Compare commits

..

11 Commits

Author SHA1 Message Date
Arseny Sher
a8a1cc5e80 Run pgindent on walproposer. 2023-09-29 13:16:48 +03:00
Arseny Sher
8a0b3d7c32 add walproposer_c top comment 2023-09-29 12:32:25 +03:00
Arthur Petukhovsky
1a19b4d2e7 Add comments 2023-09-27 13:00:27 +00:00
Arthur Petukhovsky
3bf6ecb2a6 Make walproposer 100% globals free 2023-09-27 13:00:27 +00:00
Arthur Petukhovsky
f550c26aca Cleanup walproposer.h a bit 2023-09-27 13:00:27 +00:00
Arthur Petukhovsky
2ce477b622 Cleanup imports 2023-09-27 13:00:27 +00:00
Arthur Petukhovsky
e8db6b91c5 Move safekeeper feedback handling 2023-09-27 13:00:27 +00:00
Arthur Petukhovsky
029623bfe7 Abstract away waitEvents handling 2023-09-27 13:00:27 +00:00
Arthur Petukhovsky
227eb21303 Rename walproposer_utils to neon_utils 2023-09-27 13:00:27 +00:00
Arthur Petukhovsky
0e1ff5db4c Move libpqwalproposer to walproposer_pg 2023-09-27 13:00:27 +00:00
Arthur Petukhovsky
fbaca131ca Extract pg-dependent init walproposer code 2023-09-27 13:00:27 +00:00
133 changed files with 4647 additions and 6109 deletions

View File

@@ -76,8 +76,8 @@ runs:
rm -f ${ALLURE_ZIP}
fi
env:
ALLURE_VERSION: 2.24.0
ALLURE_ZIP_SHA256: 60b1d6ce65d9ef24b23cf9c2c19fd736a123487c38e54759f1ed1a7a77353c90
ALLURE_VERSION: 2.23.1
ALLURE_ZIP_SHA256: 11141bfe727504b3fd80c0f9801eb317407fd0ac983ebb57e671f14bac4bcd86
# Potentially we could have several running build for the same key (for example, for the main branch), so we use improvised lock for this
- name: Acquire lock

View File

@@ -834,7 +834,7 @@ jobs:
run:
shell: sh -eu {0}
env:
VM_BUILDER_VERSION: v0.17.12
VM_BUILDER_VERSION: v0.17.11
steps:
- name: Checkout

View File

@@ -5,7 +5,7 @@
/libs/remote_storage/ @neondatabase/storage
/libs/safekeeper_api/ @neondatabase/safekeepers
/libs/vm_monitor/ @neondatabase/autoscaling @neondatabase/compute
/pageserver/ @neondatabase/storage
/pageserver/ @neondatabase/compute @neondatabase/storage
/pgxn/ @neondatabase/compute
/proxy/ @neondatabase/proxy
/safekeeper/ @neondatabase/safekeepers

83
Cargo.lock generated
View File

@@ -158,17 +158,6 @@ dependencies = [
"syn 1.0.109",
]
[[package]]
name = "async-channel"
version = "1.9.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "81953c529336010edd6d8e358f886d9581267795c61b19475b71314bffa46d35"
dependencies = [
"concurrent-queue",
"event-listener",
"futures-core",
]
[[package]]
name = "async-compression"
version = "0.4.0"
@@ -809,22 +798,6 @@ dependencies = [
"either",
]
[[package]]
name = "camino"
version = "1.1.6"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "c59e92b5a388f549b863a7bea62612c09f24c8393560709a54558a9abdfb3b9c"
[[package]]
name = "camino-tempfile"
version = "1.0.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d2ab15a83d13f75dbd86f082bdefd160b628476ef58d3b900a0ef74e001bb097"
dependencies = [
"camino",
"tempfile",
]
[[package]]
name = "cast"
version = "0.3.0"
@@ -1042,15 +1015,6 @@ dependencies = [
"zstd",
]
[[package]]
name = "concurrent-queue"
version = "2.3.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "f057a694a54f12365049b0958a1685bb52d567f5593b355fbf685838e873d400"
dependencies = [
"crossbeam-utils",
]
[[package]]
name = "const_format"
version = "0.2.30"
@@ -1089,7 +1053,6 @@ name = "control_plane"
version = "0.1.0"
dependencies = [
"anyhow",
"camino",
"clap",
"comfy-table",
"compute_api",
@@ -1472,12 +1435,6 @@ dependencies = [
"libc",
]
[[package]]
name = "event-listener"
version = "2.5.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "0206175f82b8d6bf6652ff7d71a1e27fd2e4efde587fd368662814d6ec1d9ce0"
[[package]]
name = "fail"
version = "0.5.1"
@@ -1823,9 +1780,18 @@ checksum = "95505c38b4572b2d910cecb0281560f54b440a19336cbbcb27bf6ce6adc6f5a8"
[[package]]
name = "hermit-abi"
version = "0.3.3"
version = "0.2.6"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d77f7ec81a6d05a3abb01ab6eb7590f6083d08449fe5a1c8b1e620283546ccb7"
checksum = "ee512640fe35acbfb4bb779db6f0d80704c2cacfa2e39b601ef3e3f47d1ae4c7"
dependencies = [
"libc",
]
[[package]]
name = "hermit-abi"
version = "0.3.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "fed44880c466736ef9a5c5b5facefb5ed0785676d0c02d612db14e54f0d84286"
[[package]]
name = "hex"
@@ -2087,7 +2053,7 @@ version = "1.0.11"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "eae7b9aee968036d54dce06cebaefd919e4472e753296daccd6d344e3e2df0c2"
dependencies = [
"hermit-abi",
"hermit-abi 0.3.1",
"libc",
"windows-sys 0.48.0",
]
@@ -2104,7 +2070,7 @@ version = "0.4.7"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "adcf93614601c8129ddf72e2d5633df827ba6551541c6d8c59520a371475be1f"
dependencies = [
"hermit-abi",
"hermit-abi 0.3.1",
"io-lifetimes",
"rustix 0.37.19",
"windows-sys 0.48.0",
@@ -2478,11 +2444,11 @@ dependencies = [
[[package]]
name = "num_cpus"
version = "1.16.0"
version = "1.15.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "4161fcb6d602d4d2081af7c3a45852d875a03dd337a6bfdd6e06407b61342a43"
checksum = "0fac9e2da13b5eb447a6ce3d392f23a29d8694bff781bf03a16cd9ac8697593b"
dependencies = [
"hermit-abi",
"hermit-abi 0.2.6",
"libc",
]
@@ -2684,7 +2650,6 @@ version = "0.1.0"
dependencies = [
"anyhow",
"bytes",
"camino",
"clap",
"git-version",
"pageserver",
@@ -2700,14 +2665,11 @@ name = "pageserver"
version = "0.1.0"
dependencies = [
"anyhow",
"async-channel",
"async-compression",
"async-stream",
"async-trait",
"byteorder",
"bytes",
"camino",
"camino-tempfile",
"chrono",
"clap",
"close_fds",
@@ -2759,6 +2721,7 @@ dependencies = [
"strum_macros",
"svg_fmt",
"sync_wrapper",
"tempfile",
"tenant_size_model",
"thiserror",
"tokio",
@@ -3283,7 +3246,6 @@ dependencies = [
"reqwest-tracing",
"routerify",
"rstest",
"rustc-hash",
"rustls",
"rustls-pemfile",
"scopeguard",
@@ -3451,8 +3413,6 @@ dependencies = [
"aws-sdk-s3",
"aws-smithy-http",
"aws-types",
"camino",
"camino-tempfile",
"hyper",
"metrics",
"once_cell",
@@ -3461,6 +3421,7 @@ dependencies = [
"scopeguard",
"serde",
"serde_json",
"tempfile",
"test-context",
"tokio",
"tokio-util",
@@ -3812,8 +3773,6 @@ dependencies = [
"async-trait",
"byteorder",
"bytes",
"camino",
"camino-tempfile",
"chrono",
"clap",
"const_format",
@@ -3842,6 +3801,7 @@ dependencies = [
"serde_with",
"signal-hook",
"storage_broker",
"tempfile",
"thiserror",
"tokio",
"tokio-io-timeout",
@@ -5140,8 +5100,6 @@ dependencies = [
"bincode",
"byteorder",
"bytes",
"camino",
"camino-tempfile",
"chrono",
"const_format",
"criterion",
@@ -5167,6 +5125,7 @@ dependencies = [
"signal-hook",
"strum",
"strum_macros",
"tempfile",
"thiserror",
"tokio",
"tokio-stream",
@@ -5240,7 +5199,6 @@ name = "wal_craft"
version = "0.1.0"
dependencies = [
"anyhow",
"camino-tempfile",
"clap",
"env_logger",
"log",
@@ -5248,6 +5206,7 @@ dependencies = [
"postgres",
"postgres_ffi",
"regex",
"tempfile",
"utils",
"workspace_hack",
]

View File

@@ -35,7 +35,6 @@ license = "Apache-2.0"
## All dependency versions, used in the project
[workspace.dependencies]
anyhow = { version = "1.0", features = ["backtrace"] }
async-channel = "1.9.0"
async-compression = { version = "0.4.0", features = ["tokio", "gzip"] }
flate2 = "1.0.26"
async-stream = "0.3"
@@ -52,7 +51,6 @@ bindgen = "0.65"
bstr = "1.0"
byteorder = "1.4"
bytes = "1.0"
camino = "1.1.6"
cfg-if = "1.0.0"
chrono = { version = "0.4", default-features = false, features = ["clock"] }
clap = { version = "4.0", features = ["derive"] }
@@ -109,7 +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"
@@ -189,7 +186,7 @@ workspace_hack = { version = "0.1", path = "./workspace_hack/" }
criterion = "0.5.1"
rcgen = "0.11"
rstest = "0.18"
camino-tempfile = "1.0.2"
tempfile = "3.4"
tonic-build = "0.9"
[patch.crates-io]

View File

@@ -615,7 +615,11 @@ RUN wget https://gitlab.com/dalibo/postgresql_anonymizer/-/archive/1.1.0/postgre
#########################################################################################
#
# Layer "rust extensions"
# This layer is used to build `pgrx` deps
# This layer is used to build `pgx` deps
#
# FIXME: This needs to be updated to latest version of 'pgrx' (it was renamed from
# 'pgx' to 'pgrx') for PostgreSQL 16. And that in turn requires bumping the pgx
# dependency on all the rust extension that depend on it, too.
#
#########################################################################################
FROM build-deps AS rust-extensions-build
@@ -631,12 +635,22 @@ USER nonroot
WORKDIR /home/nonroot
ARG PG_VERSION
RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux-gnu/rustup-init && \
RUN case "${PG_VERSION}" in \
"v14" | "v15") \
;; \
"v16") \
echo "TODO: Not yet supported for PostgreSQL 16. Need to update pgrx dependencies" && exit 0 \
;; \
*) \
echo "unexpected PostgreSQL version ${PG_VERSION}" && exit 1 \
;; \
esac && \
curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux-gnu/rustup-init && \
chmod +x rustup-init && \
./rustup-init -y --no-modify-path --profile minimal --default-toolchain stable && \
rm rustup-init && \
cargo install --locked --version 0.10.2 cargo-pgrx && \
/bin/bash -c 'cargo pgrx init --pg${PG_VERSION:1}=/usr/local/pgsql/bin/pg_config'
cargo install --locked --version 0.7.3 cargo-pgx && \
/bin/bash -c 'cargo pgx init --pg${PG_VERSION:1}=/usr/local/pgsql/bin/pg_config'
USER root
@@ -650,11 +664,23 @@ USER root
FROM rust-extensions-build AS pg-jsonschema-pg-build
ARG PG_VERSION
RUN wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v0.2.0.tar.gz -O pg_jsonschema.tar.gz && \
echo "9118fc508a6e231e7a39acaa6f066fcd79af17a5db757b47d2eefbe14f7794f0 pg_jsonschema.tar.gz" | sha256sum --check && \
# caeab60d70b2fd3ae421ec66466a3abbb37b7ee6 made on 06/03/2023
# there is no release tag yet, but we need it due to the superuser fix in the control file, switch to git tag after release >= 0.1.5
RUN case "${PG_VERSION}" in \
"v14" | "v15") \
;; \
"v16") \
echo "TODO: Not yet supported for PostgreSQL 16. Need to update pgrx dependencies" && exit 0 \
;; \
*) \
echo "unexpected PostgreSQL version \"${PG_VERSION}\"" && exit 1 \
;; \
esac && \
wget https://github.com/supabase/pg_jsonschema/archive/caeab60d70b2fd3ae421ec66466a3abbb37b7ee6.tar.gz -O pg_jsonschema.tar.gz && \
echo "54129ce2e7ee7a585648dbb4cef6d73f795d94fe72f248ac01119992518469a4 pg_jsonschema.tar.gz" | sha256sum --check && \
mkdir pg_jsonschema-src && cd pg_jsonschema-src && tar xvzf ../pg_jsonschema.tar.gz --strip-components=1 -C . && \
sed -i 's/pgrx = "0.10.2"/pgrx = { version = "0.10.2", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
cargo pgrx install --release && \
sed -i 's/pgx = "0.7.1"/pgx = { version = "0.7.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
cargo pgx install --release && \
echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_jsonschema.control
#########################################################################################
@@ -667,11 +693,26 @@ RUN wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v0.2.0.tar.
FROM rust-extensions-build AS pg-graphql-pg-build
ARG PG_VERSION
RUN wget https://github.com/supabase/pg_graphql/archive/refs/tags/v1.4.0.tar.gz -O pg_graphql.tar.gz && \
echo "bd8dc7230282b3efa9ae5baf053a54151ed0e66881c7c53750e2d0c765776edc pg_graphql.tar.gz" | sha256sum --check && \
# b4988843647450a153439be367168ed09971af85 made on 22/02/2023 (from remove-pgx-contrib-spiext branch)
# Currently pgx version bump to >= 0.7.2 causes "call to unsafe function" compliation errors in
# pgx-contrib-spiext. There is a branch that removes that dependency, so use it. It is on the
# same 1.1 version we've used before.
RUN case "${PG_VERSION}" in \
"v14" | "v15") \
;; \
"v16") \
echo "TODO: Not yet supported for PostgreSQL 16. Need to update pgrx dependencies" && exit 0 \
;; \
*) \
echo "unexpected PostgreSQL version" && exit 1 \
;; \
esac && \
wget https://github.com/yrashk/pg_graphql/archive/b4988843647450a153439be367168ed09971af85.tar.gz -O pg_graphql.tar.gz && \
echo "0c7b0e746441b2ec24187d0e03555faf935c2159e2839bddd14df6dafbc8c9bd pg_graphql.tar.gz" | sha256sum --check && \
mkdir pg_graphql-src && cd pg_graphql-src && tar xvzf ../pg_graphql.tar.gz --strip-components=1 -C . && \
sed -i 's/pgrx = "=0.10.2"/pgrx = { version = "0.10.2", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
cargo pgrx install --release && \
sed -i 's/pgx = "~0.7.1"/pgx = { version = "0.7.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
sed -i 's/pgx-tests = "~0.7.1"/pgx-tests = "0.7.3"/g' Cargo.toml && \
cargo pgx install --release && \
# it's needed to enable extension because it uses untrusted C language
sed -i 's/superuser = false/superuser = true/g' /usr/local/pgsql/share/extension/pg_graphql.control && \
echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_graphql.control
@@ -686,11 +727,21 @@ RUN wget https://github.com/supabase/pg_graphql/archive/refs/tags/v1.4.0.tar.gz
FROM rust-extensions-build AS pg-tiktoken-pg-build
ARG PG_VERSION
# 26806147b17b60763039c6a6878884c41a262318 made on 26/09/2023
RUN wget https://github.com/kelvich/pg_tiktoken/archive/26806147b17b60763039c6a6878884c41a262318.tar.gz -O pg_tiktoken.tar.gz && \
echo "e64e55aaa38c259512d3e27c572da22c4637418cf124caba904cd50944e5004e pg_tiktoken.tar.gz" | sha256sum --check && \
# 801f84f08c6881c8aa30f405fafbf00eec386a72 made on 10/03/2023
RUN case "${PG_VERSION}" in \
"v14" | "v15") \
;; \
"v16") \
echo "TODO: Not yet supported for PostgreSQL 16. Need to update pgrx dependencies" && exit 0 \
;; \
*) \
echo "unexpected PostgreSQL version" && exit 1 \
;; \
esac && \
wget https://github.com/kelvich/pg_tiktoken/archive/801f84f08c6881c8aa30f405fafbf00eec386a72.tar.gz -O pg_tiktoken.tar.gz && \
echo "52f60ac800993a49aa8c609961842b611b6b1949717b69ce2ec9117117e16e4a pg_tiktoken.tar.gz" | sha256sum --check && \
mkdir pg_tiktoken-src && cd pg_tiktoken-src && tar xvzf ../pg_tiktoken.tar.gz --strip-components=1 -C . && \
cargo pgrx install --release && \
cargo pgx install --release && \
echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_tiktoken.control
#########################################################################################
@@ -703,15 +754,21 @@ RUN wget https://github.com/kelvich/pg_tiktoken/archive/26806147b17b60763039c6a6
FROM rust-extensions-build AS pg-pgx-ulid-build
ARG PG_VERSION
RUN wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.3.tar.gz -O pgx_ulid.tar.gz && \
echo "ee5db82945d2d9f2d15597a80cf32de9dca67b897f605beb830561705f12683c pgx_ulid.tar.gz" | sha256sum --check && \
RUN case "${PG_VERSION}" in \
"v14" | "v15") \
;; \
"v16") \
echo "TODO: Not yet supported for PostgreSQL 16. Need to update pgrx dependencies" && exit 0 \
;; \
*) \
echo "unexpected PostgreSQL version" && exit 1 \
;; \
esac && \
wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.0.tar.gz -O pgx_ulid.tar.gz && \
echo "908b7358e6f846e87db508ae5349fb56a88ee6305519074b12f3d5b0ff09f791 pgx_ulid.tar.gz" | sha256sum --check && \
mkdir pgx_ulid-src && cd pgx_ulid-src && tar xvzf ../pgx_ulid.tar.gz --strip-components=1 -C . && \
echo "******************* Apply a patch for Postgres 16 support; delete in the next release ******************" && \
wget https://github.com/pksunkara/pgx_ulid/commit/f84954cf63fc8c80d964ac970d9eceed3c791196.patch && \
patch -p1 < f84954cf63fc8c80d964ac970d9eceed3c791196.patch && \
echo "********************************************************************************************************" && \
sed -i 's/pgrx = "=0.10.2"/pgrx = { version = "=0.10.2", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
cargo pgrx install --release && \
sed -i 's/pgx = "=0.7.3"/pgx = { version = "0.7.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
cargo pgx install --release && \
echo "trusted = true" >> /usr/local/pgsql/share/extension/ulid.control
#########################################################################################

View File

@@ -1039,7 +1039,7 @@ LIMIT 100",
let remote_extensions = spec
.remote_extensions
.as_ref()
.ok_or(anyhow::anyhow!("Remote extensions are not configured"))?;
.ok_or(anyhow::anyhow!("Remote extensions are not configured",))?;
info!("parse shared_preload_libraries from spec.cluster.settings");
let mut libs_vec = Vec::new();

View File

@@ -1,5 +1,5 @@
use std::sync::Arc;
use std::{thread, time::Duration};
use std::{thread, time};
use chrono::{DateTime, Utc};
use postgres::{Client, NoTls};
@@ -7,7 +7,7 @@ use tracing::{debug, info};
use crate::compute::ComputeNode;
const MONITOR_CHECK_INTERVAL: Duration = Duration::from_millis(500);
const MONITOR_CHECK_INTERVAL: u64 = 500; // milliseconds
// Spin in a loop and figure out the last activity time in the Postgres.
// Then update it in the shared state. This function never errors out.
@@ -17,12 +17,13 @@ fn watch_compute_activity(compute: &ComputeNode) {
let connstr = compute.connstr.as_str();
// Define `client` outside of the loop to reuse existing connection if it's active.
let mut client = Client::connect(connstr, NoTls);
let timeout = time::Duration::from_millis(MONITOR_CHECK_INTERVAL);
info!("watching Postgres activity at {}", connstr);
loop {
// Should be outside of the write lock to allow others to read while we sleep.
thread::sleep(MONITOR_CHECK_INTERVAL);
thread::sleep(timeout);
match &mut client {
Ok(cli) => {

View File

@@ -6,7 +6,6 @@ license.workspace = true
[dependencies]
anyhow.workspace = true
camino.workspace = true
clap.workspace = true
comfy-table.workspace = true
git-version.workspace = true

View File

@@ -1,6 +1,5 @@
use crate::{background_process, local_env::LocalEnv};
use anyhow::anyhow;
use camino::Utf8PathBuf;
use serde::{Deserialize, Serialize};
use serde_with::{serde_as, DisplayFromStr};
use std::{path::PathBuf, process::Child};
@@ -48,9 +47,8 @@ impl AttachmentService {
}
}
fn pid_file(&self) -> Utf8PathBuf {
Utf8PathBuf::from_path_buf(self.env.base_data_dir.join("attachment_service.pid"))
.expect("non-Unicode path")
fn pid_file(&self) -> PathBuf {
self.env.base_data_dir.join("attachment_service.pid")
}
pub fn start(&self) -> anyhow::Result<Child> {

View File

@@ -16,13 +16,12 @@ use std::ffi::OsStr;
use std::io::Write;
use std::os::unix::prelude::AsRawFd;
use std::os::unix::process::CommandExt;
use std::path::Path;
use std::path::{Path, PathBuf};
use std::process::{Child, Command};
use std::time::Duration;
use std::{fs, io, thread};
use anyhow::Context;
use camino::{Utf8Path, Utf8PathBuf};
use nix::errno::Errno;
use nix::fcntl::{FcntlArg, FdFlag};
use nix::sys::signal::{kill, Signal};
@@ -46,9 +45,9 @@ const NOTICE_AFTER_RETRIES: u64 = 50;
/// it itself.
pub enum InitialPidFile<'t> {
/// Create a pidfile, to allow future CLI invocations to manipulate the process.
Create(&'t Utf8Path),
Create(&'t Path),
/// The process will create the pidfile itself, need to wait for that event.
Expect(&'t Utf8Path),
Expect(&'t Path),
}
/// Start a background child process using the parameters given.
@@ -138,11 +137,7 @@ where
}
/// Stops the process, using the pid file given. Returns Ok also if the process is already not running.
pub fn stop_process(
immediate: bool,
process_name: &str,
pid_file: &Utf8Path,
) -> anyhow::Result<()> {
pub fn stop_process(immediate: bool, process_name: &str, pid_file: &Path) -> anyhow::Result<()> {
let pid = match pid_file::read(pid_file)
.with_context(|| format!("read pid_file {pid_file:?}"))?
{
@@ -257,9 +252,9 @@ fn fill_aws_secrets_vars(mut cmd: &mut Command) -> &mut Command {
/// will remain held until the cmd exits.
fn pre_exec_create_pidfile<P>(cmd: &mut Command, path: P) -> &mut Command
where
P: Into<Utf8PathBuf>,
P: Into<PathBuf>,
{
let path: Utf8PathBuf = path.into();
let path: PathBuf = path.into();
// SAFETY
// pre_exec is marked unsafe because it runs between fork and exec.
// Why is that dangerous in various ways?
@@ -316,7 +311,7 @@ where
fn process_started<F>(
pid: Pid,
pid_file_to_check: Option<&Utf8Path>,
pid_file_to_check: Option<&Path>,
status_check: &F,
) -> anyhow::Result<bool>
where

View File

@@ -7,7 +7,7 @@
//! ```
use anyhow::Context;
use camino::Utf8PathBuf;
use std::path::PathBuf;
use crate::{background_process, local_env};
@@ -30,7 +30,7 @@ pub fn start_broker_process(env: &local_env::LocalEnv) -> anyhow::Result<()> {
|| {
let url = broker.client_url();
let status_url = url.join("status").with_context(|| {
format!("Failed to append /status path to broker endpoint {url}")
format!("Failed to append /status path to broker endpoint {url}",)
})?;
let request = client
.get(status_url)
@@ -50,7 +50,6 @@ pub fn stop_broker_process(env: &local_env::LocalEnv) -> anyhow::Result<()> {
background_process::stop_process(true, "storage_broker", &storage_broker_pid_file_path(env))
}
fn storage_broker_pid_file_path(env: &local_env::LocalEnv) -> Utf8PathBuf {
Utf8PathBuf::from_path_buf(env.base_data_dir.join("storage_broker.pid"))
.expect("non-Unicode path")
fn storage_broker_pid_file_path(env: &local_env::LocalEnv) -> PathBuf {
env.base_data_dir.join("storage_broker.pid")
}

View File

@@ -14,7 +14,6 @@ use std::process::{Child, Command};
use std::{io, result};
use anyhow::{bail, Context};
use camino::Utf8PathBuf;
use pageserver_api::models::{self, TenantInfo, TimelineInfo};
use postgres_backend::AuthType;
use postgres_connection::{parse_host_port, PgConnectionConfig};
@@ -145,7 +144,7 @@ impl PageServerNode {
pub fn initialize(&self, config_overrides: &[&str]) -> anyhow::Result<()> {
// First, run `pageserver --init` and wait for it to write a config into FS and exit.
self.pageserver_init(config_overrides)
.with_context(|| format!("Failed to run init for pageserver node {}", self.conf.id))
.with_context(|| format!("Failed to run init for pageserver node {}", self.conf.id,))
}
pub fn repo_path(&self) -> PathBuf {
@@ -155,9 +154,8 @@ impl PageServerNode {
/// The pid file is created by the pageserver process, with its pid stored inside.
/// Other pageservers cannot lock the same file and overwrite it for as long as the current
/// pageserver runs. (Unless someone removes the file manually; never do that!)
fn pid_file(&self) -> Utf8PathBuf {
Utf8PathBuf::from_path_buf(self.repo_path().join("pageserver.pid"))
.expect("non-Unicode path")
fn pid_file(&self) -> PathBuf {
self.repo_path().join("pageserver.pid")
}
pub fn start(&self, config_overrides: &[&str]) -> anyhow::Result<Child> {

View File

@@ -11,7 +11,6 @@ use std::process::Child;
use std::{io, result};
use anyhow::Context;
use camino::Utf8PathBuf;
use postgres_connection::PgConnectionConfig;
use reqwest::blocking::{Client, RequestBuilder, Response};
use reqwest::{IntoUrl, Method};
@@ -98,9 +97,8 @@ impl SafekeeperNode {
SafekeeperNode::datadir_path_by_id(&self.env, self.id)
}
pub fn pid_file(&self) -> Utf8PathBuf {
Utf8PathBuf::from_path_buf(self.datadir_path().join("safekeeper.pid"))
.expect("non-Unicode path")
pub fn pid_file(&self) -> PathBuf {
self.datadir_path().join("safekeeper.pid")
}
pub fn start(&self, extra_opts: Vec<String>) -> anyhow::Result<Child> {

View File

@@ -23,7 +23,7 @@ vulnerability = "deny"
unmaintained = "warn"
yanked = "warn"
notice = "warn"
ignore = []
ignore = ["RUSTSEC-2023-0052"]
# This section is considered when running `cargo deny check licenses`
# More documentation for the licenses section can be found here:

View File

@@ -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.

View File

@@ -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]>,
}

View File

@@ -10,7 +10,6 @@ use serde_with::{serde_as, DisplayFromStr};
use strum_macros;
use utils::{
completion,
generation::Generation,
history_buffer::HistoryBufferWithDropCounter,
id::{NodeId, TenantId, TimelineId},
lsn::Lsn,
@@ -219,8 +218,6 @@ impl std::ops::Deref for TenantCreateRequest {
}
}
/// An alternative representation of `pageserver::tenant::TenantConf` with
/// simpler types.
#[derive(Serialize, Deserialize, Debug, Default)]
pub struct TenantConfig {
pub checkpoint_distance: Option<u64>,
@@ -246,39 +243,6 @@ pub struct TenantConfig {
pub gc_feedback: Option<bool>,
}
/// A flattened analog of a `pagesever::tenant::LocationMode`, which
/// lists out all possible states (and the virtual "Detached" state)
/// in a flat form rather than using rust-style enums.
#[derive(Serialize, Deserialize, Debug)]
pub enum LocationConfigMode {
AttachedSingle,
AttachedMulti,
AttachedStale,
Secondary,
Detached,
}
#[derive(Serialize, Deserialize, Debug)]
pub struct LocationConfigSecondary {
pub warm: bool,
}
/// An alternative representation of `pageserver::tenant::LocationConf`,
/// for use in external-facing APIs.
#[derive(Serialize, Deserialize, Debug)]
pub struct LocationConfig {
pub mode: LocationConfigMode,
/// If attaching, in what generation?
#[serde(default)]
pub generation: Option<Generation>,
#[serde(default)]
pub secondary_conf: Option<LocationConfigSecondary>,
// If requesting mode `Secondary`, configuration for that.
// Custom storage configuration for the tenant, if any
pub tenant_conf: TenantConfig,
}
#[serde_as]
#[derive(Serialize, Deserialize)]
#[serde(transparent)]
@@ -289,16 +253,6 @@ pub struct StatusResponse {
pub id: NodeId,
}
#[serde_as]
#[derive(Serialize, Deserialize, Debug)]
#[serde(deny_unknown_fields)]
pub struct TenantLocationConfigRequest {
#[serde_as(as = "DisplayFromStr")]
pub tenant_id: TenantId,
#[serde(flatten)]
pub config: LocationConfig, // as we have a flattened field, we should reject all unknown fields in it
}
#[serde_as]
#[derive(Serialize, Deserialize, Debug)]
#[serde(deny_unknown_fields)]

View File

@@ -12,7 +12,7 @@ log.workspace = true
once_cell.workspace = true
postgres.workspace = true
postgres_ffi.workspace = true
camino-tempfile.workspace = true
tempfile.workspace = true
workspace_hack.workspace = true

View File

@@ -1,5 +1,4 @@
use anyhow::{bail, ensure};
use camino_tempfile::{tempdir, Utf8TempDir};
use log::*;
use postgres::types::PgLsn;
use postgres::Client;
@@ -9,6 +8,7 @@ use std::cmp::Ordering;
use std::path::{Path, PathBuf};
use std::process::Command;
use std::time::{Duration, Instant};
use tempfile::{tempdir, TempDir};
macro_rules! xlog_utils_test {
($version:ident) => {
@@ -33,7 +33,7 @@ pub struct Conf {
pub struct PostgresServer {
process: std::process::Child,
_unix_socket_dir: Utf8TempDir,
_unix_socket_dir: TempDir,
client_config: postgres::Config,
}

View File

@@ -13,7 +13,6 @@ aws-types.workspace = true
aws-config.workspace = true
aws-sdk-s3.workspace = true
aws-credential-types.workspace = true
camino.workspace = true
hyper = { workspace = true, features = ["stream"] }
serde.workspace = true
serde_json.workspace = true
@@ -28,6 +27,6 @@ pin-project-lite.workspace = true
workspace_hack.workspace = true
[dev-dependencies]
camino-tempfile.workspace = true
tempfile.workspace = true
test-context.workspace = true
rand.workspace = true

View File

@@ -13,12 +13,12 @@ use std::{
collections::HashMap,
fmt::Debug,
num::{NonZeroU32, NonZeroUsize},
path::{Path, PathBuf},
pin::Pin,
sync::Arc,
};
use anyhow::{bail, Context};
use camino::{Utf8Path, Utf8PathBuf};
use serde::{Deserialize, Serialize};
use tokio::io;
@@ -52,7 +52,7 @@ const REMOTE_STORAGE_PREFIX_SEPARATOR: char = '/';
/// The prefix is an implementation detail, that allows representing local paths
/// as the remote ones, stripping the local storage prefix away.
#[derive(Debug, Clone, PartialEq, Eq, PartialOrd, Ord, Hash)]
pub struct RemotePath(Utf8PathBuf);
pub struct RemotePath(PathBuf);
impl Serialize for RemotePath {
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
@@ -69,18 +69,18 @@ impl<'de> Deserialize<'de> for RemotePath {
D: serde::Deserializer<'de>,
{
let str = String::deserialize(deserializer)?;
Ok(Self(Utf8PathBuf::from(&str)))
Ok(Self(PathBuf::from(&str)))
}
}
impl std::fmt::Display for RemotePath {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
std::fmt::Display::fmt(&self.0, f)
write!(f, "{}", self.0.display())
}
}
impl RemotePath {
pub fn new(relative_path: &Utf8Path) -> anyhow::Result<Self> {
pub fn new(relative_path: &Path) -> anyhow::Result<Self> {
anyhow::ensure!(
relative_path.is_relative(),
"Path {relative_path:?} is not relative"
@@ -89,30 +89,30 @@ impl RemotePath {
}
pub fn from_string(relative_path: &str) -> anyhow::Result<Self> {
Self::new(Utf8Path::new(relative_path))
Self::new(Path::new(relative_path))
}
pub fn with_base(&self, base_path: &Utf8Path) -> Utf8PathBuf {
pub fn with_base(&self, base_path: &Path) -> PathBuf {
base_path.join(&self.0)
}
pub fn object_name(&self) -> Option<&str> {
self.0.file_name()
self.0.file_name().and_then(|os_str| os_str.to_str())
}
pub fn join(&self, segment: &Utf8Path) -> Self {
pub fn join(&self, segment: &Path) -> Self {
Self(self.0.join(segment))
}
pub fn get_path(&self) -> &Utf8PathBuf {
pub fn get_path(&self) -> &PathBuf {
&self.0
}
pub fn extension(&self) -> Option<&str> {
self.0.extension()
self.0.extension()?.to_str()
}
pub fn strip_prefix(&self, p: &RemotePath) -> Result<&Utf8Path, std::path::StripPrefixError> {
pub fn strip_prefix(&self, p: &RemotePath) -> Result<&Path, std::path::StripPrefixError> {
self.0.strip_prefix(&p.0)
}
}
@@ -311,7 +311,7 @@ impl GenericRemoteStorage {
pub fn from_config(storage_config: &RemoteStorageConfig) -> anyhow::Result<Self> {
Ok(match &storage_config.storage {
RemoteStorageKind::LocalFs(root) => {
info!("Using fs root '{root}' as a remote storage");
info!("Using fs root '{}' as a remote storage", root.display());
Self::LocalFs(LocalFs::new(root.clone())?)
}
RemoteStorageKind::AwsS3(s3_config) => {
@@ -379,7 +379,7 @@ pub struct RemoteStorageConfig {
pub enum RemoteStorageKind {
/// Storage based on local file system.
/// Specify a root folder to place all stored files into.
LocalFs(Utf8PathBuf),
LocalFs(PathBuf),
/// AWS S3 based storage, storing all files in the S3 bucket
/// specified by the config
AwsS3(S3Config),
@@ -474,7 +474,7 @@ impl RemoteStorageConfig {
concurrency_limit,
max_keys_per_list_response,
}),
(Some(local_path), None, None) => RemoteStorageKind::LocalFs(Utf8PathBuf::from(
(Some(local_path), None, None) => RemoteStorageKind::LocalFs(PathBuf::from(
parse_toml_string("local_path", local_path)?,
)),
(Some(_), Some(_), _) => bail!("local_path and bucket_name are mutually exclusive"),
@@ -519,23 +519,23 @@ mod tests {
#[test]
fn test_object_name() {
let k = RemotePath::new(Utf8Path::new("a/b/c")).unwrap();
let k = RemotePath::new(Path::new("a/b/c")).unwrap();
assert_eq!(k.object_name(), Some("c"));
let k = RemotePath::new(Utf8Path::new("a/b/c/")).unwrap();
let k = RemotePath::new(Path::new("a/b/c/")).unwrap();
assert_eq!(k.object_name(), Some("c"));
let k = RemotePath::new(Utf8Path::new("a/")).unwrap();
let k = RemotePath::new(Path::new("a/")).unwrap();
assert_eq!(k.object_name(), Some("a"));
// XXX is it impossible to have an empty key?
let k = RemotePath::new(Utf8Path::new("")).unwrap();
let k = RemotePath::new(Path::new("")).unwrap();
assert_eq!(k.object_name(), None);
}
#[test]
fn rempte_path_cannot_be_created_from_absolute_ones() {
let err = RemotePath::new(Utf8Path::new("/")).expect_err("Should fail on absolute paths");
let err = RemotePath::new(Path::new("/")).expect_err("Should fail on absolute paths");
assert_eq!(err.to_string(), "Path \"/\" is not relative");
}
}

View File

@@ -4,10 +4,15 @@
//! This storage used in tests, but can also be used in cases when a certain persistent
//! volume is mounted to the local FS.
use std::{borrow::Cow, future::Future, io::ErrorKind, pin::Pin};
use std::{
borrow::Cow,
future::Future,
io::ErrorKind,
path::{Path, PathBuf},
pin::Pin,
};
use anyhow::{bail, ensure, Context};
use camino::{Utf8Path, Utf8PathBuf};
use tokio::{
fs,
io::{self, AsyncReadExt, AsyncSeekExt, AsyncWriteExt},
@@ -23,20 +28,20 @@ const LOCAL_FS_TEMP_FILE_SUFFIX: &str = "___temp";
#[derive(Debug, Clone)]
pub struct LocalFs {
storage_root: Utf8PathBuf,
storage_root: PathBuf,
}
impl LocalFs {
/// Attempts to create local FS storage, along with its root directory.
/// Storage root will be created (if does not exist) and transformed into an absolute path (if passed as relative).
pub fn new(mut storage_root: Utf8PathBuf) -> anyhow::Result<Self> {
pub fn new(mut storage_root: PathBuf) -> anyhow::Result<Self> {
if !storage_root.exists() {
std::fs::create_dir_all(&storage_root).with_context(|| {
format!("Failed to create all directories in the given root path {storage_root:?}")
})?;
}
if !storage_root.is_absolute() {
storage_root = storage_root.canonicalize_utf8().with_context(|| {
storage_root = storage_root.canonicalize().with_context(|| {
format!("Failed to represent path {storage_root:?} as an absolute path")
})?;
}
@@ -45,7 +50,7 @@ impl LocalFs {
}
// mirrors S3Bucket::s3_object_to_relative_path
fn local_file_to_relative_path(&self, key: Utf8PathBuf) -> RemotePath {
fn local_file_to_relative_path(&self, key: PathBuf) -> RemotePath {
let relative_path = key
.strip_prefix(&self.storage_root)
.expect("relative path must contain storage_root as prefix");
@@ -54,18 +59,22 @@ impl LocalFs {
async fn read_storage_metadata(
&self,
file_path: &Utf8Path,
file_path: &Path,
) -> anyhow::Result<Option<StorageMetadata>> {
let metadata_path = storage_metadata_path(file_path);
if metadata_path.exists() && metadata_path.is_file() {
let metadata_string = fs::read_to_string(&metadata_path).await.with_context(|| {
format!("Failed to read metadata from the local storage at '{metadata_path}'")
format!(
"Failed to read metadata from the local storage at '{}'",
metadata_path.display()
)
})?;
serde_json::from_str(&metadata_string)
.with_context(|| {
format!(
"Failed to deserialize metadata from the local storage at '{metadata_path}'",
"Failed to deserialize metadata from the local storage at '{}'",
metadata_path.display()
)
})
.map(|metadata| Some(StorageMetadata(metadata)))
@@ -162,21 +171,25 @@ impl RemoteStorage for LocalFs {
}
}
// Note that Utf8PathBuf starts_with only considers full path segments, but
// Note that PathBuf starts_with only considers full path segments, but
// object prefixes are arbitrary strings, so we need the strings for doing
// starts_with later.
let prefix = full_path.as_str();
let prefix = full_path.to_string_lossy();
let mut files = vec![];
let mut directory_queue = vec![initial_dir];
let mut directory_queue = vec![initial_dir.clone()];
while let Some(cur_folder) = directory_queue.pop() {
let mut entries = cur_folder.read_dir_utf8()?;
while let Some(Ok(entry)) = entries.next() {
let file_name = entry.file_name();
let full_file_name = cur_folder.join(file_name);
if full_file_name.as_str().starts_with(prefix) {
let mut entries = fs::read_dir(cur_folder.clone()).await?;
while let Some(entry) = entries.next_entry().await? {
let file_name: PathBuf = entry.file_name().into();
let full_file_name = cur_folder.clone().join(&file_name);
if full_file_name
.to_str()
.map(|s| s.starts_with(prefix.as_ref()))
.unwrap_or(false)
{
let file_remote_path = self.local_file_to_relative_path(full_file_name.clone());
files.push(file_remote_path);
files.push(file_remote_path.clone());
if full_file_name.is_dir() {
directory_queue.push(full_file_name);
}
@@ -217,7 +230,10 @@ impl RemoteStorage for LocalFs {
.open(&temp_file_path)
.await
.with_context(|| {
format!("Failed to open target fs destination at '{target_file_path}'")
format!(
"Failed to open target fs destination at '{}'",
target_file_path.display()
)
})?,
);
@@ -228,7 +244,8 @@ impl RemoteStorage for LocalFs {
.await
.with_context(|| {
format!(
"Failed to upload file (write temp) to the local storage at '{temp_file_path}'",
"Failed to upload file (write temp) to the local storage at '{}'",
temp_file_path.display()
)
})?;
@@ -245,7 +262,8 @@ impl RemoteStorage for LocalFs {
destination.flush().await.with_context(|| {
format!(
"Failed to upload (flush temp) file to the local storage at '{temp_file_path}'",
"Failed to upload (flush temp) file to the local storage at '{}'",
temp_file_path.display()
)
})?;
@@ -253,7 +271,8 @@ impl RemoteStorage for LocalFs {
.await
.with_context(|| {
format!(
"Failed to upload (rename) file to the local storage at '{target_file_path}'",
"Failed to upload (rename) file to the local storage at '{}'",
target_file_path.display()
)
})?;
@@ -267,7 +286,8 @@ impl RemoteStorage for LocalFs {
.await
.with_context(|| {
format!(
"Failed to write metadata to the local storage at '{storage_metadata_path}'",
"Failed to write metadata to the local storage at '{}'",
storage_metadata_path.display()
)
})?;
}
@@ -373,16 +393,16 @@ impl RemoteStorage for LocalFs {
}
}
fn storage_metadata_path(original_path: &Utf8Path) -> Utf8PathBuf {
fn storage_metadata_path(original_path: &Path) -> PathBuf {
path_with_suffix_extension(original_path, "metadata")
}
fn get_all_files<'a, P>(
directory_path: P,
recursive: bool,
) -> Pin<Box<dyn Future<Output = anyhow::Result<Vec<Utf8PathBuf>>> + Send + Sync + 'a>>
) -> Pin<Box<dyn Future<Output = anyhow::Result<Vec<PathBuf>>> + Send + Sync + 'a>>
where
P: AsRef<Utf8Path> + Send + Sync + 'a,
P: AsRef<Path> + Send + Sync + 'a,
{
Box::pin(async move {
let directory_path = directory_path.as_ref();
@@ -392,13 +412,7 @@ where
let mut dir_contents = fs::read_dir(directory_path).await?;
while let Some(dir_entry) = dir_contents.next_entry().await? {
let file_type = dir_entry.file_type().await?;
let entry_path =
Utf8PathBuf::from_path_buf(dir_entry.path()).map_err(|pb| {
anyhow::Error::msg(format!(
"non-Unicode path: {}",
pb.to_string_lossy()
))
})?;
let entry_path = dir_entry.path();
if file_type.is_symlink() {
debug!("{entry_path:?} is a symlink, skipping")
} else if file_type.is_dir() {
@@ -421,10 +435,13 @@ where
})
}
async fn create_target_directory(target_file_path: &Utf8Path) -> anyhow::Result<()> {
async fn create_target_directory(target_file_path: &Path) -> anyhow::Result<()> {
let target_dir = match target_file_path.parent() {
Some(parent_dir) => parent_dir,
None => bail!("File path '{target_file_path}' has no parent directory"),
None => bail!(
"File path '{}' has no parent directory",
target_file_path.display()
),
};
if !target_dir.exists() {
fs::create_dir_all(target_dir).await?;
@@ -432,9 +449,13 @@ async fn create_target_directory(target_file_path: &Utf8Path) -> anyhow::Result<
Ok(())
}
fn file_exists(file_path: &Utf8Path) -> anyhow::Result<bool> {
fn file_exists(file_path: &Path) -> anyhow::Result<bool> {
if file_path.exists() {
ensure!(file_path.is_file(), "file path '{file_path}' is not a file");
ensure!(
file_path.is_file(),
"file path '{}' is not a file",
file_path.display()
);
Ok(true)
} else {
Ok(false)
@@ -445,13 +466,13 @@ fn file_exists(file_path: &Utf8Path) -> anyhow::Result<bool> {
mod fs_tests {
use super::*;
use camino_tempfile::tempdir;
use std::{collections::HashMap, io::Write};
use tempfile::tempdir;
async fn read_and_assert_remote_file_contents(
storage: &LocalFs,
#[allow(clippy::ptr_arg)]
// have to use &Utf8PathBuf due to `storage.local_path` parameter requirements
// have to use &PathBuf due to `storage.local_path` parameter requirements
remote_storage_path: &RemotePath,
expected_metadata: Option<&StorageMetadata>,
) -> anyhow::Result<String> {
@@ -498,7 +519,7 @@ mod fs_tests {
async fn upload_file_negatives() -> anyhow::Result<()> {
let storage = create_storage()?;
let id = RemotePath::new(Utf8Path::new("dummy"))?;
let id = RemotePath::new(Path::new("dummy"))?;
let content = std::io::Cursor::new(b"12345");
// Check that you get an error if the size parameter doesn't match the actual
@@ -523,8 +544,7 @@ mod fs_tests {
}
fn create_storage() -> anyhow::Result<LocalFs> {
let storage_root = tempdir()?.path().to_path_buf();
LocalFs::new(storage_root)
LocalFs::new(tempdir()?.path().to_owned())
}
#[tokio::test]
@@ -541,7 +561,7 @@ mod fs_tests {
);
let non_existing_path = "somewhere/else";
match storage.download(&RemotePath::new(Utf8Path::new(non_existing_path))?).await {
match storage.download(&RemotePath::new(Path::new(non_existing_path))?).await {
Err(DownloadError::NotFound) => {} // Should get NotFound for non existing keys
other => panic!("Should get a NotFound error when downloading non-existing storage files, but got: {other:?}"),
}
@@ -755,7 +775,7 @@ mod fs_tests {
}
async fn create_file_for_upload(
path: &Utf8Path,
path: &Path,
contents: &str,
) -> anyhow::Result<(io::BufReader<fs::File>, usize)> {
std::fs::create_dir_all(path.parent().unwrap())?;

View File

@@ -47,47 +47,10 @@ pub struct S3Bucket {
bucket_name: String,
prefix_in_bucket: Option<String>,
max_keys_per_list_response: Option<i32>,
concurrency_limiter: ConcurrencyLimiter,
}
struct ConcurrencyLimiter {
// Every request to S3 can be throttled or cancelled, if a certain number of requests per second is exceeded.
// Same goes to IAM, which is queried before every S3 request, if enabled. IAM has even lower RPS threshold.
// The helps to ensure we don't exceed the thresholds.
write: Arc<Semaphore>,
read: Arc<Semaphore>,
}
impl ConcurrencyLimiter {
fn for_kind(&self, kind: RequestKind) -> &Arc<Semaphore> {
match kind {
RequestKind::Get => &self.read,
RequestKind::Put => &self.write,
RequestKind::List => &self.read,
RequestKind::Delete => &self.write,
}
}
async fn acquire(
&self,
kind: RequestKind,
) -> Result<tokio::sync::SemaphorePermit<'_>, tokio::sync::AcquireError> {
self.for_kind(kind).acquire().await
}
async fn acquire_owned(
&self,
kind: RequestKind,
) -> Result<tokio::sync::OwnedSemaphorePermit, tokio::sync::AcquireError> {
Arc::clone(self.for_kind(kind)).acquire_owned().await
}
fn new(limit: usize) -> ConcurrencyLimiter {
Self {
read: Arc::new(Semaphore::new(limit)),
write: Arc::new(Semaphore::new(limit)),
}
}
concurrency_limiter: Arc<Semaphore>,
}
#[derive(Default)]
@@ -154,7 +117,7 @@ impl S3Bucket {
bucket_name: aws_config.bucket_name.clone(),
max_keys_per_list_response: aws_config.max_keys_per_list_response,
prefix_in_bucket,
concurrency_limiter: ConcurrencyLimiter::new(aws_config.concurrency_limit.get()),
concurrency_limiter: Arc::new(Semaphore::new(aws_config.concurrency_limit.get())),
})
}
@@ -180,11 +143,12 @@ impl S3Bucket {
assert_eq!(std::path::MAIN_SEPARATOR, REMOTE_STORAGE_PREFIX_SEPARATOR);
let path_string = path
.get_path()
.as_str()
.trim_end_matches(REMOTE_STORAGE_PREFIX_SEPARATOR);
.to_string_lossy()
.trim_end_matches(REMOTE_STORAGE_PREFIX_SEPARATOR)
.to_string();
match &self.prefix_in_bucket {
Some(prefix) => prefix.clone() + "/" + path_string,
None => path_string.to_string(),
Some(prefix) => prefix.clone() + "/" + &path_string,
None => path_string,
}
}
@@ -192,7 +156,7 @@ impl S3Bucket {
let started_at = start_counting_cancelled_wait(kind);
let permit = self
.concurrency_limiter
.acquire(kind)
.acquire()
.await
.expect("semaphore is never closed");
@@ -208,7 +172,8 @@ impl S3Bucket {
let started_at = start_counting_cancelled_wait(kind);
let permit = self
.concurrency_limiter
.acquire_owned(kind)
.clone()
.acquire_owned()
.await
.expect("semaphore is never closed");
@@ -600,8 +565,8 @@ fn start_measuring_requests(
#[cfg(test)]
mod tests {
use camino::Utf8Path;
use std::num::NonZeroUsize;
use std::path::Path;
use crate::{RemotePath, S3Bucket, S3Config};
@@ -610,7 +575,7 @@ mod tests {
let all_paths = ["", "some/path", "some/path/"];
let all_paths: Vec<RemotePath> = all_paths
.iter()
.map(|x| RemotePath::new(Utf8Path::new(x)).expect("bad path"))
.map(|x| RemotePath::new(Path::new(x)).expect("bad path"))
.collect();
let prefixes = [
None,

View File

@@ -2,12 +2,11 @@ use std::collections::HashSet;
use std::env;
use std::num::{NonZeroU32, NonZeroUsize};
use std::ops::ControlFlow;
use std::path::PathBuf;
use std::path::{Path, PathBuf};
use std::sync::Arc;
use std::time::UNIX_EPOCH;
use anyhow::Context;
use camino::Utf8Path;
use once_cell::sync::OnceCell;
use remote_storage::{
GenericRemoteStorage, RemotePath, RemoteStorageConfig, RemoteStorageKind, S3Config,
@@ -56,7 +55,7 @@ async fn s3_pagination_should_work(ctx: &mut MaybeEnabledS3WithTestBlobs) -> any
let test_client = Arc::clone(&ctx.enabled.client);
let expected_remote_prefixes = ctx.remote_prefixes.clone();
let base_prefix = RemotePath::new(Utf8Path::new(ctx.enabled.base_prefix))
let base_prefix = RemotePath::new(Path::new(ctx.enabled.base_prefix))
.context("common_prefix construction")?;
let root_remote_prefixes = test_client
.list_prefixes(None)
@@ -109,7 +108,7 @@ async fn s3_list_files_works(ctx: &mut MaybeEnabledS3WithSimpleTestBlobs) -> any
};
let test_client = Arc::clone(&ctx.enabled.client);
let base_prefix =
RemotePath::new(Utf8Path::new("folder1")).context("common_prefix construction")?;
RemotePath::new(Path::new("folder1")).context("common_prefix construction")?;
let root_files = test_client
.list_files(None)
.await
@@ -130,9 +129,9 @@ async fn s3_list_files_works(ctx: &mut MaybeEnabledS3WithSimpleTestBlobs) -> any
let trim_remote_blobs: HashSet<_> = ctx
.remote_blobs
.iter()
.map(|x| x.get_path())
.map(|x| x.get_path().to_str().expect("must be valid name"))
.filter(|x| x.starts_with("folder1"))
.map(|x| RemotePath::new(x).expect("must be valid path"))
.map(|x| RemotePath::new(Path::new(x)).expect("must be valid name"))
.collect();
assert_eq!(
nested_remote_files, trim_remote_blobs,
@@ -149,9 +148,10 @@ async fn s3_delete_non_exising_works(ctx: &mut MaybeEnabledS3) -> anyhow::Result
MaybeEnabledS3::Disabled => return Ok(()),
};
let path = RemotePath::new(Utf8Path::new(
format!("{}/for_sure_there_is_nothing_there_really", ctx.base_prefix).as_str(),
))
let path = RemotePath::new(&PathBuf::from(format!(
"{}/for_sure_there_is_nothing_there_really",
ctx.base_prefix,
)))
.with_context(|| "RemotePath conversion")?;
ctx.client.delete(&path).await.expect("should succeed");
@@ -167,13 +167,13 @@ async fn s3_delete_objects_works(ctx: &mut MaybeEnabledS3) -> anyhow::Result<()>
MaybeEnabledS3::Disabled => return Ok(()),
};
let path1 = RemotePath::new(Utf8Path::new(format!("{}/path1", ctx.base_prefix).as_str()))
let path1 = RemotePath::new(&PathBuf::from(format!("{}/path1", ctx.base_prefix,)))
.with_context(|| "RemotePath conversion")?;
let path2 = RemotePath::new(Utf8Path::new(format!("{}/path2", ctx.base_prefix).as_str()))
let path2 = RemotePath::new(&PathBuf::from(format!("{}/path2", ctx.base_prefix,)))
.with_context(|| "RemotePath conversion")?;
let path3 = RemotePath::new(Utf8Path::new(format!("{}/path3", ctx.base_prefix).as_str()))
let path3 = RemotePath::new(&PathBuf::from(format!("{}/path3", ctx.base_prefix,)))
.with_context(|| "RemotePath conversion")?;
let data1 = "remote blob data1".as_bytes();
@@ -427,10 +427,10 @@ async fn upload_s3_data(
for i in 1..upload_tasks_count + 1 {
let task_client = Arc::clone(client);
upload_tasks.spawn(async move {
let prefix = format!("{base_prefix_str}/sub_prefix_{i}/");
let blob_prefix = RemotePath::new(Utf8Path::new(&prefix))
let prefix = PathBuf::from(format!("{base_prefix_str}/sub_prefix_{i}/"));
let blob_prefix = RemotePath::new(&prefix)
.with_context(|| format!("{prefix:?} to RemotePath conversion"))?;
let blob_path = blob_prefix.join(Utf8Path::new(&format!("blob_{i}")));
let blob_path = blob_prefix.join(Path::new(&format!("blob_{i}")));
debug!("Creating remote item {i} at path {blob_path:?}");
let data = format!("remote blob data {i}").into_bytes();
@@ -512,10 +512,8 @@ async fn upload_simple_s3_data(
let task_client = Arc::clone(client);
upload_tasks.spawn(async move {
let blob_path = PathBuf::from(format!("folder{}/blob_{}.txt", i / 7, i));
let blob_path = RemotePath::new(
Utf8Path::from_path(blob_path.as_path()).expect("must be valid blob path"),
)
.with_context(|| format!("{blob_path:?} to RemotePath conversion"))?;
let blob_path = RemotePath::new(&blob_path)
.with_context(|| format!("{blob_path:?} to RemotePath conversion"))?;
debug!("Creating remote item {i} at path {blob_path:?}");
let data = format!("remote blob data {i}").into_bytes();

View File

@@ -10,7 +10,6 @@ async-trait.workspace = true
anyhow.workspace = true
bincode.workspace = true
bytes.workspace = true
camino.workspace = true
chrono.workspace = true
heapless.workspace = true
hex = { workspace = true, features = ["serde"] }
@@ -54,7 +53,7 @@ byteorder.workspace = true
bytes.workspace = true
criterion.workspace = true
hex-literal.workspace = true
camino-tempfile.workspace = true
tempfile.workspace = true
[[bench]]
name = "benchmarks"

View File

@@ -2,9 +2,9 @@
use serde;
use std::fs;
use std::path::Path;
use anyhow::Result;
use camino::Utf8Path;
use jsonwebtoken::{
decode, encode, Algorithm, DecodingKey, EncodingKey, Header, TokenData, Validation,
};
@@ -65,7 +65,7 @@ impl JwtAuth {
}
}
pub fn from_key_path(key_path: &Utf8Path) -> Result<Self> {
pub fn from_key_path(key_path: &Path) -> Result<Self> {
let public_key = fs::read(key_path)?;
Ok(Self::new(DecodingKey::from_ed_pem(&public_key)?))
}

View File

@@ -1,14 +1,14 @@
use std::{
borrow::Cow,
ffi::OsStr,
fs::{self, File},
io,
path::{Path, PathBuf},
};
use camino::{Utf8Path, Utf8PathBuf};
/// Similar to [`std::fs::create_dir`], except we fsync the
/// created directory and its parent.
pub fn create_dir(path: impl AsRef<Utf8Path>) -> io::Result<()> {
pub fn create_dir(path: impl AsRef<Path>) -> io::Result<()> {
let path = path.as_ref();
fs::create_dir(path)?;
@@ -18,7 +18,7 @@ pub fn create_dir(path: impl AsRef<Utf8Path>) -> io::Result<()> {
/// Similar to [`std::fs::create_dir_all`], except we fsync all
/// newly created directories and the pre-existing parent.
pub fn create_dir_all(path: impl AsRef<Utf8Path>) -> io::Result<()> {
pub fn create_dir_all(path: impl AsRef<Path>) -> io::Result<()> {
let mut path = path.as_ref();
let mut dirs_to_create = Vec::new();
@@ -30,7 +30,7 @@ pub fn create_dir_all(path: impl AsRef<Utf8Path>) -> io::Result<()> {
Ok(_) => {
return Err(io::Error::new(
io::ErrorKind::AlreadyExists,
format!("non-directory found in path: {path}"),
format!("non-directory found in path: {}", path.display()),
));
}
Err(ref e) if e.kind() == io::ErrorKind::NotFound => {}
@@ -44,7 +44,7 @@ pub fn create_dir_all(path: impl AsRef<Utf8Path>) -> io::Result<()> {
None => {
return Err(io::Error::new(
io::ErrorKind::InvalidInput,
format!("can't find parent of path '{path}'"),
format!("can't find parent of path '{}'", path.display()).as_str(),
));
}
}
@@ -70,18 +70,21 @@ pub fn create_dir_all(path: impl AsRef<Utf8Path>) -> io::Result<()> {
/// Adds a suffix to the file(directory) name, either appending the suffix to the end of its extension,
/// or if there's no extension, creates one and puts a suffix there.
pub fn path_with_suffix_extension(
original_path: impl AsRef<Utf8Path>,
suffix: &str,
) -> Utf8PathBuf {
let new_extension = match original_path.as_ref().extension() {
pub fn path_with_suffix_extension(original_path: impl AsRef<Path>, suffix: &str) -> PathBuf {
let new_extension = match original_path
.as_ref()
.extension()
.map(OsStr::to_string_lossy)
{
Some(extension) => Cow::Owned(format!("{extension}.{suffix}")),
None => Cow::Borrowed(suffix),
};
original_path.as_ref().with_extension(new_extension)
original_path
.as_ref()
.with_extension(new_extension.as_ref())
}
pub fn fsync_file_and_parent(file_path: &Utf8Path) -> io::Result<()> {
pub fn fsync_file_and_parent(file_path: &Path) -> io::Result<()> {
let parent = file_path.parent().ok_or_else(|| {
io::Error::new(
io::ErrorKind::Other,
@@ -94,7 +97,7 @@ pub fn fsync_file_and_parent(file_path: &Utf8Path) -> io::Result<()> {
Ok(())
}
pub fn fsync(path: &Utf8Path) -> io::Result<()> {
pub fn fsync(path: &Path) -> io::Result<()> {
File::open(path)
.map_err(|e| io::Error::new(e.kind(), format!("Failed to open the file {path:?}: {e}")))
.and_then(|file| {
@@ -108,18 +111,19 @@ pub fn fsync(path: &Utf8Path) -> io::Result<()> {
.map_err(|e| io::Error::new(e.kind(), format!("Failed to fsync file {path:?}: {e}")))
}
pub async fn fsync_async(path: impl AsRef<Utf8Path>) -> Result<(), std::io::Error> {
tokio::fs::File::open(path.as_ref()).await?.sync_all().await
pub async fn fsync_async(path: impl AsRef<std::path::Path>) -> Result<(), std::io::Error> {
tokio::fs::File::open(path).await?.sync_all().await
}
#[cfg(test)]
mod tests {
use tempfile::tempdir;
use super::*;
#[test]
fn test_create_dir_fsyncd() {
let dir = camino_tempfile::tempdir().unwrap();
let dir = tempdir().unwrap();
let existing_dir_path = dir.path();
let err = create_dir(existing_dir_path).unwrap_err();
@@ -135,7 +139,7 @@ mod tests {
#[test]
fn test_create_dir_all_fsyncd() {
let dir = camino_tempfile::tempdir().unwrap();
let dir = tempdir().unwrap();
let existing_dir_path = dir.path();
create_dir_all(existing_dir_path).unwrap();
@@ -162,29 +166,29 @@ mod tests {
#[test]
fn test_path_with_suffix_extension() {
let p = Utf8PathBuf::from("/foo/bar");
let p = PathBuf::from("/foo/bar");
assert_eq!(
&path_with_suffix_extension(p, "temp").to_string(),
&path_with_suffix_extension(p, "temp").to_string_lossy(),
"/foo/bar.temp"
);
let p = Utf8PathBuf::from("/foo/bar");
let p = PathBuf::from("/foo/bar");
assert_eq!(
&path_with_suffix_extension(p, "temp.temp").to_string(),
&path_with_suffix_extension(p, "temp.temp").to_string_lossy(),
"/foo/bar.temp.temp"
);
let p = Utf8PathBuf::from("/foo/bar.baz");
let p = PathBuf::from("/foo/bar.baz");
assert_eq!(
&path_with_suffix_extension(p, "temp.temp").to_string(),
&path_with_suffix_extension(p, "temp.temp").to_string_lossy(),
"/foo/bar.baz.temp.temp"
);
let p = Utf8PathBuf::from("/foo/bar.baz");
let p = PathBuf::from("/foo/bar.baz");
assert_eq!(
&path_with_suffix_extension(p, ".temp").to_string(),
&path_with_suffix_extension(p, ".temp").to_string_lossy(),
"/foo/bar.baz..temp"
);
let p = Utf8PathBuf::from("/foo/bar/dir/");
let p = PathBuf::from("/foo/bar/dir/");
assert_eq!(
&path_with_suffix_extension(p, ".temp").to_string(),
&path_with_suffix_extension(p, ".temp").to_string_lossy(),
"/foo/bar/dir..temp"
);
}

View File

@@ -55,6 +55,8 @@ where
#[cfg(test)]
mod test {
use std::path::PathBuf;
use crate::fs_ext::{is_directory_empty, list_dir};
use super::ignore_absent_files;
@@ -63,7 +65,7 @@ mod test {
fn is_empty_dir() {
use super::PathExt;
let dir = camino_tempfile::tempdir().unwrap();
let dir = tempfile::tempdir().unwrap();
let dir_path = dir.path();
// test positive case
@@ -73,7 +75,7 @@ mod test {
);
// invoke on a file to ensure it returns an error
let file_path = dir_path.join("testfile");
let file_path: PathBuf = dir_path.join("testfile");
let f = std::fs::File::create(&file_path).unwrap();
drop(f);
assert!(file_path.is_empty_dir().is_err());
@@ -85,7 +87,7 @@ mod test {
#[tokio::test]
async fn is_empty_dir_async() {
let dir = camino_tempfile::tempdir().unwrap();
let dir = tempfile::tempdir().unwrap();
let dir_path = dir.path();
// test positive case
@@ -95,7 +97,7 @@ mod test {
);
// invoke on a file to ensure it returns an error
let file_path = dir_path.join("testfile");
let file_path: PathBuf = dir_path.join("testfile");
let f = std::fs::File::create(&file_path).unwrap();
drop(f);
assert!(is_directory_empty(&file_path).await.is_err());
@@ -107,9 +109,10 @@ mod test {
#[test]
fn ignore_absent_files_works() {
let dir = camino_tempfile::tempdir().unwrap();
let dir = tempfile::tempdir().unwrap();
let dir_path = dir.path();
let file_path = dir.path().join("testfile");
let file_path: PathBuf = dir_path.join("testfile");
ignore_absent_files(|| std::fs::remove_file(&file_path)).expect("should execute normally");
@@ -123,17 +126,17 @@ mod test {
#[tokio::test]
async fn list_dir_works() {
let dir = camino_tempfile::tempdir().unwrap();
let dir = tempfile::tempdir().unwrap();
let dir_path = dir.path();
assert!(list_dir(dir_path).await.unwrap().is_empty());
let file_path = dir_path.join("testfile");
let file_path: PathBuf = dir_path.join("testfile");
let _ = std::fs::File::create(&file_path).unwrap();
assert_eq!(&list_dir(dir_path).await.unwrap(), &["testfile"]);
let another_dir_path = dir_path.join("testdir");
let another_dir_path: PathBuf = dir_path.join("testdir");
std::fs::create_dir(another_dir_path).unwrap();
let expected = &["testdir", "testfile"];

View File

@@ -24,9 +24,6 @@ pub enum ApiError {
#[error("Precondition failed: {0}")]
PreconditionFailed(Box<str>),
#[error("Resource temporarily unavailable: {0}")]
ResourceUnavailable(String),
#[error("Shutting down")]
ShuttingDown,
@@ -62,10 +59,6 @@ impl ApiError {
"Shutting down".to_string(),
StatusCode::SERVICE_UNAVAILABLE,
),
ApiError::ResourceUnavailable(err) => HttpErrorBody::response_from_msg_and_status(
err.to_string(),
StatusCode::SERVICE_UNAVAILABLE,
),
ApiError::InternalServerError(err) => HttpErrorBody::response_from_msg_and_status(
err.to_string(),
StatusCode::INTERNAL_SERVER_ERROR,

View File

@@ -1,3 +1,4 @@
use std::ffi::OsStr;
use std::{fmt, str::FromStr};
use anyhow::Context;
@@ -214,11 +215,12 @@ pub struct TimelineId(Id);
id_newtype!(TimelineId);
impl TryFrom<Option<&str>> for TimelineId {
impl TryFrom<Option<&OsStr>> for TimelineId {
type Error = anyhow::Error;
fn try_from(value: Option<&str>) -> Result<Self, Self::Error> {
fn try_from(value: Option<&OsStr>) -> Result<Self, Self::Error> {
value
.and_then(OsStr::to_str)
.unwrap_or_default()
.parse::<TimelineId>()
.with_context(|| format!("Could not parse timeline id from {:?}", value))

View File

@@ -11,10 +11,10 @@ use std::{
io::{Read, Write},
ops::Deref,
os::unix::prelude::AsRawFd,
path::{Path, PathBuf},
};
use anyhow::Context;
use camino::{Utf8Path, Utf8PathBuf};
use nix::{errno::Errno::EAGAIN, fcntl};
use crate::crashsafe;
@@ -23,7 +23,7 @@ use crate::crashsafe;
/// Returned by [`create_exclusive`].
#[must_use]
pub struct UnwrittenLockFile {
path: Utf8PathBuf,
path: PathBuf,
file: fs::File,
}
@@ -60,7 +60,7 @@ impl UnwrittenLockFile {
///
/// It is not an error if the file already exists.
/// It is an error if the file is already locked.
pub fn create_exclusive(lock_file_path: &Utf8Path) -> anyhow::Result<UnwrittenLockFile> {
pub fn create_exclusive(lock_file_path: &Path) -> anyhow::Result<UnwrittenLockFile> {
let lock_file = fs::OpenOptions::new()
.create(true) // O_CREAT
.write(true)
@@ -101,7 +101,7 @@ pub enum LockFileRead {
/// Open & try to lock the lock file at the given `path`, returning a [handle][`LockFileRead`] to
/// inspect its content. It is not an `Err(...)` if the file does not exist or is already locked.
/// Check the [`LockFileRead`] variants for details.
pub fn read_and_hold_lock_file(path: &Utf8Path) -> anyhow::Result<LockFileRead> {
pub fn read_and_hold_lock_file(path: &Path) -> anyhow::Result<LockFileRead> {
let res = fs::OpenOptions::new().read(true).open(path);
let mut lock_file = match res {
Ok(f) => f,

View File

@@ -228,12 +228,6 @@ impl SecretString {
}
}
impl From<String> for SecretString {
fn from(s: String) -> Self {
Self(s)
}
}
impl std::fmt::Debug for SecretString {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
write!(f, "[SECRET]")

View File

@@ -1,9 +1,9 @@
#![warn(missing_docs)]
use camino::Utf8Path;
use serde::{Deserialize, Serialize};
use std::fmt;
use std::ops::{Add, AddAssign};
use std::path::Path;
use std::str::FromStr;
use std::sync::atomic::{AtomicU64, Ordering};
@@ -44,9 +44,11 @@ impl Lsn {
/// Parse an LSN from a filename in the form `0000000000000000`
pub fn from_filename<F>(filename: F) -> Result<Self, LsnParseError>
where
F: AsRef<Utf8Path>,
F: AsRef<Path>,
{
Lsn::from_hex(filename.as_ref().as_str())
let filename: &Path = filename.as_ref();
let filename = filename.to_str().ok_or(LsnParseError)?;
Lsn::from_hex(filename)
}
/// Parse an LSN from a string in the form `0000000000000000`

View File

@@ -49,10 +49,9 @@
//! At this point, `B` and `C` are running, which is hazardous.
//! Morale of the story: don't unlink pidfiles, ever.
use std::ops::Deref;
use std::{ops::Deref, path::Path};
use anyhow::Context;
use camino::Utf8Path;
use nix::unistd::Pid;
use crate::lock_file::{self, LockFileRead};
@@ -85,7 +84,7 @@ impl Deref for PidFileGuard {
/// The claim ends as soon as the returned guard object is dropped.
/// To maintain the claim for the remaining lifetime of the current process,
/// use [`std::mem::forget`] or similar.
pub fn claim_for_current_process(path: &Utf8Path) -> anyhow::Result<PidFileGuard> {
pub fn claim_for_current_process(path: &Path) -> anyhow::Result<PidFileGuard> {
let unwritten_lock_file = lock_file::create_exclusive(path).context("lock file")?;
// if any of the next steps fail, we drop the file descriptor and thereby release the lock
let guard = unwritten_lock_file
@@ -133,7 +132,7 @@ pub enum PidFileRead {
///
/// On success, this function returns a [`PidFileRead`].
/// Check its docs for a description of the meaning of its different variants.
pub fn read(pidfile: &Utf8Path) -> anyhow::Result<PidFileRead> {
pub fn read(pidfile: &Path) -> anyhow::Result<PidFileRead> {
let res = lock_file::read_and_hold_lock_file(pidfile).context("read and hold pid file")?;
let ret = match res {
LockFileRead::NotExist => PidFileRead::NotExist,

View File

@@ -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")?

View File

@@ -4,9 +4,9 @@
//! This is the "Monitor" part of the monitor binary and is the main entrypoint for
//! all functionality.
use std::fmt::Debug;
use std::sync::Arc;
use std::time::{Duration, Instant};
use std::{fmt::Debug, mem};
use anyhow::{bail, Context};
use axum::extract::ws::{Message, WebSocket};
@@ -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};
@@ -106,43 +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);
}
let mut file_cache_reserved_bytes = 0;
let mem = get_total_system_memory();
@@ -156,7 +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")?;
@@ -189,15 +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)
@@ -274,11 +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",
@@ -341,9 +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(())
@@ -409,7 +398,7 @@ impl Runner {
}
}
// we need to propagate an upscale request
request = self.dispatcher.request_upscale_events.recv(), if self.cgroup.is_some() => {
request = self.dispatcher.request_upscale_events.recv() => {
if request.is_none() {
bail!("failed to listen for upscale event from cgroup")
}

View File

@@ -12,14 +12,11 @@ testing = ["fail/failpoints"]
[dependencies]
anyhow.workspace = true
async-channel.workspace = true
async-compression.workspace = true
async-stream.workspace = true
async-trait.workspace = true
byteorder.workspace = true
bytes.workspace = true
camino.workspace = true
camino-tempfile.workspace = true
chrono = { workspace = true, features = ["serde"] }
clap = { workspace = true, features = ["string"] }
close_fds.workspace = true
@@ -83,6 +80,7 @@ enum-map.workspace = true
enumset.workspace = true
strum.workspace = true
strum_macros.workspace = true
tempfile.workspace = true
[dev-dependencies]
criterion.workspace = true

View File

@@ -25,7 +25,7 @@ fn redo_scenarios(c: &mut Criterion) {
// input to the stderr.
// utils::logging::init(utils::logging::LogFormat::Plain).unwrap();
let repo_dir = camino_tempfile::tempdir_in(env!("CARGO_TARGET_TMPDIR")).unwrap();
let repo_dir = tempfile::tempdir_in(env!("CARGO_TARGET_TMPDIR")).unwrap();
let conf = PageServerConf::dummy_conf(repo_dir.path().to_path_buf());
let conf = Box::leak(Box::new(conf));

View File

@@ -9,7 +9,6 @@ license.workspace = true
[dependencies]
anyhow.workspace = true
bytes.workspace = true
camino.workspace = true
clap = { workspace = true, features = ["string"] }
git-version.workspace = true
pageserver = { path = ".." }

View File

@@ -3,14 +3,13 @@
//! 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 camino::{Utf8Path, Utf8PathBuf};
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;
use std::ops::Range;
use std::{fs, str};
use std::{fs, path::Path, str};
use pageserver::page_cache::PAGE_SZ;
use pageserver::repository::{Key, KEY_SIZE};
@@ -99,7 +98,7 @@ 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: &Utf8Path, max_holes: usize, ctx: &RequestContext) -> Result<Vec<Hole>> {
async fn get_holes(path: &Path, max_holes: usize, ctx: &RequestContext) -> Result<Vec<Hole>> {
let file = FileBlockReader::new(VirtualFile::open(path).await?);
let summary_blk = file.read_blk(0, ctx).await?;
let actual_summary = Summary::des_prefix(summary_blk.as_ref())?;
@@ -168,9 +167,7 @@ pub(crate) async fn main(cmd: &AnalyzeLayerMapCmd) -> Result<()> {
parse_filename(&layer.file_name().into_string().unwrap())
{
if layer_file.is_delta {
let layer_path =
Utf8PathBuf::from_path_buf(layer.path()).expect("non-Unicode path");
layer_file.holes = get_holes(&layer_path, max_holes, &ctx).await?;
layer_file.holes = get_holes(&layer.path(), max_holes, &ctx).await?;
n_deltas += 1;
}
layers.push(layer_file);

View File

@@ -1,7 +1,6 @@
use std::path::{Path, PathBuf};
use anyhow::Result;
use camino::Utf8Path;
use clap::Subcommand;
use pageserver::context::{DownloadBehavior, RequestContext};
use pageserver::task_mgr::TaskKind;
@@ -48,7 +47,7 @@ pub(crate) enum LayerCmd {
}
async fn read_delta_file(path: impl AsRef<Path>, ctx: &RequestContext) -> Result<()> {
let path = Utf8Path::from_path(path.as_ref()).expect("non-Unicode path");
let path = path.as_ref();
virtual_file::init(10);
page_cache::init(100);
let file = FileBlockReader::new(VirtualFile::open(path).await?);

View File

@@ -8,7 +8,6 @@ mod draw_timeline_dir;
mod layer_map_analyzer;
mod layers;
use camino::{Utf8Path, Utf8PathBuf};
use clap::{Parser, Subcommand};
use layers::LayerCmd;
use pageserver::{
@@ -19,6 +18,7 @@ use pageserver::{
virtual_file,
};
use postgres_ffi::ControlFileData;
use std::path::{Path, PathBuf};
use utils::{lsn::Lsn, project_git_version};
project_git_version!(GIT_VERSION);
@@ -49,7 +49,7 @@ enum Commands {
#[derive(Parser)]
struct MetadataCmd {
/// Input metadata file path
metadata_path: Utf8PathBuf,
metadata_path: PathBuf,
/// Replace disk consistent Lsn
disk_consistent_lsn: Option<Lsn>,
/// Replace previous record Lsn
@@ -61,13 +61,13 @@ struct MetadataCmd {
#[derive(Parser)]
struct PrintLayerFileCmd {
/// Pageserver data path
path: Utf8PathBuf,
path: PathBuf,
}
#[derive(Parser)]
struct AnalyzeLayerMapCmd {
/// Pageserver data path
path: Utf8PathBuf,
path: PathBuf,
/// Max holes
max_holes: Option<usize>,
}
@@ -102,7 +102,7 @@ async fn main() -> anyhow::Result<()> {
Ok(())
}
fn read_pg_control_file(control_file_path: &Utf8Path) -> anyhow::Result<()> {
fn read_pg_control_file(control_file_path: &Path) -> anyhow::Result<()> {
let control_file = ControlFileData::decode(&std::fs::read(control_file_path)?)?;
println!("{control_file:?}");
let control_file_initdb = Lsn(control_file.checkPoint);
@@ -114,7 +114,7 @@ fn read_pg_control_file(control_file_path: &Utf8Path) -> anyhow::Result<()> {
Ok(())
}
async fn print_layerfile(path: &Utf8Path) -> anyhow::Result<()> {
async fn print_layerfile(path: &Path) -> anyhow::Result<()> {
// Basic initialization of things that don't change after startup
virtual_file::init(10);
page_cache::init(100);

View File

@@ -2,10 +2,9 @@
use std::env::{var, VarError};
use std::sync::Arc;
use std::{env, ops::ControlFlow, str::FromStr};
use std::{env, ops::ControlFlow, path::Path, str::FromStr};
use anyhow::{anyhow, Context};
use camino::Utf8Path;
use clap::{Arg, ArgAction, Command};
use metrics::launch_timestamp::{set_launch_timestamp_metric, LaunchTimestamp};
@@ -66,17 +65,21 @@ fn main() -> anyhow::Result<()> {
let workdir = arg_matches
.get_one::<String>("workdir")
.map(Utf8Path::new)
.unwrap_or_else(|| Utf8Path::new(".neon"));
.map(Path::new)
.unwrap_or_else(|| Path::new(".neon"));
let workdir = workdir
.canonicalize_utf8()
.with_context(|| format!("Error opening workdir '{workdir}'"))?;
.canonicalize()
.with_context(|| format!("Error opening workdir '{}'", workdir.display()))?;
let cfg_file_path = workdir.join("pageserver.toml");
// Set CWD to workdir for non-daemon modes
env::set_current_dir(&workdir)
.with_context(|| format!("Failed to set application's current dir to '{workdir}'"))?;
env::set_current_dir(&workdir).with_context(|| {
format!(
"Failed to set application's current dir to '{}'",
workdir.display()
)
})?;
let conf = match initialize_config(&cfg_file_path, arg_matches, &workdir)? {
ControlFlow::Continue(conf) => conf,
@@ -112,8 +115,12 @@ fn main() -> anyhow::Result<()> {
let tenants_path = conf.tenants_path();
if !tenants_path.exists() {
utils::crashsafe::create_dir_all(conf.tenants_path())
.with_context(|| format!("Failed to create tenants root dir at '{tenants_path}'"))?;
utils::crashsafe::create_dir_all(conf.tenants_path()).with_context(|| {
format!(
"Failed to create tenants root dir at '{}'",
tenants_path.display()
)
})?;
}
// Initialize up failpoints support
@@ -130,9 +137,9 @@ fn main() -> anyhow::Result<()> {
}
fn initialize_config(
cfg_file_path: &Utf8Path,
cfg_file_path: &Path,
arg_matches: clap::ArgMatches,
workdir: &Utf8Path,
workdir: &Path,
) -> anyhow::Result<ControlFlow<(), &'static PageServerConf>> {
let init = arg_matches.get_flag("init");
let update_config = init || arg_matches.get_flag("update-config");
@@ -140,22 +147,33 @@ fn initialize_config(
let (mut toml, config_file_exists) = if cfg_file_path.is_file() {
if init {
anyhow::bail!(
"Config file '{cfg_file_path}' already exists, cannot init it, use --update-config to update it",
"Config file '{}' already exists, cannot init it, use --update-config to update it",
cfg_file_path.display()
);
}
// Supplement the CLI arguments with the config file
let cfg_file_contents = std::fs::read_to_string(cfg_file_path)
.with_context(|| format!("Failed to read pageserver config at '{cfg_file_path}'"))?;
let cfg_file_contents = std::fs::read_to_string(cfg_file_path).with_context(|| {
format!(
"Failed to read pageserver config at '{}'",
cfg_file_path.display()
)
})?;
(
cfg_file_contents
.parse::<toml_edit::Document>()
.with_context(|| {
format!("Failed to parse '{cfg_file_path}' as pageserver config")
format!(
"Failed to parse '{}' as pageserver config",
cfg_file_path.display()
)
})?,
true,
)
} else if cfg_file_path.exists() {
anyhow::bail!("Config file '{cfg_file_path}' exists but is not a regular file");
anyhow::bail!(
"Config file '{}' exists but is not a regular file",
cfg_file_path.display()
);
} else {
// We're initializing the tenant, so there's no config file yet
(
@@ -174,7 +192,7 @@ fn initialize_config(
for (key, item) in doc.iter() {
if config_file_exists && update_config && key == "id" && toml.contains_key(key) {
anyhow::bail!("Pageserver config file exists at '{cfg_file_path}' and has node id already, it cannot be overridden");
anyhow::bail!("Pageserver config file exists at '{}' and has node id already, it cannot be overridden", cfg_file_path.display());
}
toml.insert(key, item.clone());
}
@@ -186,11 +204,18 @@ fn initialize_config(
.context("Failed to parse pageserver configuration")?;
if update_config {
info!("Writing pageserver config to '{cfg_file_path}'");
info!("Writing pageserver config to '{}'", cfg_file_path.display());
std::fs::write(cfg_file_path, toml.to_string())
.with_context(|| format!("Failed to write pageserver config to '{cfg_file_path}'"))?;
info!("Config successfully written to '{cfg_file_path}'")
std::fs::write(cfg_file_path, toml.to_string()).with_context(|| {
format!(
"Failed to write pageserver config to '{}'",
cfg_file_path.display()
)
})?;
info!(
"Config successfully written to '{}'",
cfg_file_path.display()
)
}
Ok(if init {
@@ -619,7 +644,7 @@ 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);
};

View File

@@ -16,13 +16,13 @@ use utils::logging::SecretString;
use once_cell::sync::OnceCell;
use reqwest::Url;
use std::num::NonZeroUsize;
use std::path::{Path, PathBuf};
use std::str::FromStr;
use std::sync::Arc;
use std::time::Duration;
use toml_edit;
use toml_edit::{Document, Item};
use camino::{Utf8Path, Utf8PathBuf};
use postgres_backend::AuthType;
use utils::{
id::{NodeId, TenantId, TimelineId},
@@ -37,8 +37,8 @@ use crate::tenant::{
TIMELINES_SEGMENT_NAME,
};
use crate::{
IGNORED_TENANT_FILE_NAME, METADATA_FILE_NAME, TENANT_CONFIG_NAME, TENANT_LOCATION_CONFIG_NAME,
TIMELINE_DELETE_MARK_SUFFIX, TIMELINE_UNINIT_MARK_SUFFIX,
IGNORED_TENANT_FILE_NAME, METADATA_FILE_NAME, TENANT_CONFIG_NAME, TIMELINE_DELETE_MARK_SUFFIX,
TIMELINE_UNINIT_MARK_SUFFIX,
};
pub mod defaults {
@@ -153,9 +153,9 @@ pub struct PageServerConf {
// that during unit testing, because the current directory is global
// to the process but different unit tests work on different
// repositories.
pub workdir: Utf8PathBuf,
pub workdir: PathBuf,
pub pg_distrib_dir: Utf8PathBuf,
pub pg_distrib_dir: PathBuf,
// Authentication
/// authentication method for the HTTP mgmt API
@@ -164,7 +164,7 @@ pub struct PageServerConf {
pub pg_auth_type: AuthType,
/// Path to a file containing public key for verifying JWT tokens.
/// Used for both mgmt and compute auth, if enabled.
pub auth_validation_public_key_path: Option<Utf8PathBuf>,
pub auth_validation_public_key_path: Option<PathBuf>,
pub remote_storage_config: Option<RemoteStorageConfig>,
@@ -253,15 +253,15 @@ struct PageServerConfigBuilder {
page_cache_size: BuilderValue<usize>,
max_file_descriptors: BuilderValue<usize>,
workdir: BuilderValue<Utf8PathBuf>,
workdir: BuilderValue<PathBuf>,
pg_distrib_dir: BuilderValue<Utf8PathBuf>,
pg_distrib_dir: BuilderValue<PathBuf>,
http_auth_type: BuilderValue<AuthType>,
pg_auth_type: BuilderValue<AuthType>,
//
auth_validation_public_key_path: BuilderValue<Option<Utf8PathBuf>>,
auth_validation_public_key_path: BuilderValue<Option<PathBuf>>,
remote_storage_config: BuilderValue<Option<RemoteStorageConfig>>,
id: BuilderValue<NodeId>,
@@ -305,12 +305,10 @@ impl Default for PageServerConfigBuilder {
superuser: Set(DEFAULT_SUPERUSER.to_string()),
page_cache_size: Set(DEFAULT_PAGE_CACHE_SIZE),
max_file_descriptors: Set(DEFAULT_MAX_FILE_DESCRIPTORS),
workdir: Set(Utf8PathBuf::new()),
pg_distrib_dir: Set(Utf8PathBuf::from_path_buf(
env::current_dir().expect("cannot access current directory"),
)
.expect("non-Unicode path")
.join("pg_install")),
workdir: Set(PathBuf::new()),
pg_distrib_dir: Set(env::current_dir()
.expect("cannot access current directory")
.join("pg_install")),
http_auth_type: Set(AuthType::Trust),
pg_auth_type: Set(AuthType::Trust),
auth_validation_public_key_path: Set(None),
@@ -392,11 +390,11 @@ impl PageServerConfigBuilder {
self.max_file_descriptors = BuilderValue::Set(max_file_descriptors)
}
pub fn workdir(&mut self, workdir: Utf8PathBuf) {
pub fn workdir(&mut self, workdir: PathBuf) {
self.workdir = BuilderValue::Set(workdir)
}
pub fn pg_distrib_dir(&mut self, pg_distrib_dir: Utf8PathBuf) {
pub fn pg_distrib_dir(&mut self, pg_distrib_dir: PathBuf) {
self.pg_distrib_dir = BuilderValue::Set(pg_distrib_dir)
}
@@ -410,7 +408,7 @@ impl PageServerConfigBuilder {
pub fn auth_validation_public_key_path(
&mut self,
auth_validation_public_key_path: Option<Utf8PathBuf>,
auth_validation_public_key_path: Option<PathBuf>,
) {
self.auth_validation_public_key_path = BuilderValue::Set(auth_validation_public_key_path)
}
@@ -487,10 +485,6 @@ impl PageServerConfigBuilder {
self.control_plane_api = BuilderValue::Set(api)
}
pub fn control_plane_api_token(&mut self, token: Option<SecretString>) {
self.control_plane_api_token = BuilderValue::Set(token)
}
pub fn build(self) -> anyhow::Result<PageServerConf> {
let concurrent_tenant_size_logical_size_queries = self
.concurrent_tenant_size_logical_size_queries
@@ -591,15 +585,15 @@ impl PageServerConf {
// Repository paths, relative to workdir.
//
pub fn tenants_path(&self) -> Utf8PathBuf {
pub fn tenants_path(&self) -> PathBuf {
self.workdir.join(TENANTS_SEGMENT_NAME)
}
pub fn deletion_prefix(&self) -> Utf8PathBuf {
pub fn deletion_prefix(&self) -> PathBuf {
self.workdir.join("deletion")
}
pub fn deletion_list_path(&self, sequence: u64) -> Utf8PathBuf {
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;
@@ -608,7 +602,7 @@ impl PageServerConf {
.join(format!("{sequence:016x}-{VERSION:02x}.list"))
}
pub fn deletion_header_path(&self) -> Utf8PathBuf {
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;
@@ -616,38 +610,30 @@ impl PageServerConf {
self.deletion_prefix().join(format!("header-{VERSION:02x}"))
}
pub fn tenant_path(&self, tenant_id: &TenantId) -> Utf8PathBuf {
pub fn tenant_path(&self, tenant_id: &TenantId) -> PathBuf {
self.tenants_path().join(tenant_id.to_string())
}
pub fn tenant_attaching_mark_file_path(&self, tenant_id: &TenantId) -> Utf8PathBuf {
pub fn tenant_attaching_mark_file_path(&self, tenant_id: &TenantId) -> PathBuf {
self.tenant_path(tenant_id)
.join(TENANT_ATTACHING_MARKER_FILENAME)
}
pub fn tenant_ignore_mark_file_path(&self, tenant_id: &TenantId) -> Utf8PathBuf {
pub fn tenant_ignore_mark_file_path(&self, tenant_id: &TenantId) -> PathBuf {
self.tenant_path(tenant_id).join(IGNORED_TENANT_FILE_NAME)
}
/// Points to a place in pageserver's local directory,
/// where certain tenant's tenantconf file should be located.
///
/// Legacy: superseded by tenant_location_config_path. Eventually
/// remove this function.
pub fn tenant_config_path(&self, tenant_id: &TenantId) -> Utf8PathBuf {
pub fn tenant_config_path(&self, tenant_id: &TenantId) -> PathBuf {
self.tenant_path(tenant_id).join(TENANT_CONFIG_NAME)
}
pub fn tenant_location_config_path(&self, tenant_id: &TenantId) -> Utf8PathBuf {
self.tenant_path(tenant_id)
.join(TENANT_LOCATION_CONFIG_NAME)
}
pub fn timelines_path(&self, tenant_id: &TenantId) -> Utf8PathBuf {
pub fn timelines_path(&self, tenant_id: &TenantId) -> PathBuf {
self.tenant_path(tenant_id).join(TIMELINES_SEGMENT_NAME)
}
pub fn timeline_path(&self, tenant_id: &TenantId, timeline_id: &TimelineId) -> Utf8PathBuf {
pub fn timeline_path(&self, tenant_id: &TenantId, timeline_id: &TimelineId) -> PathBuf {
self.timelines_path(tenant_id).join(timeline_id.to_string())
}
@@ -655,7 +641,7 @@ impl PageServerConf {
&self,
tenant_id: TenantId,
timeline_id: TimelineId,
) -> Utf8PathBuf {
) -> PathBuf {
path_with_suffix_extension(
self.timeline_path(&tenant_id, &timeline_id),
TIMELINE_UNINIT_MARK_SUFFIX,
@@ -666,19 +652,19 @@ impl PageServerConf {
&self,
tenant_id: TenantId,
timeline_id: TimelineId,
) -> Utf8PathBuf {
) -> PathBuf {
path_with_suffix_extension(
self.timeline_path(&tenant_id, &timeline_id),
TIMELINE_DELETE_MARK_SUFFIX,
)
}
pub fn tenant_deleted_mark_file_path(&self, tenant_id: &TenantId) -> Utf8PathBuf {
pub fn tenant_deleted_mark_file_path(&self, tenant_id: &TenantId) -> PathBuf {
self.tenant_path(tenant_id)
.join(TENANT_DELETED_MARKER_FILE_NAME)
}
pub fn traces_path(&self) -> Utf8PathBuf {
pub fn traces_path(&self) -> PathBuf {
self.workdir.join("traces")
}
@@ -687,7 +673,7 @@ impl PageServerConf {
tenant_id: &TenantId,
timeline_id: &TimelineId,
connection_id: &ConnectionId,
) -> Utf8PathBuf {
) -> PathBuf {
self.traces_path()
.join(tenant_id.to_string())
.join(timeline_id.to_string())
@@ -696,20 +682,20 @@ impl PageServerConf {
/// Points to a place in pageserver's local directory,
/// where certain timeline's metadata file should be located.
pub fn metadata_path(&self, tenant_id: &TenantId, timeline_id: &TimelineId) -> Utf8PathBuf {
pub fn metadata_path(&self, tenant_id: &TenantId, timeline_id: &TimelineId) -> PathBuf {
self.timeline_path(tenant_id, timeline_id)
.join(METADATA_FILE_NAME)
}
/// Turns storage remote path of a file into its local path.
pub fn local_path(&self, remote_path: &RemotePath) -> Utf8PathBuf {
pub fn local_path(&self, remote_path: &RemotePath) -> PathBuf {
remote_path.with_base(&self.workdir)
}
//
// Postgres distribution paths
//
pub fn pg_distrib_dir(&self, pg_version: u32) -> anyhow::Result<Utf8PathBuf> {
pub fn pg_distrib_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
let path = self.pg_distrib_dir.clone();
#[allow(clippy::manual_range_patterns)]
@@ -719,10 +705,10 @@ impl PageServerConf {
}
}
pub fn pg_bin_dir(&self, pg_version: u32) -> anyhow::Result<Utf8PathBuf> {
pub fn pg_bin_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
Ok(self.pg_distrib_dir(pg_version)?.join("bin"))
}
pub fn pg_lib_dir(&self, pg_version: u32) -> anyhow::Result<Utf8PathBuf> {
pub fn pg_lib_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
Ok(self.pg_distrib_dir(pg_version)?.join("lib"))
}
@@ -730,7 +716,7 @@ impl PageServerConf {
/// validating the input and failing on errors.
///
/// This leaves any options not present in the file in the built-in defaults.
pub fn parse_and_validate(toml: &Document, workdir: &Utf8Path) -> anyhow::Result<Self> {
pub fn parse_and_validate(toml: &Document, workdir: &Path) -> anyhow::Result<Self> {
let mut builder = PageServerConfigBuilder::default();
builder.workdir(workdir.to_owned());
@@ -749,10 +735,10 @@ impl PageServerConf {
builder.max_file_descriptors(parse_toml_u64(key, item)? as usize)
}
"pg_distrib_dir" => {
builder.pg_distrib_dir(Utf8PathBuf::from(parse_toml_string(key, item)?))
builder.pg_distrib_dir(PathBuf::from(parse_toml_string(key, item)?))
}
"auth_validation_public_key_path" => builder.auth_validation_public_key_path(Some(
Utf8PathBuf::from(parse_toml_string(key, item)?),
PathBuf::from(parse_toml_string(key, item)?),
)),
"http_auth_type" => builder.http_auth_type(parse_toml_from_str(key, item)?),
"pg_auth_type" => builder.pg_auth_type(parse_toml_from_str(key, item)?),
@@ -799,14 +785,6 @@ impl PageServerConf {
builder.control_plane_api(Some(parsed.parse().context("failed to parse control plane URL")?))
}
},
"control_plane_api_token" => {
let parsed = parse_toml_string(key, item)?;
if parsed.is_empty() {
builder.control_plane_api_token(None)
} else {
builder.control_plane_api_token(Some(parsed.into()))
}
},
_ => bail!("unrecognized pageserver option '{key}'"),
}
}
@@ -820,7 +798,8 @@ impl PageServerConf {
ensure!(
auth_validation_public_key_path.exists(),
format!(
"Can't find auth_validation_public_key at '{auth_validation_public_key_path}'",
"Can't find auth_validation_public_key at '{}'",
auth_validation_public_key_path.display()
)
);
}
@@ -936,12 +915,12 @@ impl PageServerConf {
}
#[cfg(test)]
pub fn test_repo_dir(test_name: &str) -> Utf8PathBuf {
Utf8PathBuf::from(format!("../tmp_check/test_{test_name}"))
pub fn test_repo_dir(test_name: &str) -> PathBuf {
PathBuf::from(format!("../tmp_check/test_{test_name}"))
}
pub fn dummy_conf(repo_dir: Utf8PathBuf) -> Self {
let pg_distrib_dir = Utf8PathBuf::from(env!("CARGO_MANIFEST_DIR")).join("../pg_install");
pub fn dummy_conf(repo_dir: PathBuf) -> Self {
let pg_distrib_dir = PathBuf::from(env!("CARGO_MANIFEST_DIR")).join("../pg_install");
PageServerConf {
id: NodeId(0),
@@ -1108,8 +1087,8 @@ mod tests {
num::{NonZeroU32, NonZeroUsize},
};
use camino_tempfile::{tempdir, Utf8TempDir};
use remote_storage::{RemoteStorageKind, S3Config};
use tempfile::{tempdir, TempDir};
use utils::serde_percent::Percent;
use super::*;
@@ -1148,7 +1127,8 @@ background_task_maximum_delay = '334 s'
let broker_endpoint = storage_broker::DEFAULT_ENDPOINT;
// we have to create dummy values to overcome the validation errors
let config_string = format!(
"pg_distrib_dir='{pg_distrib_dir}'\nid=10\nbroker_endpoint = '{broker_endpoint}'",
"pg_distrib_dir='{}'\nid=10\nbroker_endpoint = '{broker_endpoint}'",
pg_distrib_dir.display()
);
let toml = config_string.parse()?;
@@ -1214,7 +1194,8 @@ background_task_maximum_delay = '334 s'
let broker_endpoint = storage_broker::DEFAULT_ENDPOINT;
let config_string = format!(
"{ALL_BASE_VALUES_TOML}pg_distrib_dir='{pg_distrib_dir}'\nbroker_endpoint = '{broker_endpoint}'",
"{ALL_BASE_VALUES_TOML}pg_distrib_dir='{}'\nbroker_endpoint = '{broker_endpoint}'",
pg_distrib_dir.display()
);
let toml = config_string.parse()?;
@@ -1274,18 +1255,23 @@ background_task_maximum_delay = '334 s'
let identical_toml_declarations = &[
format!(
r#"[remote_storage]
local_path = '{local_storage_path}'"#,
local_path = '{}'"#,
local_storage_path.display()
),
format!(
"remote_storage={{local_path='{}'}}",
local_storage_path.display()
),
format!("remote_storage={{local_path='{local_storage_path}'}}"),
];
for remote_storage_config_str in identical_toml_declarations {
let config_string = format!(
r#"{ALL_BASE_VALUES_TOML}
pg_distrib_dir='{pg_distrib_dir}'
pg_distrib_dir='{}'
broker_endpoint = '{broker_endpoint}'
{remote_storage_config_str}"#,
pg_distrib_dir.display(),
);
let toml = config_string.parse()?;
@@ -1348,10 +1334,11 @@ concurrency_limit = {s3_concurrency_limit}"#
for remote_storage_config_str in identical_toml_declarations {
let config_string = format!(
r#"{ALL_BASE_VALUES_TOML}
pg_distrib_dir='{pg_distrib_dir}'
pg_distrib_dir='{}'
broker_endpoint = '{broker_endpoint}'
{remote_storage_config_str}"#,
pg_distrib_dir.display(),
);
let toml = config_string.parse()?;
@@ -1393,11 +1380,12 @@ broker_endpoint = '{broker_endpoint}'
let config_string = format!(
r#"{ALL_BASE_VALUES_TOML}
pg_distrib_dir='{pg_distrib_dir}'
pg_distrib_dir='{}'
broker_endpoint = '{broker_endpoint}'
[tenant_config]
trace_read_requests = {trace_read_requests}"#,
pg_distrib_dir.display(),
);
let toml = config_string.parse()?;
@@ -1417,7 +1405,7 @@ trace_read_requests = {trace_read_requests}"#,
let (workdir, pg_distrib_dir) = prepare_fs(&tempdir)?;
let pageserver_conf_toml = format!(
r#"pg_distrib_dir = "{pg_distrib_dir}"
r#"pg_distrib_dir = "{}"
metric_collection_endpoint = "http://sample.url"
metric_collection_interval = "10min"
id = 222
@@ -1435,6 +1423,7 @@ kind = "LayerAccessThreshold"
period = "20m"
threshold = "20m"
"#,
pg_distrib_dir.display(),
);
let toml: Document = pageserver_conf_toml.parse()?;
let conf = PageServerConf::parse_and_validate(&toml, &workdir)?;
@@ -1475,7 +1464,7 @@ threshold = "20m"
Ok(())
}
fn prepare_fs(tempdir: &Utf8TempDir) -> anyhow::Result<(Utf8PathBuf, Utf8PathBuf)> {
fn prepare_fs(tempdir: &TempDir) -> anyhow::Result<(PathBuf, PathBuf)> {
let tempdir_path = tempdir.path();
let workdir = tempdir_path.join("workdir");

View File

@@ -3,11 +3,11 @@
use crate::context::{DownloadBehavior, RequestContext};
use crate::task_mgr::{self, TaskKind, BACKGROUND_RUNTIME};
use crate::tenant::{mgr, LogicalSizeCalculationCause};
use camino::Utf8PathBuf;
use consumption_metrics::EventType;
use pageserver_api::models::TenantState;
use reqwest::Url;
use std::collections::HashMap;
use std::path::PathBuf;
use std::sync::Arc;
use std::time::{Duration, SystemTime};
use tracing::*;
@@ -41,7 +41,7 @@ pub async fn collect_metrics(
_cached_metric_collection_interval: Duration,
synthetic_size_calculation_interval: Duration,
node_id: NodeId,
local_disk_storage: Utf8PathBuf,
local_disk_storage: PathBuf,
ctx: RequestContext,
) -> anyhow::Result<()> {
if _cached_metric_collection_interval != Duration::ZERO {
@@ -68,7 +68,7 @@ pub async fn collect_metrics(
},
);
let path: Arc<Utf8PathBuf> = Arc::new(local_disk_storage);
let path: Arc<PathBuf> = Arc::new(local_disk_storage);
let cancel = task_mgr::shutdown_token();
@@ -153,7 +153,7 @@ pub async fn collect_metrics(
///
/// Cancellation safe.
async fn restore_and_reschedule(
path: &Arc<Utf8PathBuf>,
path: &Arc<PathBuf>,
metric_collection_interval: Duration,
) -> Cache {
let (cached, earlier_metric_at) = match disk_cache::read_metrics_from_disk(path.clone()).await {

View File

@@ -1,12 +1,10 @@
use anyhow::Context;
use camino::{Utf8Path, Utf8PathBuf};
use std::path::PathBuf;
use std::sync::Arc;
use super::RawMetric;
pub(super) async fn read_metrics_from_disk(
path: Arc<Utf8PathBuf>,
) -> anyhow::Result<Vec<RawMetric>> {
pub(super) async fn read_metrics_from_disk(path: Arc<PathBuf>) -> anyhow::Result<Vec<RawMetric>> {
// do not add context to each error, callsite will log with full path
let span = tracing::Span::current();
tokio::task::spawn_blocking(move || {
@@ -27,10 +25,10 @@ pub(super) async fn read_metrics_from_disk(
.and_then(|x| x)
}
fn scan_and_delete_with_same_prefix(path: &Utf8Path) -> std::io::Result<()> {
fn scan_and_delete_with_same_prefix(path: &std::path::Path) -> std::io::Result<()> {
let it = std::fs::read_dir(path.parent().expect("caller checked"))?;
let prefix = path.file_name().expect("caller checked").to_string();
let prefix = path.file_name().expect("caller checked").to_string_lossy();
for entry in it {
let entry = entry?;
@@ -64,7 +62,7 @@ fn scan_and_delete_with_same_prefix(path: &Utf8Path) -> std::io::Result<()> {
pub(super) async fn flush_metrics_to_disk(
current_metrics: &Arc<Vec<RawMetric>>,
path: &Arc<Utf8PathBuf>,
path: &Arc<PathBuf>,
) -> anyhow::Result<()> {
use std::io::Write;
@@ -83,7 +81,7 @@ pub(super) async fn flush_metrics_to_disk(
let parent = path.parent().expect("existence checked");
let file_name = path.file_name().expect("existence checked");
let mut tempfile = camino_tempfile::Builder::new()
let mut tempfile = tempfile::Builder::new()
.prefix(file_name)
.suffix(".tmp")
.tempfile_in(parent)?;

View File

@@ -86,18 +86,15 @@
//! [`RequestContext`] argument. Functions in the middle of the call chain
//! only need to pass it on.
use std::sync::{Arc, Mutex, MutexGuard};
use crate::task_mgr::TaskKind;
// The main structure of this module, see module-level comment.
#[derive(Clone)]
#[derive(Clone, Debug)]
pub struct RequestContext {
task_kind: TaskKind,
download_behavior: DownloadBehavior,
access_stats_behavior: AccessStatsBehavior,
page_content_kind: PageContentKind,
page_cache_permit: Option<Arc<crate::page_cache::PinnedSlotsPermit>>,
}
/// The kind of access to the page cache.
@@ -153,7 +150,6 @@ impl RequestContextBuilder {
download_behavior: DownloadBehavior::Download,
access_stats_behavior: AccessStatsBehavior::Update,
page_content_kind: PageContentKind::Unknown,
page_cache_permit: None,
},
}
}
@@ -167,7 +163,6 @@ impl RequestContextBuilder {
download_behavior: original.download_behavior,
access_stats_behavior: original.access_stats_behavior,
page_content_kind: original.page_content_kind,
page_cache_permit: original.page_cache_permit.clone(),
},
}
}
@@ -191,11 +186,6 @@ impl RequestContextBuilder {
self
}
pub(crate) fn page_cache_permit(mut self, p: Arc<crate::page_cache::PinnedSlotsPermit>) -> Self {
self.inner.page_cache_permit = Some(p);
self
}
pub fn build(self) -> RequestContext {
self.inner
}
@@ -296,8 +286,4 @@ impl RequestContext {
pub(crate) fn page_content_kind(&self) -> PageContentKind {
self.page_content_kind
}
pub(crate) fn permit(&self) -> Option<&crate::page_cache::PinnedSlotsPermit> {
self.page_cache_permit.as_ref().map(|p| &**p)
}
}

View File

@@ -3,6 +3,7 @@ mod list_writer;
mod validator;
use std::collections::HashMap;
use std::path::PathBuf;
use std::sync::Arc;
use std::time::Duration;
@@ -12,7 +13,6 @@ use crate::tenant::remote_timeline_client::remote_layer_path;
use crate::tenant::remote_timeline_client::remote_timeline_path;
use crate::virtual_file::VirtualFile;
use anyhow::Context;
use camino::Utf8PathBuf;
use hex::FromHex;
use remote_storage::{GenericRemoteStorage, RemotePath};
use serde::Deserialize;
@@ -336,6 +336,7 @@ impl DeletionList {
timeline_entry.extend(objects.drain(..).map(|p| {
p.strip_prefix(&timeline_remote_path)
.expect("Timeline paths always start with the timeline prefix")
.to_string_lossy()
.to_string()
}));
true
@@ -349,7 +350,7 @@ impl DeletionList {
result.extend(
timeline_layers
.into_iter()
.map(|l| timeline_remote_path.join(&Utf8PathBuf::from(l))),
.map(|l| timeline_remote_path.join(&PathBuf::from(l))),
);
}
}
@@ -726,9 +727,12 @@ impl DeletionQueue {
#[cfg(test)]
mod test {
use camino::Utf8Path;
use hex_literal::hex;
use std::{io::ErrorKind, time::Duration};
use std::{
io::ErrorKind,
path::{Path, PathBuf},
time::Duration,
};
use tracing::info;
use remote_storage::{RemoteStorageConfig, RemoteStorageKind};
@@ -760,7 +764,7 @@ mod test {
struct TestSetup {
harness: TenantHarness,
remote_fs_dir: Utf8PathBuf,
remote_fs_dir: PathBuf,
storage: GenericRemoteStorage,
mock_control_plane: MockControlPlane,
deletion_queue: DeletionQueue,
@@ -869,7 +873,7 @@ mod test {
// Set up a GenericRemoteStorage targetting a directory
let remote_fs_dir = harness.conf.workdir.join("remote_fs");
std::fs::create_dir_all(remote_fs_dir)?;
let remote_fs_dir = harness.conf.workdir.join("remote_fs").canonicalize_utf8()?;
let remote_fs_dir = std::fs::canonicalize(harness.conf.workdir.join("remote_fs"))?;
let storage_config = RemoteStorageConfig {
max_concurrent_syncs: std::num::NonZeroUsize::new(
remote_storage::DEFAULT_REMOTE_STORAGE_MAX_CONCURRENT_SYNCS,
@@ -905,7 +909,7 @@ mod test {
}
// TODO: put this in a common location so that we can share with remote_timeline_client's tests
fn assert_remote_files(expected: &[&str], remote_path: &Utf8Path) {
fn assert_remote_files(expected: &[&str], remote_path: &Path) {
let mut expected: Vec<String> = expected.iter().map(|x| String::from(*x)).collect();
expected.sort();
@@ -922,7 +926,10 @@ mod test {
unreachable!();
}
} else {
panic!("Unexpected error listing {remote_path}: {e}");
panic!(
"Unexpected error listing {}: {e}",
remote_path.to_string_lossy()
);
}
}
};
@@ -937,7 +944,7 @@ mod test {
assert_eq!(expected, found);
}
fn assert_local_files(expected: &[&str], directory: &Utf8Path) {
fn assert_local_files(expected: &[&str], directory: &Path) {
let dir = match std::fs::read_dir(directory) {
Ok(d) => d,
Err(_) => {

View File

@@ -180,7 +180,8 @@ impl ListWriter {
Ok(h) => Ok(Some(h.validated_sequence)),
Err(e) => {
warn!(
"Failed to deserialize deletion header, ignoring {header_path}: {e:#}",
"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
@@ -192,7 +193,10 @@ impl ListWriter {
}
Err(e) => {
if e.kind() == std::io::ErrorKind::NotFound {
debug!("Deletion header {header_path} not found, first start?");
debug!(
"Deletion header {} not found, first start?",
header_path.display()
);
Ok(None)
} else {
Err(anyhow::anyhow!(e))
@@ -219,7 +223,10 @@ impl ListWriter {
let mut dir = match tokio::fs::read_dir(&deletion_directory).await {
Ok(d) => d,
Err(e) => {
warn!("Failed to open deletion list directory {deletion_directory}: {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.
@@ -236,19 +243,21 @@ impl ListWriter {
let file_name = dentry.file_name();
let dentry_str = file_name.to_string_lossy();
if file_name == header_path.file_name().unwrap_or("") {
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().to_str().expect("non-Unicode path"));
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 {absolute_path}: {e:#}");
warn!(
"Failed to clean up temporary file {}: {e:#}",
absolute_path.display()
);
}
continue;
@@ -351,7 +360,7 @@ impl ListWriter {
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(),
self.conf.deletion_prefix().display()
);
metrics::DELETION_QUEUE.unexpected_errors.inc();
return;

View File

@@ -15,10 +15,10 @@
//! Deletions are passed onward to the Deleter.
use std::collections::HashMap;
use std::path::PathBuf;
use std::sync::Arc;
use std::time::Duration;
use camino::Utf8PathBuf;
use tokio_util::sync::CancellationToken;
use tracing::debug;
use tracing::info;
@@ -282,16 +282,16 @@ where
Ok(())
}
async fn cleanup_lists(&mut self, list_paths: Vec<Utf8PathBuf>) {
async fn cleanup_lists(&mut self, list_paths: Vec<PathBuf>) {
for list_path in list_paths {
debug!("Removing deletion list {list_path}");
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 {list_path}: {e:#}");
tracing::error!("Failed to delete {}: {e:#}", list_path.display());
metrics::DELETION_QUEUE.unexpected_errors.inc();
break;
}

View File

@@ -43,12 +43,12 @@
use std::{
collections::HashMap,
path::Path,
sync::Arc,
time::{Duration, SystemTime},
};
use anyhow::Context;
use camino::Utf8Path;
use remote_storage::GenericRemoteStorage;
use serde::{Deserialize, Serialize};
use tokio::time::Instant;
@@ -122,7 +122,7 @@ async fn disk_usage_eviction_task(
state: &State,
task_config: &DiskUsageEvictionTaskConfig,
storage: GenericRemoteStorage,
tenants_dir: &Utf8Path,
tenants_dir: &Path,
cancel: CancellationToken,
) {
scopeguard::defer! {
@@ -184,7 +184,7 @@ async fn disk_usage_eviction_task_iteration(
state: &State,
task_config: &DiskUsageEvictionTaskConfig,
storage: &GenericRemoteStorage,
tenants_dir: &Utf8Path,
tenants_dir: &Path,
cancel: &CancellationToken,
) -> anyhow::Result<()> {
let usage_pre = filesystem_level_usage::get(tenants_dir, task_config)
@@ -620,8 +620,9 @@ impl std::ops::Deref for TimelineKey {
}
mod filesystem_level_usage {
use std::path::Path;
use anyhow::Context;
use camino::Utf8Path;
use crate::statvfs::Statvfs;
@@ -663,7 +664,7 @@ mod filesystem_level_usage {
}
pub fn get<'a>(
tenants_dir: &Utf8Path,
tenants_dir: &Path,
config: &'a DiskUsageEvictionTaskConfig,
) -> anyhow::Result<Usage<'a>> {
let mock_config = {

View File

@@ -10,8 +10,7 @@ use hyper::StatusCode;
use hyper::{Body, Request, Response, Uri};
use metrics::launch_timestamp::LaunchTimestamp;
use pageserver_api::models::{
DownloadRemoteLayersTaskSpawnRequest, LocationConfigMode, TenantAttachRequest,
TenantLoadRequest, TenantLocationConfigRequest,
DownloadRemoteLayersTaskSpawnRequest, TenantAttachRequest, TenantLoadRequest,
};
use remote_storage::GenericRemoteStorage;
use tenant_size_model::{SizeResult, StorageModel};
@@ -30,7 +29,7 @@ use crate::deletion_queue::DeletionQueueClient;
use crate::metrics::{StorageTimeOperation, STORAGE_TIME_GLOBAL};
use crate::pgdatadir_mapping::LsnForTimestamp;
use crate::task_mgr::TaskKind;
use crate::tenant::config::{LocationConf, TenantConfOpt};
use crate::tenant::config::TenantConfOpt;
use crate::tenant::mgr::{
GetTenantError, SetNewTenantConfigError, TenantMapInsertError, TenantStateError,
};
@@ -133,7 +132,7 @@ impl From<PageReconstructError> for ApiError {
ApiError::InternalServerError(anyhow::anyhow!("request was cancelled"))
}
PageReconstructError::AncestorStopping(_) => {
ApiError::ResourceUnavailable(format!("{pre}"))
ApiError::InternalServerError(anyhow::Error::new(pre))
}
PageReconstructError::WalRedo(pre) => {
ApiError::InternalServerError(anyhow::Error::new(pre))
@@ -146,15 +145,12 @@ impl From<TenantMapInsertError> for ApiError {
fn from(tmie: TenantMapInsertError) -> ApiError {
match tmie {
TenantMapInsertError::StillInitializing | TenantMapInsertError::ShuttingDown => {
ApiError::ResourceUnavailable(format!("{tmie}"))
ApiError::InternalServerError(anyhow::Error::new(tmie))
}
TenantMapInsertError::TenantAlreadyExists(id, state) => {
ApiError::Conflict(format!("tenant {id} already exists, state: {state:?}"))
}
TenantMapInsertError::TenantExistsSecondary(id) => {
ApiError::Conflict(format!("tenant {id} already exists as secondary"))
}
TenantMapInsertError::Other(e) => ApiError::InternalServerError(e),
TenantMapInsertError::Closure(e) => ApiError::InternalServerError(e),
}
}
}
@@ -163,12 +159,6 @@ impl From<TenantStateError> for ApiError {
fn from(tse: TenantStateError) -> ApiError {
match tse {
TenantStateError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {}", tid).into()),
TenantStateError::NotActive(_) => {
ApiError::ResourceUnavailable("Tenant not yet active".into())
}
TenantStateError::IsStopping(_) => {
ApiError::ResourceUnavailable("Tenant is stopping".into())
}
_ => ApiError::InternalServerError(anyhow::Error::new(tse)),
}
}
@@ -178,17 +168,14 @@ impl From<GetTenantError> for ApiError {
fn from(tse: GetTenantError) -> ApiError {
match tse {
GetTenantError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {}", tid).into()),
GetTenantError::Broken(reason) => {
ApiError::InternalServerError(anyhow!("tenant is broken: {}", reason))
}
GetTenantError::NotActive(_) => {
e @ GetTenantError::NotActive(_) => {
// Why is this not `ApiError::NotFound`?
// Because we must be careful to never return 404 for a tenant if it does
// in fact exist locally. If we did, the caller could draw the conclusion
// that it can attach the tenant to another PS and we'd be in split-brain.
//
// (We can produce this variant only in `mgr::get_tenant(..., active=true)` calls).
ApiError::ResourceUnavailable("Tenant not yet active".into())
ApiError::InternalServerError(anyhow::Error::new(e))
}
}
}
@@ -635,9 +622,8 @@ async fn tenant_list_handler(
let response_data = mgr::list_tenants()
.instrument(info_span!("tenant_list"))
.await
.map_err(|_| {
ApiError::ResourceUnavailable("Tenant map is initializing or shutting down".to_string())
})?
.map_err(anyhow::Error::new)
.map_err(ApiError::InternalServerError)?
.iter()
.map(|(id, state)| TenantInfo {
id: *id,
@@ -1015,48 +1001,6 @@ async fn update_tenant_config_handler(
json_response(StatusCode::OK, ())
}
async fn put_tenant_location_config_handler(
mut request: Request<Body>,
_cancel: CancellationToken,
) -> Result<Response<Body>, ApiError> {
let request_data: TenantLocationConfigRequest = json_request(&mut request).await?;
let tenant_id = request_data.tenant_id;
check_permission(&request, Some(tenant_id))?;
let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
let state = get_state(&request);
let conf = state.conf;
// The `Detached` state is special, it doesn't upsert a tenant, it removes
// its local disk content and drops it from memory.
if let LocationConfigMode::Detached = request_data.config.mode {
mgr::detach_tenant(conf, tenant_id, true)
.instrument(info_span!("tenant_detach", %tenant_id))
.await?;
return json_response(StatusCode::OK, ());
}
let location_conf =
LocationConf::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
mgr::upsert_location(
state.conf,
tenant_id,
location_conf,
state.broker_client.clone(),
state.remote_storage.clone(),
state.deletion_queue_client.clone(),
&ctx,
)
.await
// TODO: badrequest assumes the caller was asking for something unreasonable, but in
// principle we might have hit something like concurrent API calls to the same tenant,
// which is not a 400 but a 409.
.map_err(ApiError::BadRequest)?;
json_response(StatusCode::OK, ())
}
/// Testing helper to transition a tenant to [`crate::tenant::TenantState::Broken`].
async fn handle_tenant_break(
r: Request<Body>,
@@ -1510,9 +1454,6 @@ pub fn make_router(
.get("/v1/tenant/:tenant_id/config", |r| {
api_handler(r, get_tenant_config_handler)
})
.put("/v1/tenant/:tenant_id/location_config", |r| {
api_handler(r, put_tenant_location_config_handler)
})
.get("/v1/tenant/:tenant_id/timeline", |r| {
api_handler(r, timeline_list_handler)
})

View File

@@ -6,7 +6,6 @@ use std::path::{Path, PathBuf};
use anyhow::{bail, ensure, Context, Result};
use bytes::Bytes;
use camino::Utf8Path;
use futures::StreamExt;
use tokio::io::{AsyncRead, AsyncReadExt};
use tokio_tar::Archive;
@@ -30,7 +29,7 @@ use postgres_ffi::{BLCKSZ, WAL_SEGMENT_SIZE};
use utils::lsn::Lsn;
// Returns checkpoint LSN from controlfile
pub fn get_lsn_from_controlfile(path: &Utf8Path) -> Result<Lsn> {
pub fn get_lsn_from_controlfile(path: &Path) -> Result<Lsn> {
// Read control file to extract the LSN
let controlfile_path = path.join("global").join("pg_control");
let controlfile = ControlFileData::decode(&std::fs::read(controlfile_path)?)?;
@@ -47,7 +46,7 @@ pub fn get_lsn_from_controlfile(path: &Utf8Path) -> Result<Lsn> {
/// cluster was not shut down cleanly.
pub async fn import_timeline_from_postgres_datadir(
tline: &Timeline,
pgdata_path: &Utf8Path,
pgdata_path: &Path,
pgdata_lsn: Lsn,
ctx: &RequestContext,
) -> Result<()> {
@@ -257,7 +256,7 @@ async fn import_slru(
/// Scan PostgreSQL WAL files in given directory and load all records between
/// 'startpoint' and 'endpoint' into the repository.
async fn import_wal(
walpath: &Utf8Path,
walpath: &Path,
tline: &Timeline,
startpoint: Lsn,
endpoint: Lsn,

View File

@@ -25,8 +25,9 @@ pub mod walredo;
pub mod failpoint_support;
use std::path::Path;
use crate::task_mgr::TaskKind;
use camino::Utf8Path;
use deletion_queue::DeletionQueue;
use tracing::info;
@@ -112,10 +113,6 @@ pub const METADATA_FILE_NAME: &str = "metadata";
/// Full path: `tenants/<tenant_id>/config`.
pub const TENANT_CONFIG_NAME: &str = "config";
/// Per-tenant configuration file.
/// Full path: `tenants/<tenant_id>/config`.
pub const TENANT_LOCATION_CONFIG_NAME: &str = "config-v1";
/// A suffix used for various temporary files. Any temporary files found in the
/// data directory at pageserver startup can be automatically removed.
pub const TEMP_FILE_SUFFIX: &str = "___temp";
@@ -135,25 +132,25 @@ pub const TIMELINE_DELETE_MARK_SUFFIX: &str = "___delete";
/// Full path: `tenants/<tenant_id>/___ignored_tenant`.
pub const IGNORED_TENANT_FILE_NAME: &str = "___ignored_tenant";
pub fn is_temporary(path: &Utf8Path) -> bool {
pub fn is_temporary(path: &Path) -> bool {
match path.file_name() {
Some(name) => name.ends_with(TEMP_FILE_SUFFIX),
Some(name) => name.to_string_lossy().ends_with(TEMP_FILE_SUFFIX),
None => false,
}
}
fn ends_with_suffix(path: &Utf8Path, suffix: &str) -> bool {
fn ends_with_suffix(path: &Path, suffix: &str) -> bool {
match path.file_name() {
Some(name) => name.ends_with(suffix),
Some(name) => name.to_string_lossy().ends_with(suffix),
None => false,
}
}
pub fn is_uninit_mark(path: &Utf8Path) -> bool {
pub fn is_uninit_mark(path: &Path) -> bool {
ends_with_suffix(path, TIMELINE_UNINIT_MARK_SUFFIX)
}
pub fn is_delete_mark(path: &Utf8Path) -> bool {
pub fn is_delete_mark(path: &Path) -> bool {
ends_with_suffix(path, TIMELINE_DELETE_MARK_SUFFIX)
}

View File

@@ -94,35 +94,15 @@ pub(crate) static READ_NUM_FS_LAYERS: Lazy<Histogram> = Lazy::new(|| {
});
// Metrics collected on operations on the storage repository.
pub(crate) struct ReconstructTimeMetrics {
ok: Histogram,
err: Histogram,
}
pub(crate) static RECONSTRUCT_TIME: Lazy<ReconstructTimeMetrics> = Lazy::new(|| {
let inner = register_histogram_vec!(
pub(crate) static RECONSTRUCT_TIME: Lazy<Histogram> = Lazy::new(|| {
register_histogram!(
"pageserver_getpage_reconstruct_seconds",
"Time spent in reconstruct_value (reconstruct a page from deltas)",
&["result"],
CRITICAL_OP_BUCKETS.into(),
)
.expect("failed to define a metric");
ReconstructTimeMetrics {
ok: inner.get_metric_with_label_values(&["ok"]).unwrap(),
err: inner.get_metric_with_label_values(&["err"]).unwrap(),
}
.expect("failed to define a metric")
});
impl ReconstructTimeMetrics {
pub(crate) fn for_result<T, E>(&self, result: &Result<T, E>) -> &Histogram {
match result {
Ok(_) => &self.ok,
Err(_) => &self.err,
}
}
}
pub(crate) static MATERIALIZED_PAGE_CACHE_HIT_DIRECT: Lazy<IntCounter> = Lazy::new(|| {
register_int_counter!(
"pageserver_materialized_cache_hits_direct_total",
@@ -284,45 +264,6 @@ pub static PAGE_CACHE_SIZE: Lazy<PageCacheSizeMetrics> = Lazy::new(|| PageCacheS
},
});
pub(crate) static PAGE_CACHE_ACQUIRE_PINNED_SLOT_TIME: Lazy<Histogram> = Lazy::new(|| {
register_histogram!(
"pageserver_page_cache_acquire_pinned_slot_seconds",
"Time spent acquiring a pinned slot in the page cache",
CRITICAL_OP_BUCKETS.into(),
)
.expect("failed to define a metric")
});
pub(crate) static PAGE_CACHE_FIND_VICTIMS_ITERS_TOTAL: Lazy<IntCounter> = Lazy::new(|| {
register_int_counter!(
"pageserver_page_cache_find_victim_iters_total",
"Counter for the number of iterations in the find_victim loop",
)
.expect("failed to define a metric")
});
static PAGE_CACHE_ERRORS: Lazy<IntCounterVec> = Lazy::new(|| {
register_int_counter_vec!(
"page_cache_errors_total",
"Number of timeouts while acquiring a pinned slot in the page cache",
&["error_kind"]
)
.expect("failed to define a metric")
});
#[derive(IntoStaticStr)]
#[strum(serialize_all = "kebab_case")]
pub(crate) enum PageCacheErrorKind {
AcquirePinnedSlotTimeout,
}
pub(crate) fn page_cache_errors_inc(error_kind: PageCacheErrorKind) {
PAGE_CACHE_ERRORS
.get_metric_with_label_values(&[error_kind.into()])
.unwrap()
.inc();
}
pub(crate) static WAIT_LSN_TIME: Lazy<Histogram> = Lazy::new(|| {
register_histogram!(
"pageserver_wait_lsn_seconds",
@@ -1875,6 +1816,7 @@ pub fn preinitialize_metrics() {
// histograms
[
&READ_NUM_FS_LAYERS,
&RECONSTRUCT_TIME,
&WAIT_LSN_TIME,
&WAL_REDO_TIME,
&WAL_REDO_WAIT_TIME,
@@ -1885,7 +1827,4 @@ pub fn preinitialize_metrics() {
.for_each(|h| {
Lazy::force(h);
});
// Custom
Lazy::force(&RECONSTRUCT_TIME);
}

View File

@@ -66,7 +66,8 @@
//! inserted to the mapping, but you must hold the write-lock on the slot until
//! the contents are valid. If you need to release the lock without initializing
//! the contents, you must remove the mapping first. We make that easy for the
//! callers with PageWriteGuard: the caller must explicitly call guard.mark_valid() after it has
//! callers with PageWriteGuard: when lock_for_write() returns an uninitialized
//! page, the caller must explicitly call guard.mark_valid() after it has
//! initialized it. If the guard is dropped without calling mark_valid(), the
//! mapping is automatically removed and the slot is marked free.
//!
@@ -74,12 +75,7 @@
use std::{
collections::{hash_map::Entry, HashMap},
convert::TryInto,
sync::{
atomic::{AtomicU64, AtomicU8, AtomicUsize, Ordering},
Arc, Weak,
},
task::Poll,
time::Duration,
sync::atomic::{AtomicU64, AtomicU8, AtomicUsize, Ordering},
};
use anyhow::Context;
@@ -169,8 +165,6 @@ struct Slot {
struct SlotInner {
key: Option<CacheKey>,
// for `coalesce_readers_permit`
permit: std::sync::Mutex<Weak<PinnedSlotsPermit>>,
buf: &'static mut [u8; PAGE_SZ],
}
@@ -213,27 +207,6 @@ impl Slot {
}
}
impl SlotInner {
/// If there is aready a reader, drop our permit and share its permit, just like we share read access.
fn coalesce_readers_permit<'c>(&self, permit: PermitKind<'c>) -> PermitKindReadGuard<'c> {
match permit {
PermitKind::CtxProvided(permit) => PermitKindReadGuard::CtxProvided(permit),
PermitKind::Acquired(permit) => {
let mut guard = self.permit.lock().unwrap();
if let Some(existing_permit) = guard.upgrade() {
drop(guard);
drop(permit);
existing_permit
} else {
let permit = Arc::new(permit);
*guard = Arc::downgrade(&permit);
permit
}
}
}
}
}
pub struct PageCache {
/// This contains the mapping from the cache key to buffer slot that currently
/// contains the page, if any.
@@ -251,52 +224,30 @@ pub struct PageCache {
/// The actual buffers with their metadata.
slots: Box<[Slot]>,
pinned_slots: Arc<tokio::sync::Semaphore>,
/// Index of the next candidate to evict, for the Clock replacement algorithm.
/// This is interpreted modulo the page cache size.
next_evict_slot: AtomicUsize,
find_victim_sender:
async_channel::Sender<(usize, tokio::sync::RwLockWriteGuard<'static, SlotInner>)>,
find_victim_waiters:
async_channel::Receiver<(usize, tokio::sync::RwLockWriteGuard<'static, SlotInner>)>,
size_metrics: &'static PageCacheSizeMetrics,
}
pub(crate) struct PinnedSlotsPermit(tokio::sync::OwnedSemaphorePermit);
enum PermitKind<'c> {
CtxProvided(&'c PinnedSlotsPermit),
Acquired(PinnedSlotsPermit),
}
enum PermitKindReadGuard<'c> {
CtxProvided(&'c PinnedSlotsPermit),
Coalesced(Arc<PinnedSlotsPermit>),
}
///
/// PageReadGuard is a "lease" on a buffer, for reading. The page is kept locked
/// until the guard is dropped.
///
pub struct PageReadGuard<'c, 'i> {
_permit: PermitKindReadGuard<'c>,
slot_guard: tokio::sync::RwLockReadGuard<'i, SlotInner>,
}
pub struct PageReadGuard<'i>(tokio::sync::RwLockReadGuard<'i, SlotInner>);
impl std::ops::Deref for PageReadGuard<'_, '_> {
impl std::ops::Deref for PageReadGuard<'_> {
type Target = [u8; PAGE_SZ];
fn deref(&self) -> &Self::Target {
self.slot_guard.buf
self.0.buf
}
}
impl AsRef<[u8; PAGE_SZ]> for PageReadGuard<'_, '_> {
impl AsRef<[u8; PAGE_SZ]> for PageReadGuard<'_> {
fn as_ref(&self) -> &[u8; PAGE_SZ] {
self.slot_guard.buf
self.0.buf
}
}
@@ -306,89 +257,76 @@ impl AsRef<[u8; PAGE_SZ]> for PageReadGuard<'_, '_> {
///
/// Counterintuitively, this is used even for a read, if the requested page is not
/// currently found in the page cache. In that case, the caller of lock_for_read()
/// is expected to fill in the page contents and call mark_valid().
pub struct PageWriteGuard<'c, 'i> {
state: PageWriteGuardState<'c, 'i>,
/// is expected to fill in the page contents and call mark_valid(). Similarly
/// lock_for_write() can return an invalid buffer that the caller is expected to
/// to initialize.
///
pub struct PageWriteGuard<'i> {
inner: tokio::sync::RwLockWriteGuard<'i, SlotInner>,
// Are the page contents currently valid?
// Used to mark pages as invalid that are assigned but not yet filled with data.
valid: bool,
}
enum PageWriteGuardState<'c, 'i> {
Invalid {
inner: tokio::sync::RwLockWriteGuard<'i, SlotInner>,
_permit: PermitKindReadGuard<'c>,
},
Downgraded,
}
impl std::ops::DerefMut for PageWriteGuard<'_, '_> {
impl std::ops::DerefMut for PageWriteGuard<'_> {
fn deref_mut(&mut self) -> &mut Self::Target {
match &mut self.state {
PageWriteGuardState::Invalid { inner, _permit } => inner.buf,
PageWriteGuardState::Downgraded => unreachable!(),
}
self.inner.buf
}
}
impl std::ops::Deref for PageWriteGuard<'_, '_> {
impl std::ops::Deref for PageWriteGuard<'_> {
type Target = [u8; PAGE_SZ];
fn deref(&self) -> &Self::Target {
match &self.state {
PageWriteGuardState::Invalid { inner, _permit } => inner.buf,
PageWriteGuardState::Downgraded => unreachable!(),
}
self.inner.buf
}
}
impl AsMut<[u8; PAGE_SZ]> for PageWriteGuard<'_, '_> {
impl AsMut<[u8; PAGE_SZ]> for PageWriteGuard<'_> {
fn as_mut(&mut self) -> &mut [u8; PAGE_SZ] {
match &mut self.state {
PageWriteGuardState::Invalid { inner, _permit } => inner.buf,
PageWriteGuardState::Downgraded => todo!(),
}
self.inner.buf
}
}
impl<'c, 'a> PageWriteGuard<'c, 'a> {
impl PageWriteGuard<'_> {
/// Mark that the buffer contents are now valid.
#[must_use]
pub fn mark_valid(mut self) -> PageReadGuard<'c, 'a> {
let prev = std::mem::replace(&mut self.state, PageWriteGuardState::Downgraded);
match prev {
PageWriteGuardState::Invalid { inner, _permit } => {
assert!(inner.key.is_some());
PageReadGuard {
_permit,
slot_guard: inner.downgrade(),
}
}
PageWriteGuardState::Downgraded => unreachable!(),
}
pub fn mark_valid(&mut self) {
assert!(self.inner.key.is_some());
assert!(
!self.valid,
"mark_valid called on a buffer that was already valid"
);
self.valid = true;
}
}
impl Drop for PageWriteGuard<'_, '_> {
impl Drop for PageWriteGuard<'_> {
///
/// If the buffer was allocated for a page that was not already in the
/// cache, but the lock_for_read/write() caller dropped the buffer without
/// initializing it, remove the mapping from the page cache.
///
fn drop(&mut self) {
match &mut self.state {
PageWriteGuardState::Invalid { inner, _permit } => {
assert!(inner.key.is_some());
let self_key = inner.key.as_ref().unwrap();
PAGE_CACHE.get().unwrap().remove_mapping(self_key);
inner.key = None;
}
PageWriteGuardState::Downgraded => {}
assert!(self.inner.key.is_some());
if !self.valid {
let self_key = self.inner.key.as_ref().unwrap();
PAGE_CACHE.get().unwrap().remove_mapping(self_key);
self.inner.key = None;
}
}
}
/// lock_for_read() return value
pub enum ReadBufResult<'c, 'a> {
Found(PageReadGuard<'c, 'a>),
NotFound(PageWriteGuard<'c, 'a>),
pub enum ReadBufResult<'a> {
Found(PageReadGuard<'a>),
NotFound(PageWriteGuard<'a>),
}
/// lock_for_write() return value
pub enum WriteBufResult<'a> {
Found(PageWriteGuard<'a>),
NotFound(PageWriteGuard<'a>),
}
impl PageCache {
@@ -410,9 +348,6 @@ impl PageCache {
lsn: Lsn,
ctx: &RequestContext,
) -> Option<(Lsn, PageReadGuard)> {
let Ok(permit) = self.try_get_pinned_slot_permit(ctx).await else {
return None;
};
crate::metrics::PAGE_CACHE
.for_ctx(ctx)
.read_accesses_materialized_page
@@ -427,10 +362,7 @@ impl PageCache {
lsn,
};
if let Some(guard) = self
.try_lock_for_read(&mut cache_key, &mut Some(permit))
.await
{
if let Some(guard) = self.try_lock_for_read(&mut cache_key).await {
if let CacheKey::MaterializedPage {
hash_key: _,
lsn: available_lsn,
@@ -460,13 +392,12 @@ impl PageCache {
/// Store an image of the given page in the cache.
///
pub async fn memorize_materialized_page(
&'static self,
&self,
tenant_id: TenantId,
timeline_id: TimelineId,
key: Key,
lsn: Lsn,
img: &[u8],
ctx: &RequestContext,
) -> anyhow::Result<()> {
let cache_key = CacheKey::MaterializedPage {
hash_key: MaterializedPageHashKey {
@@ -477,87 +408,30 @@ impl PageCache {
lsn,
};
let mut permit = Some(self.try_get_pinned_slot_permit(ctx).await?);
loop {
// First check if the key already exists in the cache.
if let Some(slot_idx) = self.search_mapping_exact(&cache_key) {
// The page was found in the mapping. Lock the slot, and re-check
// that it's still what we expected (because we don't released the mapping
// lock already, another thread could have evicted the page)
let slot = &self.slots[slot_idx];
let inner = slot.inner.write().await;
if inner.key.as_ref() == Some(&cache_key) {
slot.inc_usage_count();
debug_assert!(
{
let guard = inner.permit.lock().unwrap();
guard.upgrade().is_none()
},
"we hold a write lock, so, no one else should have a permit"
);
debug_assert_eq!(inner.buf.len(), img.len());
// We already had it in cache. Another thread must've put it there
// concurrently. Check that it had the same contents that we
// replayed.
assert!(inner.buf == img);
return Ok(());
}
match self.lock_for_write(&cache_key).await? {
WriteBufResult::Found(write_guard) => {
// We already had it in cache. Another thread must've put it there
// concurrently. Check that it had the same contents that we
// replayed.
assert!(*write_guard == img);
}
debug_assert!(permit.is_some());
// Not found. Find a victim buffer
let (slot_idx, mut inner) = self
.find_victim(permit.as_ref().unwrap())
.await
.context("Failed to find evict victim")?;
// Insert mapping for this. At this point, we may find that another
// thread did the same thing concurrently. In that case, we evicted
// our victim buffer unnecessarily. Put it into the free list and
// continue with the slot that the other thread chose.
if let Some(_existing_slot_idx) = self.try_insert_mapping(&cache_key, slot_idx) {
// TODO: put to free list
// We now just loop back to start from beginning. This is not
// optimal, we'll perform the lookup in the mapping again, which
// is not really necessary because we already got
// 'existing_slot_idx'. But this shouldn't happen often enough
// to matter much.
continue;
WriteBufResult::NotFound(mut write_guard) => {
write_guard.copy_from_slice(img);
write_guard.mark_valid();
}
// Make the slot ready
let slot = &self.slots[slot_idx];
inner.key = Some(cache_key.clone());
slot.set_usage_count(1);
// Create a write guard for the slot so we go through the expected motions.
debug_assert!(
{
let guard = inner.permit.lock().unwrap();
guard.upgrade().is_none()
},
"we hold a write lock, so, no one else should have a permit"
);
let mut write_guard = PageWriteGuard {
state: PageWriteGuardState::Invalid {
_permit: permit.take().unwrap(),
inner,
},
};
write_guard.copy_from_slice(img);
let _ = write_guard.mark_valid();
return Ok(());
}
Ok(())
}
// Section 1.2: Public interface functions for working with immutable file pages.
pub async fn read_immutable_buf<'c>(
&'static self,
pub async fn read_immutable_buf(
&self,
file_id: FileId,
blkno: u32,
ctx: &'c RequestContext,
) -> anyhow::Result<ReadBufResult<'c, 'static>> {
ctx: &RequestContext,
) -> anyhow::Result<ReadBufResult> {
let mut cache_key = CacheKey::ImmutableFilePage { file_id, blkno };
self.lock_for_read(&mut cache_key, ctx).await
@@ -571,44 +445,6 @@ impl PageCache {
// "mappings" after this section. But the routines in this section should
// not require changes.
pub(crate) async fn get_permit(&self) -> Arc<PinnedSlotsPermit> {
Arc::new(PinnedSlotsPermit(
Arc::clone(&self.pinned_slots)
.acquire_owned()
.await
.expect("the semaphore is never closed"),
))
}
async fn try_get_pinned_slot_permit<'c>(
&self,
ctx: &'c RequestContext,
) -> anyhow::Result<PermitKind<'c>> {
if let Some(permit) = ctx.permit() {
return Ok(PermitKind::CtxProvided(permit));
};
let timer = crate::metrics::PAGE_CACHE_ACQUIRE_PINNED_SLOT_TIME.start_timer();
match tokio::time::timeout(
// Choose small timeout, neon_smgr does its own retries.
// https://neondb.slack.com/archives/C04DGM6SMTM/p1694786876476869
Duration::from_secs(10),
Arc::clone(&self.pinned_slots).acquire_owned(),
)
.await
{
Ok(res) => Ok(PermitKind::Acquired(PinnedSlotsPermit(
res.expect("this semaphore is never closed"),
))),
Err(_timeout) => {
timer.stop_and_discard();
crate::metrics::page_cache_errors_inc(
crate::metrics::PageCacheErrorKind::AcquirePinnedSlotTimeout,
);
anyhow::bail!("timeout: there were page guards alive for all page cache slots")
}
}
}
/// Look up a page in the cache.
///
/// If the search criteria is not exact, *cache_key is updated with the key
@@ -618,11 +454,7 @@ impl PageCache {
///
/// If no page is found, returns None and *cache_key is left unmodified.
///
async fn try_lock_for_read<'c>(
&self,
cache_key: &mut CacheKey,
permit: &mut Option<PermitKind<'c>>,
) -> Option<PageReadGuard> {
async fn try_lock_for_read(&self, cache_key: &mut CacheKey) -> Option<PageReadGuard> {
let cache_key_orig = cache_key.clone();
if let Some(slot_idx) = self.search_mapping(cache_key) {
// The page was found in the mapping. Lock the slot, and re-check
@@ -632,10 +464,7 @@ impl PageCache {
let inner = slot.inner.read().await;
if inner.key.as_ref() == Some(cache_key) {
slot.inc_usage_count();
return Some(PageReadGuard {
_permit: inner.coalesce_readers_permit(permit.take().unwrap()),
slot_guard: inner,
});
return Some(PageReadGuard(inner));
} else {
// search_mapping might have modified the search key; restore it.
*cache_key = cache_key_orig;
@@ -674,12 +503,10 @@ impl PageCache {
/// ```
///
async fn lock_for_read(
&'static self,
&self,
cache_key: &mut CacheKey,
ctx: &RequestContext,
) -> anyhow::Result<ReadBufResult> {
let mut permit = Some(self.try_get_pinned_slot_permit(ctx).await?);
let (read_access, hit) = match cache_key {
CacheKey::MaterializedPage { .. } => {
unreachable!("Materialized pages use lookup_materialized_page")
@@ -696,21 +523,17 @@ impl PageCache {
let mut is_first_iteration = true;
loop {
// First check if the key already exists in the cache.
if let Some(read_guard) = self.try_lock_for_read(cache_key, &mut permit).await {
debug_assert!(permit.is_none());
if let Some(read_guard) = self.try_lock_for_read(cache_key).await {
if is_first_iteration {
hit.inc();
}
return Ok(ReadBufResult::Found(read_guard));
}
debug_assert!(permit.is_some());
is_first_iteration = false;
// Not found. Find a victim buffer
let (slot_idx, mut inner) = self
.find_victim(permit.as_ref().unwrap())
.await
.context("Failed to find evict victim")?;
let (slot_idx, mut inner) =
self.find_victim().context("Failed to find evict victim")?;
// Insert mapping for this. At this point, we may find that another
// thread did the same thing concurrently. In that case, we evicted
@@ -732,19 +555,70 @@ impl PageCache {
inner.key = Some(cache_key.clone());
slot.set_usage_count(1);
debug_assert!(
{
let guard = inner.permit.lock().unwrap();
guard.upgrade().is_none()
},
"we hold a write lock, so, no one else should have a permit"
);
return Ok(ReadBufResult::NotFound(PageWriteGuard {
state: PageWriteGuardState::Invalid {
_permit: permit.take().unwrap(),
inner,
},
inner,
valid: false,
}));
}
}
/// Look up a page in the cache and lock it in write mode. If it's not
/// found, returns None.
///
/// When locking a page for writing, the search criteria is always "exact".
async fn try_lock_for_write(&self, cache_key: &CacheKey) -> Option<PageWriteGuard> {
if let Some(slot_idx) = self.search_mapping_for_write(cache_key) {
// The page was found in the mapping. Lock the slot, and re-check
// that it's still what we expected (because we don't released the mapping
// lock already, another thread could have evicted the page)
let slot = &self.slots[slot_idx];
let inner = slot.inner.write().await;
if inner.key.as_ref() == Some(cache_key) {
slot.inc_usage_count();
return Some(PageWriteGuard { inner, valid: true });
}
}
None
}
/// Return a write-locked buffer for given block.
///
/// Similar to lock_for_read(), but the returned buffer is write-locked and
/// may be modified by the caller even if it's already found in the cache.
async fn lock_for_write(&self, cache_key: &CacheKey) -> anyhow::Result<WriteBufResult> {
loop {
// First check if the key already exists in the cache.
if let Some(write_guard) = self.try_lock_for_write(cache_key).await {
return Ok(WriteBufResult::Found(write_guard));
}
// Not found. Find a victim buffer
let (slot_idx, mut inner) =
self.find_victim().context("Failed to find evict victim")?;
// Insert mapping for this. At this point, we may find that another
// thread did the same thing concurrently. In that case, we evicted
// our victim buffer unnecessarily. Put it into the free list and
// continue with the slot that the other thread chose.
if let Some(_existing_slot_idx) = self.try_insert_mapping(cache_key, slot_idx) {
// TODO: put to free list
// We now just loop back to start from beginning. This is not
// optimal, we'll perform the lookup in the mapping again, which
// is not really necessary because we already got
// 'existing_slot_idx'. But this shouldn't happen often enough
// to matter much.
continue;
}
// Make the slot ready
let slot = &self.slots[slot_idx];
inner.key = Some(cache_key.clone());
slot.set_usage_count(1);
return Ok(WriteBufResult::NotFound(PageWriteGuard {
inner,
valid: false,
}));
}
}
@@ -789,7 +663,7 @@ impl PageCache {
///
/// Like 'search_mapping, but performs an "exact" search. Used for
/// allocating a new buffer.
fn search_mapping_exact(&self, key: &CacheKey) -> Option<usize> {
fn search_mapping_for_write(&self, key: &CacheKey) -> Option<usize> {
match key {
CacheKey::MaterializedPage { hash_key, lsn } => {
let map = self.materialized_page_map.read().unwrap();
@@ -895,13 +769,8 @@ impl PageCache {
/// Find a slot to evict.
///
/// On return, the slot is empty and write-locked.
async fn find_victim(
&'static self,
_permit_witness: &PinnedSlotsPermit,
) -> anyhow::Result<(usize, tokio::sync::RwLockWriteGuard<SlotInner>)> {
// Get in line.
let receiver = self.find_victim_waiters.recv();
fn find_victim(&self) -> anyhow::Result<(usize, tokio::sync::RwLockWriteGuard<SlotInner>)> {
let iter_limit = self.slots.len() * 10;
let mut iters = 0;
loop {
iters += 1;
@@ -913,8 +782,14 @@ impl PageCache {
let mut inner = match slot.inner.try_write() {
Ok(inner) => inner,
Err(_err) => {
if iters > self.slots.len() * (MAX_USAGE_COUNT as usize) {
unreachable!("find_victim_waiters prevents starvation");
// If we have looped through the whole buffer pool 10 times
// and still haven't found a victim buffer, something's wrong.
// Maybe all the buffers were in locked. That could happen in
// theory, if you have more threads holding buffers locked than
// there are buffers in the pool. In practice, with a reasonably
// large buffer pool it really shouldn't happen.
if iters > iter_limit {
anyhow::bail!("exceeded evict iter limit");
}
continue;
}
@@ -924,17 +799,7 @@ impl PageCache {
self.remove_mapping(old_key);
inner.key = None;
}
crate::metrics::PAGE_CACHE_FIND_VICTIMS_ITERS_TOTAL.inc_by(iters as u64);
self.find_victim_sender
.try_send((slot_idx, inner))
.expect("we always get in line first");
match futures::poll!(receiver) {
Poll::Ready(Ok(res)) => return Ok(res),
Poll::Ready(Err(_closed)) => unreachable!("we never close"),
Poll::Pending => {
unreachable!("we just sent to the channel and got in line earlier")
}
}
return Ok((slot_idx, inner));
}
}
}
@@ -961,26 +826,18 @@ impl PageCache {
let buf: &mut [u8; PAGE_SZ] = chunk.try_into().unwrap();
Slot {
inner: tokio::sync::RwLock::new(SlotInner {
key: None,
buf,
permit: std::sync::Mutex::new(Weak::new()),
}),
inner: tokio::sync::RwLock::new(SlotInner { key: None, buf }),
usage_count: AtomicU8::new(0),
}
})
.collect();
let (find_victim_sender, find_victim_waiters) = async_channel::bounded(num_pages);
Self {
materialized_page_map: Default::default(),
immutable_page_map: Default::default(),
slots,
next_evict_slot: AtomicUsize::new(0),
size_metrics,
pinned_slots: Arc::new(tokio::sync::Semaphore::new(num_pages)),
find_victim_sender,
find_victim_waiters,
}
}
}

View File

@@ -412,53 +412,31 @@ impl PageServerHandler {
// TODO: We could create a new per-request context here, with unique ID.
// Currently we use the same per-timeline context for all requests
let (response, span) = match neon_fe_msg {
let response = match neon_fe_msg {
PagestreamFeMessage::Exists(req) => {
let _timer = metrics.start_timer(metrics::SmgrQueryType::GetRelExists);
let span = tracing::info_span!("handle_get_rel_exists_request", rel = %req.rel, req_lsn = %req.lsn);
(
self.handle_get_rel_exists_request(&timeline, &req, &ctx)
.instrument(span.clone())
.await,
span,
)
self.handle_get_rel_exists_request(&timeline, &req, &ctx)
.await
}
PagestreamFeMessage::Nblocks(req) => {
let _timer = metrics.start_timer(metrics::SmgrQueryType::GetRelSize);
let span = tracing::info_span!("handle_get_nblocks_request", rel = %req.rel, req_lsn = %req.lsn);
(
self.handle_get_nblocks_request(&timeline, &req, &ctx)
.instrument(span.clone())
.await,
span,
)
self.handle_get_nblocks_request(&timeline, &req, &ctx).await
}
PagestreamFeMessage::GetPage(req) => {
let _timer = metrics.start_timer(metrics::SmgrQueryType::GetPageAtLsn);
let span = tracing::info_span!("handle_get_page_at_lsn_request", rel = %req.rel, blkno = %req.blkno, req_lsn = %req.lsn);
(
self.handle_get_page_at_lsn_request(&timeline, &req, &ctx)
.instrument(span.clone())
.await,
span,
)
self.handle_get_page_at_lsn_request(&timeline, &req, &ctx)
.await
}
PagestreamFeMessage::DbSize(req) => {
let _timer = metrics.start_timer(metrics::SmgrQueryType::GetDbSize);
let span = tracing::info_span!("handle_db_size_request", dbnode = %req.dbnode, req_lsn = %req.lsn);
(
self.handle_db_size_request(&timeline, &req, &ctx)
.instrument(span.clone())
.await,
span,
)
self.handle_db_size_request(&timeline, &req, &ctx).await
}
};
let response = response.unwrap_or_else(|e| {
// print the all details to the log with {:#}, but for the client the
// error message is enough
span.in_scope(|| error!("error reading relation or page version: {:#}", e));
error!("error reading relation or page version: {:?}", e);
PagestreamBeMessage::Error(PagestreamErrorResponse {
message: e.to_string(),
})
@@ -649,6 +627,7 @@ impl PageServerHandler {
Ok(lsn)
}
#[instrument(skip(self, timeline, req, ctx), fields(rel = %req.rel, req_lsn = %req.lsn))]
async fn handle_get_rel_exists_request(
&self,
timeline: &Timeline,
@@ -669,6 +648,7 @@ impl PageServerHandler {
}))
}
#[instrument(skip(self, timeline, req, ctx), fields(rel = %req.rel, req_lsn = %req.lsn))]
async fn handle_get_nblocks_request(
&self,
timeline: &Timeline,
@@ -687,6 +667,7 @@ impl PageServerHandler {
}))
}
#[instrument(skip(self, timeline, req, ctx), fields(dbnode = %req.dbnode, req_lsn = %req.lsn))]
async fn handle_db_size_request(
&self,
timeline: &Timeline,
@@ -708,6 +689,7 @@ impl PageServerHandler {
}))
}
#[instrument(skip(self, timeline, req, ctx), fields(rel = %req.rel, blkno = %req.blkno, req_lsn = %req.lsn))]
async fn handle_get_page_at_lsn_request(
&self,
timeline: &Timeline,
@@ -1283,10 +1265,7 @@ async fn get_active_tenant_with_timeout(
Ok(tenant) => tenant,
Err(e @ GetTenantError::NotFound(_)) => return Err(GetActiveTenantError::NotFound(e)),
Err(GetTenantError::NotActive(_)) => {
unreachable!("we're calling get_tenant with active_only=false")
}
Err(GetTenantError::Broken(_)) => {
unreachable!("we're calling get_tenant with active_only=false")
unreachable!("we're calling get_tenant with active=false")
}
};
let wait_time = Duration::from_secs(30);

View File

@@ -1,6 +1,6 @@
//! Wrapper around nix::sys::statvfs::Statvfs that allows for mocking.
use camino::Utf8Path;
use std::path::Path;
pub enum Statvfs {
Real(nix::sys::statvfs::Statvfs),
@@ -12,13 +12,11 @@ pub enum Statvfs {
// Sincce it should only be a problem on > 2TiB disks, let's ignore
// the problem for now and upcast to u64.
impl Statvfs {
pub fn get(tenants_dir: &Utf8Path, mocked: Option<&mock::Behavior>) -> nix::Result<Self> {
pub fn get(tenants_dir: &Path, mocked: Option<&mock::Behavior>) -> nix::Result<Self> {
if let Some(mocked) = mocked {
Ok(Statvfs::Mock(mock::get(tenants_dir, mocked)?))
} else {
Ok(Statvfs::Real(nix::sys::statvfs::statvfs(
tenants_dir.as_std_path(),
)?))
Ok(Statvfs::Real(nix::sys::statvfs::statvfs(tenants_dir)?))
}
}
@@ -57,8 +55,8 @@ impl Statvfs {
pub mod mock {
use anyhow::Context;
use camino::Utf8Path;
use regex::Regex;
use std::path::Path;
use tracing::log::info;
#[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
@@ -88,7 +86,7 @@ pub mod mock {
}
}
pub fn get(tenants_dir: &Utf8Path, behavior: &Behavior) -> nix::Result<Statvfs> {
pub fn get(tenants_dir: &Path, behavior: &Behavior) -> nix::Result<Statvfs> {
info!("running mocked statvfs");
match behavior {
@@ -121,7 +119,7 @@ pub mod mock {
}
}
fn walk_dir_disk_usage(path: &Utf8Path, name_filter: Option<&Regex>) -> anyhow::Result<u64> {
fn walk_dir_disk_usage(path: &Path, name_filter: Option<&Regex>) -> anyhow::Result<u64> {
let mut total = 0;
for entry in walkdir::WalkDir::new(path) {
let entry = entry?;

View File

@@ -12,7 +12,6 @@
//!
use anyhow::{bail, Context};
use camino::{Utf8Path, Utf8PathBuf};
use futures::FutureExt;
use pageserver_api::models::TimelineState;
use remote_storage::DownloadError;
@@ -35,6 +34,8 @@ use std::fs;
use std::fs::File;
use std::io;
use std::ops::Bound::Included;
use std::path::Path;
use std::path::PathBuf;
use std::process::Command;
use std::process::Stdio;
use std::sync::atomic::AtomicU64;
@@ -44,8 +45,6 @@ use std::sync::MutexGuard;
use std::sync::{Mutex, RwLock};
use std::time::{Duration, Instant};
use self::config::AttachedLocationConfig;
use self::config::LocationConf;
use self::config::TenantConf;
use self::delete::DeleteTenantFlow;
use self::metadata::LoadMetadataError;
@@ -66,7 +65,6 @@ use crate::metrics::{remove_tenant_metrics, TENANT_STATE_METRIC, TENANT_SYNTHETI
use crate::repository::GcResult;
use crate::task_mgr;
use crate::task_mgr::TaskKind;
use crate::tenant::config::LocationMode;
use crate::tenant::config::TenantConfOpt;
use crate::tenant::metadata::load_metadata;
pub use crate::tenant::remote_timeline_client::index::IndexPart;
@@ -163,28 +161,6 @@ pub struct TenantSharedResources {
pub deletion_queue_client: DeletionQueueClient,
}
/// A [`Tenant`] is really an _attached_ tenant. The configuration
/// for an attached tenant is a subset of the [`LocationConf`], represented
/// in this struct.
pub(super) struct AttachedTenantConf {
tenant_conf: TenantConfOpt,
location: AttachedLocationConfig,
}
impl AttachedTenantConf {
fn try_from(location_conf: LocationConf) -> anyhow::Result<Self> {
match &location_conf.mode {
LocationMode::Attached(attach_conf) => Ok(Self {
tenant_conf: location_conf.tenant_conf,
location: attach_conf.clone(),
}),
LocationMode::Secondary(_) => {
anyhow::bail!("Attempted to construct AttachedTenantConf from a LocationConf in secondary mode")
}
}
}
}
///
/// Tenant consists of multiple timelines. Keep them in a hash table.
///
@@ -202,15 +178,12 @@ pub struct Tenant {
// We keep TenantConfOpt sturct here to preserve the information
// about parameters that are not set.
// This is necessary to allow global config updates.
tenant_conf: Arc<RwLock<AttachedTenantConf>>,
tenant_conf: Arc<RwLock<TenantConfOpt>>,
tenant_id: TenantId,
/// The remote storage generation, used to protect S3 objects from split-brain.
/// Does not change over the lifetime of the [`Tenant`] object.
///
/// This duplicates the generation stored in LocationConf, but that structure is mutable:
/// this copy enforces the invariant that generatio doesn't change during a Tenant's lifetime.
generation: Generation,
timelines: Mutex<HashMap<TimelineId, Arc<Timeline>>>,
@@ -554,13 +527,14 @@ impl Tenant {
pub(crate) fn spawn_attach(
conf: &'static PageServerConf,
tenant_id: TenantId,
generation: Generation,
resources: TenantSharedResources,
attached_conf: AttachedTenantConf,
tenants: &'static tokio::sync::RwLock<TenantsMap>,
ctx: &RequestContext,
) -> anyhow::Result<Arc<Tenant>> {
// TODO dedup with spawn_load
let wal_redo_manager = Arc::new(PostgresRedoManager::new(conf, tenant_id));
let tenant_conf =
Self::load_tenant_config(conf, &tenant_id).context("load tenant config")?;
let TenantSharedResources {
broker_client,
@@ -568,12 +542,14 @@ impl Tenant {
deletion_queue_client,
} = resources;
let wal_redo_manager = Arc::new(PostgresRedoManager::new(conf, tenant_id));
let tenant = Arc::new(Tenant::new(
TenantState::Attaching,
conf,
attached_conf,
tenant_conf,
wal_redo_manager,
tenant_id,
generation,
remote_storage.clone(),
deletion_queue_client,
));
@@ -796,7 +772,7 @@ impl Tenant {
}
std::fs::remove_file(&marker_file)
.with_context(|| format!("unlink attach marker file {marker_file}"))?;
.with_context(|| format!("unlink attach marker file {}", marker_file.display()))?;
crashsafe::fsync(marker_file.parent().expect("marker file has parent dir"))
.context("fsync tenant directory after unlinking attach marker file")?;
@@ -884,9 +860,10 @@ impl Tenant {
backtrace: String::new(),
},
conf,
AttachedTenantConf::try_from(LocationConf::default()).unwrap(),
TenantConfOpt::default(),
wal_redo_manager,
tenant_id,
Generation::broken(),
None,
DeletionQueueClient::broken(),
))
@@ -905,7 +882,7 @@ impl Tenant {
pub(crate) fn spawn_load(
conf: &'static PageServerConf,
tenant_id: TenantId,
attached_conf: AttachedTenantConf,
generation: Generation,
resources: TenantSharedResources,
init_order: Option<InitializationOrder>,
tenants: &'static tokio::sync::RwLock<TenantsMap>,
@@ -913,6 +890,14 @@ impl Tenant {
) -> Arc<Tenant> {
span::debug_assert_current_span_has_tenant_id();
let tenant_conf = match Self::load_tenant_config(conf, &tenant_id) {
Ok(conf) => conf,
Err(e) => {
error!("load tenant config failed: {:?}", e);
return Tenant::create_broken_tenant(conf, tenant_id, format!("{e:#}"));
}
};
let broker_client = resources.broker_client;
let remote_storage = resources.remote_storage;
@@ -920,9 +905,10 @@ impl Tenant {
let tenant = Tenant::new(
TenantState::Loading,
conf,
attached_conf,
tenant_conf,
wal_redo_manager,
tenant_id,
generation,
remote_storage.clone(),
resources.deletion_queue_client.clone(),
);
@@ -1038,47 +1024,58 @@ impl Tenant {
let timelines_dir = self.conf.timelines_path(&self.tenant_id);
for entry in timelines_dir
.read_dir_utf8()
.context("list timelines directory for tenant")?
for entry in
std::fs::read_dir(&timelines_dir).context("list timelines directory for tenant")?
{
let entry = entry.context("read timeline dir entry")?;
let timeline_dir = entry.path();
if crate::is_temporary(timeline_dir) {
info!("Found temporary timeline directory, removing: {timeline_dir}");
if let Err(e) = std::fs::remove_dir_all(timeline_dir) {
error!("Failed to remove temporary directory '{timeline_dir}': {e:?}");
if crate::is_temporary(&timeline_dir) {
info!(
"Found temporary timeline directory, removing: {}",
timeline_dir.display()
);
if let Err(e) = std::fs::remove_dir_all(&timeline_dir) {
error!(
"Failed to remove temporary directory '{}': {:?}",
timeline_dir.display(),
e
);
}
} else if is_uninit_mark(timeline_dir) {
} else if is_uninit_mark(&timeline_dir) {
if !timeline_dir.exists() {
warn!("Timeline dir entry become invalid: {timeline_dir}");
warn!(
"Timeline dir entry become invalid: {}",
timeline_dir.display()
);
continue;
}
let timeline_uninit_mark_file = &timeline_dir;
info!(
"Found an uninit mark file {timeline_uninit_mark_file}, removing the timeline and its uninit mark",
"Found an uninit mark file {}, removing the timeline and its uninit mark",
timeline_uninit_mark_file.display()
);
let timeline_id =
TimelineId::try_from(timeline_uninit_mark_file.file_stem())
.with_context(|| {
format!(
"Could not parse timeline id out of the timeline uninit mark name {timeline_uninit_mark_file}",
let timeline_id = TimelineId::try_from(timeline_uninit_mark_file.file_stem())
.with_context(|| {
format!(
"Could not parse timeline id out of the timeline uninit mark name {}",
timeline_uninit_mark_file.display()
)
})?;
})?;
let timeline_dir = self.conf.timeline_path(&self.tenant_id, &timeline_id);
if let Err(e) =
remove_timeline_and_uninit_mark(&timeline_dir, timeline_uninit_mark_file)
{
error!("Failed to clean up uninit marked timeline: {e:?}");
}
} else if crate::is_delete_mark(timeline_dir) {
} else if crate::is_delete_mark(&timeline_dir) {
// If metadata exists, load as usual, continue deletion
let timeline_id = TimelineId::try_from(timeline_dir.file_stem())
.with_context(|| {
let timeline_id =
TimelineId::try_from(timeline_dir.file_stem()).with_context(|| {
format!(
"Could not parse timeline id out of the timeline uninit mark name {timeline_dir}",
"Could not parse timeline id out of the timeline uninit mark name {}",
timeline_dir.display()
)
})?;
@@ -1117,13 +1114,17 @@ impl Tenant {
}
} else {
if !timeline_dir.exists() {
warn!("Timeline dir entry become invalid: {timeline_dir}");
warn!(
"Timeline dir entry become invalid: {}",
timeline_dir.display()
);
continue;
}
let timeline_id = TimelineId::try_from(timeline_dir.file_name())
.with_context(|| {
let timeline_id =
TimelineId::try_from(timeline_dir.file_name()).with_context(|| {
format!(
"Could not parse timeline id out of the timeline dir name {timeline_dir}",
"Could not parse timeline id out of the timeline dir name {}",
timeline_dir.display()
)
})?;
let timeline_uninit_mark_file = self
@@ -1135,7 +1136,7 @@ impl Tenant {
"Found an uninit mark file, removing the timeline and its uninit mark",
);
if let Err(e) =
remove_timeline_and_uninit_mark(timeline_dir, &timeline_uninit_mark_file)
remove_timeline_and_uninit_mark(&timeline_dir, &timeline_uninit_mark_file)
{
error!("Failed to clean up uninit marked timeline: {e:?}");
}
@@ -1151,13 +1152,18 @@ impl Tenant {
}
let file_name = entry.file_name();
if let Ok(timeline_id) = file_name.parse::<TimelineId>() {
if let Ok(timeline_id) =
file_name.to_str().unwrap_or_default().parse::<TimelineId>()
{
let metadata = load_metadata(self.conf, &self.tenant_id, &timeline_id)
.context("failed to load metadata")?;
timelines_to_load.insert(timeline_id, metadata);
} else {
// A file or directory that doesn't look like a timeline ID
warn!("unexpected file or directory in timelines directory: {file_name}");
warn!(
"unexpected file or directory in timelines directory: {}",
file_name.to_string_lossy()
);
}
}
}
@@ -1661,15 +1667,6 @@ impl Tenant {
"Cannot run GC iteration on inactive tenant"
);
{
let conf = self.tenant_conf.read().unwrap();
if !conf.location.may_delete_layers_hint() {
info!("Skipping GC in location state {:?}", conf.location);
return Ok(GcResult::default());
}
}
self.gc_iteration_internal(target_timeline_id, horizon, pitr, ctx)
.await
}
@@ -1688,14 +1685,6 @@ impl Tenant {
"Cannot run compaction iteration on inactive tenant"
);
{
let conf = self.tenant_conf.read().unwrap();
if !conf.location.may_delete_layers_hint() || !conf.location.may_upload_layers_hint() {
info!("Skipping compaction in location state {:?}", conf.location);
return Ok(());
}
}
// Scan through the hashmap and collect a list of all the timelines,
// while holding the lock. Then drop the lock and actually perform the
// compactions. We don't want to block everything else while the
@@ -2121,7 +2110,7 @@ where
impl Tenant {
pub fn tenant_specific_overrides(&self) -> TenantConfOpt {
self.tenant_conf.read().unwrap().tenant_conf
*self.tenant_conf.read().unwrap()
}
pub fn effective_config(&self) -> TenantConf {
@@ -2130,95 +2119,84 @@ impl Tenant {
}
pub fn get_checkpoint_distance(&self) -> u64 {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.checkpoint_distance
.unwrap_or(self.conf.default_tenant_conf.checkpoint_distance)
}
pub fn get_checkpoint_timeout(&self) -> Duration {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.checkpoint_timeout
.unwrap_or(self.conf.default_tenant_conf.checkpoint_timeout)
}
pub fn get_compaction_target_size(&self) -> u64 {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.compaction_target_size
.unwrap_or(self.conf.default_tenant_conf.compaction_target_size)
}
pub fn get_compaction_period(&self) -> Duration {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.compaction_period
.unwrap_or(self.conf.default_tenant_conf.compaction_period)
}
pub fn get_compaction_threshold(&self) -> usize {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.compaction_threshold
.unwrap_or(self.conf.default_tenant_conf.compaction_threshold)
}
pub fn get_gc_horizon(&self) -> u64 {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.gc_horizon
.unwrap_or(self.conf.default_tenant_conf.gc_horizon)
}
pub fn get_gc_period(&self) -> Duration {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.gc_period
.unwrap_or(self.conf.default_tenant_conf.gc_period)
}
pub fn get_image_creation_threshold(&self) -> usize {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.image_creation_threshold
.unwrap_or(self.conf.default_tenant_conf.image_creation_threshold)
}
pub fn get_pitr_interval(&self) -> Duration {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.pitr_interval
.unwrap_or(self.conf.default_tenant_conf.pitr_interval)
}
pub fn get_trace_read_requests(&self) -> bool {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.trace_read_requests
.unwrap_or(self.conf.default_tenant_conf.trace_read_requests)
}
pub fn get_min_resident_size_override(&self) -> Option<u64> {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.min_resident_size_override
.or(self.conf.default_tenant_conf.min_resident_size_override)
}
pub fn set_new_tenant_config(&self, new_tenant_conf: TenantConfOpt) {
self.tenant_conf.write().unwrap().tenant_conf = new_tenant_conf;
// Don't hold self.timelines.lock() during the notifies.
// There's no risk of deadlock right now, but there could be if we consolidate
// mutexes in struct Timeline in the future.
let timelines = self.list_timelines();
for timeline in timelines {
timeline.tenant_conf_updated();
}
}
pub(crate) fn set_new_location_config(&self, new_conf: AttachedTenantConf) {
*self.tenant_conf.write().unwrap() = new_conf;
*self.tenant_conf.write().unwrap() = new_tenant_conf;
// Don't hold self.timelines.lock() during the notifies.
// There's no risk of deadlock right now, but there could be if we consolidate
// mutexes in struct Timeline in the future.
@@ -2288,9 +2266,10 @@ impl Tenant {
fn new(
state: TenantState,
conf: &'static PageServerConf,
attached_conf: AttachedTenantConf,
tenant_conf: TenantConfOpt,
walredo_mgr: Arc<dyn WalRedoManager + Send + Sync>,
tenant_id: TenantId,
generation: Generation,
remote_storage: Option<GenericRemoteStorage>,
deletion_queue_client: DeletionQueueClient,
) -> Tenant {
@@ -2350,12 +2329,12 @@ impl Tenant {
Tenant {
tenant_id,
generation: attached_conf.location.generation,
generation,
conf,
// using now here is good enough approximation to catch tenants with really long
// activation times.
loading_started_at: Instant::now(),
tenant_conf: Arc::new(RwLock::new(attached_conf)),
tenant_conf: Arc::new(RwLock::new(tenant_conf)),
timelines: Mutex::new(HashMap::new()),
gc_cs: tokio::sync::Mutex::new(()),
walredo_mgr,
@@ -2373,124 +2352,54 @@ impl Tenant {
pub(super) fn load_tenant_config(
conf: &'static PageServerConf,
tenant_id: &TenantId,
) -> anyhow::Result<LocationConf> {
let legacy_config_path = conf.tenant_config_path(tenant_id);
let config_path = conf.tenant_location_config_path(tenant_id);
) -> anyhow::Result<TenantConfOpt> {
let target_config_path = conf.tenant_config_path(tenant_id);
let target_config_display = target_config_path.display();
if config_path.exists() {
// New-style config takes precedence
let deserialized = Self::read_config(&config_path)?;
Ok(toml_edit::de::from_document::<LocationConf>(deserialized)?)
} else if legacy_config_path.exists() {
// Upgrade path: found an old-style configuration only
let deserialized = Self::read_config(&legacy_config_path)?;
info!("loading tenantconf from {target_config_display}");
let mut tenant_conf = TenantConfOpt::default();
for (key, item) in deserialized.iter() {
match key {
"tenant_config" => {
tenant_conf = PageServerConf::parse_toml_tenant_conf(item).with_context(|| {
format!("Failed to parse config from file '{legacy_config_path}' as pageserver config")
})?;
}
_ => bail!(
"config file {legacy_config_path} has unrecognized pageserver option '{key}'"
),
}
}
// Legacy configs are implicitly in attached state
Ok(LocationConf::attached_single(
tenant_conf,
Generation::none(),
))
} else {
// FIXME If the config file is not found, assume that we're attaching
// a detached tenant and config is passed via attach command.
// https://github.com/neondatabase/neon/issues/1555
// OR: we're loading after incomplete deletion that managed to remove config.
info!(
"tenant config not found in {} or {}",
config_path, legacy_config_path
);
Ok(LocationConf::default())
// FIXME If the config file is not found, assume that we're attaching
// a detached tenant and config is passed via attach command.
// https://github.com/neondatabase/neon/issues/1555
// OR: we're loading after incomplete deletion that managed to remove config.
if !target_config_path.exists() {
info!("tenant config not found in {target_config_display}");
return Ok(TenantConfOpt::default());
}
}
fn read_config(path: &Utf8Path) -> anyhow::Result<toml_edit::Document> {
info!("loading tenant configuration from {path}");
// load and parse file
let config = fs::read_to_string(path)
.with_context(|| format!("Failed to load config from path '{path}'"))?;
let config = fs::read_to_string(&target_config_path).with_context(|| {
format!("Failed to load config from path '{target_config_display}'")
})?;
config
.parse::<toml_edit::Document>()
.with_context(|| format!("Failed to parse config from file '{path}' as toml file"))
let toml = config.parse::<toml_edit::Document>().with_context(|| {
format!("Failed to parse config from file '{target_config_display}' as toml file")
})?;
let mut tenant_conf = TenantConfOpt::default();
for (key, item) in toml.iter() {
match key {
"tenant_config" => {
tenant_conf = PageServerConf::parse_toml_tenant_conf(item).with_context(|| {
format!("Failed to parse config from file '{target_config_display}' as pageserver config")
})?;
}
_ => bail!("config file {target_config_display} has unrecognized pageserver option '{key}'"),
}
}
Ok(tenant_conf)
}
#[tracing::instrument(skip_all, fields(%tenant_id))]
pub(super) async fn persist_tenant_config(
conf: &'static PageServerConf,
tenant_id: &TenantId,
location_conf: &LocationConf,
target_config_path: &Path,
tenant_conf: TenantConfOpt,
) -> anyhow::Result<()> {
let legacy_config_path = conf.tenant_config_path(tenant_id);
let config_path = conf.tenant_location_config_path(tenant_id);
Self::persist_tenant_config_at(tenant_id, &config_path, &legacy_config_path, location_conf)
.await
}
#[tracing::instrument(skip_all, fields(%tenant_id))]
pub(super) async fn persist_tenant_config_at(
tenant_id: &TenantId,
config_path: &Utf8Path,
legacy_config_path: &Utf8Path,
location_conf: &LocationConf,
) -> anyhow::Result<()> {
// Forward compat: write out an old-style configuration that old versions can read, in case we roll back
Self::persist_tenant_config_legacy(
tenant_id,
legacy_config_path,
&location_conf.tenant_conf,
)
.await?;
if let LocationMode::Attached(attach_conf) = &location_conf.mode {
// Once we use LocationMode, generations are mandatory. If we aren't using generations,
// then drop out after writing legacy-style config.
if attach_conf.generation.is_none() {
tracing::debug!("Running without generations, not writing new-style LocationConf");
return Ok(());
}
}
info!("persisting tenantconf to {config_path}");
let mut conf_content = r#"# This file contains a specific per-tenant's config.
# It is read in case of pageserver restart.
"#
.to_string();
// Convert the config to a toml file.
conf_content += &toml_edit::ser::to_string_pretty(&location_conf)?;
let conf_content = conf_content.as_bytes();
let temp_path = path_with_suffix_extension(config_path, TEMP_FILE_SUFFIX);
VirtualFile::crashsafe_overwrite(config_path, &temp_path, conf_content)
.await
.with_context(|| format!("write tenant {tenant_id} config to {config_path}"))?;
Ok(())
}
#[tracing::instrument(skip_all, fields(%tenant_id))]
async fn persist_tenant_config_legacy(
tenant_id: &TenantId,
target_config_path: &Utf8Path,
tenant_conf: &TenantConfOpt,
) -> anyhow::Result<()> {
info!("persisting tenantconf to {target_config_path}");
// imitate a try-block with a closure
info!("persisting tenantconf to {}", target_config_path.display());
let mut conf_content = r#"# This file contains a specific per-tenant's config.
# It is read in case of pageserver restart.
@@ -2507,7 +2416,12 @@ impl Tenant {
let temp_path = path_with_suffix_extension(target_config_path, TEMP_FILE_SUFFIX);
VirtualFile::crashsafe_overwrite(target_config_path, &temp_path, conf_content)
.await
.with_context(|| format!("write tenant {tenant_id} config to {target_config_path}"))?;
.with_context(|| {
format!(
"write tenant {tenant_id} config to {}",
target_config_path.display()
)
})?;
Ok(())
}
@@ -2874,7 +2788,10 @@ impl Tenant {
// current initdb was not run yet, so remove whatever was left from the previous runs
if initdb_path.exists() {
fs::remove_dir_all(&initdb_path).with_context(|| {
format!("Failed to remove already existing initdb directory: {initdb_path}")
format!(
"Failed to remove already existing initdb directory: {}",
initdb_path.display()
)
})?;
}
// Init temporarily repo to get bootstrap data, this creates a directory in the `initdb_path` path
@@ -2883,7 +2800,7 @@ impl Tenant {
scopeguard::defer! {
if let Err(e) = fs::remove_dir_all(&initdb_path) {
// this is unlikely, but we will remove the directory on pageserver restart or another bootstrap call
error!("Failed to remove temporary initdb directory '{initdb_path}': {e}");
error!("Failed to remove temporary initdb directory '{}': {}", initdb_path.display(), e);
}
}
let pgdata_path = &initdb_path;
@@ -3033,7 +2950,7 @@ impl Tenant {
async fn create_timeline_files(
&self,
timeline_path: &Utf8Path,
timeline_path: &Path,
new_timeline_id: &TimelineId,
new_metadata: &TimelineMetadata,
) -> anyhow::Result<()> {
@@ -3067,7 +2984,8 @@ impl Tenant {
let timeline_path = self.conf.timeline_path(&tenant_id, &timeline_id);
anyhow::ensure!(
!timeline_path.exists(),
"Timeline {timeline_path} already exists, cannot create its uninit mark file",
"Timeline {} already exists, cannot create its uninit mark file",
timeline_path.display()
);
let uninit_mark_path = self
@@ -3159,10 +3077,7 @@ impl Tenant {
}
}
fn remove_timeline_and_uninit_mark(
timeline_dir: &Utf8Path,
uninit_mark: &Utf8Path,
) -> anyhow::Result<()> {
fn remove_timeline_and_uninit_mark(timeline_dir: &Path, uninit_mark: &Path) -> anyhow::Result<()> {
fs::remove_dir_all(timeline_dir)
.or_else(|e| {
if e.kind() == std::io::ErrorKind::NotFound {
@@ -3174,10 +3089,17 @@ fn remove_timeline_and_uninit_mark(
}
})
.with_context(|| {
format!("Failed to remove unit marked timeline directory {timeline_dir}")
format!(
"Failed to remove unit marked timeline directory {}",
timeline_dir.display()
)
})?;
fs::remove_file(uninit_mark)
.with_context(|| format!("Failed to remove timeline uninit mark file {uninit_mark}"))?;
fs::remove_file(uninit_mark).with_context(|| {
format!(
"Failed to remove timeline uninit mark file {}",
uninit_mark.display()
)
})?;
Ok(())
}
@@ -3189,10 +3111,10 @@ pub(crate) enum CreateTenantFilesMode {
pub(crate) async fn create_tenant_files(
conf: &'static PageServerConf,
location_conf: &LocationConf,
tenant_conf: TenantConfOpt,
tenant_id: &TenantId,
mode: CreateTenantFilesMode,
) -> anyhow::Result<Utf8PathBuf> {
) -> anyhow::Result<PathBuf> {
let target_tenant_directory = conf.tenant_path(tenant_id);
anyhow::ensure!(
!target_tenant_directory
@@ -3203,16 +3125,22 @@ pub(crate) async fn create_tenant_files(
let temporary_tenant_dir =
path_with_suffix_extension(&target_tenant_directory, TEMP_FILE_SUFFIX);
debug!("Creating temporary directory structure in {temporary_tenant_dir}");
debug!(
"Creating temporary directory structure in {}",
temporary_tenant_dir.display()
);
// top-level dir may exist if we are creating it through CLI
crashsafe::create_dir_all(&temporary_tenant_dir).with_context(|| {
format!("could not create temporary tenant directory {temporary_tenant_dir}")
format!(
"could not create temporary tenant directory {}",
temporary_tenant_dir.display()
)
})?;
let creation_result = try_create_target_tenant_dir(
conf,
location_conf,
tenant_conf,
tenant_id,
mode,
&temporary_tenant_dir,
@@ -3238,11 +3166,11 @@ pub(crate) async fn create_tenant_files(
async fn try_create_target_tenant_dir(
conf: &'static PageServerConf,
location_conf: &LocationConf,
tenant_conf: TenantConfOpt,
tenant_id: &TenantId,
mode: CreateTenantFilesMode,
temporary_tenant_dir: &Utf8Path,
target_tenant_directory: &Utf8Path,
temporary_tenant_dir: &Path,
target_tenant_directory: &Path,
) -> Result<(), anyhow::Error> {
match mode {
CreateTenantFilesMode::Create => {} // needs no attach marker, writing tenant conf + atomic rename of dir is good enough
@@ -3268,31 +3196,20 @@ async fn try_create_target_tenant_dir(
temporary_tenant_dir,
)
.with_context(|| format!("resolve tenant {tenant_id} temporary timelines dir"))?;
let temporary_legacy_tenant_config_path = rebase_directory(
let temporary_tenant_config_path = rebase_directory(
&conf.tenant_config_path(tenant_id),
target_tenant_directory,
temporary_tenant_dir,
)
.with_context(|| format!("resolve tenant {tenant_id} temporary config path"))?;
let temporary_tenant_config_path = rebase_directory(
&conf.tenant_location_config_path(tenant_id),
target_tenant_directory,
temporary_tenant_dir,
)
.with_context(|| format!("resolve tenant {tenant_id} temporary config path"))?;
Tenant::persist_tenant_config_at(
tenant_id,
&temporary_tenant_config_path,
&temporary_legacy_tenant_config_path,
location_conf,
)
.await?;
Tenant::persist_tenant_config(tenant_id, &temporary_tenant_config_path, tenant_conf).await?;
crashsafe::create_dir(&temporary_tenant_timelines_dir).with_context(|| {
format!(
"create tenant {} temporary timelines directory {}",
tenant_id, temporary_tenant_timelines_dir,
tenant_id,
temporary_tenant_timelines_dir.display()
)
})?;
fail::fail_point!("tenant-creation-before-tmp-rename", |_| {
@@ -3307,34 +3224,35 @@ async fn try_create_target_tenant_dir(
fs::rename(temporary_tenant_dir, target_tenant_directory).with_context(|| {
format!(
"move tenant {} temporary directory {} into the permanent one {}",
tenant_id, temporary_tenant_dir, target_tenant_directory
tenant_id,
temporary_tenant_dir.display(),
target_tenant_directory.display()
)
})?;
let target_dir_parent = target_tenant_directory.parent().with_context(|| {
format!(
"get tenant {} dir parent for {}",
tenant_id, target_tenant_directory,
tenant_id,
target_tenant_directory.display()
)
})?;
crashsafe::fsync(target_dir_parent).with_context(|| {
format!(
"fsync renamed directory's parent {} for tenant {}",
target_dir_parent, tenant_id,
target_dir_parent.display(),
tenant_id,
)
})?;
Ok(())
}
fn rebase_directory(
original_path: &Utf8Path,
base: &Utf8Path,
new_base: &Utf8Path,
) -> anyhow::Result<Utf8PathBuf> {
fn rebase_directory(original_path: &Path, base: &Path, new_base: &Path) -> anyhow::Result<PathBuf> {
let relative_path = original_path.strip_prefix(base).with_context(|| {
format!(
"Failed to strip base prefix '{}' off path '{}'",
base, original_path
base.display(),
original_path.display()
)
})?;
Ok(new_base.join(relative_path))
@@ -3344,18 +3262,20 @@ fn rebase_directory(
/// to get bootstrap data for timeline initialization.
fn run_initdb(
conf: &'static PageServerConf,
initdb_target_dir: &Utf8Path,
initdb_target_dir: &Path,
pg_version: u32,
) -> anyhow::Result<()> {
let initdb_bin_path = conf.pg_bin_dir(pg_version)?.join("initdb");
let initdb_lib_dir = conf.pg_lib_dir(pg_version)?;
info!(
"running {} in {}, libdir: {}",
initdb_bin_path, initdb_target_dir, initdb_lib_dir,
initdb_bin_path.display(),
initdb_target_dir.display(),
initdb_lib_dir.display(),
);
let initdb_output = Command::new(&initdb_bin_path)
.args(["-D", initdb_target_dir.as_ref()])
.args(["-D", &initdb_target_dir.to_string_lossy()])
.args(["-U", &conf.superuser])
.args(["-E", "utf8"])
.arg("--no-instructions")
@@ -3370,7 +3290,8 @@ fn run_initdb(
.with_context(|| {
format!(
"failed to execute {} at target dir {}",
initdb_bin_path, initdb_target_dir,
initdb_bin_path.display(),
initdb_target_dir.display()
)
})?;
if !initdb_output.status.success() {
@@ -3390,7 +3311,7 @@ impl Drop for Tenant {
}
/// Dump contents of a layer file to stdout.
pub async fn dump_layerfile_from_path(
path: &Utf8Path,
path: &Path,
verbose: bool,
ctx: &RequestContext,
) -> anyhow::Result<()> {
@@ -3423,8 +3344,8 @@ pub async fn dump_layerfile_from_path(
pub mod harness {
use bytes::{Bytes, BytesMut};
use once_cell::sync::OnceCell;
use std::fs;
use std::sync::Arc;
use std::{fs, path::PathBuf};
use utils::logging;
use utils::lsn::Lsn;
@@ -3489,7 +3410,7 @@ pub mod harness {
pub tenant_id: TenantId,
pub generation: Generation,
pub remote_storage: GenericRemoteStorage,
pub remote_fs_dir: Utf8PathBuf,
pub remote_fs_dir: PathBuf,
pub deletion_queue: MockDeletionQueue,
}
@@ -3568,13 +3489,10 @@ pub mod harness {
let tenant = Arc::new(Tenant::new(
TenantState::Loading,
self.conf,
AttachedTenantConf::try_from(LocationConf::attached_single(
TenantConfOpt::from(self.tenant_conf),
self.generation,
))
.unwrap(),
TenantConfOpt::from(self.tenant_conf),
walredo_mgr,
self.tenant_id,
self.generation,
Some(self.remote_storage.clone()),
self.deletion_queue.new_client(),
));
@@ -3591,7 +3509,7 @@ pub mod harness {
Ok(tenant)
}
pub fn timeline_path(&self, timeline_id: &TimelineId) -> Utf8PathBuf {
pub fn timeline_path(&self, timeline_id: &TimelineId) -> PathBuf {
self.conf.timeline_path(&self.tenant_id, timeline_id)
}
}

View File

@@ -20,10 +20,10 @@ use std::io::{Error, ErrorKind};
impl<'a> BlockCursor<'a> {
/// Read a blob into a new buffer.
pub async fn read_blob<'c>(
pub async fn read_blob(
&self,
offset: u64,
ctx: &'c RequestContext,
ctx: &RequestContext,
) -> Result<Vec<u8>, std::io::Error> {
let mut buf = Vec::new();
self.read_blob_into_buf(offset, &mut buf, ctx).await?;
@@ -31,11 +31,11 @@ impl<'a> BlockCursor<'a> {
}
/// Read blob into the given buffer. Any previous contents in the buffer
/// are overwritten.
pub async fn read_blob_into_buf<'c>(
pub async fn read_blob_into_buf(
&self,
offset: u64,
dstbuf: &mut Vec<u8>,
ctx: &'c RequestContext,
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;
@@ -238,14 +238,14 @@ mod tests {
use rand::{Rng, SeedableRng};
async fn round_trip_test<const BUFFERED: bool>(blobs: &[Vec<u8>]) -> Result<(), Error> {
let temp_dir = camino_tempfile::tempdir()?;
let pathbuf = temp_dir.path().join("file");
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();
{
let file = VirtualFile::create(pathbuf.as_path()).await?;
let file = VirtualFile::create(&path).await?;
let mut wtr = BlobWriter::<BUFFERED>::new(file, 0);
for blob in blobs.iter() {
let offs = wtr.write_blob(blob).await?;
@@ -258,7 +258,7 @@ mod tests {
wtr.flush_buffer().await?;
}
let file = VirtualFile::open(pathbuf.as_path()).await?;
let file = VirtualFile::open(&path).await?;
let rdr = BlockReaderRef::VirtualFile(&file);
let rdr = BlockCursor::new(rdr);
for (idx, (blob, offset)) in blobs.iter().zip(offsets.iter()).enumerate() {

View File

@@ -34,27 +34,27 @@ where
}
/// Reference to an in-memory copy of an immutable on-disk block.
pub enum BlockLease<'c, 'a> {
PageReadGuard(PageReadGuard<'c, 'static>),
pub enum BlockLease<'a> {
PageReadGuard(PageReadGuard<'static>),
EphemeralFileMutableTail(&'a [u8; PAGE_SZ]),
#[cfg(test)]
Arc(std::sync::Arc<[u8; PAGE_SZ]>),
}
impl<'c, 'a> From<PageReadGuard<'c, 'a>> for BlockLease<'c, 'a> {
fn from(value: PageReadGuard<'c, 'a>) -> BlockLease<'c, 'a> {
impl From<PageReadGuard<'static>> for BlockLease<'static> {
fn from(value: PageReadGuard<'static>) -> BlockLease<'static> {
BlockLease::PageReadGuard(value)
}
}
#[cfg(test)]
impl<'c, 'a> From<std::sync::Arc<[u8; PAGE_SZ]>> for BlockLease<'c, 'a> {
impl<'a> From<std::sync::Arc<[u8; PAGE_SZ]>> for BlockLease<'a> {
fn from(value: std::sync::Arc<[u8; PAGE_SZ]>) -> Self {
BlockLease::Arc(value)
}
}
impl<'c, 'a> Deref for BlockLease<'c, 'a> {
impl<'a> Deref for BlockLease<'a> {
type Target = [u8; PAGE_SZ];
fn deref(&self) -> &Self::Target {
@@ -83,11 +83,11 @@ pub(crate) enum BlockReaderRef<'a> {
impl<'a> BlockReaderRef<'a> {
#[inline(always)]
async fn read_blk<'c>(
async fn read_blk(
&self,
blknum: u32,
ctx: &'c RequestContext,
) -> Result<BlockLease<'c, '_>, std::io::Error> {
ctx: &RequestContext,
) -> Result<BlockLease, std::io::Error> {
use BlockReaderRef::*;
match self {
FileBlockReader(r) => r.read_blk(blknum, ctx).await,
@@ -141,11 +141,11 @@ 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<'c>(
pub async fn read_blk(
&self,
blknum: u32,
ctx: &'c RequestContext,
) -> Result<BlockLease<'c, '_>, std::io::Error> {
ctx: &RequestContext,
) -> Result<BlockLease, std::io::Error> {
self.reader.read_blk(blknum, ctx).await
}
}
@@ -180,27 +180,32 @@ 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<'c>(
pub async fn read_blk(
&self,
blknum: u32,
ctx: &'c RequestContext,
) -> Result<BlockLease<'c, 'static>, std::io::Error> {
ctx: &RequestContext,
) -> Result<BlockLease, std::io::Error> {
let cache = page_cache::get();
match cache
.read_immutable_buf(self.file_id, blknum, ctx)
.await
.map_err(|e| {
std::io::Error::new(
std::io::ErrorKind::Other,
format!("Failed to read immutable buf: {e:#}"),
)
})? {
ReadBufResult::Found(guard) => Ok(guard.into()),
ReadBufResult::NotFound(mut write_guard) => {
// Read the page from disk into the buffer
self.fill_buffer(write_guard.deref_mut(), blknum).await?;
Ok(write_guard.mark_valid().into())
}
loop {
match cache
.read_immutable_buf(self.file_id, blknum, ctx)
.await
.map_err(|e| {
std::io::Error::new(
std::io::ErrorKind::Other,
format!("Failed to read immutable buf: {e:#}"),
)
})? {
ReadBufResult::Found(guard) => break Ok(guard.into()),
ReadBufResult::NotFound(mut write_guard) => {
// Read the page from disk into the buffer
self.fill_buffer(write_guard.deref_mut(), blknum).await?;
write_guard.mark_valid();
// Swap for read lock
continue;
}
};
}
}
}

View File

@@ -13,7 +13,6 @@ use pageserver_api::models;
use serde::{Deserialize, Serialize};
use std::num::NonZeroU64;
use std::time::Duration;
use utils::generation::Generation;
pub mod defaults {
// FIXME: This current value is very low. I would imagine something like 1 GB or 10 GB
@@ -45,211 +44,7 @@ pub mod defaults {
pub const DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD: &str = "24 hour";
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub(crate) enum AttachmentMode {
/// Our generation is current as far as we know, and as far as we know we are the only attached
/// pageserver. This is the "normal" attachment mode.
Single,
/// Our generation number is current as far as we know, but we are advised that another
/// pageserver is still attached, and therefore to avoid executing deletions. This is
/// the attachment mode of a pagesever that is the destination of a migration.
Multi,
/// Our generation number is superseded, or about to be superseded. We are advised
/// to avoid remote storage writes if possible, and to avoid sending billing data. This
/// is the attachment mode of a pageserver that is the origin of a migration.
Stale,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub(crate) struct AttachedLocationConfig {
pub(crate) generation: Generation,
pub(crate) attach_mode: AttachmentMode,
// TODO: add a flag to override AttachmentMode's policies under
// disk pressure (i.e. unblock uploads under disk pressure in Stale
// state, unblock deletions after timeout in Multi state)
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub(crate) struct SecondaryLocationConfig {
/// If true, keep the local cache warm by polling remote storage
pub(crate) warm: bool,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub(crate) enum LocationMode {
Attached(AttachedLocationConfig),
Secondary(SecondaryLocationConfig),
}
/// Per-tenant, per-pageserver configuration. All pageservers use the same TenantConf,
/// but have distinct LocationConf.
#[derive(Clone, PartialEq, Eq, Serialize, Deserialize)]
pub(crate) struct LocationConf {
/// The location-specific part of the configuration, describes the operating
/// mode of this pageserver for this tenant.
pub(crate) mode: LocationMode,
/// The pan-cluster tenant configuration, the same on all locations
pub(crate) tenant_conf: TenantConfOpt,
}
impl std::fmt::Debug for LocationConf {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
match &self.mode {
LocationMode::Attached(conf) => {
write!(
f,
"Attached {:?}, gen={:?}",
conf.attach_mode, conf.generation
)
}
LocationMode::Secondary(conf) => {
write!(f, "Secondary, warm={}", conf.warm)
}
}
}
}
impl AttachedLocationConfig {
/// Consult attachment mode to determine whether we are currently permitted
/// to delete layers. This is only advisory, not required for data safety.
/// See [`AttachmentMode`] for more context.
pub(crate) fn may_delete_layers_hint(&self) -> bool {
// TODO: add an override for disk pressure in AttachedLocationConfig,
// and respect it here.
match &self.attach_mode {
AttachmentMode::Single => true,
AttachmentMode::Multi | AttachmentMode::Stale => {
// In Multi mode we avoid doing deletions because some other
// attached pageserver might get 404 while trying to read
// a layer we delete which is still referenced in their metadata.
//
// In Stale mode, we avoid doing deletions because we expect
// that they would ultimately fail validation in the deletion
// queue due to our stale generation.
false
}
}
}
/// Whether we are currently hinted that it is worthwhile to upload layers.
/// This is only advisory, not required for data safety.
/// See [`AttachmentMode`] for more context.
pub(crate) fn may_upload_layers_hint(&self) -> bool {
// TODO: add an override for disk pressure in AttachedLocationConfig,
// and respect it here.
match &self.attach_mode {
AttachmentMode::Single | AttachmentMode::Multi => true,
AttachmentMode::Stale => {
// In Stale mode, we avoid doing uploads because we expect that
// our replacement pageserver will already have started its own
// IndexPart that will never reference layers we upload: it is
// wasteful.
false
}
}
}
}
impl LocationConf {
/// For use when loading from a legacy configuration: presence of a tenant
/// implies it is in AttachmentMode::Single, which used to be the only
/// possible state. This function should eventually be removed.
pub(crate) fn attached_single(tenant_conf: TenantConfOpt, generation: Generation) -> Self {
Self {
mode: LocationMode::Attached(AttachedLocationConfig {
generation,
attach_mode: AttachmentMode::Single,
}),
tenant_conf,
}
}
/// For use when attaching/re-attaching: update the generation stored in this
/// structure. If we were in a secondary state, promote to attached (posession
/// of a fresh generation implies this).
pub(crate) fn attach_in_generation(&mut self, generation: Generation) {
match &mut self.mode {
LocationMode::Attached(attach_conf) => {
attach_conf.generation = generation;
}
LocationMode::Secondary(_) => {
// We are promoted to attached by the control plane's re-attach response
self.mode = LocationMode::Attached(AttachedLocationConfig {
generation,
attach_mode: AttachmentMode::Single,
})
}
}
}
pub(crate) fn try_from(conf: &'_ models::LocationConfig) -> anyhow::Result<Self> {
let tenant_conf = TenantConfOpt::try_from(&conf.tenant_conf)?;
fn get_generation(conf: &'_ models::LocationConfig) -> Result<Generation, anyhow::Error> {
conf.generation
.ok_or_else(|| anyhow::anyhow!("Generation must be set when attaching"))
}
let mode = match &conf.mode {
models::LocationConfigMode::AttachedMulti => {
LocationMode::Attached(AttachedLocationConfig {
generation: get_generation(conf)?,
attach_mode: AttachmentMode::Multi,
})
}
models::LocationConfigMode::AttachedSingle => {
LocationMode::Attached(AttachedLocationConfig {
generation: get_generation(conf)?,
attach_mode: AttachmentMode::Single,
})
}
models::LocationConfigMode::AttachedStale => {
LocationMode::Attached(AttachedLocationConfig {
generation: get_generation(conf)?,
attach_mode: AttachmentMode::Stale,
})
}
models::LocationConfigMode::Secondary => {
anyhow::ensure!(conf.generation.is_none());
let warm = conf
.secondary_conf
.as_ref()
.map(|c| c.warm)
.unwrap_or(false);
LocationMode::Secondary(SecondaryLocationConfig { warm })
}
models::LocationConfigMode::Detached => {
// Should not have been called: API code should translate this mode
// into a detach rather than trying to decode it as a LocationConf
return Err(anyhow::anyhow!("Cannot decode a Detached configuration"));
}
};
Ok(Self { mode, tenant_conf })
}
}
impl Default for LocationConf {
// TODO: this should be removed once tenant loading can guarantee that we are never
// loading from a directory without a configuration.
// => tech debt since https://github.com/neondatabase/neon/issues/1555
fn default() -> Self {
Self {
mode: LocationMode::Attached(AttachedLocationConfig {
generation: Generation::none(),
attach_mode: AttachmentMode::Single,
}),
tenant_conf: TenantConfOpt::default(),
}
}
}
/// A tenant's calcuated configuration, which is the result of merging a
/// tenant's TenantConfOpt with the global TenantConf from PageServerConf.
///
/// For storing and transmitting individual tenant's configuration, see
/// TenantConfOpt.
/// Per-tenant configuration options
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
pub struct TenantConf {
// Flush out an inmemory layer, if it's holding WAL older than this

View File

@@ -1,7 +1,9 @@
use std::sync::Arc;
use std::{
path::{Path, PathBuf},
sync::Arc,
};
use anyhow::Context;
use camino::{Utf8Path, Utf8PathBuf};
use pageserver_api::models::TenantState;
use remote_storage::{DownloadError, GenericRemoteStorage, RemotePath};
use tokio::sync::OwnedMutexGuard;
@@ -60,7 +62,7 @@ fn remote_tenant_delete_mark_path(
.context("Failed to strip workdir prefix")
.and_then(RemotePath::new)
.context("tenant path")?;
Ok(tenant_remote_path.join(Utf8Path::new("deleted")))
Ok(tenant_remote_path.join(Path::new("deleted")))
}
async fn create_remote_delete_mark(
@@ -146,7 +148,7 @@ async fn schedule_ordered_timeline_deletions(
Ok(already_running_deletions)
}
async fn ensure_timelines_dir_empty(timelines_path: &Utf8Path) -> Result<(), DeleteTenantError> {
async fn ensure_timelines_dir_empty(timelines_path: &Path) -> Result<(), DeleteTenantError> {
// Assert timelines dir is empty.
if !fs_ext::is_directory_empty(timelines_path).await? {
// Display first 10 items in directory
@@ -186,18 +188,20 @@ async fn cleanup_remaining_fs_traces(
conf: &PageServerConf,
tenant_id: &TenantId,
) -> Result<(), DeleteTenantError> {
let rm = |p: Utf8PathBuf, is_dir: bool| async move {
let rm = |p: PathBuf, is_dir: bool| async move {
if is_dir {
tokio::fs::remove_dir(&p).await
} else {
tokio::fs::remove_file(&p).await
}
.or_else(fs_ext::ignore_not_found)
.with_context(|| format!("failed to delete {p}"))
.with_context(|| {
let to_display = p.display();
format!("failed to delete {to_display}")
})
};
rm(conf.tenant_config_path(tenant_id), false).await?;
rm(conf.tenant_location_config_path(tenant_id), false).await?;
fail::fail_point!("tenant-delete-before-remove-timelines-dir", |_| {
Err(anyhow::anyhow!(

View File

@@ -6,11 +6,11 @@ use crate::context::RequestContext;
use crate::page_cache::{self, PAGE_SZ};
use crate::tenant::block_io::{BlockCursor, BlockLease, BlockReader};
use crate::virtual_file::VirtualFile;
use camino::Utf8PathBuf;
use std::cmp::min;
use std::fs::OpenOptions;
use std::io::{self, ErrorKind};
use std::ops::DerefMut;
use std::path::PathBuf;
use std::sync::atomic::AtomicU64;
use tracing::*;
use utils::id::{TenantId, TimelineId};
@@ -40,9 +40,7 @@ impl EphemeralFile {
let filename = conf
.timeline_path(&tenant_id, &timeline_id)
.join(Utf8PathBuf::from(format!(
"ephemeral-{filename_disambiguator}"
)));
.join(PathBuf::from(format!("ephemeral-{filename_disambiguator}")));
let file = VirtualFile::open_with_options(
&filename,
@@ -64,40 +62,46 @@ impl EphemeralFile {
self.len
}
pub(crate) async fn read_blk<'c>(
pub(crate) async fn read_blk(
&self,
blknum: u32,
ctx: &'c RequestContext,
) -> Result<BlockLease<'c, '_>, io::Error> {
ctx: &RequestContext,
) -> 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();
match cache
.read_immutable_buf(self.page_cache_file_id, blknum, ctx)
.await
.map_err(|e| {
std::io::Error::new(
std::io::ErrorKind::Other,
// order path before error because error is anyhow::Error => might have many contexts
format!(
"ephemeral file: read immutable page #{}: {}: {:#}",
blknum, self.file.path, e,
),
)
})? {
page_cache::ReadBufResult::Found(guard) => {
return Ok(BlockLease::PageReadGuard(guard))
}
page_cache::ReadBufResult::NotFound(mut write_guard) => {
let buf: &mut [u8] = write_guard.deref_mut();
debug_assert_eq!(buf.len(), PAGE_SZ);
self.file
.read_exact_at(&mut buf[..], blknum as u64 * PAGE_SZ as u64)
.await?;
let read_guard = write_guard.mark_valid();
return Ok(BlockLease::PageReadGuard(read_guard));
}
};
loop {
match cache
.read_immutable_buf(self.page_cache_file_id, blknum, ctx)
.await
.map_err(|e| {
std::io::Error::new(
std::io::ErrorKind::Other,
// order path before error because error is anyhow::Error => might have many contexts
format!(
"ephemeral file: read immutable page #{}: {}: {:#}",
blknum,
self.file.path.display(),
e,
),
)
})? {
page_cache::ReadBufResult::Found(guard) => {
return Ok(BlockLease::PageReadGuard(guard))
}
page_cache::ReadBufResult::NotFound(mut write_guard) => {
let buf: &mut [u8] = write_guard.deref_mut();
debug_assert_eq!(buf.len(), PAGE_SZ);
self.file
.read_exact_at(&mut buf[..], blknum as u64 * PAGE_SZ as u64)
.await?;
write_guard.mark_valid();
// Swap for read lock
continue;
}
};
}
} else {
debug_assert_eq!(blknum as u64, self.len / PAGE_SZ as u64);
Ok(BlockLease::EphemeralFileMutableTail(&self.mutable_tail))
@@ -167,7 +171,7 @@ impl EphemeralFile {
let buf: &mut [u8] = write_guard.deref_mut();
debug_assert_eq!(buf.len(), PAGE_SZ);
buf.copy_from_slice(&self.ephemeral_file.mutable_tail);
let _ = write_guard.mark_valid();
write_guard.mark_valid();
// pre-warm successful
}
Err(e) => {
@@ -191,7 +195,7 @@ impl EphemeralFile {
"ephemeral_file: write_blob: write-back full tail blk #{}: {:#}: {}",
self.blknum,
e,
self.ephemeral_file.file.path,
self.ephemeral_file.file.path.display(),
),
));
}
@@ -254,7 +258,8 @@ impl Drop for EphemeralFile {
// not found files might also be related to https://github.com/neondatabase/neon/issues/2442
error!(
"could not remove ephemeral file '{}': {}",
self.file.path, e
self.file.path.display(),
e
);
}
}

View File

@@ -1,9 +1,10 @@
//! This module acts as a switchboard to access different repositories managed by this
//! page server.
use camino::{Utf8Path, Utf8PathBuf};
use rand::{distributions::Alphanumeric, Rng};
use std::collections::{hash_map, HashMap};
use std::ffi::OsStr;
use std::path::{Path, PathBuf};
use std::sync::Arc;
use tokio::fs;
@@ -24,11 +25,9 @@ use crate::control_plane_client::{
};
use crate::deletion_queue::DeletionQueueClient;
use crate::task_mgr::{self, TaskKind};
use crate::tenant::config::{LocationConf, LocationMode, TenantConfOpt};
use crate::tenant::config::TenantConfOpt;
use crate::tenant::delete::DeleteTenantFlow;
use crate::tenant::{
create_tenant_files, AttachedTenantConf, CreateTenantFilesMode, Tenant, TenantState,
};
use crate::tenant::{create_tenant_files, CreateTenantFilesMode, Tenant, TenantState};
use crate::{InitializationOrder, IGNORED_TENANT_FILE_NAME, TEMP_FILE_SUFFIX};
use utils::crashsafe::path_with_suffix_extension;
@@ -40,39 +39,6 @@ use super::delete::DeleteTenantError;
use super::timeline::delete::DeleteTimelineFlow;
use super::TenantSharedResources;
/// For a tenant that appears in TenantsMap, it may either be
/// - `Attached`: has a full Tenant object, is elegible to service
/// reads and ingest WAL.
/// - `Secondary`: is only keeping a local cache warm.
///
/// Secondary is a totally distinct state rather than being a mode of a `Tenant`, because
/// that way we avoid having to carefully switch a tenant's ingestion etc on and off during
/// its lifetime, and we can preserve some important safety invariants like `Tenant` always
/// having a properly acquired generation (Secondary doesn't need a generation)
#[derive(Clone)]
pub enum TenantSlot {
Attached(Arc<Tenant>),
Secondary,
}
impl TenantSlot {
/// Return the `Tenant` in this slot if attached, else None
fn get_attached(&self) -> Option<&Arc<Tenant>> {
match self {
Self::Attached(t) => Some(t),
Self::Secondary => None,
}
}
/// Consume self and return the `Tenant` that was in this slot if attached, else None
fn into_attached(self) -> Option<Arc<Tenant>> {
match self {
Self::Attached(t) => Some(t),
Self::Secondary => None,
}
}
}
/// The tenants known to the pageserver.
/// The enum variants are used to distinguish the different states that the pageserver can be in.
pub(crate) enum TenantsMap {
@@ -80,27 +46,14 @@ pub(crate) enum TenantsMap {
Initializing,
/// [`init_tenant_mgr`] is done, all on-disk tenants have been loaded.
/// New tenants can be added using [`tenant_map_insert`].
Open(HashMap<TenantId, TenantSlot>),
Open(HashMap<TenantId, Arc<Tenant>>),
/// The pageserver has entered shutdown mode via [`shutdown_all_tenants`].
/// Existing tenants are still accessible, but no new tenants can be created.
ShuttingDown(HashMap<TenantId, TenantSlot>),
ShuttingDown(HashMap<TenantId, Arc<Tenant>>),
}
impl TenantsMap {
/// Convenience function for typical usage, where we want to get a `Tenant` object, for
/// working with attached tenants. If the TenantId is in the map but in Secondary state,
/// None is returned.
pub(crate) fn get(&self, tenant_id: &TenantId) -> Option<&Arc<Tenant>> {
match self {
TenantsMap::Initializing => None,
TenantsMap::Open(m) | TenantsMap::ShuttingDown(m) => {
m.get(tenant_id).and_then(TenantSlot::get_attached)
}
}
}
/// Get the contents of the map at this tenant ID, even if it is in secondary state.
pub(crate) fn get_slot(&self, tenant_id: &TenantId) -> Option<&TenantSlot> {
match self {
TenantsMap::Initializing => None,
TenantsMap::Open(m) | TenantsMap::ShuttingDown(m) => m.get(tenant_id),
@@ -109,9 +62,7 @@ impl TenantsMap {
pub(crate) fn remove(&mut self, tenant_id: &TenantId) -> Option<Arc<Tenant>> {
match self {
TenantsMap::Initializing => None,
TenantsMap::Open(m) | TenantsMap::ShuttingDown(m) => {
m.remove(tenant_id).and_then(TenantSlot::into_attached)
}
TenantsMap::Open(m) | TenantsMap::ShuttingDown(m) => m.remove(tenant_id),
}
}
}
@@ -122,12 +73,12 @@ impl TenantsMap {
///
/// This is pageserver-specific, as it relies on future processes after a crash to check
/// for TEMP_FILE_SUFFIX when loading things.
async fn safe_remove_tenant_dir_all(path: impl AsRef<Utf8Path>) -> std::io::Result<()> {
async fn safe_remove_tenant_dir_all(path: impl AsRef<Path>) -> std::io::Result<()> {
let tmp_path = safe_rename_tenant_dir(path).await?;
fs::remove_dir_all(tmp_path).await
}
async fn safe_rename_tenant_dir(path: impl AsRef<Utf8Path>) -> std::io::Result<Utf8PathBuf> {
async fn safe_rename_tenant_dir(path: impl AsRef<Path>) -> std::io::Result<PathBuf> {
let parent = path
.as_ref()
.parent()
@@ -144,7 +95,7 @@ async fn safe_rename_tenant_dir(path: impl AsRef<Utf8Path>) -> std::io::Result<U
.collect::<String>()
+ TEMP_FILE_SUFFIX;
let tmp_path = path_with_suffix_extension(&path, &rand_suffix);
fs::rename(path.as_ref(), &tmp_path).await?;
fs::rename(&path, &tmp_path).await?;
fs::File::open(parent).await?.sync_all().await?;
Ok(tmp_path)
}
@@ -179,15 +130,10 @@ pub async fn init_tenant_mgr(
// 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.
//
// Must only do this if remote storage is enabled, otherwise deletion queue
// is not running and channel push will fail.
if resources.remote_storage.is_some() {
resources
.deletion_queue_client
.recover(result.clone())
.await?;
}
resources
.deletion_queue_client
.recover(result.clone())
.await?;
Some(result)
} else {
@@ -195,25 +141,29 @@ pub async fn init_tenant_mgr(
None
};
let mut dir_entries = tenants_dir
.read_dir_utf8()
let mut dir_entries = fs::read_dir(&tenants_dir)
.await
.with_context(|| format!("Failed to list tenants dir {tenants_dir:?}"))?;
let ctx = RequestContext::todo_child(TaskKind::Startup, DownloadBehavior::Warn);
loop {
match dir_entries.next() {
None => break,
Some(Ok(dir_entry)) => {
let tenant_dir_path = dir_entry.path().to_path_buf();
match dir_entries.next_entry().await {
Ok(None) => break,
Ok(Some(dir_entry)) => {
let tenant_dir_path = dir_entry.path();
if crate::is_temporary(&tenant_dir_path) {
info!("Found temporary tenant directory, removing: {tenant_dir_path}");
info!(
"Found temporary tenant directory, removing: {}",
tenant_dir_path.display()
);
// No need to use safe_remove_tenant_dir_all because this is already
// a temporary path
if let Err(e) = fs::remove_dir_all(&tenant_dir_path).await {
error!(
"Failed to remove temporary directory '{}': {:?}",
tenant_dir_path, e
tenant_dir_path.display(),
e
);
}
} else {
@@ -228,7 +178,7 @@ pub async fn init_tenant_mgr(
if let Err(e) = fs::remove_dir(&tenant_dir_path).await {
error!(
"Failed to remove empty tenant directory '{}': {e:#}",
tenant_dir_path
tenant_dir_path.display()
)
}
continue;
@@ -242,6 +192,7 @@ pub async fn init_tenant_mgr(
let tenant_id = match tenant_dir_path
.file_name()
.and_then(OsStr::to_str)
.unwrap_or_default()
.parse::<TenantId>()
{
@@ -249,65 +200,26 @@ pub async fn init_tenant_mgr(
Err(_) => {
warn!(
"Invalid tenant path (garbage in our repo directory?): {}",
tenant_dir_path
tenant_dir_path.display()
);
continue;
}
};
// Try loading the location configuration
let mut location_conf = match Tenant::load_tenant_config(conf, &tenant_id)
.context("load tenant config")
{
Ok(c) => c,
Err(e) => {
warn!("Marking tenant broken, failed to {e:#}");
tenants.insert(
tenant_id,
TenantSlot::Attached(Tenant::create_broken_tenant(
conf,
tenant_id,
"error loading tenant location configuration".to_string(),
)),
);
continue;
}
};
let generation = if let Some(generations) = &tenant_generations {
// We have a generation map: treat it as the authority for whether
// this tenant is really attached.
if let Some(gen) = generations.get(&tenant_id) {
*gen
} else {
match &location_conf.mode {
LocationMode::Secondary(_) => {
// We do not require the control plane's permission for secondary mode
// tenants, because they do no remote writes and hence require no
// generation number
info!("Loaded tenant {tenant_id} in secondary mode");
tenants.insert(tenant_id, TenantSlot::Secondary);
}
LocationMode::Attached(_) => {
// TODO: augment re-attach API to enable the control plane to
// instruct us about secondary attachments. That way, instead of throwing
// away local state, we can gracefully fall back to secondary here, if the control
// plane tells us so.
// (https://github.com/neondatabase/neon/issues/5377)
info!("Detaching tenant {tenant_id}, control plane omitted it in re-attach response");
if let Err(e) =
safe_remove_tenant_dir_all(&tenant_dir_path).await
{
error!(
"Failed to remove detached tenant directory '{}': {:?}",
tenant_dir_path, e
);
}
}
};
info!("Detaching tenant {tenant_id}, control plane omitted it in re-attach response");
if let Err(e) = safe_remove_tenant_dir_all(&tenant_dir_path).await {
error!(
"Failed to remove detached tenant directory '{}': {:?}",
tenant_dir_path.display(),
e
);
}
continue;
}
} else {
@@ -315,28 +227,23 @@ pub async fn init_tenant_mgr(
// on local disk may activate
info!(
"Starting tenant {} in legacy mode, no generation",
tenant_dir_path
tenant_dir_path.display()
);
Generation::none()
};
// Presence of a generation number implies attachment: attach the tenant
// if it wasn't already, and apply the generation number.
location_conf.attach_in_generation(generation);
Tenant::persist_tenant_config(conf, &tenant_id, &location_conf).await?;
match schedule_local_tenant_processing(
conf,
tenant_id,
&tenant_dir_path,
AttachedTenantConf::try_from(location_conf)?,
generation,
resources.clone(),
Some(init_order.clone()),
&TENANTS,
&ctx,
) {
Ok(tenant) => {
tenants.insert(tenant.tenant_id(), TenantSlot::Attached(tenant));
tenants.insert(tenant.tenant_id(), tenant);
}
Err(e) => {
error!("Failed to collect tenant files from dir {tenants_dir:?} for entry {dir_entry:?}, reason: {e:#}");
@@ -344,7 +251,7 @@ pub async fn init_tenant_mgr(
}
}
}
Some(Err(e)) => {
Err(e) => {
// On error, print it, but continue with the other tenants. If we error out
// here, the pageserver startup fails altogether, causing outage for *all*
// tenants. That seems worse.
@@ -367,8 +274,8 @@ pub async fn init_tenant_mgr(
pub(crate) fn schedule_local_tenant_processing(
conf: &'static PageServerConf,
tenant_id: TenantId,
tenant_path: &Utf8Path,
location_conf: AttachedTenantConf,
tenant_path: &Path,
generation: Generation,
resources: TenantSharedResources,
init_order: Option<InitializationOrder>,
tenants: &'static tokio::sync::RwLock<TenantsMap>,
@@ -405,7 +312,7 @@ pub(crate) fn schedule_local_tenant_processing(
"attaching mark file present but no remote storage configured".to_string(),
)
} else {
match Tenant::spawn_attach(conf, tenant_id, resources, location_conf, tenants, ctx) {
match Tenant::spawn_attach(conf, tenant_id, generation, resources, tenants, ctx) {
Ok(tenant) => tenant,
Err(e) => {
error!("Failed to spawn_attach tenant {tenant_id}, reason: {e:#}");
@@ -417,13 +324,7 @@ pub(crate) fn schedule_local_tenant_processing(
info!("tenant {tenant_id} is assumed to be loadable, starting load operation");
// Start loading the tenant into memory. It will initially be in Loading state.
Tenant::spawn_load(
conf,
tenant_id,
location_conf,
resources,
init_order,
tenants,
ctx,
conf, tenant_id, generation, resources, init_order, tenants, ctx,
)
};
Ok(tenant)
@@ -479,16 +380,7 @@ async fn shutdown_all_tenants0(tenants: &tokio::sync::RwLock<TenantsMap>) {
let res = {
let (_guard, shutdown_progress) = completion::channel();
match tenant {
TenantSlot::Attached(t) => {
t.shutdown(shutdown_progress, freeze_and_flush).await
}
TenantSlot::Secondary => {
// TODO: once secondary mode downloads are implemented,
// ensure they have all stopped before we reach this point.
Ok(())
}
}
tenant.shutdown(shutdown_progress, freeze_and_flush).await
};
if let Err(other_progress) = res {
@@ -561,19 +453,16 @@ pub async fn create_tenant(
ctx: &RequestContext,
) -> Result<Arc<Tenant>, TenantMapInsertError> {
tenant_map_insert(tenant_id, || async {
let location_conf = LocationConf::attached_single(tenant_conf, generation);
// We're holding the tenants lock in write mode while doing local IO.
// If this section ever becomes contentious, introduce a new `TenantState::Creating`
// and do the work in that state.
let tenant_directory = super::create_tenant_files(conf, &location_conf, &tenant_id, CreateTenantFilesMode::Create).await?;
let tenant_directory = super::create_tenant_files(conf, tenant_conf, &tenant_id, CreateTenantFilesMode::Create).await?;
// TODO: tenant directory remains on disk if we bail out from here on.
// See https://github.com/neondatabase/neon/issues/4233
let created_tenant =
schedule_local_tenant_processing(conf, tenant_id, &tenant_directory,
AttachedTenantConf::try_from(location_conf)?, resources, None, &TENANTS, ctx)?;
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
@@ -602,137 +491,20 @@ pub async fn set_new_tenant_config(
info!("configuring tenant {tenant_id}");
let tenant = get_tenant(tenant_id, true).await?;
// This is a legacy API that only operates on attached tenants: the preferred
// API to use is the location_config/ endpoint, which lets the caller provide
// the full LocationConf.
let location_conf = LocationConf::attached_single(new_tenant_conf, tenant.generation);
Tenant::persist_tenant_config(conf, &tenant_id, &location_conf)
let tenant_config_path = conf.tenant_config_path(&tenant_id);
Tenant::persist_tenant_config(&tenant_id, &tenant_config_path, new_tenant_conf)
.await
.map_err(SetNewTenantConfigError::Persist)?;
tenant.set_new_tenant_config(new_tenant_conf);
Ok(())
}
#[instrument(skip_all, fields(tenant_id, new_location_config))]
pub(crate) async fn upsert_location(
conf: &'static PageServerConf,
tenant_id: TenantId,
new_location_config: LocationConf,
broker_client: storage_broker::BrokerClientChannel,
remote_storage: Option<GenericRemoteStorage>,
deletion_queue_client: DeletionQueueClient,
ctx: &RequestContext,
) -> Result<(), anyhow::Error> {
info!("configuring tenant location {tenant_id} to state {new_location_config:?}");
let mut existing_tenant = match get_tenant(tenant_id, false).await {
Ok(t) => Some(t),
Err(GetTenantError::NotFound(_)) => None,
Err(e) => anyhow::bail!(e),
};
// If we need to shut down a Tenant, do that first
let shutdown_tenant = match (&new_location_config.mode, &existing_tenant) {
(LocationMode::Secondary(_), Some(t)) => Some(t),
(LocationMode::Attached(attach_conf), Some(t)) => {
if attach_conf.generation != t.generation {
Some(t)
} else {
None
}
}
_ => None,
};
// TODO: currently we risk concurrent operations interfering with the tenant
// while we await shutdown, but we also should not hold the TenantsMap lock
// across the whole operation. Before we start using this function in production,
// a follow-on change will revise how concurrency is handled in TenantsMap.
// (https://github.com/neondatabase/neon/issues/5378)
if let Some(tenant) = shutdown_tenant {
let (_guard, progress) = utils::completion::channel();
info!("Shutting down attached tenant");
match tenant.shutdown(progress, false).await {
Ok(()) => {}
Err(barrier) => {
info!("Shutdown already in progress, waiting for it to complete");
barrier.wait().await;
}
}
existing_tenant = None;
}
if let Some(tenant) = existing_tenant {
// Update the existing tenant
Tenant::persist_tenant_config(conf, &tenant_id, &new_location_config)
.await
.map_err(SetNewTenantConfigError::Persist)?;
tenant.set_new_location_config(AttachedTenantConf::try_from(new_location_config)?);
} else {
// Upsert a fresh TenantSlot into TenantsMap. Do it within the map write lock,
// and re-check that the state of anything we are replacing is as expected.
tenant_map_upsert_slot(tenant_id, |old_value| async move {
if let Some(TenantSlot::Attached(t)) = old_value {
if !matches!(t.current_state(), TenantState::Stopping { .. }) {
anyhow::bail!("Tenant state changed during location configuration update");
}
}
let new_slot = match &new_location_config.mode {
LocationMode::Secondary(_) => TenantSlot::Secondary,
LocationMode::Attached(_attach_config) => {
// Do a schedule_local_tenant_processing
// FIXME: should avoid doing this disk I/O inside the TenantsMap lock,
// we have the same problem in load_tenant/attach_tenant. Probably
// need a lock in TenantSlot to fix this.
Tenant::persist_tenant_config(conf, &tenant_id, &new_location_config)
.await
.map_err(SetNewTenantConfigError::Persist)?;
let tenant_path = conf.tenant_path(&tenant_id);
let resources = TenantSharedResources {
broker_client,
remote_storage,
deletion_queue_client,
};
let new_tenant = schedule_local_tenant_processing(
conf,
tenant_id,
&tenant_path,
AttachedTenantConf::try_from(new_location_config)?,
resources,
None,
&TENANTS,
ctx,
)
.with_context(|| {
format!("Failed to schedule tenant processing in path {tenant_path:?}")
})?;
TenantSlot::Attached(new_tenant)
}
};
Ok(new_slot)
})
.await?;
}
Ok(())
}
#[derive(Debug, thiserror::Error)]
pub enum GetTenantError {
#[error("Tenant {0} not found")]
NotFound(TenantId),
#[error("Tenant {0} is not active")]
NotActive(TenantId),
/// Broken is logically a subset of NotActive, but a distinct error is useful as
/// NotActive is usually a retryable state for API purposes, whereas Broken
/// is a stuck error state
#[error("Tenant is broken: {0}")]
Broken(String),
}
/// Gets the tenant from the in-memory data, erroring if it's absent or is not fitting to the query.
@@ -747,20 +519,10 @@ pub async fn get_tenant(
let tenant = m
.get(&tenant_id)
.ok_or(GetTenantError::NotFound(tenant_id))?;
match tenant.current_state() {
TenantState::Broken {
reason,
backtrace: _,
} if active_only => Err(GetTenantError::Broken(reason)),
TenantState::Active => Ok(Arc::clone(tenant)),
_ => {
if active_only {
Err(GetTenantError::NotActive(tenant_id))
} else {
Ok(Arc::clone(tenant))
}
}
if active_only && !tenant.is_active() {
Err(GetTenantError::NotActive(tenant_id))
} else {
Ok(Arc::clone(tenant))
}
}
@@ -833,7 +595,7 @@ async fn detach_tenant0(
tenants: &tokio::sync::RwLock<TenantsMap>,
tenant_id: TenantId,
detach_ignored: bool,
) -> Result<Utf8PathBuf, TenantStateError> {
) -> Result<PathBuf, TenantStateError> {
let tenant_dir_rename_operation = |tenant_id_to_clean| async move {
let local_tenant_directory = conf.tenant_path(&tenant_id_to_clean);
safe_rename_tenant_dir(&local_tenant_directory)
@@ -882,12 +644,7 @@ pub async fn load_tenant(
remote_storage,
deletion_queue_client
};
let mut location_conf = Tenant::load_tenant_config(conf, &tenant_id).map_err( TenantMapInsertError::Other)?;
location_conf.attach_in_generation(generation);
Tenant::persist_tenant_config(conf, &tenant_id, &location_conf).await?;
let new_tenant = schedule_local_tenant_processing(conf, tenant_id, &tenant_path, AttachedTenantConf::try_from(location_conf)?, resources, None, &TENANTS, ctx)
let new_tenant = schedule_local_tenant_processing(conf, tenant_id, &tenant_path, generation, resources, None, &TENANTS, ctx)
.with_context(|| {
format!("Failed to schedule tenant processing in path {tenant_path:?}")
})?;
@@ -940,10 +697,7 @@ pub async fn list_tenants() -> Result<Vec<(TenantId, TenantState)>, TenantMapLis
TenantsMap::Open(m) | TenantsMap::ShuttingDown(m) => m,
};
Ok(m.iter()
.filter_map(|(id, tenant)| match tenant {
TenantSlot::Attached(tenant) => Some((*id, tenant.current_state())),
TenantSlot::Secondary => None,
})
.map(|(id, tenant)| (*id, tenant.current_state()))
.collect())
}
@@ -960,8 +714,7 @@ pub async fn attach_tenant(
ctx: &RequestContext,
) -> Result<(), TenantMapInsertError> {
tenant_map_insert(tenant_id, || async {
let location_conf = LocationConf::attached_single(tenant_conf, generation);
let tenant_dir = create_tenant_files(conf, &location_conf, &tenant_id, CreateTenantFilesMode::Attach).await?;
let tenant_dir = create_tenant_files(conf, tenant_conf, &tenant_id, CreateTenantFilesMode::Attach).await?;
// TODO: tenant directory remains on disk if we bail out from here on.
// See https://github.com/neondatabase/neon/issues/4233
@@ -972,7 +725,8 @@ 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 attached_tenant = schedule_local_tenant_processing(conf, tenant_id, &tenant_dir, AttachedTenantConf::try_from(location_conf)?, resources, None, &TENANTS, ctx)?;
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
@@ -995,10 +749,8 @@ pub enum TenantMapInsertError {
ShuttingDown,
#[error("tenant {0} already exists, state: {1:?}")]
TenantAlreadyExists(TenantId, TenantState),
#[error("tenant {0} already exists in secondary state")]
TenantExistsSecondary(TenantId),
#[error(transparent)]
Other(#[from] anyhow::Error),
Closure(#[from] anyhow::Error),
}
/// Give the given closure access to the tenants map entry for the given `tenant_id`, iff that
@@ -1022,47 +774,20 @@ where
TenantsMap::Open(m) => m,
};
match m.entry(tenant_id) {
hash_map::Entry::Occupied(e) => match e.get() {
TenantSlot::Attached(t) => Err(TenantMapInsertError::TenantAlreadyExists(
tenant_id,
t.current_state(),
)),
TenantSlot::Secondary => Err(TenantMapInsertError::TenantExistsSecondary(tenant_id)),
},
hash_map::Entry::Occupied(e) => Err(TenantMapInsertError::TenantAlreadyExists(
tenant_id,
e.get().current_state(),
)),
hash_map::Entry::Vacant(v) => match insert_fn().await {
Ok(tenant) => {
v.insert(TenantSlot::Attached(tenant.clone()));
v.insert(tenant.clone());
Ok(tenant)
}
Err(e) => Err(TenantMapInsertError::Other(e)),
Err(e) => Err(TenantMapInsertError::Closure(e)),
},
}
}
async fn tenant_map_upsert_slot<'a, F, R>(
tenant_id: TenantId,
upsert_fn: F,
) -> Result<(), TenantMapInsertError>
where
F: FnOnce(Option<TenantSlot>) -> R,
R: std::future::Future<Output = anyhow::Result<TenantSlot>>,
{
let mut guard = TENANTS.write().await;
let m = match &mut *guard {
TenantsMap::Initializing => return Err(TenantMapInsertError::StillInitializing),
TenantsMap::ShuttingDown(_) => return Err(TenantMapInsertError::ShuttingDown),
TenantsMap::Open(m) => m,
};
match upsert_fn(m.remove(&tenant_id)).await {
Ok(upsert_val) => {
m.insert(tenant_id, upsert_val);
Ok(())
}
Err(e) => Err(TenantMapInsertError::Other(e)),
}
}
/// Stops and removes the tenant from memory, if it's not [`TenantState::Stopping`] already, bails otherwise.
/// Allows to remove other tenant resources manually, via `tenant_cleanup`.
/// If the cleanup fails, tenant will stay in memory in [`TenantState::Broken`] state, and another removal
@@ -1082,40 +807,28 @@ where
// tenant-wde cleanup operations may take some time (removing the entire tenant directory), we want to
// avoid holding the lock for the entire process.
let tenant = {
match tenants
tenants
.write()
.await
.get_slot(&tenant_id)
.get(&tenant_id)
.cloned()
.ok_or(TenantStateError::NotFound(tenant_id))?
{
TenantSlot::Attached(t) => Some(t.clone()),
TenantSlot::Secondary => None,
}
};
// allow pageserver shutdown to await for our completion
let (_guard, progress) = completion::channel();
// If the tenant was attached, shut it down gracefully. For secondary
// locations this part is not necessary
match tenant {
Some(attached_tenant) => {
// whenever we remove a tenant from memory, we don't want to flush and wait for upload
let freeze_and_flush = false;
// whenever we remove a tenant from memory, we don't want to flush and wait for upload
let freeze_and_flush = false;
// shutdown is sure to transition tenant to stopping, and wait for all tasks to complete, so
// that we can continue safely to cleanup.
match attached_tenant.shutdown(progress, freeze_and_flush).await {
Ok(()) => {}
Err(_other) => {
// if pageserver shutdown or other detach/ignore is already ongoing, we don't want to
// wait for it but return an error right away because these are distinct requests.
return Err(TenantStateError::IsStopping(tenant_id));
}
}
}
None => {
// Nothing to wait on when not attached, proceed.
// shutdown is sure to transition tenant to stopping, and wait for all tasks to complete, so
// that we can continue safely to cleanup.
match tenant.shutdown(progress, freeze_and_flush).await {
Ok(()) => {}
Err(_other) => {
// if pageserver shutdown or other detach/ignore is already ongoing, we don't want to
// wait for it but return an error right away because these are distinct requests.
return Err(TenantStateError::IsStopping(tenant_id));
}
}
@@ -1206,8 +919,6 @@ mod tests {
use std::sync::Arc;
use tracing::{info_span, Instrument};
use crate::tenant::mgr::TenantSlot;
use super::{super::harness::TenantHarness, TenantsMap};
#[tokio::test(start_paused = true)]
@@ -1229,7 +940,7 @@ mod tests {
// tenant harness configures the logging and we cannot escape it
let _e = info_span!("testing", tenant_id = %id).entered();
let tenants = HashMap::from([(id, TenantSlot::Attached(t.clone()))]);
let tenants = HashMap::from([(id, t.clone())]);
let tenants = Arc::new(tokio::sync::RwLock::new(TenantsMap::Open(tenants)));
let (until_cleanup_completed, can_complete_cleanup) = utils::completion::channel();

View File

@@ -1,17 +1,16 @@
use std::{
io,
path::{Path, PathBuf},
sync::atomic::{AtomicUsize, Ordering},
};
use camino::{Utf8Path, Utf8PathBuf};
fn fsync_path(path: &Utf8Path) -> io::Result<()> {
fn fsync_path(path: &Path) -> io::Result<()> {
// TODO use VirtualFile::fsync_all once we fully go async.
let file = std::fs::File::open(path)?;
file.sync_all()
}
fn parallel_worker(paths: &[Utf8PathBuf], next_path_idx: &AtomicUsize) -> io::Result<()> {
fn parallel_worker(paths: &[PathBuf], next_path_idx: &AtomicUsize) -> io::Result<()> {
while let Some(path) = paths.get(next_path_idx.fetch_add(1, Ordering::Relaxed)) {
fsync_path(path)?;
}
@@ -19,7 +18,7 @@ fn parallel_worker(paths: &[Utf8PathBuf], next_path_idx: &AtomicUsize) -> io::Re
Ok(())
}
fn fsync_in_thread_pool(paths: &[Utf8PathBuf]) -> io::Result<()> {
fn fsync_in_thread_pool(paths: &[PathBuf]) -> io::Result<()> {
// TODO: remove this function in favor of `par_fsync_async` once we asyncify everything.
/// Use at most this number of threads.
@@ -48,7 +47,7 @@ fn fsync_in_thread_pool(paths: &[Utf8PathBuf]) -> io::Result<()> {
}
/// Parallel fsync all files. Can be used in non-async context as it is using rayon thread pool.
pub fn par_fsync(paths: &[Utf8PathBuf]) -> io::Result<()> {
pub fn par_fsync(paths: &[PathBuf]) -> io::Result<()> {
if paths.len() == 1 {
fsync_path(&paths[0])?;
return Ok(());
@@ -59,7 +58,7 @@ pub fn par_fsync(paths: &[Utf8PathBuf]) -> io::Result<()> {
/// Parallel fsync asynchronously. If number of files are less than PARALLEL_PATH_THRESHOLD, fsync is done in the current
/// execution thread. Otherwise, we will spawn_blocking and run it in tokio.
pub async fn par_fsync_async(paths: &[Utf8PathBuf]) -> io::Result<()> {
pub async fn par_fsync_async(paths: &[PathBuf]) -> io::Result<()> {
const MAX_CONCURRENT_FSYNC: usize = 64;
let mut next = paths.iter().peekable();
let mut js = tokio::task::JoinSet::new();

View File

@@ -209,7 +209,6 @@ pub mod index;
mod upload;
use anyhow::Context;
use camino::Utf8Path;
use chrono::{NaiveDateTime, Utc};
// re-export these
pub use download::{is_temp_download_file, list_remote_timelines};
@@ -220,6 +219,7 @@ use utils::backoff::{
};
use std::collections::{HashMap, VecDeque};
use std::path::{Path, PathBuf};
use std::sync::atomic::{AtomicU32, Ordering};
use std::sync::{Arc, Mutex};
@@ -924,7 +924,7 @@ impl RemoteTimelineClient {
))?
});
let index_file_path = timeline_storage_path.join(Utf8Path::new(IndexPart::FILE_NAME));
let index_file_path = timeline_storage_path.join(Path::new(IndexPart::FILE_NAME));
debug!("enqueuing index part deletion");
self.deletion_queue_client
@@ -1409,7 +1409,7 @@ pub fn remote_timelines_path(tenant_id: &TenantId) -> RemotePath {
}
pub fn remote_timeline_path(tenant_id: &TenantId, timeline_id: &TimelineId) -> RemotePath {
remote_timelines_path(tenant_id).join(Utf8Path::new(&timeline_id.to_string()))
remote_timelines_path(tenant_id).join(&PathBuf::from(timeline_id.to_string()))
}
pub fn remote_layer_path(
@@ -1452,7 +1452,14 @@ pub(crate) fn parse_remote_index_path(path: RemotePath) -> Option<Generation> {
}
};
match file_name.split_once('-') {
let file_name_str = match file_name.to_str() {
Some(s) => s,
None => {
tracing::warn!("Malformed index key {:?}", path);
return None;
}
};
match file_name_str.split_once('-') {
Some((_, gen_suffix)) => Generation::parse_suffix(gen_suffix),
None => None,
}
@@ -1464,16 +1471,20 @@ pub(crate) fn parse_remote_index_path(path: RemotePath) -> Option<Generation> {
/// Errors if the path provided does not start from pageserver's workdir.
pub fn remote_path(
conf: &PageServerConf,
local_path: &Utf8Path,
local_path: &Path,
generation: Generation,
) -> anyhow::Result<RemotePath> {
let stripped = local_path
.strip_prefix(&conf.workdir)
.context("Failed to strip workdir prefix")?;
let suffixed = format!("{0}{1}", stripped, generation.get_suffix());
let suffixed = format!(
"{0}{1}",
stripped.to_string_lossy(),
generation.get_suffix()
);
RemotePath::new(Utf8Path::new(&suffixed)).with_context(|| {
RemotePath::new(&PathBuf::from(suffixed)).with_context(|| {
format!(
"to resolve remote part of path {:?} for base {:?}",
local_path, conf.workdir
@@ -1493,7 +1504,7 @@ mod tests {
DEFAULT_PG_VERSION,
};
use std::collections::HashSet;
use std::{collections::HashSet, path::Path};
use utils::lsn::Lsn;
pub(super) fn dummy_contents(name: &str) -> Vec<u8> {
@@ -1527,7 +1538,7 @@ mod tests {
assert_eq!(avec, bvec);
}
fn assert_remote_files(expected: &[&str], remote_path: &Utf8Path, generation: Generation) {
fn assert_remote_files(expected: &[&str], remote_path: &Path, generation: Generation) {
let mut expected: Vec<String> = expected
.iter()
.map(|x| format!("{}{}", x, generation.get_suffix()))
@@ -1646,12 +1657,12 @@ mod tests {
let timeline_path = harness.timeline_path(&TIMELINE_ID);
println!("workdir: {}", harness.conf.workdir);
println!("workdir: {}", harness.conf.workdir.display());
let remote_timeline_dir = harness
.remote_fs_dir
.join(timeline_path.strip_prefix(&harness.conf.workdir).unwrap());
println!("remote_timeline_dir: {remote_timeline_dir}");
println!("remote_timeline_dir: {}", remote_timeline_dir.display());
let generation = harness.generation;
@@ -1898,7 +1909,7 @@ mod tests {
let index_path = test_state.harness.remote_fs_dir.join(
remote_index_path(&test_state.harness.tenant_id, &TIMELINE_ID, generation).get_path(),
);
eprintln!("Writing {index_path}");
eprintln!("Writing {}", index_path.display());
std::fs::write(&index_path, index_part_bytes).unwrap();
example_index_part
}

View File

@@ -5,10 +5,10 @@
use std::collections::HashSet;
use std::future::Future;
use std::path::Path;
use std::time::Duration;
use anyhow::{anyhow, Context};
use camino::Utf8Path;
use tokio::fs;
use tokio::io::AsyncWriteExt;
use tokio_util::sync::CancellationToken;
@@ -74,7 +74,12 @@ pub async fn download_layer_file<'a>(
// TODO: this doesn't use the cached fd for some reason?
let mut destination_file = fs::File::create(&temp_file_path)
.await
.with_context(|| format!("create a destination file for layer '{temp_file_path}'"))
.with_context(|| {
format!(
"create a destination file for layer '{}'",
temp_file_path.display()
)
})
.map_err(DownloadError::Other)?;
let mut download = storage
.download(&remote_path)
@@ -116,7 +121,7 @@ pub async fn download_layer_file<'a>(
destination_file
.flush()
.await
.with_context(|| format!("flush source file at {temp_file_path}"))
.with_context(|| format!("flush source file at {}", temp_file_path.display()))
.map_err(DownloadError::Other)?;
let expected = layer_metadata.file_size();
@@ -130,7 +135,12 @@ pub async fn download_layer_file<'a>(
destination_file
.sync_all()
.await
.with_context(|| format!("failed to fsync source file at {temp_file_path}"))
.with_context(|| {
format!(
"failed to fsync source file at {}",
temp_file_path.display()
)
})
.map_err(DownloadError::Other)?;
drop(destination_file);
@@ -142,23 +152,27 @@ pub async fn download_layer_file<'a>(
fs::rename(&temp_file_path, &local_path)
.await
.with_context(|| format!("rename download layer file to {local_path}"))
.with_context(|| format!("rename download layer file to {}", local_path.display(),))
.map_err(DownloadError::Other)?;
crashsafe::fsync_async(&local_path)
.await
.with_context(|| format!("fsync layer file {local_path}"))
.with_context(|| format!("fsync layer file {}", local_path.display(),))
.map_err(DownloadError::Other)?;
tracing::debug!("download complete: {local_path}");
tracing::debug!("download complete: {}", local_path.display());
Ok(bytes_amount)
}
const TEMP_DOWNLOAD_EXTENSION: &str = "temp_download";
pub fn is_temp_download_file(path: &Utf8Path) -> bool {
let extension = path.extension();
pub fn is_temp_download_file(path: &Path) -> bool {
let extension = path.extension().map(|pname| {
pname
.to_str()
.expect("paths passed to this function must be valid Rust strings")
});
match extension {
Some(TEMP_DOWNLOAD_EXTENSION) => true,
Some(_) => false,

View File

@@ -1,9 +1,8 @@
//! Helper functions to upload files to remote storage with a RemoteStorage
use anyhow::{bail, Context};
use camino::Utf8Path;
use fail::fail_point;
use std::io::ErrorKind;
use std::{io::ErrorKind, path::Path};
use tokio::fs;
use super::Generation;
@@ -51,7 +50,7 @@ pub(super) async fn upload_index_part<'a>(
pub(super) async fn upload_timeline_layer<'a>(
conf: &'static PageServerConf,
storage: &'a GenericRemoteStorage,
source_path: &'a Utf8Path,
source_path: &'a Path,
known_metadata: &'a LayerFileMetadata,
generation: Generation,
) -> anyhow::Result<()> {
@@ -69,7 +68,7 @@ pub(super) async fn upload_timeline_layer<'a>(
// upload. However, a nonexistent file can also be indicative of
// something worse, like when a file is scheduled for upload before
// it has been written to disk yet.
info!(path = %source_path, "File to upload doesn't exist. Likely the file has been deleted and an upload is not required any more.");
info!(path = %source_path.display(), "File to upload doesn't exist. Likely the file has been deleted and an upload is not required any more.");
return Ok(());
}
Err(e) => {
@@ -94,7 +93,7 @@ pub(super) async fn upload_timeline_layer<'a>(
storage
.upload(source_file, fs_size, &storage_path, None)
.await
.with_context(|| format!("upload layer from local path '{source_path}'"))?;
.with_context(|| format!("upload layer from local path '{}'", source_path.display()))?;
Ok(())
}

View File

@@ -14,7 +14,6 @@ use crate::task_mgr::TaskKind;
use crate::walrecord::NeonWalRecord;
use anyhow::Result;
use bytes::Bytes;
use camino::Utf8PathBuf;
use enum_map::EnumMap;
use enumset::EnumSet;
use once_cell::sync::Lazy;
@@ -23,6 +22,7 @@ use pageserver_api::models::{
HistoricLayerInfo, LayerResidenceEvent, LayerResidenceEventReason, LayerResidenceStatus,
};
use std::ops::Range;
use std::path::PathBuf;
use std::sync::{Arc, Mutex};
use std::time::{Duration, SystemTime, UNIX_EPOCH};
use tracing::warn;
@@ -378,7 +378,7 @@ pub trait PersistentLayer: Layer + AsLayerDesc {
// Path to the layer file in the local filesystem.
// `None` for `RemoteLayer`.
fn local_path(&self) -> Option<Utf8PathBuf>;
fn local_path(&self) -> Option<PathBuf>;
/// Permanently remove this layer from disk.
fn delete_resident_layer_file(&self) -> Result<()>;
@@ -456,7 +456,7 @@ pub mod tests {
/// config. In that case, we use the Path variant to hold the full path to the file on
/// disk.
enum PathOrConf {
Path(Utf8PathBuf),
Path(PathBuf),
Conf(&'static PageServerConf),
}

View File

@@ -41,7 +41,6 @@ use crate::virtual_file::VirtualFile;
use crate::{walrecord, TEMP_FILE_SUFFIX};
use crate::{DELTA_FILE_MAGIC, STORAGE_FORMAT_VERSION};
use anyhow::{bail, ensure, Context, Result};
use camino::{Utf8Path, Utf8PathBuf};
use pageserver_api::models::{HistoricLayerInfo, LayerAccessKind};
use rand::{distributions::Alphanumeric, Rng};
use serde::{Deserialize, Serialize};
@@ -49,6 +48,7 @@ use std::fs::{self, File};
use std::io::SeekFrom;
use std::ops::Range;
use std::os::unix::fs::FileExt;
use std::path::{Path, PathBuf};
use std::sync::Arc;
use tokio::sync::OnceCell;
use tracing::*;
@@ -267,7 +267,7 @@ impl PersistentLayer for DeltaLayer {
Some(self)
}
fn local_path(&self) -> Option<Utf8PathBuf> {
fn local_path(&self) -> Option<PathBuf> {
self.local_path()
}
@@ -374,7 +374,7 @@ impl DeltaLayer {
.await
}
pub(crate) fn local_path(&self) -> Option<Utf8PathBuf> {
pub(crate) fn local_path(&self) -> Option<PathBuf> {
Some(self.path())
}
@@ -409,7 +409,7 @@ impl DeltaLayer {
tenant_id: &TenantId,
timeline_id: &TimelineId,
fname: &DeltaFileName,
) -> Utf8PathBuf {
) -> PathBuf {
match path_or_conf {
PathOrConf::Path(path) => path.clone(),
PathOrConf::Conf(conf) => conf
@@ -424,7 +424,7 @@ impl DeltaLayer {
timeline_id: &TimelineId,
key_start: Key,
lsn_range: &Range<Lsn>,
) -> Utf8PathBuf {
) -> PathBuf {
let rand_string: String = rand::thread_rng()
.sample_iter(&Alphanumeric)
.take(8)
@@ -455,7 +455,7 @@ impl DeltaLayer {
self.inner
.get_or_try_init(|| self.load_inner(ctx))
.await
.with_context(|| format!("Failed to load delta layer {}", self.path()))
.with_context(|| format!("Failed to load delta layer {}", self.path().display()))
}
async fn load_inner(&self, ctx: &RequestContext) -> Result<Arc<DeltaLayerInner>> {
@@ -471,7 +471,7 @@ impl DeltaLayer {
if let PathOrConf::Path(ref path) = self.path_or_conf {
// not production code
let actual_filename = path.file_name().unwrap().to_owned();
let actual_filename = path.file_name().unwrap().to_str().unwrap().to_owned();
let expected_filename = self.filename().file_name();
if actual_filename != expected_filename {
@@ -510,7 +510,7 @@ impl DeltaLayer {
/// Create a DeltaLayer struct representing an existing file on disk.
///
/// This variant is only used for debugging purposes, by the 'pagectl' binary.
pub fn new_for_path(path: &Utf8Path, file: File) -> Result<Self> {
pub fn new_for_path(path: &Path, file: File) -> Result<Self> {
let mut summary_buf = Vec::new();
summary_buf.resize(PAGE_SZ, 0);
file.read_exact_at(&mut summary_buf, 0)?;
@@ -538,7 +538,7 @@ impl DeltaLayer {
self.desc.delta_file_name()
}
/// Path to the layer file in pageserver workdir.
pub fn path(&self) -> Utf8PathBuf {
pub fn path(&self) -> PathBuf {
Self::path_for(
&self.path_or_conf,
&self.desc.tenant_id,
@@ -549,7 +549,7 @@ impl DeltaLayer {
/// Loads all keys stored in the layer. Returns key, lsn, value size and value reference.
///
/// The value can be obtained via the [`ValueRef::load`] function.
pub(crate) async fn load_keys<'c>(&self, ctx: &RequestContext) -> Result<Vec<DeltaEntry<'_>>> {
pub(crate) async fn load_keys(&self, ctx: &RequestContext) -> Result<Vec<DeltaEntry<'_>>> {
let inner = self
.load(LayerAccessKind::KeyIter, ctx)
.await
@@ -573,7 +573,7 @@ impl DeltaLayer {
///
struct DeltaLayerWriterInner {
conf: &'static PageServerConf,
pub path: Utf8PathBuf,
pub path: PathBuf,
timeline_id: TimelineId,
tenant_id: TenantId,
@@ -711,7 +711,7 @@ impl DeltaLayerWriterInner {
ensure!(
metadata.len() <= S3_UPLOAD_LIMIT,
"Created delta layer file at {} of size {} above limit {S3_UPLOAD_LIMIT}!",
file.path,
file.path.display(),
metadata.len()
);
@@ -748,7 +748,7 @@ impl DeltaLayerWriterInner {
);
std::fs::rename(self.path, &final_path)?;
trace!("created delta layer {final_path}");
trace!("created delta layer {}", final_path.display());
Ok(layer)
}
@@ -847,13 +847,13 @@ impl Drop for DeltaLayerWriter {
impl DeltaLayerInner {
pub(super) async fn load(
path: &Utf8Path,
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}'"))?;
.with_context(|| format!("Failed to open file '{}'", path.display()))?;
let file = FileBlockReader::new(file);
let summary_blk = file.read_blk(0, ctx).await?;
@@ -933,12 +933,15 @@ impl DeltaLayerInner {
.read_blob_into_buf(pos, &mut buf, ctx)
.await
.with_context(|| {
format!("Failed to read blob from virtual file {}", file.file.path)
format!(
"Failed to read blob from virtual file {}",
file.file.path.display()
)
})?;
let val = Value::des(&buf).with_context(|| {
format!(
"Failed to deserialize file blob from virtual file {}",
file.file.path
file.file.path.display()
)
})?;
match val {
@@ -1038,9 +1041,9 @@ pub struct ValueRef<'a> {
reader: BlockCursor<'a>,
}
impl<'c, 'a> ValueRef<'a> {
impl<'a> ValueRef<'a> {
/// Loads the value from disk
pub async fn load(&self, ctx: &'c RequestContext) -> Result<Value> {
pub async fn load(&self, ctx: &RequestContext) -> 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 val = Value::des(&buf)?;
@@ -1051,11 +1054,11 @@ impl<'c, 'a> ValueRef<'a> {
pub(crate) struct Adapter<T>(T);
impl<T: AsRef<DeltaLayerInner>> Adapter<T> {
pub(crate) async fn read_blk<'c>(
pub(crate) async fn read_blk(
&self,
blknum: u32,
ctx: &'c RequestContext,
) -> Result<BlockLease<'c, '_>, std::io::Error> {
ctx: &RequestContext,
) -> Result<BlockLease, std::io::Error> {
self.0.as_ref().file.read_blk(blknum, ctx).await
}
}

View File

@@ -37,7 +37,6 @@ use crate::virtual_file::VirtualFile;
use crate::{IMAGE_FILE_MAGIC, STORAGE_FORMAT_VERSION, TEMP_FILE_SUFFIX};
use anyhow::{bail, ensure, Context, Result};
use bytes::Bytes;
use camino::{Utf8Path, Utf8PathBuf};
use hex;
use pageserver_api::models::{HistoricLayerInfo, LayerAccessKind};
use rand::{distributions::Alphanumeric, Rng};
@@ -46,6 +45,7 @@ use std::fs::{self, File};
use std::io::SeekFrom;
use std::ops::Range;
use std::os::unix::prelude::FileExt;
use std::path::{Path, PathBuf};
use tokio::sync::OnceCell;
use tracing::*;
@@ -195,7 +195,7 @@ impl AsLayerDesc for ImageLayer {
}
impl PersistentLayer for ImageLayer {
fn local_path(&self) -> Option<Utf8PathBuf> {
fn local_path(&self) -> Option<PathBuf> {
self.local_path()
}
@@ -269,10 +269,10 @@ impl ImageLayer {
.get_value_reconstruct_data(key, reconstruct_state, ctx)
.await
// FIXME: makes no sense to dump paths
.with_context(|| format!("read {}", self.path()))
.with_context(|| format!("read {}", self.path().display()))
}
pub(crate) fn local_path(&self) -> Option<Utf8PathBuf> {
pub(crate) fn local_path(&self) -> Option<PathBuf> {
Some(self.path())
}
@@ -304,7 +304,7 @@ impl ImageLayer {
timeline_id: TimelineId,
tenant_id: TenantId,
fname: &ImageFileName,
) -> Utf8PathBuf {
) -> PathBuf {
match path_or_conf {
PathOrConf::Path(path) => path.to_path_buf(),
PathOrConf::Conf(conf) => conf
@@ -318,7 +318,7 @@ impl ImageLayer {
timeline_id: TimelineId,
tenant_id: TenantId,
fname: &ImageFileName,
) -> Utf8PathBuf {
) -> PathBuf {
let rand_string: String = rand::thread_rng()
.sample_iter(&Alphanumeric)
.take(8)
@@ -342,7 +342,7 @@ impl ImageLayer {
self.inner
.get_or_try_init(|| self.load_inner(ctx))
.await
.with_context(|| format!("Failed to load image layer {}", self.path()))
.with_context(|| format!("Failed to load image layer {}", self.path().display()))
}
async fn load_inner(&self, ctx: &RequestContext) -> Result<ImageLayerInner> {
@@ -359,7 +359,7 @@ impl ImageLayer {
if let PathOrConf::Path(ref path) = self.path_or_conf {
// not production code
let actual_filename = path.file_name().unwrap().to_owned();
let actual_filename = path.file_name().unwrap().to_str().unwrap().to_owned();
let expected_filename = self.filename().file_name();
if actual_filename != expected_filename {
@@ -399,7 +399,7 @@ impl ImageLayer {
/// Create an ImageLayer struct representing an existing file on disk.
///
/// This variant is only used for debugging purposes, by the 'pagectl' binary.
pub fn new_for_path(path: &Utf8Path, file: File) -> Result<ImageLayer> {
pub fn new_for_path(path: &Path, file: File) -> Result<ImageLayer> {
let mut summary_buf = Vec::new();
summary_buf.resize(PAGE_SZ, 0);
file.read_exact_at(&mut summary_buf, 0)?;
@@ -427,7 +427,7 @@ impl ImageLayer {
}
/// Path to the layer file in pageserver workdir.
pub fn path(&self) -> Utf8PathBuf {
pub fn path(&self) -> PathBuf {
Self::path_for(
&self.path_or_conf,
self.desc.timeline_id,
@@ -439,14 +439,14 @@ impl ImageLayer {
impl ImageLayerInner {
pub(super) async fn load(
path: &Utf8Path,
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))?;
.with_context(|| format!("Failed to open file '{}'", path.display()))?;
let file = FileBlockReader::new(file);
let summary_blk = file.read_blk(0, ctx).await?;
let actual_summary = Summary::des_prefix(summary_blk.as_ref())?;
@@ -526,7 +526,7 @@ impl ImageLayerInner {
///
struct ImageLayerWriterInner {
conf: &'static PageServerConf,
path: Utf8PathBuf,
path: PathBuf,
timeline_id: TimelineId,
tenant_id: TenantId,
key_range: Range<Key>,
@@ -558,7 +558,7 @@ impl ImageLayerWriterInner {
lsn,
},
);
info!("new image layer {path}");
info!("new image layer {}", path.display());
let mut file = VirtualFile::open_with_options(
&path,
std::fs::OpenOptions::new().write(true).create_new(true),
@@ -685,7 +685,7 @@ impl ImageLayerWriterInner {
);
std::fs::rename(self.path, final_path)?;
trace!("created image layer {}", layer.path());
trace!("created image layer {}", layer.path().display());
Ok(layer)
}

View File

@@ -8,9 +8,9 @@ use crate::tenant::remote_timeline_client::index::LayerFileMetadata;
use crate::tenant::storage_layer::{Layer, ValueReconstructResult, ValueReconstructState};
use crate::tenant::timeline::layer_manager::LayerManager;
use anyhow::{bail, Result};
use camino::Utf8PathBuf;
use pageserver_api::models::HistoricLayerInfo;
use std::ops::Range;
use std::path::PathBuf;
use std::sync::Arc;
use utils::{
@@ -92,7 +92,7 @@ impl AsLayerDesc for RemoteLayer {
}
impl PersistentLayer for RemoteLayer {
fn local_path(&self) -> Option<Utf8PathBuf> {
fn local_path(&self) -> Option<PathBuf> {
None
}

View File

@@ -9,7 +9,6 @@ mod walreceiver;
use anyhow::{anyhow, bail, ensure, Context, Result};
use bytes::Bytes;
use camino::{Utf8Path, Utf8PathBuf};
use fail::fail_point;
use futures::StreamExt;
use itertools::Itertools;
@@ -30,6 +29,7 @@ use utils::id::TenantTimelineId;
use std::cmp::{max, min, Ordering};
use std::collections::{BinaryHeap, HashMap, HashSet};
use std::ops::{Deref, Range};
use std::path::{Path, PathBuf};
use std::pin::pin;
use std::sync::atomic::Ordering as AtomicOrdering;
use std::sync::{Arc, Mutex, RwLock, Weak};
@@ -56,7 +56,7 @@ use crate::config::PageServerConf;
use crate::keyspace::{KeyPartitioning, KeySpace, KeySpaceRandomAccum};
use crate::metrics::{
TimelineMetrics, MATERIALIZED_PAGE_CACHE_HIT, MATERIALIZED_PAGE_CACHE_HIT_DIRECT,
UNEXPECTED_ONDEMAND_DOWNLOADS,
RECONSTRUCT_TIME, UNEXPECTED_ONDEMAND_DOWNLOADS,
};
use crate::pgdatadir_mapping::LsnForTimestamp;
use crate::pgdatadir_mapping::{is_rel_fsm_block_key, is_rel_vm_block_key};
@@ -91,12 +91,12 @@ use self::logical_size::LogicalSize;
use self::walreceiver::{WalReceiver, WalReceiverConf};
use super::config::TenantConf;
use super::debug_assert_current_span_has_tenant_and_timeline_id;
use super::remote_timeline_client::index::IndexPart;
use super::remote_timeline_client::RemoteTimelineClient;
use super::storage_layer::{
AsLayerDesc, DeltaLayer, ImageLayer, LayerAccessStatsReset, PersistentLayerDesc,
};
use super::{debug_assert_current_span_has_tenant_and_timeline_id, AttachedTenantConf};
#[derive(Debug, PartialEq, Eq, Clone, Copy)]
pub(super) enum FlushLoopState {
@@ -149,7 +149,7 @@ pub struct TimelineResources {
pub struct Timeline {
conf: &'static PageServerConf,
tenant_conf: Arc<RwLock<AttachedTenantConf>>,
tenant_conf: Arc<RwLock<TenantConfOpt>>,
myself: Weak<Self>,
@@ -158,9 +158,6 @@ pub struct Timeline {
/// The generation of the tenant that instantiated us: this is used for safety when writing remote objects.
/// Never changes for the lifetime of this [`Timeline`] object.
///
/// This duplicates the generation stored in LocationConf, but that structure is mutable:
/// this copy enforces the invariant that generatio doesn't change during a Tenant's lifetime.
generation: Generation,
pub pg_version: u32,
@@ -499,39 +496,13 @@ impl Timeline {
};
let timer = crate::metrics::GET_RECONSTRUCT_DATA_TIME.start_timer();
let path = self
.get_reconstruct_data(key, lsn, &mut reconstruct_state, ctx)
self.get_reconstruct_data(key, lsn, &mut reconstruct_state, ctx)
.await?;
timer.stop_and_record();
let start = Instant::now();
let res = self.reconstruct_value(key, lsn, reconstruct_state, ctx).await;
let elapsed = start.elapsed();
crate::metrics::RECONSTRUCT_TIME
.for_result(&res)
.observe(elapsed.as_secs_f64());
if cfg!(feature = "testing") && res.is_err() {
// it can only be walredo issue
use std::fmt::Write;
let mut msg = String::new();
path.into_iter().for_each(|(res, cont_lsn, layer)| {
writeln!(
msg,
"- layer traversal: result {res:?}, cont_lsn {cont_lsn}, layer: {}",
layer(),
)
.expect("string grows")
});
// this is to rule out or provide evidence that we could in some cases read a duplicate
// walrecord
tracing::info!("walredo failed, path:\n{msg}");
}
res
RECONSTRUCT_TIME
.observe_closure_duration(|| self.reconstruct_value(key, lsn, reconstruct_state))
.await
}
/// Get last or prev record separately. Same as get_last_record_rlsn().last/prev.
@@ -1381,42 +1352,42 @@ const REPARTITION_FREQ_IN_CHECKPOINT_DISTANCE: u64 = 10;
// Private functions
impl Timeline {
fn get_checkpoint_distance(&self) -> u64 {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.checkpoint_distance
.unwrap_or(self.conf.default_tenant_conf.checkpoint_distance)
}
fn get_checkpoint_timeout(&self) -> Duration {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.checkpoint_timeout
.unwrap_or(self.conf.default_tenant_conf.checkpoint_timeout)
}
fn get_compaction_target_size(&self) -> u64 {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.compaction_target_size
.unwrap_or(self.conf.default_tenant_conf.compaction_target_size)
}
fn get_compaction_threshold(&self) -> usize {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.compaction_threshold
.unwrap_or(self.conf.default_tenant_conf.compaction_threshold)
}
fn get_image_creation_threshold(&self) -> usize {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.image_creation_threshold
.unwrap_or(self.conf.default_tenant_conf.image_creation_threshold)
}
fn get_eviction_policy(&self) -> EvictionPolicy {
let tenant_conf = self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.eviction_policy
.unwrap_or(self.conf.default_tenant_conf.eviction_policy)
@@ -1432,7 +1403,7 @@ impl Timeline {
}
fn get_gc_feedback(&self) -> bool {
let tenant_conf = &self.tenant_conf.read().unwrap().tenant_conf;
let tenant_conf = self.tenant_conf.read().unwrap();
tenant_conf
.gc_feedback
.unwrap_or(self.conf.default_tenant_conf.gc_feedback)
@@ -1445,7 +1416,7 @@ impl Timeline {
// The threshold is embedded in the metric. So, we need to update it.
{
let new_threshold = Self::get_evictions_low_residence_duration_metric_threshold(
&self.tenant_conf.read().unwrap().tenant_conf,
&self.tenant_conf.read().unwrap(),
&self.conf.default_tenant_conf,
);
let tenant_id_str = self.tenant_id.to_string();
@@ -1464,7 +1435,7 @@ impl Timeline {
#[allow(clippy::too_many_arguments)]
pub(super) fn new(
conf: &'static PageServerConf,
tenant_conf: Arc<RwLock<AttachedTenantConf>>,
tenant_conf: Arc<RwLock<TenantConfOpt>>,
metadata: &TimelineMetadata,
ancestor: Option<Arc<Timeline>>,
timeline_id: TimelineId,
@@ -1487,7 +1458,7 @@ impl Timeline {
let evictions_low_residence_duration_metric_threshold =
Self::get_evictions_low_residence_duration_metric_threshold(
&tenant_conf_guard.tenant_conf,
&tenant_conf_guard,
&conf.default_tenant_conf,
);
drop(tenant_conf_guard);
@@ -1652,15 +1623,12 @@ impl Timeline {
let tenant_conf_guard = self.tenant_conf.read().unwrap();
let wal_connect_timeout = tenant_conf_guard
.tenant_conf
.walreceiver_connect_timeout
.unwrap_or(self.conf.default_tenant_conf.walreceiver_connect_timeout);
let lagging_wal_timeout = tenant_conf_guard
.tenant_conf
.lagging_wal_timeout
.unwrap_or(self.conf.default_tenant_conf.lagging_wal_timeout);
let max_lsn_wal_lag = tenant_conf_guard
.tenant_conf
.max_lsn_wal_lag
.unwrap_or(self.conf.default_tenant_conf.max_lsn_wal_lag);
drop(tenant_conf_guard);
@@ -1742,7 +1710,7 @@ impl Timeline {
Discovered::Temporary(name) => (name, "temporary timeline file"),
Discovered::TemporaryDownload(name) => (name, "temporary download"),
};
path.push(Utf8Path::new(&name));
path.push(name);
init::cleanup(&path, kind)?;
path.pop();
}
@@ -2223,10 +2191,10 @@ impl TraversalLayerExt for Arc<dyn PersistentLayer> {
let timeline_id = self.layer_desc().timeline_id;
match self.local_path() {
Some(local_path) => {
debug_assert!(local_path.to_string().contains(&format!("{}", timeline_id)),
debug_assert!(local_path.to_str().unwrap().contains(&format!("{}", timeline_id)),
"need timeline ID to uniquely identify the layer when traversal crosses ancestor boundary",
);
format!("{local_path}")
format!("{}", local_path.display())
}
None => {
format!("remote {}/{self}", timeline_id)
@@ -2256,7 +2224,7 @@ impl Timeline {
request_lsn: Lsn,
reconstruct_state: &mut ValueReconstructState,
ctx: &RequestContext,
) -> Result<Vec<TraversalPathItem>, PageReconstructError> {
) -> Result<(), PageReconstructError> {
// Start from the current timeline.
let mut timeline_owned;
let mut timeline = self;
@@ -2287,12 +2255,12 @@ impl Timeline {
// The function should have updated 'state'
//info!("CALLED for {} at {}: {:?} with {} records, cached {}", key, cont_lsn, result, reconstruct_state.records.len(), cached_lsn);
match result {
ValueReconstructResult::Complete => return Ok(traversal_path),
ValueReconstructResult::Complete => return Ok(()),
ValueReconstructResult::Continue => {
// If we reached an earlier cached page image, we're done.
if cont_lsn == cached_lsn + 1 {
MATERIALIZED_PAGE_CACHE_HIT.inc_by(1);
return Ok(traversal_path);
return Ok(());
}
if prev_lsn <= cont_lsn {
// Didn't make any progress in last iteration. Error out to avoid
@@ -3728,11 +3696,6 @@ impl Timeline {
});
writer.as_mut().unwrap().put_value(key, lsn, value).await?;
if !new_layers.is_empty() {
fail_point!("after-timeline-compacted-first-L1");
}
prev_key = Some(key);
}
if let Some(writer) = writer {
@@ -3754,7 +3717,7 @@ impl Timeline {
);
}
}
let mut layer_paths: Vec<Utf8PathBuf> = new_layers.iter().map(|l| l.path()).collect();
let mut layer_paths: Vec<PathBuf> = new_layers.iter().map(|l| l.path()).collect();
// Fsync all the layer files and directory using multiple threads to
// minimize latency.
@@ -3864,7 +3827,10 @@ impl Timeline {
let new_delta_path = l.path();
let metadata = new_delta_path.metadata().with_context(|| {
format!("read file metadata for new created layer {new_delta_path}")
format!(
"read file metadata for new created layer {}",
new_delta_path.display()
)
})?;
if let Some(remote_client) = &self.remote_client {
@@ -3887,7 +3853,6 @@ impl Timeline {
);
let l = l as Arc<dyn PersistentLayer>;
if guard.contains(&l) {
tracing::error!(layer=%l, "duplicated L1 layer");
duplicated_layers.insert(l.layer_desc().key());
} else {
if LayerMap::is_l0(l.layer_desc()) {
@@ -4279,7 +4244,6 @@ impl Timeline {
key: Key,
request_lsn: Lsn,
mut data: ValueReconstructState,
ctx: &RequestContext,
) -> Result<Bytes, PageReconstructError> {
// Perform WAL redo if needed
data.records.reverse();
@@ -4343,7 +4307,6 @@ impl Timeline {
key,
last_rec_lsn,
&img,
ctx,
)
.await
.context("Materialized page memoization failed")
@@ -4801,10 +4764,11 @@ fn is_send() {
/// Add a suffix to a layer file's name: .{num}.old
/// Uses the first available num (starts at 0)
fn rename_to_backup(path: &Utf8Path) -> anyhow::Result<()> {
fn rename_to_backup(path: &Path) -> anyhow::Result<()> {
let filename = path
.file_name()
.ok_or_else(|| anyhow!("Path {path} don't have a file name"))?;
.ok_or_else(|| anyhow!("Path {} don't have a file name", path.display()))?
.to_string_lossy();
let mut new_path = path.to_owned();
for i in 0u32.. {

View File

@@ -16,7 +16,7 @@
use std::{
collections::HashMap,
ops::ControlFlow,
sync::{Arc, Mutex},
sync::Arc,
time::{Duration, SystemTime},
};
@@ -25,7 +25,7 @@ use tokio_util::sync::CancellationToken;
use tracing::{debug, error, info, info_span, instrument, warn, Instrument};
use crate::{
context::{DownloadBehavior, RequestContext, RequestContextBuilder},
context::{DownloadBehavior, RequestContext},
task_mgr::{self, TaskKind, BACKGROUND_RUNTIME},
tenant::{
config::{EvictionPolicy, EvictionPolicyLayerAccessThreshold},
@@ -397,14 +397,9 @@ impl Timeline {
}
}
let permit = crate::page_cache::get().get_permit().await;
let ctx = RequestContextBuilder::extend(ctx)
.page_cache_permit(permit)
.build();
// imitiate repartiting on first compactation
if let Err(e) = self
.collect_keyspace(lsn, &ctx)
.collect_keyspace(lsn, ctx)
.instrument(info_span!("collect_keyspace"))
.await
{

View File

@@ -12,8 +12,7 @@ use crate::{
METADATA_FILE_NAME,
};
use anyhow::Context;
use camino::Utf8Path;
use std::{collections::HashMap, str::FromStr};
use std::{collections::HashMap, ffi::OsString, path::Path, str::FromStr};
use utils::lsn::Lsn;
/// Identified files in the timeline directory.
@@ -21,43 +20,46 @@ pub(super) enum Discovered {
/// The only one we care about
Layer(LayerFileName, u64),
/// Old ephmeral files from previous launches, should be removed
Ephemeral(String),
Ephemeral(OsString),
/// Old temporary timeline files, unsure what these really are, should be removed
Temporary(String),
Temporary(OsString),
/// Temporary on-demand download files, should be removed
TemporaryDownload(String),
TemporaryDownload(OsString),
/// "metadata" file we persist locally and include in `index_part.json`
Metadata,
/// Backup file from previously future layers
IgnoredBackup,
/// Unrecognized, warn about these
Unknown(String),
Unknown(OsString),
}
/// Scans the timeline directory for interesting files.
pub(super) fn scan_timeline_dir(path: &Utf8Path) -> anyhow::Result<Vec<Discovered>> {
pub(super) fn scan_timeline_dir(path: &Path) -> anyhow::Result<Vec<Discovered>> {
let mut ret = Vec::new();
for direntry in path.read_dir_utf8()? {
for direntry in std::fs::read_dir(path)? {
let direntry = direntry?;
let file_name = direntry.file_name().to_string();
let direntry_path = direntry.path();
let file_name = direntry.file_name();
let discovered = match LayerFileName::from_str(&file_name) {
let fname = file_name.to_string_lossy();
let discovered = match LayerFileName::from_str(&fname) {
Ok(file_name) => {
let file_size = direntry.metadata()?.len();
Discovered::Layer(file_name, file_size)
}
Err(_) => {
if file_name == METADATA_FILE_NAME {
if fname == METADATA_FILE_NAME {
Discovered::Metadata
} else if file_name.ends_with(".old") {
} else if fname.ends_with(".old") {
// ignore these
Discovered::IgnoredBackup
} else if remote_timeline_client::is_temp_download_file(direntry.path()) {
} else if remote_timeline_client::is_temp_download_file(&direntry_path) {
Discovered::TemporaryDownload(file_name)
} else if is_ephemeral_file(&file_name) {
} else if is_ephemeral_file(&fname) {
Discovered::Ephemeral(file_name)
} else if is_temporary(direntry.path()) {
} else if is_temporary(&direntry_path) {
Discovered::Temporary(file_name)
} else {
Discovered::Unknown(file_name)
@@ -160,14 +162,15 @@ pub(super) fn reconcile(
.collect::<Vec<_>>()
}
pub(super) fn cleanup(path: &Utf8Path, kind: &str) -> anyhow::Result<()> {
pub(super) fn cleanup(path: &Path, kind: &str) -> anyhow::Result<()> {
let file_name = path.file_name().expect("must be file path");
tracing::debug!(kind, ?file_name, "cleaning up");
std::fs::remove_file(path).with_context(|| format!("failed to remove {kind} at {path}"))
std::fs::remove_file(path)
.with_context(|| format!("failed to remove {kind} at {}", path.display()))
}
pub(super) fn cleanup_local_file_for_remote(
path: &Utf8Path,
path: &Path,
local: &LayerFileMetadata,
remote: &LayerFileMetadata,
) -> anyhow::Result<()> {
@@ -179,7 +182,8 @@ pub(super) fn cleanup_local_file_for_remote(
if let Err(err) = crate::tenant::timeline::rename_to_backup(path) {
assert!(
path.exists(),
"we would leave the local_layer without a file if this does not hold: {path}",
"we would leave the local_layer without a file if this does not hold: {}",
path.display()
);
Err(err)
} else {
@@ -188,7 +192,7 @@ pub(super) fn cleanup_local_file_for_remote(
}
pub(super) fn cleanup_future_layer(
path: &Utf8Path,
path: &Path,
name: &LayerFileName,
disk_consistent_lsn: Lsn,
) -> anyhow::Result<()> {

View File

@@ -1,7 +1,6 @@
use std::{collections::hash_map::Entry, fs, sync::Arc};
use std::{collections::hash_map::Entry, fs, path::PathBuf, sync::Arc};
use anyhow::Context;
use camino::Utf8PathBuf;
use tracing::{error, info, info_span, warn};
use utils::{crashsafe, fs_ext, id::TimelineId, lsn::Lsn};
@@ -156,12 +155,12 @@ pub(crate) fn cleanup_timeline_directory(uninit_mark: TimelineUninitMark) {
#[must_use]
pub(crate) struct TimelineUninitMark {
uninit_mark_deleted: bool,
uninit_mark_path: Utf8PathBuf,
pub(crate) timeline_path: Utf8PathBuf,
uninit_mark_path: PathBuf,
pub(crate) timeline_path: PathBuf,
}
impl TimelineUninitMark {
pub(crate) fn new(uninit_mark_path: Utf8PathBuf, timeline_path: Utf8PathBuf) -> Self {
pub(crate) fn new(uninit_mark_path: PathBuf, timeline_path: PathBuf) -> Self {
Self {
uninit_mark_deleted: false,
uninit_mark_path,
@@ -198,13 +197,14 @@ impl Drop for TimelineUninitMark {
if self.timeline_path.exists() {
error!(
"Uninit mark {} is not removed, timeline {} stays uninitialized",
self.uninit_mark_path, self.timeline_path
self.uninit_mark_path.display(),
self.timeline_path.display()
)
} else {
// unblock later timeline creation attempts
warn!(
"Removing intermediate uninit mark file {}",
self.uninit_mark_path
self.uninit_mark_path.display()
);
if let Err(e) = self.delete_mark_file_if_present() {
error!("Failed to remove the uninit mark file: {e}")

View File

@@ -253,7 +253,7 @@ impl std::fmt::Display for UploadOp {
write!(f, "UploadMetadata(lsn: {})", lsn)
}
UploadOp::Delete(delete) => {
write!(f, "Delete({} layers)", delete.layers.len())
write!(f, "Delete({} layers)", delete.layers.len(),)
}
UploadOp::Barrier(_) => write!(f, "Barrier"),
}

View File

@@ -1,8 +1,8 @@
use bytes::Bytes;
use camino::Utf8PathBuf;
use std::{
fs::{create_dir_all, File},
io::{BufWriter, Write},
path::PathBuf,
};
pub struct Tracer {
@@ -16,7 +16,7 @@ impl Drop for Tracer {
}
impl Tracer {
pub fn new(path: Utf8PathBuf) -> Self {
pub fn new(path: PathBuf) -> Self {
let parent = path.parent().expect("failed to parse parent path");
create_dir_all(parent).expect("failed to create trace dir");

View File

@@ -12,11 +12,11 @@
//!
use crate::metrics::{StorageIoOperation, STORAGE_IO_SIZE, STORAGE_IO_TIME_METRIC};
use crate::tenant::TENANTS_SEGMENT_NAME;
use camino::{Utf8Path, Utf8PathBuf};
use once_cell::sync::OnceCell;
use std::fs::{self, File, OpenOptions};
use std::io::{Error, ErrorKind, Seek, SeekFrom};
use std::os::unix::fs::FileExt;
use std::path::{Path, PathBuf};
use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
use std::sync::{RwLock, RwLockWriteGuard};
@@ -51,7 +51,7 @@ pub struct VirtualFile {
/// if a new file is created, we only pass the create flag when it's initially
/// opened, in the VirtualFile::create() function, and strip the flag before
/// storing it here.
pub path: Utf8PathBuf,
pub path: PathBuf,
open_options: OpenOptions,
// These are strings becase we only use them for metrics, and those expect strings.
@@ -177,19 +177,19 @@ impl OpenFiles {
pub enum CrashsafeOverwriteError {
#[error("final path has no parent dir")]
FinalPathHasNoParentDir,
#[error("remove tempfile")]
#[error("remove tempfile: {0}")]
RemovePreviousTempfile(#[source] std::io::Error),
#[error("create tempfile")]
#[error("create tempfile: {0}")]
CreateTempfile(#[source] std::io::Error),
#[error("write tempfile")]
#[error("write tempfile: {0}")]
WriteContents(#[source] std::io::Error),
#[error("sync tempfile")]
#[error("sync tempfile: {0}")]
SyncTempfile(#[source] std::io::Error),
#[error("rename tempfile to final path")]
#[error("rename tempfile to final path: {0}")]
RenameTempfileToFinalPath(#[source] std::io::Error),
#[error("open final path parent dir")]
#[error("open final path parent dir: {0}")]
OpenFinalPathParentDir(#[source] std::io::Error),
#[error("sync final path parent dir")]
#[error("sync final path parent dir: {0}")]
SyncFinalPathParentDir(#[source] std::io::Error),
}
impl CrashsafeOverwriteError {
@@ -210,13 +210,13 @@ impl CrashsafeOverwriteError {
impl VirtualFile {
/// Open a file in read-only mode. Like File::open.
pub async fn open(path: &Utf8Path) -> Result<VirtualFile, std::io::Error> {
pub async fn open(path: &Path) -> Result<VirtualFile, std::io::Error> {
Self::open_with_options(path, OpenOptions::new().read(true)).await
}
/// Create a new file for writing. If the file exists, it will be truncated.
/// Like File::create.
pub async fn create(path: &Utf8Path) -> Result<VirtualFile, std::io::Error> {
pub async fn create(path: &Path) -> Result<VirtualFile, std::io::Error> {
Self::open_with_options(
path,
OpenOptions::new().write(true).create(true).truncate(true),
@@ -230,10 +230,10 @@ impl VirtualFile {
/// they will be applied also when the file is subsequently re-opened, not only
/// on the first time. Make sure that's sane!
pub async fn open_with_options(
path: &Utf8Path,
path: &Path,
open_options: &OpenOptions,
) -> Result<VirtualFile, std::io::Error> {
let path_str = path.to_string();
let path_str = path.to_string_lossy();
let parts = path_str.split('/').collect::<Vec<&str>>();
let tenant_id;
let timeline_id;
@@ -281,8 +281,8 @@ impl VirtualFile {
/// atomic, a crash during the write operation will never leave behind a
/// partially written file.
pub async fn crashsafe_overwrite(
final_path: &Utf8Path,
tmp_path: &Utf8Path,
final_path: &Path,
tmp_path: &Path,
content: &[u8],
) -> Result<(), CrashsafeOverwriteError> {
let Some(final_path_parent) = final_path.parent() else {
@@ -544,7 +544,7 @@ impl VirtualFile {
pub(crate) async fn read_blk(
&self,
blknum: u32,
) -> Result<crate::tenant::block_io::BlockLease<'_, '_>, std::io::Error> {
) -> Result<crate::tenant::block_io::BlockLease<'_>, std::io::Error> {
use crate::page_cache::PAGE_SZ;
let mut buf = [0; PAGE_SZ];
self.read_exact_at(&mut buf, blknum as u64 * (PAGE_SZ as u64))
@@ -734,7 +734,7 @@ mod tests {
async fn test_files<OF, FT>(testname: &str, openfunc: OF) -> Result<(), Error>
where
OF: Fn(Utf8PathBuf, OpenOptions) -> FT,
OF: Fn(PathBuf, OpenOptions) -> FT,
FT: Future<Output = Result<MaybeVirtualFile, std::io::Error>>,
{
let testdir = crate::config::PageServerConf::test_repo_dir(testname);

View File

@@ -38,9 +38,6 @@ use tracing::*;
use utils::crashsafe::path_with_suffix_extension;
use utils::{bin_ser::BeSer, id::TenantId, lsn::Lsn, nonblock::set_nonblock};
#[cfg(feature = "testing")]
use std::sync::atomic::{AtomicUsize, Ordering};
use crate::metrics::{
WAL_REDO_BYTES_HISTOGRAM, WAL_REDO_RECORDS_HISTOGRAM, WAL_REDO_RECORD_COUNTER, WAL_REDO_TIME,
WAL_REDO_WAIT_TIME,
@@ -116,9 +113,6 @@ struct ProcessOutput {
pub struct PostgresRedoManager {
tenant_id: TenantId,
conf: &'static PageServerConf,
/// Counter to separate same sized walredo inputs failing at the same millisecond.
#[cfg(feature = "testing")]
dump_sequence: AtomicUsize,
stdout: Mutex<Option<ProcessOutput>>,
stdin: Mutex<Option<ProcessInput>>,
@@ -230,8 +224,6 @@ impl PostgresRedoManager {
PostgresRedoManager {
tenant_id,
conf,
#[cfg(feature = "testing")]
dump_sequence: AtomicUsize::default(),
stdin: Mutex::new(None),
stdout: Mutex::new(None),
stderr: Mutex::new(None),
@@ -298,27 +290,25 @@ impl PostgresRedoManager {
WAL_REDO_BYTES_HISTOGRAM.observe(nbytes as f64);
debug!(
"postgres applied {} WAL records ({} bytes) in {} us to reconstruct page image at LSN {}",
len,
nbytes,
duration.as_micros(),
lsn
);
"postgres applied {} WAL records ({} bytes) in {} us to reconstruct page image at LSN {}",
len,
nbytes,
duration.as_micros(),
lsn
);
// If something went wrong, don't try to reuse the process. Kill it, and
// next request will launch a new one.
if let Err(e) = result.as_ref() {
if result.is_err() {
error!(
n_attempts,
"error applying {} WAL records {}..{} ({} bytes) to base image with LSN {} to reconstruct page image at LSN {}: {}",
records.len(),
records.first().map(|p| p.0).unwrap_or(Lsn(0)),
records.last().map(|p| p.0).unwrap_or(Lsn(0)),
nbytes,
base_img_lsn,
lsn,
utils::error::report_compact_sources(e),
);
"error applying {} WAL records {}..{} ({} bytes) to base image with LSN {} to reconstruct page image at LSN {}",
records.len(),
records.first().map(|p| p.0).unwrap_or(Lsn(0)),
records.last().map(|p| p.0).unwrap_or(Lsn(0)),
nbytes,
base_img_lsn,
lsn
);
// self.stdin only holds stdin & stderr as_raw_fd().
// Dropping it as part of take() doesn't close them.
// The owning objects (ChildStdout and ChildStderr) are stored in
@@ -335,8 +325,6 @@ impl PostgresRedoManager {
if let Some(proc) = self.stdin.lock().unwrap().take() {
proc.child.kill_and_wait();
}
} else if n_attempts != 0 {
info!(n_attempts, "retried walredo succeeded");
}
n_attempts += 1;
if n_attempts > MAX_RETRY_ATTEMPTS || result.is_ok() {
@@ -754,7 +742,7 @@ impl PostgresRedoManager {
#[instrument(skip_all, fields(tenant_id=%self.tenant_id, pid=%input.as_ref().unwrap().child.id()))]
fn apply_wal_records(
&self,
input: MutexGuard<Option<ProcessInput>>,
mut input: MutexGuard<Option<ProcessInput>>,
tag: BufferTag,
base_img: &Option<Bytes>,
records: &[(Lsn, NeonWalRecord)],
@@ -791,23 +779,6 @@ impl PostgresRedoManager {
build_get_page_msg(tag, &mut writebuf);
WAL_REDO_RECORD_COUNTER.inc_by(records.len() as u64);
let res = self.apply_wal_records0(&writebuf, input, wal_redo_timeout);
if res.is_err() {
// not all of these can be caused by this particular input, however these are so rare
// in tests so capture all.
self.record_and_log(&writebuf);
}
res
}
fn apply_wal_records0(
&self,
writebuf: &[u8],
mut input: MutexGuard<Option<ProcessInput>>,
wal_redo_timeout: Duration,
) -> Result<Bytes, std::io::Error> {
let proc = input.as_mut().unwrap();
let mut nwrite = 0usize;
let stdout_fd = proc.stdout_fd;
@@ -1013,38 +984,6 @@ impl PostgresRedoManager {
}
Ok(res)
}
#[cfg(feature = "testing")]
fn record_and_log(&self, writebuf: &[u8]) {
let millis = std::time::SystemTime::now()
.duration_since(std::time::SystemTime::UNIX_EPOCH)
.unwrap()
.as_millis();
let seq = self.dump_sequence.fetch_add(1, Ordering::Relaxed);
// these files will be collected to an allure report
let filename = format!("walredo-{millis}-{}-{seq}.walredo", writebuf.len());
let path = self.conf.tenant_path(&self.tenant_id).join(&filename);
let res = std::fs::OpenOptions::new()
.write(true)
.create_new(true)
.read(true)
.open(path)
.and_then(|mut f| f.write_all(writebuf));
// trip up allowed_errors
if let Err(e) = res {
tracing::error!(target=%filename, length=writebuf.len(), "failed to write out the walredo errored input: {e}");
} else {
tracing::error!(filename, "erroring walredo input saved");
}
}
#[cfg(not(feature = "testing"))]
fn record_and_log(&self, _: &[u8]) {}
}
/// Wrapper type around `std::process::Child` which guarantees that the child
@@ -1278,13 +1217,13 @@ mod tests {
struct RedoHarness {
// underscored because unused, except for removal at drop
_repo_dir: camino_tempfile::Utf8TempDir,
_repo_dir: tempfile::TempDir,
manager: PostgresRedoManager,
}
impl RedoHarness {
fn new() -> anyhow::Result<Self> {
let repo_dir = camino_tempfile::tempdir()?;
let repo_dir = tempfile::tempdir()?;
let conf = PageServerConf::dummy_conf(repo_dir.path().to_path_buf());
let conf = Box::leak(Box::new(conf));
let tenant_id = TenantId::generate();

View File

@@ -7,12 +7,12 @@ OBJS = \
extension_server.o \
file_cache.o \
libpagestore.o \
libpqwalproposer.o \
neon.o \
neon_utils.o \
pagestore_smgr.o \
relsize_cache.o \
walproposer.o \
walproposer_utils.o \
walproposer_pg.o \
control_plane_connector.o
PG_CPPFLAGS = -I$(libpq_srcdir)

View File

@@ -741,13 +741,6 @@ NeonProcessUtility(
break;
case T_DropdbStmt:
HandleDropDb(castNode(DropdbStmt, parseTree));
/*
* We do this here to hack around the fact that Postgres performs the drop
* INSIDE of standard_ProcessUtility, which means that if we try to
* abort the drop normally it'll be too late. DROP DATABASE can't be inside
* of a transaction block anyway, so this should be fine to do.
*/
NeonXactCallback(XACT_EVENT_PRE_COMMIT, NULL);
break;
case T_CreateRoleStmt:
HandleCreateRole(castNode(CreateRoleStmt, parseTree));

View File

@@ -14,6 +14,7 @@
*/
#include <sys/file.h>
#include <sys/statvfs.h>
#include <unistd.h>
#include <fcntl.h>
@@ -37,6 +38,9 @@
#include "storage/fd.h"
#include "storage/pg_shmem.h"
#include "storage/buf_internals.h"
#include "storage/procsignal.h"
#include "postmaster/bgworker.h"
#include "postmaster/interrupt.h"
/*
* Local file cache is used to temporary store relations pages in local file system.
@@ -62,6 +66,9 @@
#define SIZE_MB_TO_CHUNKS(size) ((uint32)((size) * MB / BLCKSZ / BLOCKS_PER_CHUNK))
#define MAX_MONITOR_INTERVAL_USEC 1000000 /* 1 second */
#define MAX_DISK_WRITE_RATE 1000 /* MB/sec */
typedef struct FileCacheEntry
{
BufferTag key;
@@ -84,12 +91,14 @@ static int lfc_desc = 0;
static LWLockId lfc_lock;
static int lfc_max_size;
static int lfc_size_limit;
static int lfc_free_space_watermark;
static char* lfc_path;
static FileCacheControl* lfc_ctl;
static shmem_startup_hook_type prev_shmem_startup_hook;
#if PG_VERSION_NUM>=150000
static shmem_request_hook_type prev_shmem_request_hook;
#endif
static int lfc_shrinking_factor; /* power of two by which local cache size will be shrinked when lfc_free_space_watermark is reached */
void FileCacheMonitorMain(Datum main_arg);
@@ -245,6 +254,80 @@ lfc_change_limit_hook(int newval, void *extra)
LWLockRelease(lfc_lock);
}
/*
* Local file system state monitor check available free space.
* If it is lower than lfc_free_space_watermark then we shrink size of local cache
* but throwing away least recently accessed chunks.
* First time low space watermark is reached cache size is divided by two,
* second time by four,... Finally we remove all chunks from local cache.
*
* Please notice that we are not changing lfc_cache_size: it is used to be adjusted by autoscaler.
* We only throw away cached chunks but do not prevent from filling cache by new chunks.
*
* Interval of poooling cache state is calculated as minimal time needed to consume lfc_free_space_watermark
* disk space with maximal possible disk write speed (1Gb/sec). But not larger than 1 second.
* Calling statvfs each second should not add any noticeable overhead.
*/
void
FileCacheMonitorMain(Datum main_arg)
{
/*
* Choose file system state monitor interval so that space can not be exosted
* during this period but not longer than MAX_MONITOR_INTERVAL (10 sec)
*/
uint64 monitor_interval = Min(MAX_MONITOR_INTERVAL_USEC, lfc_free_space_watermark*MB/MAX_DISK_WRITE_RATE);
/* Establish signal handlers. */
pqsignal(SIGUSR1, procsignal_sigusr1_handler);
pqsignal(SIGHUP, SignalHandlerForConfigReload);
pqsignal(SIGTERM, SignalHandlerForShutdownRequest);
BackgroundWorkerUnblockSignals();
/* Periodically dump buffers until terminated. */
while (!ShutdownRequestPending)
{
if (lfc_size_limit != 0)
{
struct statvfs sfs;
if (statvfs(lfc_path, &sfs) < 0)
{
elog(WARNING, "Failed to obtain status of %s: %m", lfc_path);
}
else
{
if (sfs.f_bavail*sfs.f_bsize < lfc_free_space_watermark*MB)
{
if (lfc_shrinking_factor < 31) {
lfc_shrinking_factor += 1;
}
lfc_change_limit_hook(lfc_size_limit >> lfc_shrinking_factor, NULL);
}
else
lfc_shrinking_factor = 0; /* reset to initial value */
}
}
pg_usleep(monitor_interval);
}
}
static void
lfc_register_free_space_monitor(void)
{
BackgroundWorker bgw;
memset(&bgw, 0, sizeof(bgw));
bgw.bgw_flags = BGWORKER_SHMEM_ACCESS;
bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
snprintf(bgw.bgw_library_name, BGW_MAXLEN, "neon");
snprintf(bgw.bgw_function_name, BGW_MAXLEN, "FileCacheMonitorMain");
snprintf(bgw.bgw_name, BGW_MAXLEN, "Local free space monitor");
snprintf(bgw.bgw_type, BGW_MAXLEN, "Local free space monitor");
bgw.bgw_restart_time = 5;
bgw.bgw_notify_pid = 0;
bgw.bgw_main_arg = (Datum) 0;
RegisterBackgroundWorker(&bgw);
}
void
lfc_init(void)
{
@@ -281,6 +364,19 @@ lfc_init(void)
lfc_change_limit_hook,
NULL);
DefineCustomIntVariable("neon.free_space_watermark",
"Minimal free space in local file system after reaching which local file cache will be truncated",
NULL,
&lfc_free_space_watermark,
1024, /* 1GB */
0,
INT_MAX,
PGC_SIGHUP,
GUC_UNIT_MB,
NULL,
NULL,
NULL);
DefineCustomStringVariable("neon.file_cache_path",
"Path to local file cache (can be raw device)",
NULL,
@@ -295,6 +391,9 @@ lfc_init(void)
if (lfc_max_size == 0)
return;
if (lfc_free_space_watermark != 0)
lfc_register_free_space_monitor();
prev_shmem_startup_hook = shmem_startup_hook;
shmem_startup_hook = lfc_shmem_startup;
#if PG_VERSION_NUM>=150000

View File

@@ -30,7 +30,7 @@
#include "neon.h"
#include "walproposer.h"
#include "walproposer_utils.h"
#include "neon_utils.h"
#define PageStoreTrace DEBUG5

View File

@@ -1,424 +0,0 @@
#include "postgres.h"
#include "libpq-fe.h"
#include "neon.h"
#include "walproposer.h"
/* Header in walproposer.h -- Wrapper struct to abstract away the libpq connection */
struct WalProposerConn
{
PGconn *pg_conn;
bool is_nonblocking; /* whether the connection is non-blocking */
char *recvbuf; /* last received data from
* walprop_async_read */
};
/* Helper function */
static bool
ensure_nonblocking_status(WalProposerConn *conn, bool is_nonblocking)
{
/* If we're already correctly blocking or nonblocking, all good */
if (is_nonblocking == conn->is_nonblocking)
return true;
/* Otherwise, set it appropriately */
if (PQsetnonblocking(conn->pg_conn, is_nonblocking) == -1)
return false;
conn->is_nonblocking = is_nonblocking;
return true;
}
/* Exported function definitions */
char *
walprop_error_message(WalProposerConn *conn)
{
return PQerrorMessage(conn->pg_conn);
}
WalProposerConnStatusType
walprop_status(WalProposerConn *conn)
{
switch (PQstatus(conn->pg_conn))
{
case CONNECTION_OK:
return WP_CONNECTION_OK;
case CONNECTION_BAD:
return WP_CONNECTION_BAD;
default:
return WP_CONNECTION_IN_PROGRESS;
}
}
WalProposerConn *
walprop_connect_start(char *conninfo, char *password)
{
WalProposerConn *conn;
PGconn *pg_conn;
const char *keywords[3];
const char *values[3];
int n;
/*
* Connect using the given connection string. If the
* NEON_AUTH_TOKEN environment variable was set, use that as
* the password.
*
* The connection options are parsed in the order they're given, so
* when we set the password before the connection string, the
* connection string can override the password from the env variable.
* Seems useful, although we don't currently use that capability
* anywhere.
*/
n = 0;
if (password)
{
keywords[n] = "password";
values[n] = password;
n++;
}
keywords[n] = "dbname";
values[n] = conninfo;
n++;
keywords[n] = NULL;
values[n] = NULL;
n++;
pg_conn = PQconnectStartParams(keywords, values, 1);
/*
* Allocation of a PQconn can fail, and will return NULL. We want to fully
* replicate the behavior of PQconnectStart here.
*/
if (!pg_conn)
return NULL;
/*
* And in theory this allocation can fail as well, but it's incredibly
* unlikely if we just successfully allocated a PGconn.
*
* palloc will exit on failure though, so there's not much we could do if
* it *did* fail.
*/
conn = palloc(sizeof(WalProposerConn));
conn->pg_conn = pg_conn;
conn->is_nonblocking = false; /* connections always start in blocking
* mode */
conn->recvbuf = NULL;
return conn;
}
WalProposerConnectPollStatusType
walprop_connect_poll(WalProposerConn *conn)
{
WalProposerConnectPollStatusType return_val;
switch (PQconnectPoll(conn->pg_conn))
{
case PGRES_POLLING_FAILED:
return_val = WP_CONN_POLLING_FAILED;
break;
case PGRES_POLLING_READING:
return_val = WP_CONN_POLLING_READING;
break;
case PGRES_POLLING_WRITING:
return_val = WP_CONN_POLLING_WRITING;
break;
case PGRES_POLLING_OK:
return_val = WP_CONN_POLLING_OK;
break;
/*
* There's a comment at its source about this constant being
* unused. We'll expect it's never returned.
*/
case PGRES_POLLING_ACTIVE:
elog(FATAL, "Unexpected PGRES_POLLING_ACTIVE returned from PQconnectPoll");
/*
* This return is never actually reached, but it's here to make
* the compiler happy
*/
return WP_CONN_POLLING_FAILED;
default:
Assert(false);
return_val = WP_CONN_POLLING_FAILED; /* keep the compiler quiet */
}
return return_val;
}
bool
walprop_send_query(WalProposerConn *conn, char *query)
{
/*
* We need to be in blocking mode for sending the query to run without
* requiring a call to PQflush
*/
if (!ensure_nonblocking_status(conn, false))
return false;
/* PQsendQuery returns 1 on success, 0 on failure */
if (!PQsendQuery(conn->pg_conn, query))
return false;
return true;
}
WalProposerExecStatusType
walprop_get_query_result(WalProposerConn *conn)
{
PGresult *result;
WalProposerExecStatusType return_val;
/* Marker variable if we need to log an unexpected success result */
char *unexpected_success = NULL;
/* Consume any input that we might be missing */
if (!PQconsumeInput(conn->pg_conn))
return WP_EXEC_FAILED;
if (PQisBusy(conn->pg_conn))
return WP_EXEC_NEEDS_INPUT;
result = PQgetResult(conn->pg_conn);
/*
* PQgetResult returns NULL only if getting the result was successful &
* there's no more of the result to get.
*/
if (!result)
{
elog(WARNING, "[libpqwalproposer] Unexpected successful end of command results");
return WP_EXEC_UNEXPECTED_SUCCESS;
}
/* Helper macro to reduce boilerplate */
#define UNEXPECTED_SUCCESS(msg) \
return_val = WP_EXEC_UNEXPECTED_SUCCESS; \
unexpected_success = msg; \
break;
switch (PQresultStatus(result))
{
/* "true" success case */
case PGRES_COPY_BOTH:
return_val = WP_EXEC_SUCCESS_COPYBOTH;
break;
/* Unexpected success case */
case PGRES_EMPTY_QUERY:
UNEXPECTED_SUCCESS("empty query return");
case PGRES_COMMAND_OK:
UNEXPECTED_SUCCESS("data-less command end");
case PGRES_TUPLES_OK:
UNEXPECTED_SUCCESS("tuples return");
case PGRES_COPY_OUT:
UNEXPECTED_SUCCESS("'Copy Out' response");
case PGRES_COPY_IN:
UNEXPECTED_SUCCESS("'Copy In' response");
case PGRES_SINGLE_TUPLE:
UNEXPECTED_SUCCESS("single tuple return");
case PGRES_PIPELINE_SYNC:
UNEXPECTED_SUCCESS("pipeline sync point");
/* Failure cases */
case PGRES_BAD_RESPONSE:
case PGRES_NONFATAL_ERROR:
case PGRES_FATAL_ERROR:
case PGRES_PIPELINE_ABORTED:
return_val = WP_EXEC_FAILED;
break;
default:
Assert(false);
return_val = WP_EXEC_FAILED; /* keep the compiler quiet */
}
if (unexpected_success)
elog(WARNING, "[libpqwalproposer] Unexpected successful %s", unexpected_success);
return return_val;
}
pgsocket
walprop_socket(WalProposerConn *conn)
{
return PQsocket(conn->pg_conn);
}
int
walprop_flush(WalProposerConn *conn)
{
return (PQflush(conn->pg_conn));
}
void
walprop_finish(WalProposerConn *conn)
{
if (conn->recvbuf != NULL)
PQfreemem(conn->recvbuf);
PQfinish(conn->pg_conn);
pfree(conn);
}
/*
* Receive a message from the safekeeper.
*
* On success, the data is placed in *buf. It is valid until the next call
* to this function.
*/
PGAsyncReadResult
walprop_async_read(WalProposerConn *conn, char **buf, int *amount)
{
int result;
if (conn->recvbuf != NULL)
{
PQfreemem(conn->recvbuf);
conn->recvbuf = NULL;
}
/* Call PQconsumeInput so that we have the data we need */
if (!PQconsumeInput(conn->pg_conn))
{
*amount = 0;
*buf = NULL;
return PG_ASYNC_READ_FAIL;
}
/*
* The docs for PQgetCopyData list the return values as: 0 if the copy is
* still in progress, but no "complete row" is available -1 if the copy is
* done -2 if an error occurred (> 0) if it was successful; that value is
* the amount transferred.
*
* The protocol we use between walproposer and safekeeper means that we
* *usually* wouldn't expect to see that the copy is done, but this can
* sometimes be triggered by the server returning an ErrorResponse (which
* also happens to have the effect that the copy is done).
*/
switch (result = PQgetCopyData(conn->pg_conn, &conn->recvbuf, true))
{
case 0:
*amount = 0;
*buf = NULL;
return PG_ASYNC_READ_TRY_AGAIN;
case -1:
{
/*
* If we get -1, it's probably because of a server error; the
* safekeeper won't normally send a CopyDone message.
*
* We can check PQgetResult to make sure that the server
* failed; it'll always result in PGRES_FATAL_ERROR
*/
ExecStatusType status = PQresultStatus(PQgetResult(conn->pg_conn));
if (status != PGRES_FATAL_ERROR)
elog(FATAL, "unexpected result status %d after failed PQgetCopyData", status);
/*
* If there was actually an error, it'll be properly reported
* by calls to PQerrorMessage -- we don't have to do anything
* else
*/
*amount = 0;
*buf = NULL;
return PG_ASYNC_READ_FAIL;
}
case -2:
*amount = 0;
*buf = NULL;
return PG_ASYNC_READ_FAIL;
default:
/* Positive values indicate the size of the returned result */
*amount = result;
*buf = conn->recvbuf;
return PG_ASYNC_READ_SUCCESS;
}
}
PGAsyncWriteResult
walprop_async_write(WalProposerConn *conn, void const *buf, size_t size)
{
int result;
/* If we aren't in non-blocking mode, switch to it. */
if (!ensure_nonblocking_status(conn, true))
return PG_ASYNC_WRITE_FAIL;
/*
* The docs for PQputcopyData list the return values as: 1 if the data was
* queued, 0 if it was not queued because of full buffers, or -1 if an
* error occurred
*/
result = PQputCopyData(conn->pg_conn, buf, size);
/*
* We won't get a result of zero because walproposer always empties the
* connection's buffers before sending more
*/
Assert(result != 0);
switch (result)
{
case 1:
/* good -- continue */
break;
case -1:
return PG_ASYNC_WRITE_FAIL;
default:
elog(FATAL, "invalid return %d from PQputCopyData", result);
}
/*
* After queueing the data, we still need to flush to get it to send. This
* might take multiple tries, but we don't want to wait around until it's
* done.
*
* PQflush has the following returns (directly quoting the docs): 0 if
* sucessful, 1 if it was unable to send all the data in the send queue
* yet -1 if it failed for some reason
*/
switch (result = PQflush(conn->pg_conn))
{
case 0:
return PG_ASYNC_WRITE_SUCCESS;
case 1:
return PG_ASYNC_WRITE_TRY_FLUSH;
case -1:
return PG_ASYNC_WRITE_FAIL;
default:
elog(FATAL, "invalid return %d from PQflush", result);
}
}
/*
* This function is very similar to walprop_async_write. For more
* information, refer to the comments there.
*/
bool
walprop_blocking_write(WalProposerConn *conn, void const *buf, size_t size)
{
int result;
/* If we are in non-blocking mode, switch out of it. */
if (!ensure_nonblocking_status(conn, false))
return false;
if ((result = PQputCopyData(conn->pg_conn, buf, size)) == -1)
return false;
Assert(result == 1);
/* Because the connection is non-blocking, flushing returns 0 or -1 */
if ((result = PQflush(conn->pg_conn)) == -1)
return false;
Assert(result == 0);
return true;
}

View File

@@ -18,6 +18,10 @@ extern char *neon_auth_token;
extern char *neon_timeline;
extern char *neon_tenant;
extern char *wal_acceptors_list;
extern int wal_acceptor_reconnect_timeout;
extern int wal_acceptor_connection_timeout;
extern void pg_init_libpagestore(void);
extern void pg_init_walproposer(void);
@@ -30,4 +34,10 @@ extern void pg_init_extension_server(void);
extern bool neon_redo_read_buffer_filter(XLogReaderState *record, uint8 block_id);
extern bool (*old_redo_read_buffer_filter) (XLogReaderState *record, uint8 block_id);
extern uint64 BackpressureThrottlingTime(void);
extern void replication_feedback_get_lsns(XLogRecPtr *writeLsn, XLogRecPtr *flushLsn, XLogRecPtr *applyLsn);
extern void PGDLLEXPORT WalProposerSync(int argc, char *argv[]);
extern void PGDLLEXPORT WalProposerMain(Datum main_arg);
#endif /* NEON_H */

116
pgxn/neon/neon_utils.c Normal file
View File

@@ -0,0 +1,116 @@
#include "postgres.h"
#include "access/timeline.h"
#include "access/xlogutils.h"
#include "common/logging.h"
#include "common/ip.h"
#include "funcapi.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "miscadmin.h"
#include "postmaster/interrupt.h"
#include "replication/slot.h"
#include "replication/walsender_private.h"
#include "storage/ipc.h"
#include "utils/builtins.h"
#include "utils/ps_status.h"
#include "libpq-fe.h"
#include <netinet/tcp.h>
#include <unistd.h>
#if PG_VERSION_NUM >= 150000
#include "access/xlogutils.h"
#include "access/xlogrecovery.h"
#endif
#if PG_MAJORVERSION_NUM >= 16
#include "utils/guc.h"
#endif
/*
* Convert a character which represents a hexadecimal digit to an integer.
*
* Returns -1 if the character is not a hexadecimal digit.
*/
int
HexDecodeChar(char c)
{
if (c >= '0' && c <= '9')
return c - '0';
if (c >= 'a' && c <= 'f')
return c - 'a' + 10;
if (c >= 'A' && c <= 'F')
return c - 'A' + 10;
return -1;
}
/*
* Decode a hex string into a byte string, 2 hex chars per byte.
*
* Returns false if invalid characters are encountered; otherwise true.
*/
bool
HexDecodeString(uint8 *result, char *input, int nbytes)
{
int i;
for (i = 0; i < nbytes; ++i)
{
int n1 = HexDecodeChar(input[i * 2]);
int n2 = HexDecodeChar(input[i * 2 + 1]);
if (n1 < 0 || n2 < 0)
return false;
result[i] = n1 * 16 + n2;
}
return true;
}
/* --------------------------------
* pq_getmsgint32_le - get a binary 4-byte int from a message buffer in native (LE) order
* --------------------------------
*/
uint32
pq_getmsgint32_le(StringInfo msg)
{
uint32 n32;
pq_copymsgbytes(msg, (char *) &n32, sizeof(n32));
return n32;
}
/* --------------------------------
* pq_getmsgint64 - get a binary 8-byte int from a message buffer in native (LE) order
* --------------------------------
*/
uint64
pq_getmsgint64_le(StringInfo msg)
{
uint64 n64;
pq_copymsgbytes(msg, (char *) &n64, sizeof(n64));
return n64;
}
/* append a binary [u]int32 to a StringInfo buffer in native (LE) order */
void
pq_sendint32_le(StringInfo buf, uint32 i)
{
enlargeStringInfo(buf, sizeof(uint32));
memcpy(buf->data + buf->len, &i, sizeof(uint32));
buf->len += sizeof(uint32);
}
/* append a binary [u]int64 to a StringInfo buffer in native (LE) order */
void
pq_sendint64_le(StringInfo buf, uint64 i)
{
enlargeStringInfo(buf, sizeof(uint64));
memcpy(buf->data + buf->len, &i, sizeof(uint64));
buf->len += sizeof(uint64);
}

12
pgxn/neon/neon_utils.h Normal file
View File

@@ -0,0 +1,12 @@
#ifndef __NEON_UTILS_H__
#define __NEON_UTILS_H__
#include "postgres.h"
bool HexDecodeString(uint8 *result, char *input, int nbytes);
uint32 pq_getmsgint32_le(StringInfo msg);
uint64 pq_getmsgint64_le(StringInfo msg);
void pq_sendint32_le(StringInfo buf, uint32 i);
void pq_sendint64_le(StringInfo buf, uint64 i);
#endif /* __NEON_UTILS_H__ */

File diff suppressed because it is too large Load Diff

View File

@@ -1,8 +1,8 @@
#ifndef __NEON_WALPROPOSER_H__
#define __NEON_WALPROPOSER_H__
#include "access/xlogdefs.h"
#include "postgres.h"
#include "access/xlogdefs.h"
#include "port.h"
#include "access/xlog_internal.h"
#include "access/transam.h"
@@ -16,29 +16,15 @@
#define MAX_SAFEKEEPERS 32
#define MAX_SEND_SIZE (XLOG_BLCKSZ * 16) /* max size of a single* WAL
* message */
#define XLOG_HDR_SIZE (1 + 8 * 3) /* 'w' + startPos + walEnd + timestamp */
#define XLOG_HDR_START_POS 1 /* offset of start position in wal sender*
* message header */
#define XLOG_HDR_END_POS (1 + 8) /* offset of end position in wal sender*
* message header */
/*
* In the spirit of WL_SOCKET_READABLE and others, this corresponds to no events having occurred,
* because all WL_* events are given flags equal to some (1 << i), starting from i = 0
*/
#define WL_NO_EVENTS 0
extern char *wal_acceptors_list;
extern int wal_acceptor_reconnect_timeout;
extern int wal_acceptor_connection_timeout;
extern bool am_wal_proposer;
struct WalProposerConn; /* Defined in libpqwalproposer */
struct WalProposerConn; /* Defined in implementation (walprop_pg.c) */
typedef struct WalProposerConn WalProposerConn;
struct WalMessage;
typedef struct WalMessage WalMessage;
/* Possible return values from ReadPGAsync */
typedef enum
{
@@ -52,7 +38,7 @@ typedef enum
PG_ASYNC_READ_TRY_AGAIN,
/* Reading failed. Check PQerrorMessage(conn) */
PG_ASYNC_READ_FAIL,
} PGAsyncReadResult;
} PGAsyncReadResult;
/* Possible return values from WritePGAsync */
typedef enum
@@ -71,7 +57,7 @@ typedef enum
PG_ASYNC_WRITE_TRY_FLUSH,
/* Writing failed. Check PQerrorMessage(conn) */
PG_ASYNC_WRITE_FAIL,
} PGAsyncWriteResult;
} PGAsyncWriteResult;
/*
* WAL safekeeper state, which is used to wait for some event.
@@ -147,7 +133,7 @@ typedef enum
* to read.
*/
SS_ACTIVE,
} SafekeeperState;
} SafekeeperState;
/* Consensus logical timestamp. */
typedef uint64 term_t;
@@ -171,12 +157,12 @@ typedef struct ProposerGreeting
uint8 tenant_id[16];
TimeLineID timeline;
uint32 walSegSize;
} ProposerGreeting;
} ProposerGreeting;
typedef struct AcceptorProposerMessage
{
uint64 tag;
} AcceptorProposerMessage;
} AcceptorProposerMessage;
/*
* Acceptor -> Proposer initial response: the highest term acceptor voted for.
@@ -186,7 +172,7 @@ typedef struct AcceptorGreeting
AcceptorProposerMessage apm;
term_t term;
NNodeId nodeId;
} AcceptorGreeting;
} AcceptorGreeting;
/*
* Proposer -> Acceptor vote request.
@@ -196,20 +182,20 @@ typedef struct VoteRequest
uint64 tag;
term_t term;
pg_uuid_t proposerId; /* for monitoring/debugging */
} VoteRequest;
} VoteRequest;
/* Element of term switching chain. */
typedef struct TermSwitchEntry
{
term_t term;
XLogRecPtr lsn;
} TermSwitchEntry;
} TermSwitchEntry;
typedef struct TermHistory
{
uint32 n_entries;
TermSwitchEntry *entries;
} TermHistory;
} TermHistory;
/* Vote itself, sent from safekeeper to proposer */
typedef struct VoteResponse
@@ -227,7 +213,7 @@ typedef struct VoteResponse
* recovery of some safekeeper */
TermHistory termHistory;
XLogRecPtr timelineStartLsn; /* timeline globally starts at this LSN */
} VoteResponse;
} VoteResponse;
/*
* Proposer -> Acceptor message announcing proposer is elected and communicating
@@ -243,7 +229,7 @@ typedef struct ProposerElected
TermHistory *termHistory;
/* timeline globally starts at this LSN */
XLogRecPtr timelineStartLsn;
} ProposerElected;
} ProposerElected;
/*
* Header of request with WAL message sent from proposer to safekeeper.
@@ -268,7 +254,7 @@ typedef struct AppendRequestHeader
*/
XLogRecPtr truncateLsn;
pg_uuid_t proposerId; /* for monitoring/debugging */
} AppendRequestHeader;
} AppendRequestHeader;
/*
* Hot standby feedback received from replica
@@ -278,7 +264,7 @@ typedef struct HotStandbyFeedback
TimestampTz ts;
FullTransactionId xmin;
FullTransactionId catalog_xmin;
} HotStandbyFeedback;
} HotStandbyFeedback;
typedef struct PageserverFeedback
{
@@ -289,7 +275,7 @@ typedef struct PageserverFeedback
XLogRecPtr disk_consistent_lsn;
XLogRecPtr remote_consistent_lsn;
TimestampTz replytime;
} PageserverFeedback;
} PageserverFeedback;
typedef struct WalproposerShmemState
{
@@ -297,7 +283,7 @@ typedef struct WalproposerShmemState
PageserverFeedback feedback;
term_t mineLastElectedTerm;
pg_atomic_uint64 backpressureThrottlingTime;
} WalproposerShmemState;
} WalproposerShmemState;
/*
* Report safekeeper state to proposer
@@ -321,17 +307,22 @@ typedef struct AppendResponse
/* and custom neon feedback. */
/* This part of the message is extensible. */
PageserverFeedback rf;
} AppendResponse;
} AppendResponse;
/* PageserverFeedback is extensible part of the message that is parsed separately */
/* Other fields are fixed part */
#define APPENDRESPONSE_FIXEDPART_SIZE offsetof(AppendResponse, rf)
struct WalProposer;
typedef struct WalProposer WalProposer;
/*
* Descriptor of safekeeper
*/
typedef struct Safekeeper
{
WalProposer *wp;
char const *host;
char const *port;
@@ -340,7 +331,7 @@ typedef struct Safekeeper
*
* May contain private information like password and should not be logged.
*/
char conninfo[MAXCONNINFO];
char conninfo[MAXCONNINFO];
/*
* postgres protocol connection to the WAL acceptor
@@ -373,27 +364,12 @@ typedef struct Safekeeper
int eventPos; /* position in wait event set. Equal to -1 if*
* no event */
SafekeeperState state; /* safekeeper state machine state */
TimestampTz latestMsgReceivedAt; /* when latest msg is received */
TimestampTz latestMsgReceivedAt; /* when latest msg is received */
AcceptorGreeting greetResponse; /* acceptor greeting */
VoteResponse voteResponse; /* the vote */
AppendResponse appendResponse; /* feedback for master */
} Safekeeper;
extern void PGDLLEXPORT WalProposerSync(int argc, char *argv[]);
extern void PGDLLEXPORT WalProposerMain(Datum main_arg);
extern void WalProposerBroadcast(XLogRecPtr startpos, XLogRecPtr endpos);
extern void WalProposerPoll(void);
extern void ParsePageserverFeedbackMessage(StringInfo reply_message,
PageserverFeedback *rf);
extern void StartProposerReplication(StartReplicationCmd *cmd);
extern Size WalproposerShmemSize(void);
extern bool WalproposerShmemInit(void);
extern void replication_feedback_set(PageserverFeedback *rf);
extern void replication_feedback_get_lsns(XLogRecPtr *writeLsn, XLogRecPtr *flushLsn, XLogRecPtr *applyLsn);
/* libpqwalproposer hooks & helper type */
/* Re-exported PostgresPollingStatusType */
typedef enum
{
@@ -406,7 +382,7 @@ typedef enum
* 'libpq-fe.h' still has PGRES_POLLING_ACTIVE, but says it's unused.
* We've removed it here to avoid clutter.
*/
} WalProposerConnectPollStatusType;
} WalProposerConnectPollStatusType;
/* Re-exported and modified ExecStatusType */
typedef enum
@@ -431,7 +407,7 @@ typedef enum
WP_EXEC_NEEDS_INPUT,
/* Catch-all failure. Check PQerrorMessage. */
WP_EXEC_FAILED,
} WalProposerExecStatusType;
} WalProposerExecStatusType;
/* Re-exported ConnStatusType */
typedef enum
@@ -445,67 +421,252 @@ typedef enum
* that extra functionality, so we collect them into a single tag here.
*/
WP_CONNECTION_IN_PROGRESS,
} WalProposerConnStatusType;
/* Re-exported PQerrorMessage */
extern char *walprop_error_message(WalProposerConn *conn);
/* Re-exported PQstatus */
extern WalProposerConnStatusType walprop_status(WalProposerConn *conn);
/* Re-exported PQconnectStart */
extern WalProposerConn * walprop_connect_start(char *conninfo, char *password);
/* Re-exported PQconectPoll */
extern WalProposerConnectPollStatusType walprop_connect_poll(WalProposerConn *conn);
/* Blocking wrapper around PQsendQuery */
extern bool walprop_send_query(WalProposerConn *conn, char *query);
/* Wrapper around PQconsumeInput + PQisBusy + PQgetResult */
extern WalProposerExecStatusType walprop_get_query_result(WalProposerConn *conn);
/* Re-exported PQsocket */
extern pgsocket walprop_socket(WalProposerConn *conn);
/* Wrapper around PQconsumeInput (if socket's read-ready) + PQflush */
extern int walprop_flush(WalProposerConn *conn);
/* Re-exported PQfinish */
extern void walprop_finish(WalProposerConn *conn);
} WalProposerConnStatusType;
/*
* Ergonomic wrapper around PGgetCopyData
*
* Reads a CopyData block from a safekeeper, setting *amount to the number
* of bytes returned.
*
* This function is allowed to assume certain properties specific to the
* protocol with the safekeepers, so it should not be used as-is for any
* other purpose.
*
* Note: If possible, using <AsyncRead> is generally preferred, because it
* performs a bit of extra checking work that's always required and is normally
* somewhat verbose.
* Collection of hooks for walproposer, to call postgres functions,
* read WAL and send it over the network.
*/
extern PGAsyncReadResult walprop_async_read(WalProposerConn *conn, char **buf, int *amount);
typedef struct walproposer_api
{
/*
* Get WalproposerShmemState. This is used to store information about last
* elected term.
*/
WalproposerShmemState *(*get_shmem_state) (void);
/*
* Start receiving notifications about new WAL. This is an infinite loop
* which calls WalProposerBroadcast() and WalProposerPoll() to send the
* WAL.
*/
void (*start_streaming) (WalProposer *wp, XLogRecPtr startpos);
/* Get pointer to the latest available WAL. */
XLogRecPtr (*get_flush_rec_ptr) (void);
/* Get current time. */
TimestampTz (*get_current_timestamp) (void);
/* Get postgres timeline. */
TimeLineID (*get_timeline_id) (void);
/* Current error message, aka PQerrorMessage. */
char *(*conn_error_message) (WalProposerConn *conn);
/* Connection status, aka PQstatus. */
WalProposerConnStatusType (*conn_status) (WalProposerConn *conn);
/* Start the connection, aka PQconnectStart. */
WalProposerConn *(*conn_connect_start) (char *conninfo);
/* Poll an asynchronous connection, aka PQconnectPoll. */
WalProposerConnectPollStatusType (*conn_connect_poll) (WalProposerConn *conn);
/* Send a blocking SQL query, aka PQsendQuery. */
bool (*conn_send_query) (WalProposerConn *conn, char *query);
/* Read the query result, aka PQgetResult. */
WalProposerExecStatusType (*conn_get_query_result) (WalProposerConn *conn);
/* Flush buffer to the network, aka PQflush. */
int (*conn_flush) (WalProposerConn *conn);
/* Close the connection, aka PQfinish. */
void (*conn_finish) (WalProposerConn *conn);
/* Try to read CopyData message, aka PQgetCopyData. */
PGAsyncReadResult (*conn_async_read) (WalProposerConn *conn, char **buf, int *amount);
/* Try to write CopyData message, aka PQputCopyData. */
PGAsyncWriteResult (*conn_async_write) (WalProposerConn *conn, void const *buf, size_t size);
/* Blocking CopyData write, aka PQputCopyData + PQflush. */
bool (*conn_blocking_write) (WalProposerConn *conn, void const *buf, size_t size);
/* Download WAL from startpos to endpos and make it available locally. */
bool (*recovery_download) (Safekeeper *sk, TimeLineID timeline, XLogRecPtr startpos, XLogRecPtr endpos);
/* Read WAL from disk to buf. */
void (*wal_read) (XLogReaderState *state, char *buf, XLogRecPtr startptr, Size count);
/* Allocate WAL reader. */
XLogReaderState *(*wal_reader_allocate) (void);
/* Deallocate event set. */
void (*free_event_set) (void);
/* Initialize event set. */
void (*init_event_set) (int n_safekeepers);
/* Update events for an existing safekeeper connection. */
void (*update_event_set) (Safekeeper *sk, uint32 events);
/* Add a new safekeeper connection to the event set. */
void (*add_safekeeper_event_set) (Safekeeper *sk, uint32 events);
/*
* Wait until some event happens: - timeout is reached - socket event for
* safekeeper connection - new WAL is available
*
* Returns 0 if timeout is reached, 1 if some event happened. Updates
* events mask to indicate events and sets sk to the safekeeper which has
* an event.
*/
int (*wait_event_set) (long timeout, Safekeeper **sk, uint32 *events);
/* Read random bytes. */
bool (*strong_random) (void *buf, size_t len);
/*
* Get a basebackup LSN. Used to cross-validate with the latest available
* LSN on the safekeepers.
*/
XLogRecPtr (*get_redo_start_lsn) (void);
/*
* Finish sync safekeepers with the given LSN. This function should not
* return and should exit the program.
*/
void (*finish_sync_safekeepers) (XLogRecPtr lsn);
/*
* Called after every new message from the safekeeper. Used to propagate
* backpressure feedback and to confirm WAL persistence (has been commited
* on the quorum of safekeepers).
*/
void (*process_safekeeper_feedback) (WalProposer *wp, XLogRecPtr commitLsn);
/*
* Called on peer_horizon_lsn updates. Used to advance replication slot
* and to free up disk space by deleting unnecessary WAL.
*/
void (*confirm_wal_streamed) (XLogRecPtr lsn);
} walproposer_api;
/*
* Ergonomic wrapper around PQputCopyData + PQflush
*
* Starts to write a CopyData block to a safekeeper.
*
* For information on the meaning of return codes, refer to PGAsyncWriteResult.
* Configuration of the WAL proposer.
*/
extern PGAsyncWriteResult walprop_async_write(WalProposerConn *conn, void const *buf, size_t size);
typedef struct WalProposerConfig
{
/* hex-encoded TenantId cstr */
char *neon_tenant;
/* hex-encoded TimelineId cstr */
char *neon_timeline;
/*
* Comma-separated list of safekeepers, in the following format:
* host1:port1,host2:port2,host3:port3
*
* This cstr should be editable.
*/
char *safekeepers_list;
/*
* WalProposer reconnects to offline safekeepers once in this interval.
* Time is in milliseconds.
*/
int safekeeper_reconnect_timeout;
/*
* WalProposer terminates the connection if it doesn't receive any message
* from the safekeeper in this interval. Time is in milliseconds.
*/
int safekeeper_connection_timeout;
/*
* WAL segment size. Will be passed to safekeepers in greet request. Also
* used to detect page headers.
*/
int wal_segment_size;
/*
* If safekeeper was started in sync mode, walproposer will not subscribe
* for new WAL and will exit when quorum of safekeepers will be synced to
* the latest available LSN.
*/
bool syncSafekeepers;
/* Will be passed to safekeepers in greet request. */
uint64 systemId;
} WalProposerConfig;
/*
* Blocking equivalent to walprop_async_write_fn
*
* Returns 'true' if successful, 'false' on failure.
* WAL proposer state.
*/
extern bool walprop_blocking_write(WalProposerConn *conn, void const *buf, size_t size);
typedef struct WalProposer
{
WalProposerConfig *config;
int n_safekeepers;
extern uint64 BackpressureThrottlingTime(void);
/* (n_safekeepers / 2) + 1 */
int quorum;
Safekeeper safekeeper[MAX_SAFEKEEPERS];
/* WAL has been generated up to this point */
XLogRecPtr availableLsn;
/* last commitLsn broadcasted to safekeepers */
XLogRecPtr lastSentCommitLsn;
ProposerGreeting greetRequest;
/* Vote request for safekeeper */
VoteRequest voteRequest;
/*
* Minimal LSN which may be needed for recovery of some safekeeper,
* record-aligned (first record which might not yet received by someone).
*/
XLogRecPtr truncateLsn;
/*
* Term of the proposer. We want our term to be highest and unique, so we
* collect terms from safekeepers quorum, choose max and +1. After that
* our term is fixed and must not change. If we observe that some
* safekeeper has higher term, it means that we have another running
* compute, so we must stop immediately.
*/
term_t propTerm;
/* term history of the proposer */
TermHistory propTermHistory;
/* epoch start lsn of the proposer */
XLogRecPtr propEpochStartLsn;
/* Most advanced acceptor epoch */
term_t donorEpoch;
/* Most advanced acceptor */
int donor;
/* timeline globally starts at this LSN */
XLogRecPtr timelineStartLsn;
/* number of votes collected from safekeepers */
int n_votes;
/* number of successful connections over the lifetime of walproposer */
int n_connected;
/*
* Timestamp of the last reconnection attempt. Related to
* config->safekeeper_reconnect_timeout
*/
TimestampTz last_reconnect_attempt;
walproposer_api api;
} WalProposer;
extern WalProposer *WalProposerCreate(WalProposerConfig *config, walproposer_api api);
extern void WalProposerStart(WalProposer *wp);
extern void WalProposerBroadcast(WalProposer *wp, XLogRecPtr startpos, XLogRecPtr endpos);
extern void WalProposerPoll(WalProposer *wp);
extern void ParsePageserverFeedbackMessage(StringInfo reply_message,
PageserverFeedback *rf);
#endif /* __NEON_WALPROPOSER_H__ */

1667
pgxn/neon/walproposer_pg.c Normal file

File diff suppressed because it is too large Load Diff

View File

@@ -1,659 +0,0 @@
#include "postgres.h"
#include "access/timeline.h"
#include "access/xlogutils.h"
#include "common/logging.h"
#include "common/ip.h"
#include "funcapi.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "miscadmin.h"
#include "postmaster/interrupt.h"
#include "replication/slot.h"
#include "walproposer_utils.h"
#include "replication/walsender_private.h"
#include "storage/ipc.h"
#include "utils/builtins.h"
#include "utils/ps_status.h"
#include "libpq-fe.h"
#include <netinet/tcp.h>
#include <unistd.h>
#if PG_VERSION_NUM >= 150000
#include "access/xlogutils.h"
#include "access/xlogrecovery.h"
#endif
#if PG_MAJORVERSION_NUM >= 16
#include "utils/guc.h"
#endif
/*
* These variables are used similarly to openLogFile/SegNo,
* but for walproposer to write the XLOG during recovery. walpropFileTLI is the TimeLineID
* corresponding the filename of walpropFile.
*/
static int walpropFile = -1;
static TimeLineID walpropFileTLI = 0;
static XLogSegNo walpropSegNo = 0;
/* START cloned file-local variables and functions from walsender.c */
/*
* How far have we sent WAL already? This is also advertised in
* MyWalSnd->sentPtr. (Actually, this is the next WAL location to send.)
*/
static XLogRecPtr sentPtr = InvalidXLogRecPtr;
static void WalSndLoop(void);
static void XLogBroadcastWalProposer(void);
/* END cloned file-level variables and functions from walsender.c */
int
CompareLsn(const void *a, const void *b)
{
XLogRecPtr lsn1 = *((const XLogRecPtr *) a);
XLogRecPtr lsn2 = *((const XLogRecPtr *) b);
if (lsn1 < lsn2)
return -1;
else if (lsn1 == lsn2)
return 0;
else
return 1;
}
/* Returns a human-readable string corresonding to the SafekeeperState
*
* The string should not be freed.
*
* The strings are intended to be used as a prefix to "state", e.g.:
*
* elog(LOG, "currently in %s state", FormatSafekeeperState(sk->state));
*
* If this sort of phrasing doesn't fit the message, instead use something like:
*
* elog(LOG, "currently in state [%s]", FormatSafekeeperState(sk->state));
*/
char *
FormatSafekeeperState(SafekeeperState state)
{
char *return_val = NULL;
switch (state)
{
case SS_OFFLINE:
return_val = "offline";
break;
case SS_CONNECTING_READ:
case SS_CONNECTING_WRITE:
return_val = "connecting";
break;
case SS_WAIT_EXEC_RESULT:
return_val = "receiving query result";
break;
case SS_HANDSHAKE_RECV:
return_val = "handshake (receiving)";
break;
case SS_VOTING:
return_val = "voting";
break;
case SS_WAIT_VERDICT:
return_val = "wait-for-verdict";
break;
case SS_SEND_ELECTED_FLUSH:
return_val = "send-announcement-flush";
break;
case SS_IDLE:
return_val = "idle";
break;
case SS_ACTIVE:
return_val = "active";
break;
}
Assert(return_val != NULL);
return return_val;
}
/* Asserts that the provided events are expected for given safekeeper's state */
void
AssertEventsOkForState(uint32 events, Safekeeper *sk)
{
uint32 expected = SafekeeperStateDesiredEvents(sk->state);
/*
* The events are in-line with what we're expecting, under two conditions:
* (a) if we aren't expecting anything, `events` has no read- or
* write-ready component. (b) if we are expecting something, there's
* overlap (i.e. `events & expected != 0`)
*/
bool events_ok_for_state; /* long name so the `Assert` is more
* clear later */
if (expected == WL_NO_EVENTS)
events_ok_for_state = ((events & (WL_SOCKET_READABLE | WL_SOCKET_WRITEABLE)) == 0);
else
events_ok_for_state = ((events & expected) != 0);
if (!events_ok_for_state)
{
/*
* To give a descriptive message in the case of failure, we use elog
* and then an assertion that's guaranteed to fail.
*/
elog(WARNING, "events %s mismatched for safekeeper %s:%s in state [%s]",
FormatEvents(events), sk->host, sk->port, FormatSafekeeperState(sk->state));
Assert(events_ok_for_state);
}
}
/* Returns the set of events a safekeeper in this state should be waiting on
*
* This will return WL_NO_EVENTS (= 0) for some events. */
uint32
SafekeeperStateDesiredEvents(SafekeeperState state)
{
uint32 result = WL_NO_EVENTS;
/* If the state doesn't have a modifier, we can check the base state */
switch (state)
{
/* Connecting states say what they want in the name */
case SS_CONNECTING_READ:
result = WL_SOCKET_READABLE;
break;
case SS_CONNECTING_WRITE:
result = WL_SOCKET_WRITEABLE;
break;
/* Reading states need the socket to be read-ready to continue */
case SS_WAIT_EXEC_RESULT:
case SS_HANDSHAKE_RECV:
case SS_WAIT_VERDICT:
result = WL_SOCKET_READABLE;
break;
/*
* Idle states use read-readiness as a sign that the connection
* has been disconnected.
*/
case SS_VOTING:
case SS_IDLE:
result = WL_SOCKET_READABLE;
break;
/*
* Flush states require write-ready for flushing. Active state
* does both reading and writing.
*
* TODO: SS_ACTIVE sometimes doesn't need to be write-ready. We
* should check sk->flushWrite here to set WL_SOCKET_WRITEABLE.
*/
case SS_SEND_ELECTED_FLUSH:
case SS_ACTIVE:
result = WL_SOCKET_READABLE | WL_SOCKET_WRITEABLE;
break;
/* The offline state expects no events. */
case SS_OFFLINE:
result = WL_NO_EVENTS;
break;
default:
Assert(false);
break;
}
return result;
}
/* Returns a human-readable string corresponding to the event set
*
* If the events do not correspond to something set as the `events` field of a `WaitEvent`, the
* returned string may be meaingless.
*
* The string should not be freed. It should also not be expected to remain the same between
* function calls. */
char *
FormatEvents(uint32 events)
{
static char return_str[8];
/* Helper variable to check if there's extra bits */
uint32 all_flags = WL_LATCH_SET
| WL_SOCKET_READABLE
| WL_SOCKET_WRITEABLE
| WL_TIMEOUT
| WL_POSTMASTER_DEATH
| WL_EXIT_ON_PM_DEATH
| WL_SOCKET_CONNECTED;
/*
* The formatting here isn't supposed to be *particularly* useful -- it's
* just to give an sense of what events have been triggered without
* needing to remember your powers of two.
*/
return_str[0] = (events & WL_LATCH_SET) ? 'L' : '_';
return_str[1] = (events & WL_SOCKET_READABLE) ? 'R' : '_';
return_str[2] = (events & WL_SOCKET_WRITEABLE) ? 'W' : '_';
return_str[3] = (events & WL_TIMEOUT) ? 'T' : '_';
return_str[4] = (events & WL_POSTMASTER_DEATH) ? 'D' : '_';
return_str[5] = (events & WL_EXIT_ON_PM_DEATH) ? 'E' : '_';
return_str[5] = (events & WL_SOCKET_CONNECTED) ? 'C' : '_';
if (events & (~all_flags))
{
elog(WARNING, "Event formatting found unexpected component %d",
events & (~all_flags));
return_str[6] = '*';
return_str[7] = '\0';
}
else
return_str[6] = '\0';
return (char *) &return_str;
}
/*
* Convert a character which represents a hexadecimal digit to an integer.
*
* Returns -1 if the character is not a hexadecimal digit.
*/
static int
HexDecodeChar(char c)
{
if (c >= '0' && c <= '9')
return c - '0';
if (c >= 'a' && c <= 'f')
return c - 'a' + 10;
if (c >= 'A' && c <= 'F')
return c - 'A' + 10;
return -1;
}
/*
* Decode a hex string into a byte string, 2 hex chars per byte.
*
* Returns false if invalid characters are encountered; otherwise true.
*/
bool
HexDecodeString(uint8 *result, char *input, int nbytes)
{
int i;
for (i = 0; i < nbytes; ++i)
{
int n1 = HexDecodeChar(input[i * 2]);
int n2 = HexDecodeChar(input[i * 2 + 1]);
if (n1 < 0 || n2 < 0)
return false;
result[i] = n1 * 16 + n2;
}
return true;
}
/* --------------------------------
* pq_getmsgint32_le - get a binary 4-byte int from a message buffer in native (LE) order
* --------------------------------
*/
uint32
pq_getmsgint32_le(StringInfo msg)
{
uint32 n32;
pq_copymsgbytes(msg, (char *) &n32, sizeof(n32));
return n32;
}
/* --------------------------------
* pq_getmsgint64 - get a binary 8-byte int from a message buffer in native (LE) order
* --------------------------------
*/
uint64
pq_getmsgint64_le(StringInfo msg)
{
uint64 n64;
pq_copymsgbytes(msg, (char *) &n64, sizeof(n64));
return n64;
}
/* append a binary [u]int32 to a StringInfo buffer in native (LE) order */
void
pq_sendint32_le(StringInfo buf, uint32 i)
{
enlargeStringInfo(buf, sizeof(uint32));
memcpy(buf->data + buf->len, &i, sizeof(uint32));
buf->len += sizeof(uint32);
}
/* append a binary [u]int64 to a StringInfo buffer in native (LE) order */
void
pq_sendint64_le(StringInfo buf, uint64 i)
{
enlargeStringInfo(buf, sizeof(uint64));
memcpy(buf->data + buf->len, &i, sizeof(uint64));
buf->len += sizeof(uint64);
}
/*
* Write XLOG data to disk.
*/
void
XLogWalPropWrite(char *buf, Size nbytes, XLogRecPtr recptr)
{
int startoff;
int byteswritten;
while (nbytes > 0)
{
int segbytes;
/* Close the current segment if it's completed */
if (walpropFile >= 0 && !XLByteInSeg(recptr, walpropSegNo, wal_segment_size))
XLogWalPropClose(recptr);
if (walpropFile < 0)
{
#if PG_VERSION_NUM >= 150000
/* FIXME Is it ok to use hardcoded value here? */
TimeLineID tli = 1;
#else
bool use_existent = true;
#endif
/* Create/use new log file */
XLByteToSeg(recptr, walpropSegNo, wal_segment_size);
#if PG_VERSION_NUM >= 150000
walpropFile = XLogFileInit(walpropSegNo, tli);
walpropFileTLI = tli;
#else
walpropFile = XLogFileInit(walpropSegNo, &use_existent, false);
walpropFileTLI = ThisTimeLineID;
#endif
}
/* Calculate the start offset of the received logs */
startoff = XLogSegmentOffset(recptr, wal_segment_size);
if (startoff + nbytes > wal_segment_size)
segbytes = wal_segment_size - startoff;
else
segbytes = nbytes;
/* OK to write the logs */
errno = 0;
byteswritten = pg_pwrite(walpropFile, buf, segbytes, (off_t) startoff);
if (byteswritten <= 0)
{
char xlogfname[MAXFNAMELEN];
int save_errno;
/* if write didn't set errno, assume no disk space */
if (errno == 0)
errno = ENOSPC;
save_errno = errno;
XLogFileName(xlogfname, walpropFileTLI, walpropSegNo, wal_segment_size);
errno = save_errno;
ereport(PANIC,
(errcode_for_file_access(),
errmsg("could not write to log segment %s "
"at offset %u, length %lu: %m",
xlogfname, startoff, (unsigned long) segbytes)));
}
/* Update state for write */
recptr += byteswritten;
nbytes -= byteswritten;
buf += byteswritten;
}
/*
* Close the current segment if it's fully written up in the last cycle of
* the loop.
*/
if (walpropFile >= 0 && !XLByteInSeg(recptr, walpropSegNo, wal_segment_size))
{
XLogWalPropClose(recptr);
}
}
/*
* Close the current segment.
*/
void
XLogWalPropClose(XLogRecPtr recptr)
{
Assert(walpropFile >= 0 && !XLByteInSeg(recptr, walpropSegNo, wal_segment_size));
if (close(walpropFile) != 0)
{
char xlogfname[MAXFNAMELEN];
XLogFileName(xlogfname, walpropFileTLI, walpropSegNo, wal_segment_size);
ereport(PANIC,
(errcode_for_file_access(),
errmsg("could not close log segment %s: %m",
xlogfname)));
}
walpropFile = -1;
}
/* START of cloned functions from walsender.c */
/*
* Subscribe for new WAL and stream it in the loop to safekeepers.
*
* At the moment, this never returns, but an ereport(ERROR) will take us back
* to the main loop.
*/
void
StartProposerReplication(StartReplicationCmd *cmd)
{
XLogRecPtr FlushPtr;
TimeLineID currTLI;
#if PG_VERSION_NUM < 150000
if (ThisTimeLineID == 0)
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("IDENTIFY_SYSTEM has not been run before START_REPLICATION")));
#endif
/*
* We assume here that we're logging enough information in the WAL for
* log-shipping, since this is checked in PostmasterMain().
*
* NOTE: wal_level can only change at shutdown, so in most cases it is
* difficult for there to be WAL data that we can still see that was
* written at wal_level='minimal'.
*/
if (cmd->slotname)
{
ReplicationSlotAcquire(cmd->slotname, true);
if (SlotIsLogical(MyReplicationSlot))
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("cannot use a logical replication slot for physical replication")));
/*
* We don't need to verify the slot's restart_lsn here; instead we
* rely on the caller requesting the starting point to use. If the
* WAL segment doesn't exist, we'll fail later.
*/
}
/*
* Select the timeline. If it was given explicitly by the client, use
* that. Otherwise use the timeline of the last replayed record, which is
* kept in ThisTimeLineID.
*
* Neon doesn't currently use PG Timelines, but it may in the future, so
* we keep this code around to lighten the load for when we need it.
*/
#if PG_VERSION_NUM >= 150000
FlushPtr = GetFlushRecPtr(&currTLI);
#else
FlushPtr = GetFlushRecPtr();
currTLI = ThisTimeLineID;
#endif
/*
* When we first start replication the standby will be behind the
* primary. For some applications, for example synchronous
* replication, it is important to have a clear state for this initial
* catchup mode, so we can trigger actions when we change streaming
* state later. We may stay in this state for a long time, which is
* exactly why we want to be able to monitor whether or not we are
* still here.
*/
WalSndSetState(WALSNDSTATE_CATCHUP);
/*
* Don't allow a request to stream from a future point in WAL that
* hasn't been flushed to disk in this server yet.
*/
if (FlushPtr < cmd->startpoint)
{
ereport(ERROR,
(errmsg("requested starting point %X/%X is ahead of the WAL flush position of this server %X/%X",
LSN_FORMAT_ARGS(cmd->startpoint),
LSN_FORMAT_ARGS(FlushPtr))));
}
/* Start streaming from the requested point */
sentPtr = cmd->startpoint;
/* Initialize shared memory status, too */
SpinLockAcquire(&MyWalSnd->mutex);
MyWalSnd->sentPtr = sentPtr;
SpinLockRelease(&MyWalSnd->mutex);
SyncRepInitConfig();
/* Infinite send loop, never returns */
WalSndLoop();
WalSndSetState(WALSNDSTATE_STARTUP);
if (cmd->slotname)
ReplicationSlotRelease();
}
/*
* Main loop that waits for LSN updates and calls the walproposer.
* Synchronous replication sets latch in WalSndWakeup at walsender.c
*/
static void
WalSndLoop(void)
{
/* Clear any already-pending wakeups */
ResetLatch(MyLatch);
for (;;)
{
CHECK_FOR_INTERRUPTS();
XLogBroadcastWalProposer();
if (MyWalSnd->state == WALSNDSTATE_CATCHUP)
WalSndSetState(WALSNDSTATE_STREAMING);
WalProposerPoll();
}
}
/*
* Notify walproposer about the new WAL position.
*/
static void
XLogBroadcastWalProposer(void)
{
XLogRecPtr startptr;
XLogRecPtr endptr;
/* Start from the last sent position */
startptr = sentPtr;
/*
* Streaming the current timeline on a primary.
*
* Attempt to send all data that's already been written out and
* fsync'd to disk. We cannot go further than what's been written out
* given the current implementation of WALRead(). And in any case
* it's unsafe to send WAL that is not securely down to disk on the
* primary: if the primary subsequently crashes and restarts, standbys
* must not have applied any WAL that got lost on the primary.
*/
#if PG_VERSION_NUM >= 150000
endptr = GetFlushRecPtr(NULL);
#else
endptr = GetFlushRecPtr();
#endif
/*
* Record the current system time as an approximation of the time at which
* this WAL location was written for the purposes of lag tracking.
*
* In theory we could make XLogFlush() record a time in shmem whenever WAL
* is flushed and we could get that time as well as the LSN when we call
* GetFlushRecPtr() above (and likewise for the cascading standby
* equivalent), but rather than putting any new code into the hot WAL path
* it seems good enough to capture the time here. We should reach this
* after XLogFlush() runs WalSndWakeupProcessRequests(), and although that
* may take some time, we read the WAL flush pointer and take the time
* very close to together here so that we'll get a later position if it is
* still moving.
*
* Because LagTrackerWrite ignores samples when the LSN hasn't advanced,
* this gives us a cheap approximation for the WAL flush time for this
* LSN.
*
* Note that the LSN is not necessarily the LSN for the data contained in
* the present message; it's the end of the WAL, which might be further
* ahead. All the lag tracking machinery cares about is finding out when
* that arbitrary LSN is eventually reported as written, flushed and
* applied, so that it can measure the elapsed time.
*/
LagTrackerWrite(endptr, GetCurrentTimestamp());
/* Do we have any work to do? */
Assert(startptr <= endptr);
if (endptr <= startptr)
return;
WalProposerBroadcast(startptr, endptr);
sentPtr = endptr;
/* Update shared memory status */
{
WalSnd *walsnd = MyWalSnd;
SpinLockAcquire(&walsnd->mutex);
walsnd->sentPtr = sentPtr;
SpinLockRelease(&walsnd->mutex);
}
/* Report progress of XLOG streaming in PS display */
if (update_process_title)
{
char activitymsg[50];
snprintf(activitymsg, sizeof(activitymsg), "streaming %X/%X",
LSN_FORMAT_ARGS(sentPtr));
set_ps_display(activitymsg);
}
}

View File

@@ -1,19 +0,0 @@
#ifndef __NEON_WALPROPOSER_UTILS_H__
#define __NEON_WALPROPOSER_UTILS_H__
#include "walproposer.h"
int CompareLsn(const void *a, const void *b);
char *FormatSafekeeperState(SafekeeperState state);
void AssertEventsOkForState(uint32 events, Safekeeper *sk);
uint32 SafekeeperStateDesiredEvents(SafekeeperState state);
char *FormatEvents(uint32 events);
bool HexDecodeString(uint8 *result, char *input, int nbytes);
uint32 pq_getmsgint32_le(StringInfo msg);
uint64 pq_getmsgint64_le(StringInfo msg);
void pq_sendint32_le(StringInfo buf, uint32 i);
void pq_sendint64_le(StringInfo buf, uint64 i);
void XLogWalPropWrite(char *buf, Size nbytes, XLogRecPtr recptr);
void XLogWalPropClose(XLogRecPtr recptr);
#endif /* __NEON_WALPROPOSER_UTILS_H__ */

72
poetry.lock generated
View File

@@ -2415,18 +2415,18 @@ files = [
[[package]]
name = "urllib3"
version = "1.26.17"
version = "1.26.11"
description = "HTTP library with thread-safe connection pooling, file post, and more."
optional = false
python-versions = ">=2.7, !=3.0.*, !=3.1.*, !=3.2.*, !=3.3.*, !=3.4.*, !=3.5.*"
python-versions = ">=2.7, !=3.0.*, !=3.1.*, !=3.2.*, !=3.3.*, !=3.4.*, !=3.5.*, <4"
files = [
{file = "urllib3-1.26.17-py2.py3-none-any.whl", hash = "sha256:94a757d178c9be92ef5539b8840d48dc9cf1b2709c9d6b588232a055c524458b"},
{file = "urllib3-1.26.17.tar.gz", hash = "sha256:24d6a242c28d29af46c3fae832c36db3bbebcc533dd1bb549172cd739c82df21"},
{file = "urllib3-1.26.11-py2.py3-none-any.whl", hash = "sha256:c33ccba33c819596124764c23a97d25f32b28433ba0dedeb77d873a38722c9bc"},
{file = "urllib3-1.26.11.tar.gz", hash = "sha256:ea6e8fb210b19d950fab93b60c9009226c63a28808bc8386e05301e25883ac0a"},
]
[package.extras]
brotli = ["brotli (==1.0.9)", "brotli (>=1.0.9)", "brotlicffi (>=0.8.0)", "brotlipy (>=0.6.0)"]
secure = ["certifi", "cryptography (>=1.3.4)", "idna (>=2.0.0)", "ipaddress", "pyOpenSSL (>=0.14)", "urllib3-secure-extra"]
brotli = ["brotli (>=1.0.9)", "brotlicffi (>=0.8.0)", "brotlipy (>=0.6.0)"]
secure = ["certifi", "cryptography (>=1.3.4)", "idna (>=2.0.0)", "ipaddress", "pyOpenSSL (>=0.14)"]
socks = ["PySocks (>=1.5.6,!=1.5.7,<2.0)"]
[[package]]
@@ -2648,65 +2648,7 @@ files = [
docs = ["jaraco.packaging (>=9)", "jaraco.tidelift (>=1.4)", "rst.linker (>=1.9)", "sphinx"]
testing = ["func-timeout", "jaraco.itertools", "pytest (>=6)", "pytest-black (>=0.3.7)", "pytest-checkdocs (>=2.4)", "pytest-cov", "pytest-enabler (>=1.3)", "pytest-flake8", "pytest-mypy (>=0.9.1)"]
[[package]]
name = "zstandard"
version = "0.21.0"
description = "Zstandard bindings for Python"
optional = false
python-versions = ">=3.7"
files = [
{file = "zstandard-0.21.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:649a67643257e3b2cff1c0a73130609679a5673bf389564bc6d4b164d822a7ce"},
{file = "zstandard-0.21.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:144a4fe4be2e747bf9c646deab212666e39048faa4372abb6a250dab0f347a29"},
{file = "zstandard-0.21.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:b72060402524ab91e075881f6b6b3f37ab715663313030d0ce983da44960a86f"},
{file = "zstandard-0.21.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:8257752b97134477fb4e413529edaa04fc0457361d304c1319573de00ba796b1"},
{file = "zstandard-0.21.0-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.manylinux_2_12_i686.manylinux2010_i686.whl", hash = "sha256:c053b7c4cbf71cc26808ed67ae955836232f7638444d709bfc302d3e499364fa"},
{file = "zstandard-0.21.0-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:2769730c13638e08b7a983b32cb67775650024632cd0476bf1ba0e6360f5ac7d"},
{file = "zstandard-0.21.0-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:7d3bc4de588b987f3934ca79140e226785d7b5e47e31756761e48644a45a6766"},
{file = "zstandard-0.21.0-cp310-cp310-win32.whl", hash = "sha256:67829fdb82e7393ca68e543894cd0581a79243cc4ec74a836c305c70a5943f07"},
{file = "zstandard-0.21.0-cp310-cp310-win_amd64.whl", hash = "sha256:e6048a287f8d2d6e8bc67f6b42a766c61923641dd4022b7fd3f7439e17ba5a4d"},
{file = "zstandard-0.21.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:7f2afab2c727b6a3d466faee6974a7dad0d9991241c498e7317e5ccf53dbc766"},
{file = "zstandard-0.21.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:ff0852da2abe86326b20abae912d0367878dd0854b8931897d44cfeb18985472"},
{file = "zstandard-0.21.0-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:d12fa383e315b62630bd407477d750ec96a0f438447d0e6e496ab67b8b451d39"},
{file = "zstandard-0.21.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:f1b9703fe2e6b6811886c44052647df7c37478af1b4a1a9078585806f42e5b15"},
{file = "zstandard-0.21.0-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:df28aa5c241f59a7ab524f8ad8bb75d9a23f7ed9d501b0fed6d40ec3064784e8"},
{file = "zstandard-0.21.0-cp311-cp311-win32.whl", hash = "sha256:0aad6090ac164a9d237d096c8af241b8dcd015524ac6dbec1330092dba151657"},
{file = "zstandard-0.21.0-cp311-cp311-win_amd64.whl", hash = "sha256:48b6233b5c4cacb7afb0ee6b4f91820afbb6c0e3ae0fa10abbc20000acdf4f11"},
{file = "zstandard-0.21.0-cp37-cp37m-macosx_10_9_x86_64.whl", hash = "sha256:e7d560ce14fd209db6adacce8908244503a009c6c39eee0c10f138996cd66d3e"},
{file = "zstandard-0.21.0-cp37-cp37m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:1e6e131a4df2eb6f64961cea6f979cdff22d6e0d5516feb0d09492c8fd36f3bc"},
{file = "zstandard-0.21.0-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:e1e0c62a67ff425927898cf43da2cf6b852289ebcc2054514ea9bf121bec10a5"},
{file = "zstandard-0.21.0-cp37-cp37m-manylinux_2_5_i686.manylinux1_i686.manylinux_2_12_i686.manylinux2010_i686.whl", hash = "sha256:1545fb9cb93e043351d0cb2ee73fa0ab32e61298968667bb924aac166278c3fc"},
{file = "zstandard-0.21.0-cp37-cp37m-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:fe6c821eb6870f81d73bf10e5deed80edcac1e63fbc40610e61f340723fd5f7c"},
{file = "zstandard-0.21.0-cp37-cp37m-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:ddb086ea3b915e50f6604be93f4f64f168d3fc3cef3585bb9a375d5834392d4f"},
{file = "zstandard-0.21.0-cp37-cp37m-win32.whl", hash = "sha256:57ac078ad7333c9db7a74804684099c4c77f98971c151cee18d17a12649bc25c"},
{file = "zstandard-0.21.0-cp37-cp37m-win_amd64.whl", hash = "sha256:1243b01fb7926a5a0417120c57d4c28b25a0200284af0525fddba812d575f605"},
{file = "zstandard-0.21.0-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:ea68b1ba4f9678ac3d3e370d96442a6332d431e5050223626bdce748692226ea"},
{file = "zstandard-0.21.0-cp38-cp38-macosx_11_0_arm64.whl", hash = "sha256:8070c1cdb4587a8aa038638acda3bd97c43c59e1e31705f2766d5576b329e97c"},
{file = "zstandard-0.21.0-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:4af612c96599b17e4930fe58bffd6514e6c25509d120f4eae6031b7595912f85"},
{file = "zstandard-0.21.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:cff891e37b167bc477f35562cda1248acc115dbafbea4f3af54ec70821090965"},
{file = "zstandard-0.21.0-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.manylinux_2_12_i686.manylinux2010_i686.whl", hash = "sha256:a9fec02ce2b38e8b2e86079ff0b912445495e8ab0b137f9c0505f88ad0d61296"},
{file = "zstandard-0.21.0-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:0bdbe350691dec3078b187b8304e6a9c4d9db3eb2d50ab5b1d748533e746d099"},
{file = "zstandard-0.21.0-cp38-cp38-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:b69cccd06a4a0a1d9fb3ec9a97600055cf03030ed7048d4bcb88c574f7895773"},
{file = "zstandard-0.21.0-cp38-cp38-win32.whl", hash = "sha256:9980489f066a391c5572bc7dc471e903fb134e0b0001ea9b1d3eff85af0a6f1b"},
{file = "zstandard-0.21.0-cp38-cp38-win_amd64.whl", hash = "sha256:0e1e94a9d9e35dc04bf90055e914077c80b1e0c15454cc5419e82529d3e70728"},
{file = "zstandard-0.21.0-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:d2d61675b2a73edcef5e327e38eb62bdfc89009960f0e3991eae5cc3d54718de"},
{file = "zstandard-0.21.0-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:25fbfef672ad798afab12e8fd204d122fca3bc8e2dcb0a2ba73bf0a0ac0f5f07"},
{file = "zstandard-0.21.0-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:62957069a7c2626ae80023998757e27bd28d933b165c487ab6f83ad3337f773d"},
{file = "zstandard-0.21.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:14e10ed461e4807471075d4b7a2af51f5234c8f1e2a0c1d37d5ca49aaaad49e8"},
{file = "zstandard-0.21.0-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.manylinux_2_12_i686.manylinux2010_i686.whl", hash = "sha256:9cff89a036c639a6a9299bf19e16bfb9ac7def9a7634c52c257166db09d950e7"},
{file = "zstandard-0.21.0-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:52b2b5e3e7670bd25835e0e0730a236f2b0df87672d99d3bf4bf87248aa659fb"},
{file = "zstandard-0.21.0-cp39-cp39-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:b1367da0dde8ae5040ef0413fb57b5baeac39d8931c70536d5f013b11d3fc3a5"},
{file = "zstandard-0.21.0-cp39-cp39-win32.whl", hash = "sha256:db62cbe7a965e68ad2217a056107cc43d41764c66c895be05cf9c8b19578ce9c"},
{file = "zstandard-0.21.0-cp39-cp39-win_amd64.whl", hash = "sha256:a8d200617d5c876221304b0e3fe43307adde291b4a897e7b0617a61611dfff6a"},
{file = "zstandard-0.21.0.tar.gz", hash = "sha256:f08e3a10d01a247877e4cb61a82a319ea746c356a3786558bed2481e6c405546"},
]
[package.dependencies]
cffi = {version = ">=1.11", markers = "platform_python_implementation == \"PyPy\""}
[package.extras]
cffi = ["cffi (>=1.11)"]
[metadata]
lock-version = "2.0"
python-versions = "^3.9"
content-hash = "c5981d8d7c2deadd47c823bc35f86f830c8e320b653d2d3718bade1f4d2dabca"
content-hash = "c40f62277e788011920f4edb6f7392046ee440f792a104c903097415def9a916"

View File

@@ -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

View File

@@ -160,19 +160,6 @@ impl BackendType<'_, ClientCredentials<'_>> {
Test(_) => Some("test".to_owned()),
}
}
/// Get username from the credentials.
pub fn get_user(&self) -> &str {
use BackendType::*;
match self {
Console(_, creds) => creds.user,
Postgres(_, creds) => creds.user,
Link(_) => "link",
Test(_) => "test",
}
}
/// Authenticate the client via the requested backend, possibly using credentials.
#[tracing::instrument(fields(allow_cleartext = allow_cleartext), skip_all)]
pub async fn authenticate(

View File

@@ -168,11 +168,6 @@ async fn task_main(
.instrument(tracing::info_span!("handle_client", ?session_id))
);
}
Some(Err(e)) = connections.join_next(), if !connections.is_empty() => {
if !e.is_panic() && !e.is_cancelled() {
warn!("unexpected error from joined connection task: {e:?}");
}
}
_ = cancellation_token.cancelled() => {
drop(listener);
break;

View File

@@ -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())
}
}

Some files were not shown because too many files have changed in this diff Show More