
-        help="Id of the tenant",
-    )
-    parser.add_argument(
-        "--timeline-id",
-        dest="timeline",
-        required=True,
-        help="Id of the timeline",
-    )
-    parser.add_argument(
-        "--lsn",
-        dest="lsn",
-        required=True,
-        help="LSN to take the basebackup at",
-    )
-    parser.add_argument(
-        "--pageserver-connstr",
-        dest="pageserver_connstr",
-        required=True,
-        help="libpq connection string of the pageserver",
-    )
-    parser.add_argument(
-        "--output",
-        dest="output_path",
-        required=True,
-        help="output path to write the basebackup to",
-    )
-    args = parser.parse_args()
-    main(args)

test_runner/fixtures/compare_fixtures.py

@@ -105,8 +105,6 @@ class NeonCompare(PgCompare):
         self._pg_bin = pg_bin
         self.pageserver_http_client = self.env.pageserver.http_client()
 
-        # note that neon_simple_env now uses LOCAL_FS remote storage
-
         # Create tenant
         tenant_conf: Dict[str, str] = {}
         if False:  # TODO add pytest setting for this

test_runner/fixtures/neon_fixtures.py

@@ -460,11 +460,9 @@ class NeonEnvBuilder:
         ), "Unexpectedly instantiated from outside a test function"
         self.test_name = test_name
 
-    def init_configs(self, default_remote_storage_if_missing: bool = True) -> NeonEnv:
+    def init_configs(self) -> NeonEnv:
         # Cannot create more than one environment from one builder
         assert self.env is None, "environment already initialized"
-        if default_remote_storage_if_missing and self.pageserver_remote_storage is None:
-            self.enable_pageserver_remote_storage(RemoteStorageKind.LOCAL_FS)
         self.env = NeonEnv(self)
         return self.env
 
@@ -472,19 +470,8 @@ class NeonEnvBuilder:
         assert self.env is not None, "environment is not already initialized, call init() first"
         self.env.start()
 
-    def init_start(
-        self,
-        initial_tenant_conf: Optional[Dict[str, str]] = None,
-        default_remote_storage_if_missing: bool = True,
-    ) -> NeonEnv:
-        """
-        Default way to create and start NeonEnv. Also creates the initial_tenant with root initial_timeline.
-
-        To avoid creating initial_tenant, call init_configs to setup the environment.
-
-        Configuring pageserver with remote storage is now the default. There will be a warning if pageserver is created without one.
-        """
-        env = self.init_configs(default_remote_storage_if_missing=default_remote_storage_if_missing)
+    def init_start(self, initial_tenant_conf: Optional[Dict[str, str]] = None) -> NeonEnv:
+        env = self.init_configs()
         self.start()
 
         # Prepare the default branch to start the postgres on later.
@@ -559,7 +546,7 @@ class NeonEnvBuilder:
         user: RemoteStorageUser,
         bucket_name: Optional[str] = None,
         bucket_region: Optional[str] = None,
-    ) -> RemoteStorage:
+    ) -> Optional[RemoteStorage]:
         ret = kind.configure(
             self.repo_dir,
             self.mock_s3_server,
@@ -902,8 +889,6 @@ def _shared_simple_env(
     """
     # Internal fixture backing the `neon_simple_env` fixture. If TEST_SHARED_FIXTURES
      is set, this is shared by all tests using `neon_simple_env`.
-
-    This fixture will use RemoteStorageKind.LOCAL_FS with pageserver.
     """
 
     if os.environ.get("TEST_SHARED_FIXTURES") is None:
@@ -1496,16 +1481,6 @@ class NeonAttachmentService:
             self.running = False
         return self
 
-    def attach_hook(self, tenant_id: TenantId, pageserver_id: int) -> int:
-        response = requests.post(
-            f"{self.env.control_plane_api}/attach_hook",
-            json={"tenant_id": str(tenant_id), "pageserver_id": pageserver_id},
-        )
-        response.raise_for_status()
-        gen = response.json()["gen"]
-        assert isinstance(gen, int)
-        return gen
-
     def __enter__(self) -> "NeonAttachmentService":
         return self
 
@@ -1714,7 +1689,12 @@ class NeonPageserver(PgProtocol):
         to call into the pageserver HTTP client.
         """
         if self.env.attachment_service is not None:
-            generation = self.env.attachment_service.attach_hook(tenant_id, self.id)
+            response = requests.post(
+                f"{self.env.control_plane_api}/attach_hook",
+                json={"tenant_id": str(tenant_id), "pageserver_id": self.id},
+            )
+            response.raise_for_status()
+            generation = response.json()["gen"]
         else:
             generation = None
 

test_runner/fixtures/pageserver/http.py

@@ -620,8 +620,3 @@ class PageserverHttpClient(requests.Session):
             },
         )
         self.verbose_error(res)
-
-    def deletion_queue_flush(self, execute: bool = False):
-        self.put(
-            f"http://localhost:{self.port}/v1/deletion_queue/flush?execute={'true' if execute else 'false'}"
-        ).raise_for_status()

test_runner/fixtures/pageserver/utils.py

@@ -267,7 +267,7 @@ def assert_prefix_not_empty(neon_env_builder: "NeonEnvBuilder", prefix: Optional
 
 
 def list_prefix(
-    neon_env_builder: "NeonEnvBuilder", prefix: Optional[str] = None, delimiter: str = "/"
+    neon_env_builder: "NeonEnvBuilder", prefix: Optional[str] = None
 ) -> ListObjectsV2OutputTypeDef:
     """
     Note that this function takes into account prefix_in_bucket.
@@ -287,7 +287,7 @@ def list_prefix(
 
     # Note that this doesnt use pagination, so list is not guaranteed to be exhaustive.
     response = remote.client.list_objects_v2(
-        Delimiter=delimiter,
+        Delimiter="/",
         Bucket=remote.bucket_name,
         Prefix=prefix,
     )

test_runner/fixtures/remote_storage.py

@@ -202,6 +202,9 @@ class RemoteStorageKind(str, enum.Enum):
     LOCAL_FS = "local_fs"
     MOCK_S3 = "mock_s3"
     REAL_S3 = "real_s3"
+    # Pass to tests that are generic to remote storage
+    # to ensure the test pass with or without the remote storage
+    NOOP = "noop"
 
     def configure(
         self,
@@ -212,7 +215,10 @@ class RemoteStorageKind(str, enum.Enum):
         user: RemoteStorageUser,
         bucket_name: Optional[str] = None,
         bucket_region: Optional[str] = None,
-    ) -> RemoteStorage:
+    ) -> Optional[RemoteStorage]:
+        if self == RemoteStorageKind.NOOP:
+            return None
+
         if self == RemoteStorageKind.LOCAL_FS:
             return LocalFsStorage(LocalFsStorage.component_path(repo_dir, user))
 

test_runner/regress/test_broken_timeline.py

@@ -4,12 +4,7 @@ from typing import List, Tuple
 
 import pytest
 from fixtures.log_helper import log
-from fixtures.neon_fixtures import (
-    Endpoint,
-    NeonEnv,
-    NeonEnvBuilder,
-    wait_for_last_flush_lsn,
-)
+from fixtures.neon_fixtures import Endpoint, NeonEnv, NeonEnvBuilder
 from fixtures.types import TenantId, TimelineId
 
 
@@ -31,18 +26,17 @@ def test_broken_timeline(neon_env_builder: NeonEnvBuilder):
 
     tenant_timelines: List[Tuple[TenantId, TimelineId, Endpoint]] = []
 
-    for _ in range(3):
+    for _ in range(4):
         tenant_id, timeline_id = env.neon_cli.create_tenant()
 
         endpoint = env.endpoints.create_start("main", tenant_id=tenant_id)
         with endpoint.cursor() as cur:
             cur.execute("CREATE TABLE t(key int primary key, value text)")
             cur.execute("INSERT INTO t SELECT generate_series(1,100), 'payload'")
-            wait_for_last_flush_lsn(env, endpoint, tenant_id, timeline_id)
         endpoint.stop()
         tenant_timelines.append((tenant_id, timeline_id, endpoint))
 
-    # Stop the pageserver -- this has to be not immediate or we need to wait for uploads
+    # Stop the pageserver
     env.pageserver.stop()
 
     # Leave the first timeline alone, but corrupt the others in different ways
@@ -51,21 +45,30 @@ def test_broken_timeline(neon_env_builder: NeonEnvBuilder):
 
     (tenant1, timeline1, pg1) = tenant_timelines[1]
     metadata_path = f"{env.pageserver.workdir}/tenants/{tenant1}/timelines/{timeline1}/metadata"
-    with open(metadata_path, "w") as f:
-        f.write("overwritten with garbage!")
+    f = open(metadata_path, "w")
+    f.write("overwritten with garbage!")
+    f.close()
     log.info(f"Timeline {tenant1}/{timeline1} got its metadata spoiled")
 
     (tenant2, timeline2, pg2) = tenant_timelines[2]
     timeline_path = f"{env.pageserver.workdir}/tenants/{tenant2}/timelines/{timeline2}/"
+    for filename in os.listdir(timeline_path):
+        if filename.startswith("00000"):
+            # Looks like a layer file. Remove it
+            os.remove(f"{timeline_path}/{filename}")
+    log.info(
+        f"Timeline {tenant2}/{timeline2} got its layer files removed (no remote storage enabled)"
+    )
+
+    (tenant3, timeline3, pg3) = tenant_timelines[3]
+    timeline_path = f"{env.pageserver.workdir}/tenants/{tenant3}/timelines/{timeline3}/"
     for filename in os.listdir(timeline_path):
         if filename.startswith("00000"):
             # Looks like a layer file. Corrupt it
-            p = f"{timeline_path}/{filename}"
-            size = os.path.getsize(p)
-            with open(p, "wb") as f:
-                f.truncate(0)
-                f.truncate(size)
-    log.info(f"Timeline {tenant2}/{timeline2} got its local layer files spoiled")
+            f = open(f"{timeline_path}/{filename}", "w")
+            f.write("overwritten with garbage!")
+            f.close()
+    log.info(f"Timeline {tenant3}/{timeline3} got its layer files spoiled")
 
     env.pageserver.start()
 
@@ -84,13 +87,22 @@ def test_broken_timeline(neon_env_builder: NeonEnvBuilder):
         f"As expected, compute startup failed eagerly for timeline with corrupt metadata: {err}"
     )
 
-    # Second timeline will fail during basebackup, because the local layer file is corrupt.
+    # Second timeline has no ancestors, only the metadata file and no layer files locally,
+    # and we don't have the remote storage enabled. It is loaded into memory, but getting
+    # the basebackup from it will fail.
+    with pytest.raises(
+        Exception, match=f"Tenant {tenant2} will not become active. Current state: Broken"
+    ) as err:
+        pg2.start()
+    log.info(f"As expected, compute startup failed for timeline with missing layers: {err}")
+
+    # Third timeline will also fail during basebackup, because the layer file is corrupt.
     # It will fail when we try to read (and reconstruct) a page from it, ergo the error message.
     # (We don't check layer file contents on startup, when loading the timeline)
     with pytest.raises(Exception, match="Failed to load delta layer") as err:
-        pg2.start()
+        pg3.start()
     log.info(
-        f"As expected, compute startup failed for timeline {tenant2}/{timeline2} with corrupt layers: {err}"
+        f"As expected, compute startup failed for timeline {tenant3}/{timeline3} with corrupt layers: {err}"
     )
 
 

test_runner/regress/test_disk_usage_eviction.py

@@ -74,13 +74,11 @@ class EvictionEnv:
     pgbench_init_lsns: Dict[TenantId, Lsn]
 
     def timelines_du(self) -> Tuple[int, int, int]:
-        return poor_mans_du(
-            self.neon_env, [(tid, tlid) for tid, tlid in self.timelines], verbose=False
-        )
+        return poor_mans_du(self.neon_env, [(tid, tlid) for tid, tlid in self.timelines])
 
     def du_by_timeline(self) -> Dict[Tuple[TenantId, TimelineId], int]:
         return {
-            (tid, tlid): poor_mans_du(self.neon_env, [(tid, tlid)], verbose=True)[0]
+            (tid, tlid): poor_mans_du(self.neon_env, [(tid, tlid)])[0]
             for tid, tlid in self.timelines
         }
 
@@ -91,21 +89,7 @@ class EvictionEnv:
         """
         lsn = self.pgbench_init_lsns[tenant_id]
         with self.neon_env.endpoints.create_start("main", tenant_id=tenant_id, lsn=lsn) as endpoint:
-            # instead of using pgbench --select-only which does point selects,
-            # run full table scans for all tables
-            with endpoint.connect() as conn:
-                cur = conn.cursor()
-
-                tables_cols = {
-                    "pgbench_accounts": "abalance",
-                    "pgbench_tellers": "tbalance",
-                    "pgbench_branches": "bbalance",
-                    "pgbench_history": "delta",
-                }
-
-                for table, column in tables_cols.items():
-                    cur.execute(f"select avg({column}) from {table}")
-                    _avg = cur.fetchone()
+            self.pg_bin.run(["pgbench", "-S", endpoint.connstr()])
 
     def pageserver_start_with_disk_usage_eviction(
         self, period, max_usage_pct, min_avail_bytes, mock_behavior
@@ -143,19 +127,6 @@ class EvictionEnv:
         self.neon_env.pageserver.allowed_errors.append(".*WARN.* disk usage still high.*")
 
 
-def human_bytes(amt: float) -> str:
-    suffixes = ["", "Ki", "Mi", "Gi"]
-
-    last = suffixes[-1]
-
-    for name in suffixes:
-        if amt < 1024 or name == last:
-            return f"{int(round(amt))} {name}B"
-        amt = amt / 1024
-
-    raise RuntimeError("unreachable")
-
-
 @pytest.fixture
 def eviction_env(request, neon_env_builder: NeonEnvBuilder, pg_bin: PgBin) -> EvictionEnv:
     """
@@ -244,12 +215,8 @@ def test_broken_tenants_are_skipped(eviction_env: EvictionEnv):
 
     healthy_tenant_id, healthy_timeline_id = env.timelines[1]
 
-    broken_size_pre, _, _ = poor_mans_du(
-        env.neon_env, [(broken_tenant_id, broken_timeline_id)], verbose=True
-    )
-    healthy_size_pre, _, _ = poor_mans_du(
-        env.neon_env, [(healthy_tenant_id, healthy_timeline_id)], verbose=True
-    )
+    broken_size_pre, _, _ = poor_mans_du(env.neon_env, [(broken_tenant_id, broken_timeline_id)])
+    healthy_size_pre, _, _ = poor_mans_du(env.neon_env, [(healthy_tenant_id, healthy_timeline_id)])
 
     # try to evict everything, then validate that broken tenant wasn't touched
     target = broken_size_pre + healthy_size_pre
@@ -257,12 +224,8 @@ def test_broken_tenants_are_skipped(eviction_env: EvictionEnv):
     response = env.pageserver_http.disk_usage_eviction_run({"evict_bytes": target})
     log.info(f"{response}")
 
-    broken_size_post, _, _ = poor_mans_du(
-        env.neon_env, [(broken_tenant_id, broken_timeline_id)], verbose=True
-    )
-    healthy_size_post, _, _ = poor_mans_du(
-        env.neon_env, [(healthy_tenant_id, healthy_timeline_id)], verbose=True
-    )
+    broken_size_post, _, _ = poor_mans_du(env.neon_env, [(broken_tenant_id, broken_timeline_id)])
+    healthy_size_post, _, _ = poor_mans_du(env.neon_env, [(healthy_tenant_id, healthy_timeline_id)])
 
     assert broken_size_pre == broken_size_post, "broken tenant should not be touched"
     assert healthy_size_post < healthy_size_pre
@@ -403,16 +366,18 @@ def test_partial_evict_tenant(eviction_env: EvictionEnv):
     du_by_timeline = env.du_by_timeline()
 
     # pick any tenant
-    [warm, cold] = list(du_by_timeline.keys())
-    (tenant_id, timeline_id) = warm
+    [our_tenant, other_tenant] = list(du_by_timeline.keys())
+    (tenant_id, timeline_id) = our_tenant
 
-    # make picked tenant more recently used than the other one
+    # make our tenant more recently used than the other one
     env.warm_up_tenant(tenant_id)
 
     # Build up enough pressure to require evictions from both tenants,
     # but not enough to fall into global LRU.
-    # So, set target to all occupied space, except 2*env.layer_size per tenant
-    target = du_by_timeline[cold] + (du_by_timeline[warm] // 2) - 2 * 2 * env.layer_size
+    # So, set target to all occipied space, except 2*env.layer_size per tenant
+    target = (
+        du_by_timeline[other_tenant] + (du_by_timeline[our_tenant] // 2) - 2 * 2 * env.layer_size
+    )
     response = ps_http.disk_usage_eviction_run({"evict_bytes": target})
     log.info(f"{response}")
 
@@ -427,33 +392,22 @@ def test_partial_evict_tenant(eviction_env: EvictionEnv):
             later_tenant_usage < du_by_timeline[tenant]
         ), "all tenants should have lost some layers"
 
-    warm_size = later_du_by_timeline[warm]
-
-    # bounds for warmed_size
-    warm_lower = 0.5 * du_by_timeline[warm]
-
-    # We don't know exactly whether the cold tenant needs 2 or just 1 env.layer_size wiggle room.
-    # So, check for up to 3 here.
-    warm_upper = warm_lower + 3 * env.layer_size
-
-    cold_size = later_du_by_timeline[cold]
-    cold_upper = 2 * env.layer_size
-
-    log.info(
-        f"expecting for warm tenant: {human_bytes(warm_lower)} < {human_bytes(warm_size)} < {human_bytes(warm_upper)}"
-    )
-    log.info(f"expecting for cold tenant: {human_bytes(cold_size)} < {human_bytes(cold_upper)}")
-
-    assert warm_size > warm_lower, "warmed up tenant should be at about half size (lower)"
-    assert warm_size < warm_upper, "warmed up tenant should be at about half size (upper)"
-
     assert (
-        cold_size < cold_upper
-    ), "the cold tenant should be evicted to its min_resident_size, i.e., max layer file size"
+        later_du_by_timeline[our_tenant] > 0.5 * du_by_timeline[our_tenant]
+    ), "our warmed up tenant should be at about half capacity, part 1"
+    assert (
+        # We don't know exactly whether the cold tenant needs 2 or just 1 env.layer_size wiggle room.
+        # So, check for up to 3 here.
+        later_du_by_timeline[our_tenant]
+        < 0.5 * du_by_timeline[our_tenant] + 3 * env.layer_size
+    ), "our warmed up tenant should be at about half capacity, part 2"
+    assert (
+        later_du_by_timeline[other_tenant] < 2 * env.layer_size
+    ), "the other tenant should be evicted to is min_resident_size, i.e., max layer file size"
 
 
 def poor_mans_du(
-    env: NeonEnv, timelines: list[Tuple[TenantId, TimelineId]], verbose: bool = False
+    env: NeonEnv, timelines: list[Tuple[TenantId, TimelineId]]
 ) -> Tuple[int, int, int]:
     """
     Disk usage, largest, smallest layer for layer files over the given (tenant, timeline) tuples;
@@ -476,11 +430,9 @@ def poor_mans_du(
                 smallest_layer = min(smallest_layer, size)
             else:
                 smallest_layer = size
-            if verbose:
-                log.info(f"{tenant_id}/{timeline_id} => {file.name} {size} ({human_bytes(size)})")
+            log.info(f"{tenant_id}/{timeline_id} => {file.name} {size}")
 
-        if verbose:
-            log.info(f"{tenant_id}/{timeline_id}: sum {total} ({human_bytes(total)})")
+        log.info(f"{tenant_id}/{timeline_id}: sum {total}")
         total_on_disk += total
 
     assert smallest_layer is not None or total_on_disk == 0 and largest_layer == 0

test_runner/regress/test_lfc_resize.py

@@ -1,44 +0,0 @@
-import threading
-import time
-
-import pytest
-from fixtures.log_helper import log
-from fixtures.neon_fixtures import NeonEnv, PgBin
-
-
-#
-# Test branching, when a transaction is in prepared state
-#
-@pytest.mark.timeout(600)
-def test_lfc_resize(neon_simple_env: NeonEnv, pg_bin: PgBin):
-    env = neon_simple_env
-    env.neon_cli.create_branch("test_lfc_resize", "empty")
-    endpoint = env.endpoints.create_start(
-        "test_lfc_resize",
-        config_lines=[
-            "neon.file_cache_path='file.cache'",
-            "neon.max_file_cache_size=1GB",
-            "neon.file_cache_size_limit=1GB",
-        ],
-    )
-    n_resize = 10
-    scale = 10
-    log.info("postgres is running on 'test_lfc_resize' branch")
-
-    def run_pgbench(connstr: str):
-        log.info(f"Start a pgbench workload on pg {connstr}")
-        pg_bin.run_capture(["pgbench", "-i", f"-s{scale}", connstr])
-        pg_bin.run_capture(["pgbench", "-c4", f"-T{n_resize}", "-Mprepared", connstr])
-
-    thread = threading.Thread(target=run_pgbench, args=(endpoint.connstr(),), daemon=True)
-    thread.start()
-
-    conn = endpoint.connect()
-    cur = conn.cursor()
-
-    for i in range(n_resize):
-        cur.execute(f"alter system set neon.file_cache_size_limit='{i*10}MB'")
-        cur.execute("select pg_reload_conf()")
-        time.sleep(1)
-
-    thread.join()

test_runner/regress/test_pageserver_generations.py

@@ -1,352 +0,0 @@
-"""
-
-Tests in this module exercise the pageserver's behavior around generation numbers,
-as defined in docs/rfcs/025-generation-numbers.md.  Briefly, the behaviors we require
-of the pageserver are:
-- Do not start a tenant without a generation number if control_plane_api is set
-- Remote objects must be suffixed with generation
-- Deletions may only be executed after validating generation
-- Updates to remote_consistent_lsn may only be made visible after validating generation
-"""
-
-
-import re
-import time
-from typing import Optional
-
-import pytest
-from fixtures.log_helper import log
-from fixtures.neon_fixtures import (
-    NeonEnv,
-    NeonEnvBuilder,
-    PgBin,
-    last_flush_lsn_upload,
-    wait_for_last_flush_lsn,
-)
-from fixtures.pageserver.utils import list_prefix
-from fixtures.remote_storage import (
-    RemoteStorageKind,
-)
-from fixtures.types import TenantId, TimelineId
-from fixtures.utils import print_gc_result, wait_until
-
-# A tenant configuration that is convenient for generating uploads and deletions
-# without a large amount of postgres traffic.
-TENANT_CONF = {
-    # small checkpointing and compaction targets to ensure we generate many upload operations
-    "checkpoint_distance": f"{128 * 1024}",
-    "compaction_threshold": "1",
-    "compaction_target_size": f"{128 * 1024}",
-    # no PITR horizon, we specify the horizon when we request on-demand GC
-    "pitr_interval": "0s",
-    # disable background compaction and GC. We invoke it manually when we want it to happen.
-    "gc_period": "0s",
-    "compaction_period": "0s",
-    # create image layers eagerly, so that GC can remove some layers
-    "image_creation_threshold": "1",
-}
-
-
-def generate_uploads_and_deletions(
-    env: NeonEnv,
-    *,
-    init: bool = True,
-    tenant_id: Optional[TenantId] = None,
-    timeline_id: Optional[TimelineId] = None,
-    data: Optional[str] = None,
-):
-    """
-    Using the environment's default tenant + timeline, generate a load pattern
-    that results in some uploads and some deletions to remote storage.
-    """
-
-    if tenant_id is None:
-        tenant_id = env.initial_tenant
-    assert tenant_id is not None
-
-    if timeline_id is None:
-        timeline_id = env.initial_timeline
-    assert timeline_id is not None
-
-    ps_http = env.pageserver.http_client()
-
-    with env.endpoints.create_start("main", tenant_id=tenant_id) as endpoint:
-        if init:
-            endpoint.safe_psql("CREATE TABLE foo (id INTEGER PRIMARY KEY, val text)")
-            last_flush_lsn_upload(env, endpoint, tenant_id, timeline_id)
-
-        def churn(data):
-            endpoint.safe_psql_many(
-                [
-                    f"""
-                INSERT INTO foo (id, val)
-                SELECT g, '{data}'
-                FROM generate_series(1, 20000) g
-                ON CONFLICT (id) DO UPDATE
-                SET val = EXCLUDED.val
-                """,
-                    # to ensure that GC can actually remove some layers
-                    "VACUUM foo",
-                ]
-            )
-            assert tenant_id is not None
-            assert timeline_id is not None
-            wait_for_last_flush_lsn(env, endpoint, tenant_id, timeline_id)
-            ps_http.timeline_checkpoint(tenant_id, timeline_id)
-
-        # Compaction should generate some GC-elegible layers
-        for i in range(0, 2):
-            churn(f"{i if data is None else data}")
-
-        gc_result = ps_http.timeline_gc(tenant_id, timeline_id, 0)
-        print_gc_result(gc_result)
-        assert gc_result["layers_removed"] > 0
-
-
-def get_metric_or_0(ps_http, metric: str) -> int:
-    v = ps_http.get_metric_value(metric)
-    return 0 if v is None else int(v)
-
-
-def get_deletion_queue_executed(ps_http) -> int:
-    return get_metric_or_0(ps_http, "pageserver_deletion_queue_executed_total")
-
-
-def get_deletion_queue_submitted(ps_http) -> int:
-    return get_metric_or_0(ps_http, "pageserver_deletion_queue_submitted_total")
-
-
-def get_deletion_queue_dropped(ps_http) -> int:
-    return get_metric_or_0(ps_http, "pageserver_deletion_queue_dropped_total")
-
-
-def get_deletion_queue_unexpected_errors(ps_http) -> int:
-    return get_metric_or_0(ps_http, "pageserver_deletion_queue_unexpected_errors_total")
-
-
-def get_deletion_queue_dropped_lsn_updates(ps_http) -> int:
-    return get_metric_or_0(ps_http, "pageserver_deletion_queue_dropped_lsn_updates_total")
-
-
-def get_deletion_queue_depth(ps_http) -> int:
-    """
-    Queue depth if at least one deletion has been submitted, else None
-    """
-    submitted = get_deletion_queue_submitted(ps_http)
-    executed = get_deletion_queue_executed(ps_http)
-    dropped = get_deletion_queue_dropped(ps_http)
-    depth = submitted - executed - dropped
-    log.info(f"get_deletion_queue_depth: {depth} ({submitted} - {executed} - {dropped})")
-
-    assert depth >= 0
-    return int(depth)
-
-
-def assert_deletion_queue(ps_http, size_fn) -> None:
-    v = get_deletion_queue_depth(ps_http)
-    assert v is not None
-    assert size_fn(v) is True
-
-
-def test_generations_upgrade(neon_env_builder: NeonEnvBuilder):
-    """
-    Validate behavior when a pageserver is run without generation support enabled,
-    then started again after activating it:
-    - Before upgrade, no objects should have generation suffixes
-    - After upgrade, the bucket should contain a mixture.
-    - In both cases, postgres I/O should work.
-    """
-    neon_env_builder.enable_generations = True
-    neon_env_builder.enable_pageserver_remote_storage(
-        RemoteStorageKind.MOCK_S3,
-    )
-
-    env = neon_env_builder.init_configs()
-    env.broker.try_start()
-    for sk in env.safekeepers:
-        sk.start()
-    assert env.attachment_service is not None
-    env.attachment_service.start()
-
-    env.pageserver.start(overrides=('--pageserver-config-override=control_plane_api=""',))
-
-    env.neon_cli.create_tenant(
-        tenant_id=env.initial_tenant, conf=TENANT_CONF, timeline_id=env.initial_timeline
-    )
-    generate_uploads_and_deletions(env)
-
-    def parse_generation_suffix(key):
-        m = re.match(".+-([0-9a-zA-Z]{8})$", key)
-        if m is None:
-            return None
-        else:
-            log.info(f"match: {m}")
-            log.info(f"group: {m.group(1)}")
-            return int(m.group(1), 16)
-
-    pre_upgrade_keys = list(
-        [o["Key"] for o in list_prefix(neon_env_builder, delimiter="")["Contents"]]
-    )
-    for key in pre_upgrade_keys:
-        assert parse_generation_suffix(key) is None
-
-    env.pageserver.stop()
-
-    # Starting without the override that disabled control_plane_api
-    env.pageserver.start()
-
-    generate_uploads_and_deletions(env, init=False)
-
-    legacy_objects: list[str] = []
-    suffixed_objects = []
-    post_upgrade_keys = list(
-        [o["Key"] for o in list_prefix(neon_env_builder, delimiter="")["Contents"]]
-    )
-    for key in post_upgrade_keys:
-        log.info(f"post-upgrade key: {key}")
-        if parse_generation_suffix(key) is not None:
-            suffixed_objects.append(key)
-        else:
-            legacy_objects.append(key)
-
-    # Bucket now contains a mixture of suffixed and non-suffixed objects
-    assert len(suffixed_objects) > 0
-    assert len(legacy_objects) > 0
-
-    assert get_deletion_queue_unexpected_errors(env.pageserver.http_client()) == 0
-
-
-def test_deferred_deletion(neon_env_builder: NeonEnvBuilder):
-    neon_env_builder.enable_generations = True
-    neon_env_builder.enable_pageserver_remote_storage(
-        RemoteStorageKind.MOCK_S3,
-    )
-    env = neon_env_builder.init_start(initial_tenant_conf=TENANT_CONF)
-    assert env.attachment_service is not None
-
-    some_other_pageserver = 1234
-    ps_http = env.pageserver.http_client()
-
-    generate_uploads_and_deletions(env)
-
-    # Flush: pending deletions should all complete
-    assert_deletion_queue(ps_http, lambda n: n > 0)
-    ps_http.deletion_queue_flush(execute=True)
-    assert_deletion_queue(ps_http, lambda n: n == 0)
-    assert get_deletion_queue_dropped(ps_http) == 0
-
-    # Our visible remote_consistent_lsn should match projected
-    timeline = ps_http.timeline_detail(env.initial_tenant, env.initial_timeline)
-    assert timeline["remote_consistent_lsn"] == timeline["remote_consistent_lsn_visible"]
-    assert get_deletion_queue_dropped_lsn_updates(ps_http) == 0
-
-    env.pageserver.allowed_errors.extend(
-        [".*Dropped remote consistent LSN updates.*", ".*Dropping stale deletions.*"]
-    )
-
-    # Now advance the generation in the control plane: subsequent validations
-    # from the running pageserver will fail.  No more deletions should happen.
-    env.attachment_service.attach_hook(env.initial_tenant, some_other_pageserver)
-    generate_uploads_and_deletions(env, init=False)
-
-    assert_deletion_queue(ps_http, lambda n: n > 0)
-    queue_depth_before = get_deletion_queue_depth(ps_http)
-    executed_before = get_deletion_queue_executed(ps_http)
-    ps_http.deletion_queue_flush(execute=True)
-
-    # Queue drains to zero because we dropped deletions
-    assert_deletion_queue(ps_http, lambda n: n == 0)
-    # The executed counter has not incremented
-    assert get_deletion_queue_executed(ps_http) == executed_before
-    # The dropped counter has incremented to consume all of the deletions that were previously enqueued
-    assert get_deletion_queue_dropped(ps_http) == queue_depth_before
-
-    # Flush to S3 and see that remote_consistent_lsn does not advance: it cannot
-    # because generation validation fails.
-    timeline = ps_http.timeline_detail(env.initial_tenant, env.initial_timeline)
-    assert timeline["remote_consistent_lsn"] != timeline["remote_consistent_lsn_visible"]
-    assert get_deletion_queue_dropped_lsn_updates(ps_http) > 0
-
-    # TODO: list bucket and confirm all objects have a generation suffix.
-
-    assert get_deletion_queue_unexpected_errors(ps_http) == 0
-
-
-@pytest.mark.parametrize("keep_attachment", [True, False])
-def test_deletion_queue_recovery(
-    neon_env_builder: NeonEnvBuilder, pg_bin: PgBin, keep_attachment: bool
-):
-    """
-    :param keep_attachment: If true, we re-attach after restart.  Else, we act as if some other
-    node took the attachment while we were restarting.
-    """
-    neon_env_builder.enable_generations = True
-    neon_env_builder.enable_pageserver_remote_storage(
-        RemoteStorageKind.MOCK_S3,
-    )
-    env = neon_env_builder.init_start(initial_tenant_conf=TENANT_CONF)
-
-    ps_http = env.pageserver.http_client()
-
-    # Prevent deletion lists from being executed, to build up some backlog of deletions
-    ps_http.configure_failpoints(
-        [
-            ("deletion-queue-before-execute", "return"),
-        ]
-    )
-
-    generate_uploads_and_deletions(env)
-
-    # There should be entries in the deletion queue
-    assert_deletion_queue(ps_http, lambda n: n > 0)
-    ps_http.deletion_queue_flush()
-    before_restart_depth = get_deletion_queue_depth(ps_http)
-
-    assert get_deletion_queue_unexpected_errors(ps_http) == 0
-    assert get_deletion_queue_dropped_lsn_updates(ps_http) == 0
-
-    log.info(f"Restarting pageserver with {before_restart_depth} deletions enqueued")
-    env.pageserver.stop(immediate=True)
-
-    if not keep_attachment:
-        some_other_pageserver = 101010
-        assert env.attachment_service is not None
-        env.attachment_service.attach_hook(env.initial_tenant, some_other_pageserver)
-
-    env.pageserver.start()
-
-    def assert_deletions_submitted(n: int):
-        assert ps_http.get_metric_value("pageserver_deletion_queue_submitted_total") == n
-
-    # After restart, issue a flush to kick the deletion frontend to do recovery.
-    # It should recover all the operations we submitted before the restart.
-    ps_http.deletion_queue_flush(execute=False)
-    wait_until(20, 0.25, lambda: assert_deletions_submitted(before_restart_depth))
-
-    # The queue should drain through completely if we flush it
-    ps_http.deletion_queue_flush(execute=True)
-    wait_until(10, 1, lambda: assert_deletion_queue(ps_http, lambda n: n == 0))
-
-    if keep_attachment:
-        # If we kept the attachment, then our pre-restart deletions should have executed
-        # successfully
-        assert get_deletion_queue_executed(ps_http) == before_restart_depth
-    else:
-        # If we lost the attachment, we should have dropped our pre-restart deletions.
-        assert get_deletion_queue_dropped(ps_http) == before_restart_depth
-        env.pageserver.allowed_errors.extend([".*Dropping stale deletions.*"])
-
-    assert get_deletion_queue_unexpected_errors(ps_http) == 0
-    assert get_deletion_queue_dropped_lsn_updates(ps_http) == 0
-
-    # Restart again
-    env.pageserver.stop(immediate=True)
-    env.pageserver.start()
-
-    # No deletion lists should be recovered: this demonstrates that deletion lists
-    # were cleaned up after being executed or dropped in the previous process lifetime.
-    time.sleep(1)
-    assert_deletion_queue(ps_http, lambda n: n == 0)
-
-    assert get_deletion_queue_unexpected_errors(ps_http) == 0
-    assert get_deletion_queue_dropped_lsn_updates(ps_http) == 0

test_runner/regress/test_pageserver_metric_collection.py

@@ -5,6 +5,7 @@ from pathlib import Path
 from queue import SimpleQueue
 from typing import Any, Dict, Set
 
+import pytest
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import (
     NeonEnvBuilder,
@@ -16,13 +17,15 @@ from pytest_httpserver import HTTPServer
 from werkzeug.wrappers.request import Request
 from werkzeug.wrappers.response import Response
 
-# TODO: collect all of the env setup *AFTER* removal of RemoteStorageKind.NOOP
-
 
+@pytest.mark.parametrize(
+    "remote_storage_kind", [RemoteStorageKind.NOOP, RemoteStorageKind.LOCAL_FS]
+)
 def test_metric_collection(
     httpserver: HTTPServer,
     neon_env_builder: NeonEnvBuilder,
     httpserver_listen_address,
+    remote_storage_kind: RemoteStorageKind,
 ):
     (host, port) = httpserver_listen_address
     metric_collection_endpoint = f"http://{host}:{port}/billing/api/v1/usage_events"
@@ -52,7 +55,7 @@ def test_metric_collection(
         synthetic_size_calculation_interval="3s"
         """
 
-    neon_env_builder.enable_pageserver_remote_storage(RemoteStorageKind.LOCAL_FS)
+    neon_env_builder.enable_pageserver_remote_storage(remote_storage_kind)
 
     log.info(f"test_metric_collection endpoint is {metric_collection_endpoint}")
 
@@ -65,14 +68,6 @@ def test_metric_collection(
     env = neon_env_builder.init_start(initial_tenant_conf={"pitr_interval": "0 sec"})
     # httpserver is shut down before pageserver during passing run
     env.pageserver.allowed_errors.append(".*metrics endpoint refused the sent metrics*")
-    # we have a fast rate of calculation, these can happen at shutdown
-    env.pageserver.allowed_errors.append(
-        ".*synthetic_size_worker:calculate_synthetic_size.*:gather_size_inputs.*: failed to calculate logical size at .*: cancelled.*"
-    )
-    env.pageserver.allowed_errors.append(
-        ".*synthetic_size_worker: failed to calculate synthetic size for tenant .*: failed to calculate some logical_sizes"
-    )
-
     tenant_id = env.initial_tenant
     timeline_id = env.initial_timeline
     endpoint = env.endpoints.create_start("main", tenant_id=tenant_id)
@@ -103,14 +98,17 @@ def test_metric_collection(
             total += sample[2]
         return int(total)
 
-    # upload some data to remote storage
-    wait_for_last_flush_lsn(env, endpoint, tenant_id, timeline_id)
-    pageserver_http = env.pageserver.http_client()
-    pageserver_http.timeline_checkpoint(tenant_id, timeline_id)
-    pageserver_http.timeline_gc(tenant_id, timeline_id, 10000)
+    remote_uploaded = 0
 
-    remote_uploaded = get_num_remote_ops("index", "upload")
-    assert remote_uploaded > 0
+    # upload some data to remote storage
+    if remote_storage_kind == RemoteStorageKind.LOCAL_FS:
+        wait_for_last_flush_lsn(env, endpoint, tenant_id, timeline_id)
+        pageserver_http = env.pageserver.http_client()
+        pageserver_http.timeline_checkpoint(tenant_id, timeline_id)
+        pageserver_http.timeline_gc(tenant_id, timeline_id, 10000)
+
+        remote_uploaded = get_num_remote_ops("index", "upload")
+        assert remote_uploaded > 0
 
     # we expect uploads at 1Hz, on busy runners this could be too optimistic,
     # so give 5s we only want to get the following upload after "ready" value.
@@ -213,14 +211,6 @@ def test_metric_collection_cleans_up_tempfile(
 
     # httpserver is shut down before pageserver during passing run
     env.pageserver.allowed_errors.append(".*metrics endpoint refused the sent metrics*")
-    # we have a fast rate of calculation, these can happen at shutdown
-    env.pageserver.allowed_errors.append(
-        ".*synthetic_size_worker:calculate_synthetic_size.*:gather_size_inputs.*: failed to calculate logical size at .*: cancelled.*"
-    )
-    env.pageserver.allowed_errors.append(
-        ".*synthetic_size_worker: failed to calculate synthetic size for tenant .*: failed to calculate some logical_sizes"
-    )
-
     tenant_id = env.initial_tenant
     timeline_id = env.initial_timeline
     endpoint = env.endpoints.create_start("main", tenant_id=tenant_id)

test_runner/regress/test_tenant_delete.py

@@ -30,7 +30,9 @@ from fixtures.types import TenantId
 from fixtures.utils import run_pg_bench_small
 
 
-@pytest.mark.parametrize("remote_storage_kind", available_remote_storages())
+@pytest.mark.parametrize(
+    "remote_storage_kind", [RemoteStorageKind.NOOP, *available_remote_storages()]
+)
 def test_tenant_delete_smoke(
     neon_env_builder: NeonEnvBuilder,
     remote_storage_kind: RemoteStorageKind,
@@ -41,12 +43,6 @@ def test_tenant_delete_smoke(
     neon_env_builder.enable_pageserver_remote_storage(remote_storage_kind)
 
     env = neon_env_builder.init_start()
-    env.pageserver.allowed_errors.extend(
-        [
-            # The deletion queue will complain when it encounters simulated S3 errors
-            ".*deletion executor: DeleteObjects request failed.*",
-        ]
-    )
 
     # lucky race with stopping from flushing a layer we fail to schedule any uploads
     env.pageserver.allowed_errors.append(
@@ -142,12 +138,18 @@ FAILPOINTS_BEFORE_BACKGROUND = [
 def combinations():
     result = []
 
-    remotes = [RemoteStorageKind.MOCK_S3]
+    remotes = [RemoteStorageKind.NOOP, RemoteStorageKind.MOCK_S3]
     if os.getenv("ENABLE_REAL_S3_REMOTE_STORAGE"):
         remotes.append(RemoteStorageKind.REAL_S3)
 
     for remote_storage_kind in remotes:
         for delete_failpoint in FAILPOINTS:
+            if remote_storage_kind is RemoteStorageKind.NOOP and delete_failpoint in (
+                "timeline-delete-before-index-delete",
+            ):
+                # the above failpoint are not relevant for config without remote storage
+                continue
+
             # Simulate failures for only one type of remote storage
             # to avoid log pollution and make tests run faster
             if remote_storage_kind is RemoteStorageKind.MOCK_S3:
@@ -193,32 +195,27 @@ def test_delete_tenant_exercise_crash_safety_failpoints(
         ]
     )
 
-    if simulate_failures:
-        env.pageserver.allowed_errors.extend(
-            [
-                # The deletion queue will complain when it encounters simulated S3 errors
-                ".*deletion executor: DeleteObjects request failed.*",
-            ]
-        )
-
     ps_http = env.pageserver.http_client()
 
     timeline_id = env.neon_cli.create_timeline("delete", tenant_id=tenant_id)
     with env.endpoints.create_start("delete", tenant_id=tenant_id) as endpoint:
         # generate enough layers
         run_pg_bench_small(pg_bin, endpoint.connstr())
-        last_flush_lsn_upload(env, endpoint, tenant_id, timeline_id)
+        if remote_storage_kind is RemoteStorageKind.NOOP:
+            wait_for_last_flush_lsn(env, endpoint, tenant_id, timeline_id)
+        else:
+            last_flush_lsn_upload(env, endpoint, tenant_id, timeline_id)
 
-        if remote_storage_kind in available_s3_storages():
-            assert_prefix_not_empty(
-                neon_env_builder,
-                prefix="/".join(
-                    (
-                        "tenants",
-                        str(tenant_id),
-                    )
-                ),
-            )
+            if remote_storage_kind in available_s3_storages():
+                assert_prefix_not_empty(
+                    neon_env_builder,
+                    prefix="/".join(
+                        (
+                            "tenants",
+                            str(tenant_id),
+                        )
+                    ),
+                )
 
     ps_http.configure_failpoints((failpoint, "return"))
 
@@ -249,7 +246,12 @@ def test_delete_tenant_exercise_crash_safety_failpoints(
         env.pageserver.stop()
         env.pageserver.start()
 
-        if failpoint in (
+        if (
+            remote_storage_kind is RemoteStorageKind.NOOP
+            and failpoint == "tenant-delete-before-create-local-mark"
+        ):
+            tenant_delete_wait_completed(ps_http, tenant_id, iterations=iterations)
+        elif failpoint in (
             "tenant-delete-before-shutdown",
             "tenant-delete-before-create-remote-mark",
         ):
@@ -381,7 +383,6 @@ def test_tenant_delete_is_resumed_on_attach(
     assert not tenant_path.exists()
 
     if remote_storage_kind in available_s3_storages():
-        ps_http.deletion_queue_flush(execute=True)
         assert_prefix_empty(
             neon_env_builder,
             prefix="/".join(

test_runner/regress/test_tenant_detach.py

@@ -519,8 +519,11 @@ def test_detach_while_attaching(
 # * restart the pageserver and verify that ignored tenant is still not loaded
 # * `load` the same tenant
 # * ensure that it's status is `Active` and it's present in pageserver's memory with all timelines
-def test_ignored_tenant_reattach(neon_env_builder: NeonEnvBuilder):
-    neon_env_builder.enable_pageserver_remote_storage(RemoteStorageKind.MOCK_S3)
+@pytest.mark.parametrize("remote_storage_kind", [RemoteStorageKind.NOOP, RemoteStorageKind.MOCK_S3])
+def test_ignored_tenant_reattach(
+    neon_env_builder: NeonEnvBuilder, remote_storage_kind: RemoteStorageKind
+):
+    neon_env_builder.enable_pageserver_remote_storage(remote_storage_kind)
     env = neon_env_builder.init_start()
     pageserver_http = env.pageserver.http_client()
 

test_runner/regress/test_tenants.py

@@ -12,6 +12,7 @@ from fixtures.log_helper import log
 from fixtures.metrics import (
     PAGESERVER_GLOBAL_METRICS,
     PAGESERVER_PER_TENANT_METRICS,
+    PAGESERVER_PER_TENANT_REMOTE_TIMELINE_CLIENT_METRICS,
     parse_metrics,
 )
 from fixtures.neon_fixtures import (
@@ -231,10 +232,17 @@ def test_metrics_normal_work(neon_env_builder: NeonEnvBuilder):
         assert value
 
 
-def test_pageserver_metrics_removed_after_detach(neon_env_builder: NeonEnvBuilder):
+@pytest.mark.parametrize(
+    "remote_storage_kind",
+    # exercise both the code paths where remote_storage=None and remote_storage=Some(...)
+    [RemoteStorageKind.NOOP, RemoteStorageKind.MOCK_S3],
+)
+def test_pageserver_metrics_removed_after_detach(
+    neon_env_builder: NeonEnvBuilder, remote_storage_kind: RemoteStorageKind
+):
     """Tests that when a tenant is detached, the tenant specific metrics are not left behind"""
 
-    neon_env_builder.enable_pageserver_remote_storage(RemoteStorageKind.MOCK_S3)
+    neon_env_builder.enable_pageserver_remote_storage(remote_storage_kind)
 
     neon_env_builder.num_safekeepers = 3
 
@@ -274,6 +282,9 @@ def test_pageserver_metrics_removed_after_detach(neon_env_builder: NeonEnvBuilde
     for tenant in [tenant_1, tenant_2]:
         pre_detach_samples = set([x.name for x in get_ps_metric_samples_for_tenant(tenant)])
         expected = set(PAGESERVER_PER_TENANT_METRICS)
+        if remote_storage_kind == RemoteStorageKind.NOOP:
+            # if there's no remote storage configured, we don't expose the remote timeline client metrics
+            expected -= set(PAGESERVER_PER_TENANT_REMOTE_TIMELINE_CLIENT_METRICS)
         assert pre_detach_samples == expected
 
         env.pageserver.http_client().tenant_detach(tenant)
@@ -283,7 +294,9 @@ def test_pageserver_metrics_removed_after_detach(neon_env_builder: NeonEnvBuilde
 
 
 # Check that empty tenants work with or without the remote storage
-@pytest.mark.parametrize("remote_storage_kind", available_remote_storages())
+@pytest.mark.parametrize(
+    "remote_storage_kind", available_remote_storages() + [RemoteStorageKind.NOOP]
+)
 def test_pageserver_with_empty_tenants(
     neon_env_builder: NeonEnvBuilder, remote_storage_kind: RemoteStorageKind
 ):

test_runner/regress/test_timeline_delete.py

@@ -12,6 +12,7 @@ from fixtures.neon_fixtures import (
     NeonEnvBuilder,
     PgBin,
     last_flush_lsn_upload,
+    wait_for_last_flush_lsn,
 )
 from fixtures.pageserver.http import PageserverApiException
 from fixtures.pageserver.utils import (
@@ -144,12 +145,19 @@ DELETE_FAILPOINTS = [
 def combinations():
     result = []
 
-    remotes = [RemoteStorageKind.MOCK_S3]
+    remotes = [RemoteStorageKind.NOOP, RemoteStorageKind.MOCK_S3]
     if os.getenv("ENABLE_REAL_S3_REMOTE_STORAGE"):
         remotes.append(RemoteStorageKind.REAL_S3)
 
     for remote_storage_kind in remotes:
         for delete_failpoint in DELETE_FAILPOINTS:
+            if remote_storage_kind == RemoteStorageKind.NOOP and delete_failpoint in (
+                "timeline-delete-before-index-delete",
+                "timeline-delete-after-index-delete",
+            ):
+                # the above failpoints are not relevant for config without remote storage
+                continue
+
             result.append((remote_storage_kind, delete_failpoint))
     return result
 
@@ -197,21 +205,23 @@ def test_delete_timeline_exercise_crash_safety_failpoints(
     with env.endpoints.create_start("delete") as endpoint:
         # generate enough layers
         run_pg_bench_small(pg_bin, endpoint.connstr())
+        if remote_storage_kind is RemoteStorageKind.NOOP:
+            wait_for_last_flush_lsn(env, endpoint, env.initial_tenant, timeline_id)
+        else:
+            last_flush_lsn_upload(env, endpoint, env.initial_tenant, timeline_id)
 
-        last_flush_lsn_upload(env, endpoint, env.initial_tenant, timeline_id)
-
-        if remote_storage_kind in available_s3_storages():
-            assert_prefix_not_empty(
-                neon_env_builder,
-                prefix="/".join(
-                    (
-                        "tenants",
-                        str(env.initial_tenant),
-                        "timelines",
-                        str(timeline_id),
-                    )
-                ),
-            )
+            if remote_storage_kind in available_s3_storages():
+                assert_prefix_not_empty(
+                    neon_env_builder,
+                    prefix="/".join(
+                        (
+                            "tenants",
+                            str(env.initial_tenant),
+                            "timelines",
+                            str(timeline_id),
+                        )
+                    ),
+                )
 
     env.pageserver.allowed_errors.append(f".*{timeline_id}.*failpoint: {failpoint}")
     # It appears when we stopped flush loop during deletion and then pageserver is stopped
@@ -797,8 +807,6 @@ def test_delete_orphaned_objects(
     reason = timeline_info["state"]["Broken"]["reason"]
     assert reason.endswith(f"failpoint: {failpoint}"), reason
 
-    ps_http.deletion_queue_flush(execute=True)
-
     for orphan in orphans:
         assert not orphan.exists()
         assert env.pageserver.log_contains(

test_runner/regress/test_timeline_size.py

@@ -301,8 +301,12 @@ def test_timeline_initial_logical_size_calculation_cancellation(
     # message emitted by the code behind failpoint "timeline-calculate-logical-size-check-dir-exists"
 
 
-def test_timeline_physical_size_init(neon_env_builder: NeonEnvBuilder):
-    neon_env_builder.enable_pageserver_remote_storage(RemoteStorageKind.LOCAL_FS)
+@pytest.mark.parametrize("remote_storage_kind", [None, RemoteStorageKind.LOCAL_FS])
+def test_timeline_physical_size_init(
+    neon_env_builder: NeonEnvBuilder, remote_storage_kind: Optional[RemoteStorageKind]
+):
+    if remote_storage_kind is not None:
+        neon_env_builder.enable_pageserver_remote_storage(remote_storage_kind)
 
     env = neon_env_builder.init_start()
 
@@ -333,12 +337,17 @@ def test_timeline_physical_size_init(neon_env_builder: NeonEnvBuilder):
     )
 
     assert_physical_size_invariants(
-        get_physical_size_values(env, env.initial_tenant, new_timeline_id),
+        get_physical_size_values(env, env.initial_tenant, new_timeline_id, remote_storage_kind),
+        remote_storage_kind,
     )
 
 
-def test_timeline_physical_size_post_checkpoint(neon_env_builder: NeonEnvBuilder):
-    neon_env_builder.enable_pageserver_remote_storage(RemoteStorageKind.LOCAL_FS)
+@pytest.mark.parametrize("remote_storage_kind", [None, RemoteStorageKind.LOCAL_FS])
+def test_timeline_physical_size_post_checkpoint(
+    neon_env_builder: NeonEnvBuilder, remote_storage_kind: Optional[RemoteStorageKind]
+):
+    if remote_storage_kind is not None:
+        neon_env_builder.enable_pageserver_remote_storage(remote_storage_kind)
 
     env = neon_env_builder.init_start()
 
@@ -360,14 +369,19 @@ def test_timeline_physical_size_post_checkpoint(neon_env_builder: NeonEnvBuilder
 
     def check():
         assert_physical_size_invariants(
-            get_physical_size_values(env, env.initial_tenant, new_timeline_id),
+            get_physical_size_values(env, env.initial_tenant, new_timeline_id, remote_storage_kind),
+            remote_storage_kind,
         )
 
     wait_until(10, 1, check)
 
 
-def test_timeline_physical_size_post_compaction(neon_env_builder: NeonEnvBuilder):
-    neon_env_builder.enable_pageserver_remote_storage(RemoteStorageKind.LOCAL_FS)
+@pytest.mark.parametrize("remote_storage_kind", [None, RemoteStorageKind.LOCAL_FS])
+def test_timeline_physical_size_post_compaction(
+    neon_env_builder: NeonEnvBuilder, remote_storage_kind: Optional[RemoteStorageKind]
+):
+    if remote_storage_kind is not None:
+        neon_env_builder.enable_pageserver_remote_storage(remote_storage_kind)
 
     # Disable background compaction as we don't want it to happen after `get_physical_size` request
     # and before checking the expected size on disk, which makes the assertion failed
@@ -406,15 +420,21 @@ def test_timeline_physical_size_post_compaction(neon_env_builder: NeonEnvBuilder
     pageserver_http.timeline_checkpoint(env.initial_tenant, new_timeline_id)
     pageserver_http.timeline_compact(env.initial_tenant, new_timeline_id)
 
-    wait_for_upload_queue_empty(pageserver_http, env.initial_tenant, new_timeline_id)
+    if remote_storage_kind is not None:
+        wait_for_upload_queue_empty(pageserver_http, env.initial_tenant, new_timeline_id)
 
     assert_physical_size_invariants(
-        get_physical_size_values(env, env.initial_tenant, new_timeline_id),
+        get_physical_size_values(env, env.initial_tenant, new_timeline_id, remote_storage_kind),
+        remote_storage_kind,
     )
 
 
-def test_timeline_physical_size_post_gc(neon_env_builder: NeonEnvBuilder):
-    neon_env_builder.enable_pageserver_remote_storage(RemoteStorageKind.LOCAL_FS)
+@pytest.mark.parametrize("remote_storage_kind", [None, RemoteStorageKind.LOCAL_FS])
+def test_timeline_physical_size_post_gc(
+    neon_env_builder: NeonEnvBuilder, remote_storage_kind: Optional[RemoteStorageKind]
+):
+    if remote_storage_kind is not None:
+        neon_env_builder.enable_pageserver_remote_storage(remote_storage_kind)
 
     # Disable background compaction and GC as we don't want it to happen after `get_physical_size` request
     # and before checking the expected size on disk, which makes the assertion failed
@@ -451,10 +471,12 @@ def test_timeline_physical_size_post_gc(neon_env_builder: NeonEnvBuilder):
     pageserver_http.timeline_checkpoint(env.initial_tenant, new_timeline_id)
     pageserver_http.timeline_gc(env.initial_tenant, new_timeline_id, gc_horizon=None)
 
-    wait_for_upload_queue_empty(pageserver_http, env.initial_tenant, new_timeline_id)
+    if remote_storage_kind is not None:
+        wait_for_upload_queue_empty(pageserver_http, env.initial_tenant, new_timeline_id)
 
     assert_physical_size_invariants(
-        get_physical_size_values(env, env.initial_tenant, new_timeline_id),
+        get_physical_size_values(env, env.initial_tenant, new_timeline_id, remote_storage_kind),
+        remote_storage_kind,
     )
 
 
@@ -538,10 +560,14 @@ def test_timeline_size_metrics(
     assert math.isclose(dbsize_sum, tl_logical_size_metric, abs_tol=2 * 1024 * 1024)
 
 
-def test_tenant_physical_size(neon_env_builder: NeonEnvBuilder):
+@pytest.mark.parametrize("remote_storage_kind", [None, RemoteStorageKind.LOCAL_FS])
+def test_tenant_physical_size(
+    neon_env_builder: NeonEnvBuilder, remote_storage_kind: Optional[RemoteStorageKind]
+):
     random.seed(100)
 
-    neon_env_builder.enable_pageserver_remote_storage(RemoteStorageKind.LOCAL_FS)
+    if remote_storage_kind is not None:
+        neon_env_builder.enable_pageserver_remote_storage(remote_storage_kind)
 
     env = neon_env_builder.init_start()
 
@@ -549,10 +575,12 @@ def test_tenant_physical_size(neon_env_builder: NeonEnvBuilder):
     client = env.pageserver.http_client()
 
     tenant, timeline = env.neon_cli.create_tenant()
+    if remote_storage_kind is not None:
+        wait_for_upload_queue_empty(pageserver_http, tenant, timeline)
 
     def get_timeline_resident_physical_size(timeline: TimelineId):
-        sizes = get_physical_size_values(env, tenant, timeline)
-        assert_physical_size_invariants(sizes)
+        sizes = get_physical_size_values(env, tenant, timeline, remote_storage_kind)
+        assert_physical_size_invariants(sizes, remote_storage_kind)
         return sizes.prometheus_resident_physical
 
     timeline_total_resident_physical_size = get_timeline_resident_physical_size(timeline)
@@ -572,7 +600,8 @@ def test_tenant_physical_size(neon_env_builder: NeonEnvBuilder):
         wait_for_last_flush_lsn(env, endpoint, tenant, timeline)
         pageserver_http.timeline_checkpoint(tenant, timeline)
 
-        wait_for_upload_queue_empty(pageserver_http, tenant, timeline)
+        if remote_storage_kind is not None:
+            wait_for_upload_queue_empty(pageserver_http, tenant, timeline)
 
         timeline_total_resident_physical_size += get_timeline_resident_physical_size(timeline)
 
@@ -601,6 +630,7 @@ def get_physical_size_values(
     env: NeonEnv,
     tenant_id: TenantId,
     timeline_id: TimelineId,
+    remote_storage_kind: Optional[RemoteStorageKind],
 ) -> TimelinePhysicalSizeValues:
     res = TimelinePhysicalSizeValues()
 
@@ -616,9 +646,12 @@ def get_physical_size_values(
     res.prometheus_resident_physical = metrics.query_one(
         "pageserver_resident_physical_size", metrics_filter
     ).value
-    res.prometheus_remote_physical = metrics.query_one(
-        "pageserver_remote_physical_size", metrics_filter
-    ).value
+    if remote_storage_kind is not None:
+        res.prometheus_remote_physical = metrics.query_one(
+            "pageserver_remote_physical_size", metrics_filter
+        ).value
+    else:
+        res.prometheus_remote_physical = None
 
     detail = client.timeline_detail(
         tenant_id, timeline_id, include_timeline_dir_layer_file_size_sum=True
@@ -631,15 +664,20 @@ def get_physical_size_values(
     return res
 
 
-def assert_physical_size_invariants(sizes: TimelinePhysicalSizeValues):
+def assert_physical_size_invariants(
+    sizes: TimelinePhysicalSizeValues, remote_storage_kind: Optional[RemoteStorageKind]
+):
     # resident phyiscal size is defined as
     assert sizes.python_timelinedir_layerfiles_physical == sizes.prometheus_resident_physical
     assert sizes.python_timelinedir_layerfiles_physical == sizes.layer_map_file_size_sum
 
     # we don't do layer eviction, so, all layers are resident
     assert sizes.api_current_physical == sizes.prometheus_resident_physical
-    assert sizes.prometheus_resident_physical == sizes.prometheus_remote_physical
-    # XXX would be nice to assert layer file physical storage utilization here as well, but we can only do that for LocalFS
+    if remote_storage_kind is not None:
+        assert sizes.prometheus_resident_physical == sizes.prometheus_remote_physical
+        # XXX would be nice to assert layer file physical storage utilization here as well, but we can only do that for LocalFS
+    else:
+        assert sizes.prometheus_remote_physical is None
 
 
 # Timeline logical size initialization is an asynchronous background task that runs once,

vendor/revisions.json

@@ -1,5 +1,5 @@
 {
-    "postgres-v16": "389ce36b4b3da7aa654a25e1b3f10b641319a87f",
-    "postgres-v15": "74cfe3e681836747a31fdbd47bdd14b3d81b0772",
+    "postgres-v16": "7a50f139c6269454ab9260c7a9752874b9089943",
+    "postgres-v15": "026d6b093d49e25cec44dd04598152329ceac027",
     "postgres-v14": "5d5cfee12783f0989a9c9fe13bb40b5585812568"
 }

workspace_hack/Cargo.toml

@@ -64,7 +64,6 @@ toml_edit = { version = "0.19", features = ["serde"] }
 tower = { version = "0.4", features = ["balance", "buffer", "limit", "retry", "timeout", "util"] }
 tracing = { version = "0.1", features = ["log"] }
 tracing-core = { version = "0.1" }
-tungstenite = { version = "0.20" }
 url = { version = "2", features = ["serde"] }
 uuid = { version = "1", features = ["serde", "v4"] }