Compare commits

..

10 Commits

Author SHA1 Message Date
Vlad Lazar
950df2668c fixup: hakari 2025-07-30 16:10:41 +01:00
Vlad Lazar
d9774e2c67 fixup: revert lock file changes 2025-07-30 10:27:25 +01:00
Vlad Lazar
dc25f1de90 Merge remote-tracking branch 'origin/main' into vlad/port-storcon-persistence 2025-07-29 19:06:18 +01:00
Vlad Lazar
2f9cc9a11e tests: remove tests that depend on libpq 2025-07-29 17:35:47 +01:00
Vlad Lazar
e92ed85e9a Merge remote-tracking branch 'origin/main' into vlad/port-storcon-persistence 2025-07-29 10:45:21 +01:00
Vlad Lazar
07396d3fc7 fixup: massage imports for tests 2025-07-29 10:09:50 +01:00
Vlad Lazar
d5cd2de3cf fixup: allow unused code and fix blocking thread count 2025-07-29 10:09:50 +01:00
Vlad Lazar
bab09fffb6 storcon: add persistence configuration options 2025-07-25 19:00:28 +01:00
Vlad Lazar
7bd73eba72 storcon: cut over to the hadron impl for establishing connection 2025-07-25 19:00:28 +01:00
Vlad Lazar
a1924e72ad storcon: port over hadron persistence changes 2025-07-25 19:00:22 +01:00
153 changed files with 2079 additions and 8893 deletions

1
.gitignore vendored
View File

@@ -15,7 +15,6 @@ neon.iml
/.neon
/integration_tests/.neon
compaction-suite-results.*
pgxn/neon/communicator/communicator_bindings.h
docker-compose/docker-compose-parallel.yml
# Coverage

108
Cargo.lock generated
View File

@@ -259,17 +259,6 @@ version = "1.1.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "a8ab6b55fe97976e46f91ddbed8d147d966475dc29b2032757ba47e02376fbc3"
[[package]]
name = "atomic_enum"
version = "0.3.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "99e1aca718ea7b89985790c94aad72d77533063fe00bc497bb79a7c2dae6a661"
dependencies = [
"proc-macro2",
"quote",
"syn 2.0.100",
]
[[package]]
name = "autocfg"
version = "1.1.0"
@@ -1307,30 +1296,13 @@ dependencies = [
name = "communicator"
version = "0.1.0"
dependencies = [
"atomic_enum",
"axum",
"bytes",
"cbindgen",
"clashmap",
"http 1.3.1",
"libc",
"measured",
"metrics",
"neon-shmem",
"nix 0.30.1",
"pageserver_api",
"pageserver_client_grpc",
"pageserver_page_api",
"prometheus",
"prost 0.13.5",
"strum_macros",
"thiserror 1.0.69",
"tokio",
"tokio-pipe",
"tonic",
"tracing",
"tracing-subscriber",
"uring-common",
"utils",
"workspace_hack",
]
@@ -1671,9 +1643,9 @@ dependencies = [
[[package]]
name = "crossbeam-utils"
version = "0.8.21"
version = "0.8.19"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d0a5c400df2834b80a4c3327b3aad3a4c4cd4de0629063962b03235697506a28"
checksum = "248e3bacc7dc6baa3b21e405ee045c3047101a49145e7e9eca583ab4c2ca5345"
[[package]]
name = "crossterm"
@@ -2389,12 +2361,6 @@ version = "1.0.7"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "3f9eec918d3f24069decb9af1554cad7c880e2da24a9afd88aca000531ab82c1"
[[package]]
name = "foldhash"
version = "0.1.5"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d9c4f5dac5e15c24eb999c26181a6ca40b39fe946cbe4c263c7209467bc83af2"
[[package]]
name = "form_urlencoded"
version = "1.2.1"
@@ -2776,16 +2742,6 @@ version = "0.15.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "bf151400ff0baff5465007dd2f3e717f3fe502074ca563069ce3a6629d07b289"
[[package]]
name = "hashbrown"
version = "0.15.4"
source = "git+https://github.com/quantumish/hashbrown.git?rev=6610e6d#6610e6d2b1f288ef7b0709a3efefbc846395dc5e"
dependencies = [
"allocator-api2",
"equivalent",
"foldhash",
]
[[package]]
name = "hashlink"
version = "0.9.1"
@@ -3866,7 +3822,7 @@ dependencies = [
"prometheus",
"rand 0.9.1",
"rand_distr",
"twox-hash 1.6.3",
"twox-hash",
]
[[package]]
@@ -3972,21 +3928,15 @@ checksum = "e5ce46fe64a9d73be07dcbe690a38ce1b293be448fd8ce1e6c1b8062c9f72c6a"
name = "neon-shmem"
version = "0.1.0"
dependencies = [
"ahash",
"criterion",
"hashbrown 0.15.4",
"libc",
"lock_api",
"nix 0.30.1",
"rand 0.9.1",
"rand_distr",
"rustc-hash 2.1.1",
"seahash",
"tempfile",
"thiserror 1.0.69",
"twox-hash 2.1.1",
"workspace_hack",
"xxhash-rust",
]
[[package]]
@@ -4441,16 +4391,13 @@ version = "0.1.0"
dependencies = [
"anyhow",
"async-trait",
"axum",
"bytes",
"camino",
"clap",
"futures",
"hdrhistogram",
"http 1.3.1",
"humantime",
"humantime-serde",
"metrics",
"pageserver_api",
"pageserver_client",
"pageserver_client_grpc",
@@ -4540,7 +4487,6 @@ dependencies = [
"pageserver_client",
"pageserver_compaction",
"pageserver_page_api",
"peekable",
"pem",
"pin-project-lite",
"postgres-protocol",
@@ -4554,7 +4500,6 @@ dependencies = [
"pprof",
"pq_proto",
"procfs",
"prost 0.13.5",
"rand 0.9.1",
"range-set-blaze",
"regex",
@@ -4591,7 +4536,7 @@ dependencies = [
"tower 0.5.2",
"tracing",
"tracing-utils",
"twox-hash 1.6.3",
"twox-hash",
"url",
"utils",
"uuid",
@@ -4803,7 +4748,7 @@ dependencies = [
"paste",
"seq-macro",
"thrift",
"twox-hash 1.6.3",
"twox-hash",
"zstd",
"zstd-sys",
]
@@ -4849,15 +4794,6 @@ dependencies = [
"sha2",
]
[[package]]
name = "peekable"
version = "0.3.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "225f9651e475709164f871dc2f5724956be59cb9edb055372ffeeab01ec2d20b"
dependencies = [
"smallvec",
]
[[package]]
name = "pem"
version = "3.0.3"
@@ -6557,12 +6493,6 @@ version = "0.4.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "621e3680f3e07db4c9c2c3fb07c6223ab2fab2e54bd3c04c3ae037990f428c32"
[[package]]
name = "seahash"
version = "4.1.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "1c107b6f4780854c8b126e228ea8869f4d7b71260f962fefb57b996b8959ba6b"
[[package]]
name = "sec1"
version = "0.3.0"
@@ -7716,16 +7646,6 @@ dependencies = [
"syn 2.0.100",
]
[[package]]
name = "tokio-pipe"
version = "0.2.12"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "f213a84bffbd61b8fa0ba8a044b4bbe35d471d0b518867181e82bd5c15542784"
dependencies = [
"libc",
"tokio",
]
[[package]]
name = "tokio-postgres"
version = "0.7.10"
@@ -8263,15 +8183,6 @@ dependencies = [
"static_assertions",
]
[[package]]
name = "twox-hash"
version = "2.1.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "8b907da542cbced5261bd3256de1b3a1bf340a3d37f93425a07362a1d687de56"
dependencies = [
"rand 0.9.1",
]
[[package]]
name = "typed-json"
version = "0.1.1"
@@ -9102,8 +9013,8 @@ dependencies = [
"clap",
"clap_builder",
"const-oid",
"criterion",
"crossbeam-epoch",
"crossbeam-utils",
"crypto-bigint 0.5.5",
"der 0.7.8",
"deranged",
@@ -9146,6 +9057,7 @@ dependencies = [
"num-iter",
"num-rational",
"num-traits",
"once_cell",
"p256 0.13.2",
"parquet",
"portable-atomic",
@@ -9254,12 +9166,6 @@ version = "0.13.5"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "4d25c75bf9ea12c4040a97f829154768bbbce366287e2dc044af160cd79a13fd"
[[package]]
name = "xxhash-rust"
version = "0.8.15"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "fdd20c5420375476fbd4394763288da7eb0cc0b8c11deed431a91562af7335d3"
[[package]]
name = "yansi"
version = "1.0.1"

View File

@@ -93,7 +93,6 @@ clap = { version = "4.0", features = ["derive", "env"] }
clashmap = { version = "1.0", features = ["raw-api"] }
comfy-table = "7.1"
const_format = "0.2"
crossbeam-utils = "0.8.21"
crc32c = "0.6"
diatomic-waker = { version = "0.2.3" }
either = "1.8"
@@ -153,7 +152,6 @@ parquet = { version = "53", default-features = false, features = ["zstd"] }
parquet_derive = "53"
pbkdf2 = { version = "0.12.1", features = ["simple", "std"] }
pem = "3.0.3"
peekable = "0.3.0"
pin-project-lite = "0.2"
pprof = { version = "0.14", features = ["criterion", "flamegraph", "frame-pointer", "prost-codec"] }
procfs = "0.16"
@@ -192,7 +190,6 @@ smallvec = "1.11"
smol_str = { version = "0.2.0", features = ["serde"] }
socket2 = "0.5"
spki = "0.7.3"
spin = "0.9.8"
strum = "0.26"
strum_macros = "0.26"
"subtle" = "2.5.0"
@@ -204,6 +201,7 @@ thiserror = "1.0"
tikv-jemallocator = { version = "0.6", features = ["profiling", "stats", "unprefixed_malloc_on_supported_platforms"] }
tikv-jemalloc-ctl = { version = "0.6", features = ["stats"] }
tokio = { version = "1.43.1", features = ["macros"] }
tokio-epoll-uring = { git = "https://github.com/neondatabase/tokio-epoll-uring.git" , branch = "main" }
tokio-io-timeout = "1.2.0"
tokio-postgres-rustls = "0.12.0"
tokio-rustls = { version = "0.26.0", default-features = false, features = ["tls12", "ring"]}
@@ -244,9 +242,6 @@ zeroize = "1.8"
env_logger = "0.11"
log = "0.4"
tokio-epoll-uring = { git = "https://github.com/neondatabase/tokio-epoll-uring.git" , branch = "main" }
uring-common = { git = "https://github.com/neondatabase/tokio-epoll-uring.git" , branch = "main" }
## Libraries from neondatabase/ git forks, ideally with changes to be upstreamed
postgres = { git = "https://github.com/neondatabase/rust-postgres.git", branch = "neon" }
postgres-protocol = { git = "https://github.com/neondatabase/rust-postgres.git", branch = "neon" }

View File

@@ -1 +1 @@
SELECT num_requested AS checkpoints_req FROM pg_catalog.pg_stat_checkpointer;
SELECT num_requested AS checkpoints_req FROM pg_stat_checkpointer;

View File

@@ -1 +1 @@
SELECT checkpoints_req FROM pg_catalog.pg_stat_bgwriter;
SELECT checkpoints_req FROM pg_stat_bgwriter;

View File

@@ -1 +1 @@
SELECT checkpoints_timed FROM pg_catalog.pg_stat_bgwriter;
SELECT checkpoints_timed FROM pg_stat_bgwriter;

View File

@@ -1 +1 @@
SELECT (neon.backpressure_throttling_time()::pg_catalog.float8 / 1000000) AS throttled;
SELECT (neon.backpressure_throttling_time()::float8 / 1000000) AS throttled;

View File

@@ -1,4 +1,4 @@
SELECT CASE
WHEN pg_catalog.pg_is_in_recovery() THEN (pg_catalog.pg_last_wal_replay_lsn() - '0/0')::pg_catalog.FLOAT8
ELSE (pg_catalog.pg_current_wal_lsn() - '0/0')::pg_catalog.FLOAT8
WHEN pg_catalog.pg_is_in_recovery() THEN (pg_last_wal_replay_lsn() - '0/0')::FLOAT8
ELSE (pg_current_wal_lsn() - '0/0')::FLOAT8
END AS lsn;

View File

@@ -1,7 +1,7 @@
SELECT
(SELECT setting FROM pg_catalog.pg_settings WHERE name = 'neon.timeline_id') AS timeline_id,
(SELECT setting FROM pg_settings WHERE name = 'neon.timeline_id') AS timeline_id,
-- Postgres creates temporary snapshot files of the form %X-%X.snap.%d.tmp.
-- These temporary snapshot files are renamed to the actual snapshot files
-- after they are completely built. We only WAL-log the completely built
-- snapshot files
(SELECT COUNT(*) FROM pg_catalog.pg_ls_dir('pg_logical/snapshots') AS name WHERE name LIKE '%.snap') AS num_logical_snapshot_files;
(SELECT COUNT(*) FROM pg_ls_dir('pg_logical/snapshots') AS name WHERE name LIKE '%.snap') AS num_logical_snapshot_files;

View File

@@ -1,7 +1,7 @@
SELECT
(SELECT pg_catalog.current_setting('neon.timeline_id')) AS timeline_id,
(SELECT current_setting('neon.timeline_id')) AS timeline_id,
-- Postgres creates temporary snapshot files of the form %X-%X.snap.%d.tmp.
-- These temporary snapshot files are renamed to the actual snapshot files
-- after they are completely built. We only WAL-log the completely built
-- snapshot files
(SELECT COALESCE(pg_catalog.sum(size), 0) FROM pg_catalog.pg_ls_logicalsnapdir() WHERE name LIKE '%.snap') AS logical_snapshots_bytes;
(SELECT COALESCE(sum(size), 0) FROM pg_ls_logicalsnapdir() WHERE name LIKE '%.snap') AS logical_snapshots_bytes;

View File

@@ -1,9 +1,9 @@
SELECT
(SELECT setting FROM pg_catalog.pg_settings WHERE name = 'neon.timeline_id') AS timeline_id,
(SELECT setting FROM pg_settings WHERE name = 'neon.timeline_id') AS timeline_id,
-- Postgres creates temporary snapshot files of the form %X-%X.snap.%d.tmp.
-- These temporary snapshot files are renamed to the actual snapshot files
-- after they are completely built. We only WAL-log the completely built
-- snapshot files
(SELECT COALESCE(pg_catalog.sum((pg_catalog.pg_stat_file('pg_logical/snapshots/' || name, missing_ok => true)).size), 0)
FROM (SELECT * FROM pg_catalog.pg_ls_dir('pg_logical/snapshots') WHERE pg_ls_dir LIKE '%.snap') AS name
(SELECT COALESCE(sum((pg_stat_file('pg_logical/snapshots/' || name, missing_ok => true)).size), 0)
FROM (SELECT * FROM pg_ls_dir('pg_logical/snapshots') WHERE pg_ls_dir LIKE '%.snap') AS name
) AS logical_snapshots_bytes;

View File

@@ -1 +1 @@
SELECT pg_catalog.current_setting('max_connections') AS max_connections;
SELECT current_setting('max_connections') as max_connections;

View File

@@ -1,4 +1,4 @@
SELECT datname database_name,
pg_catalog.age(datfrozenxid) frozen_xid_age
FROM pg_catalog.pg_database
age(datfrozenxid) frozen_xid_age
FROM pg_database
ORDER BY frozen_xid_age DESC LIMIT 10;

View File

@@ -1,4 +1,4 @@
SELECT datname database_name,
pg_catalog.mxid_age(datminmxid) min_mxid_age
FROM pg_catalog.pg_database
mxid_age(datminmxid) min_mxid_age
FROM pg_database
ORDER BY min_mxid_age DESC LIMIT 10;

View File

@@ -1,4 +1,4 @@
SELECT CASE
WHEN pg_catalog.pg_is_in_recovery() THEN (pg_catalog.pg_last_wal_receive_lsn() - '0/0')::pg_catalog.FLOAT8
WHEN pg_catalog.pg_is_in_recovery() THEN (pg_last_wal_receive_lsn() - '0/0')::FLOAT8
ELSE 0
END AS lsn;

View File

@@ -1 +1 @@
SELECT subenabled::pg_catalog.text AS enabled, pg_catalog.count(*) AS subscriptions_count FROM pg_catalog.pg_subscription GROUP BY subenabled;
SELECT subenabled::text AS enabled, count(*) AS subscriptions_count FROM pg_subscription GROUP BY subenabled;

View File

@@ -1 +1 @@
SELECT datname, state, pg_catalog.count(*) AS count FROM pg_catalog.pg_stat_activity WHERE state <> '' GROUP BY datname, state;
SELECT datname, state, count(*) AS count FROM pg_stat_activity WHERE state <> '' GROUP BY datname, state;

View File

@@ -1,5 +1,5 @@
SELECT pg_catalog.sum(pg_catalog.pg_database_size(datname)) AS total
FROM pg_catalog.pg_database
SELECT sum(pg_database_size(datname)) AS total
FROM pg_database
-- Ignore invalid databases, as we will likely have problems with
-- getting their size from the Pageserver.
WHERE datconnlimit != -2;

View File

@@ -3,6 +3,6 @@
-- minutes.
SELECT
x::pg_catalog.text AS duration_seconds,
x::text as duration_seconds,
neon.approximate_working_set_size_seconds(x) AS size
FROM (SELECT generate_series * 60 AS x FROM generate_series(1, 60)) AS t (x);

View File

@@ -3,6 +3,6 @@
SELECT
x AS duration,
neon.approximate_working_set_size_seconds(extract('epoch' FROM x::pg_catalog.interval)::pg_catalog.int4) AS size FROM (
neon.approximate_working_set_size_seconds(extract('epoch' FROM x::interval)::int) AS size FROM (
VALUES ('5m'), ('15m'), ('1h')
) AS t (x);

View File

@@ -1 +1 @@
SELECT pg_catalog.pg_size_bytes(pg_catalog.current_setting('neon.file_cache_size_limit')) AS lfc_cache_size_limit;
SELECT pg_size_bytes(current_setting('neon.file_cache_size_limit')) AS lfc_cache_size_limit;

View File

@@ -1,3 +1,3 @@
SELECT slot_name, (restart_lsn - '0/0')::pg_catalog.FLOAT8 AS restart_lsn
FROM pg_catalog.pg_replication_slots
SELECT slot_name, (restart_lsn - '0/0')::FLOAT8 as restart_lsn
FROM pg_replication_slots
WHERE slot_type = 'logical';

View File

@@ -1 +1 @@
SELECT setting::pg_catalog.int4 AS max_cluster_size FROM pg_catalog.pg_settings WHERE name = 'neon.max_cluster_size';
SELECT setting::int AS max_cluster_size FROM pg_settings WHERE name = 'neon.max_cluster_size';

View File

@@ -1,13 +1,13 @@
-- We export stats for 10 non-system databases. Without this limit it is too
-- easy to abuse the system by creating lots of databases.
SELECT pg_catalog.pg_database_size(datname) AS db_size,
SELECT pg_database_size(datname) AS db_size,
deadlocks,
tup_inserted AS inserted,
tup_updated AS updated,
tup_deleted AS deleted,
datname
FROM pg_catalog.pg_stat_database
FROM pg_stat_database
WHERE datname IN (
SELECT datname FROM pg_database
-- Ignore invalid databases, as we will likely have problems with

View File

@@ -3,4 +3,4 @@
-- replay LSN may have advanced past the receive LSN we are using for the
-- calculation.
SELECT GREATEST(0, pg_catalog.pg_wal_lsn_diff(pg_catalog.pg_last_wal_receive_lsn(), pg_catalog.pg_last_wal_replay_lsn())) AS replication_delay_bytes;
SELECT GREATEST(0, pg_wal_lsn_diff(pg_last_wal_receive_lsn(), pg_last_wal_replay_lsn())) AS replication_delay_bytes;

View File

@@ -1,5 +1,5 @@
SELECT
CASE
WHEN pg_catalog.pg_last_wal_receive_lsn() = pg_catalog.pg_last_wal_replay_lsn() THEN 0
ELSE GREATEST(0, EXTRACT (EPOCH FROM pg_catalog.now() - pg_catalog.pg_last_xact_replay_timestamp()))
WHEN pg_last_wal_receive_lsn() = pg_last_wal_replay_lsn() THEN 0
ELSE GREATEST(0, EXTRACT (EPOCH FROM now() - pg_last_xact_replay_timestamp()))
END AS replication_delay_seconds;

View File

@@ -1,10 +1,10 @@
SELECT
slot_name,
pg_catalog.pg_wal_lsn_diff(
pg_wal_lsn_diff(
CASE
WHEN pg_catalog.pg_is_in_recovery() THEN pg_catalog.pg_last_wal_replay_lsn()
ELSE pg_catalog.pg_current_wal_lsn()
WHEN pg_is_in_recovery() THEN pg_last_wal_replay_lsn()
ELSE pg_current_wal_lsn()
END,
restart_lsn)::pg_catalog.FLOAT8 AS retained_wal
FROM pg_catalog.pg_replication_slots
restart_lsn)::FLOAT8 AS retained_wal
FROM pg_replication_slots
WHERE active = false;

View File

@@ -4,4 +4,4 @@ SELECT
WHEN wal_status = 'lost' THEN 1
ELSE 0
END AS wal_is_lost
FROM pg_catalog.pg_replication_slots;
FROM pg_replication_slots;

View File

@@ -279,7 +279,7 @@ fn main() -> Result<()> {
config,
)?;
let exit_code = compute_node.run().context("running compute node")?;
let exit_code = compute_node.run()?;
scenario.teardown();

View File

@@ -24,9 +24,9 @@ pub async fn check_writability(compute: &ComputeNode) -> Result<()> {
});
let query = "
INSERT INTO public.health_check VALUES (1, pg_catalog.now())
INSERT INTO health_check VALUES (1, now())
ON CONFLICT (id) DO UPDATE
SET updated_at = pg_catalog.now();";
SET updated_at = now();";
match client.simple_query(query).await {
Result::Ok(result) => {

View File

@@ -7,7 +7,7 @@ use compute_api::responses::{
};
use compute_api::spec::{
ComputeAudit, ComputeFeature, ComputeMode, ComputeSpec, ExtVersion, GenericOption,
PageserverConnectionInfo, PageserverProtocol, PgIdent, Role,
PageserverProtocol, PgIdent, Role,
};
use futures::StreamExt;
use futures::future::join_all;
@@ -38,7 +38,7 @@ use utils::id::{TenantId, TimelineId};
use utils::lsn::Lsn;
use utils::measured_stream::MeasuredReader;
use utils::pid_file;
use utils::shard::{ShardIndex, ShardNumber, ShardStripeSize};
use utils::shard::{ShardCount, ShardIndex, ShardNumber};
use crate::configurator::launch_configurator;
use crate::disk_quota::set_disk_quota;
@@ -249,7 +249,7 @@ pub struct ParsedSpec {
pub spec: ComputeSpec,
pub tenant_id: TenantId,
pub timeline_id: TimelineId,
pub pageserver_conninfo: PageserverConnectionInfo,
pub pageserver_connstr: String,
pub safekeeper_connstrings: Vec<String>,
pub storage_auth_token: Option<String>,
/// k8s dns name and port
@@ -297,47 +297,25 @@ impl ParsedSpec {
}
impl TryFrom<ComputeSpec> for ParsedSpec {
type Error = anyhow::Error;
fn try_from(spec: ComputeSpec) -> Result<Self, anyhow::Error> {
type Error = String;
fn try_from(spec: ComputeSpec) -> Result<Self, String> {
// Extract the options from the spec file that are needed to connect to
// the storage system.
//
// In compute specs generated by old control plane versions, the spec file might
// be missing the `pageserver_connection_info` field. In that case, we need to dig
// the pageserver connection info from the `pageserver_connstr` field instead, or
// if that's missing too, from the GUC in the cluster.settings field.
let mut pageserver_conninfo = spec.pageserver_connection_info.clone();
if pageserver_conninfo.is_none() {
if let Some(pageserver_connstr_field) = &spec.pageserver_connstring {
pageserver_conninfo = Some(PageserverConnectionInfo::from_connstr(
pageserver_connstr_field,
spec.shard_stripe_size,
)?);
}
}
if pageserver_conninfo.is_none() {
if let Some(guc) = spec.cluster.settings.find("neon.pageserver_connstring") {
let stripe_size = if let Some(guc) = spec.cluster.settings.find("neon.stripe_size")
{
Some(ShardStripeSize(u32::from_str(&guc)?))
} else {
None
};
pageserver_conninfo =
Some(PageserverConnectionInfo::from_connstr(&guc, stripe_size)?);
}
}
let pageserver_conninfo = pageserver_conninfo.ok_or(anyhow::anyhow!(
"pageserver connection information should be provided"
))?;
// Similarly for safekeeper connection strings
// For backwards-compatibility, the top-level fields in the spec file
// may be empty. In that case, we need to dig them from the GUCs in the
// cluster.settings field.
let pageserver_connstr = spec
.pageserver_connstring
.clone()
.or_else(|| spec.cluster.settings.find("neon.pageserver_connstring"))
.ok_or("pageserver connstr should be provided")?;
let safekeeper_connstrings = if spec.safekeeper_connstrings.is_empty() {
if matches!(spec.mode, ComputeMode::Primary) {
spec.cluster
.settings
.find("neon.safekeepers")
.ok_or(anyhow::anyhow!("safekeeper connstrings should be provided"))?
.ok_or("safekeeper connstrings should be provided")?
.split(',')
.map(|str| str.to_string())
.collect()
@@ -352,22 +330,22 @@ impl TryFrom<ComputeSpec> for ParsedSpec {
let tenant_id: TenantId = if let Some(tenant_id) = spec.tenant_id {
tenant_id
} else {
let guc = spec
.cluster
spec.cluster
.settings
.find("neon.tenant_id")
.ok_or(anyhow::anyhow!("tenant id should be provided"))?;
TenantId::from_str(&guc).context("invalid tenant id")?
.ok_or("tenant id should be provided")
.map(|s| TenantId::from_str(&s))?
.or(Err("invalid tenant id"))?
};
let timeline_id: TimelineId = if let Some(timeline_id) = spec.timeline_id {
timeline_id
} else {
let guc = spec
.cluster
spec.cluster
.settings
.find("neon.timeline_id")
.ok_or(anyhow::anyhow!("timeline id should be provided"))?;
TimelineId::from_str(&guc).context(anyhow::anyhow!("invalid timeline id"))?
.ok_or("timeline id should be provided")
.map(|s| TimelineId::from_str(&s))?
.or(Err("invalid timeline id"))?
};
let endpoint_storage_addr: Option<String> = spec
@@ -381,7 +359,7 @@ impl TryFrom<ComputeSpec> for ParsedSpec {
let res = ParsedSpec {
spec,
pageserver_conninfo,
pageserver_connstr,
safekeeper_connstrings,
storage_auth_token,
tenant_id,
@@ -391,7 +369,7 @@ impl TryFrom<ComputeSpec> for ParsedSpec {
};
// Now check validity of the parsed specification
res.validate().map_err(anyhow::Error::msg)?;
res.validate()?;
Ok(res)
}
}
@@ -583,7 +561,7 @@ impl ComputeNode {
// that can affect `compute_ctl` and prevent it from properly configuring the database schema.
// Unset them via connection string options before connecting to the database.
// N.B. keep it in sync with `ZENITH_OPTIONS` in `get_maintenance_client()`.
const EXTRA_OPTIONS: &str = "-c role=cloud_admin -c default_transaction_read_only=off -c search_path='' -c statement_timeout=0 -c pgaudit.log=none";
const EXTRA_OPTIONS: &str = "-c role=cloud_admin -c default_transaction_read_only=off -c search_path=public -c statement_timeout=0 -c pgaudit.log=none";
let options = match conn_conf.get_options() {
// Allow the control plane to override any options set by the
// compute
@@ -1217,10 +1195,12 @@ impl ComputeNode {
fn try_get_basebackup(&self, compute_state: &ComputeState, lsn: Lsn) -> Result<()> {
let spec = compute_state.pspec.as_ref().expect("spec must be set");
let shard0_connstr = spec.pageserver_connstr.split(',').next().unwrap();
let started = Instant::now();
let (connected, size) = match spec.pageserver_conninfo.prefer_protocol {
PageserverProtocol::Grpc => self.try_get_basebackup_grpc(spec, lsn)?,
let (connected, size) = match PageserverProtocol::from_connstring(shard0_connstr)? {
PageserverProtocol::Libpq => self.try_get_basebackup_libpq(spec, lsn)?,
PageserverProtocol::Grpc => self.try_get_basebackup_grpc(spec, lsn)?,
};
self.fix_zenith_signal_neon_signal()?;
@@ -1258,20 +1238,23 @@ impl ComputeNode {
/// Fetches a basebackup via gRPC. The connstring must use grpc://. Returns the timestamp when
/// the connection was established, and the (compressed) size of the basebackup.
fn try_get_basebackup_grpc(&self, spec: &ParsedSpec, lsn: Lsn) -> Result<(Instant, usize)> {
let shard0_index = ShardIndex {
shard_number: ShardNumber(0),
shard_count: spec.pageserver_conninfo.shard_count,
let shard0_connstr = spec
.pageserver_connstr
.split(',')
.next()
.unwrap()
.to_string();
let shard_index = match spec.pageserver_connstr.split(',').count() as u8 {
0 | 1 => ShardIndex::unsharded(),
count => ShardIndex::new(ShardNumber(0), ShardCount(count)),
};
let shard0_url = spec
.pageserver_conninfo
.shard_url(ShardNumber(0), PageserverProtocol::Grpc)?
.to_owned();
let (reader, connected) = tokio::runtime::Handle::current().block_on(async move {
let mut client = page_api::Client::connect(
shard0_url,
shard0_connstr,
spec.tenant_id,
spec.timeline_id,
shard0_index,
shard_index,
spec.storage_auth_token.clone(),
None, // NB: base backups use payload compression
)
@@ -1303,9 +1286,7 @@ impl ComputeNode {
/// Fetches a basebackup via libpq. The connstring must use postgresql://. Returns the timestamp
/// when the connection was established, and the (compressed) size of the basebackup.
fn try_get_basebackup_libpq(&self, spec: &ParsedSpec, lsn: Lsn) -> Result<(Instant, usize)> {
let shard0_connstr = spec
.pageserver_conninfo
.shard_url(ShardNumber(0), PageserverProtocol::Libpq)?;
let shard0_connstr = spec.pageserver_connstr.split(',').next().unwrap();
let mut config = postgres::Config::from_str(shard0_connstr)?;
// Use the storage auth token from the config file, if given.
@@ -1392,7 +1373,10 @@ impl ComputeNode {
return result;
}
Err(ref e) if attempts < max_attempts => {
warn!("Failed to get basebackup: {e:?} (attempt {attempts}/{max_attempts})");
warn!(
"Failed to get basebackup: {} (attempt {}/{})",
e, attempts, max_attempts
);
std::thread::sleep(std::time::Duration::from_millis(retry_period_ms as u64));
retry_period_ms *= 1.5;
}
@@ -1605,8 +1589,16 @@ impl ComputeNode {
}
};
self.get_basebackup(compute_state, lsn)
.with_context(|| format!("failed to get basebackup@{lsn}"))?;
info!(
"getting basebackup@{} from pageserver {}",
lsn, &pspec.pageserver_connstr
);
self.get_basebackup(compute_state, lsn).with_context(|| {
format!(
"failed to get basebackup@{} from pageserver {}",
lsn, &pspec.pageserver_connstr
)
})?;
if let Some(settings) = databricks_settings {
copy_tls_certificates(
@@ -1884,7 +1876,7 @@ impl ComputeNode {
// It doesn't matter what were the options before, here we just want
// to connect and create a new superuser role.
const ZENITH_OPTIONS: &str = "-c role=zenith_admin -c default_transaction_read_only=off -c search_path='' -c statement_timeout=0";
const ZENITH_OPTIONS: &str = "-c role=zenith_admin -c default_transaction_read_only=off -c search_path=public -c statement_timeout=0";
zenith_admin_conf.options(ZENITH_OPTIONS);
let mut client =
@@ -2339,13 +2331,13 @@ impl ComputeNode {
let result = client
.simple_query(
"SELECT
pg_catalog.row_to_json(pss)
row_to_json(pg_stat_statements)
FROM
public.pg_stat_statements pss
pg_stat_statements
WHERE
pss.userid != 'cloud_admin'::pg_catalog.regrole::pg_catalog.oid
userid != 'cloud_admin'::regrole::oid
ORDER BY
(pss.mean_exec_time + pss.mean_plan_time) DESC
(mean_exec_time + mean_plan_time) DESC
LIMIT 100",
)
.await;
@@ -2473,11 +2465,11 @@ LIMIT 100",
// check the role grants first - to gracefully handle read-replicas.
let select = "SELECT privilege_type
FROM pg_catalog.pg_namespace
JOIN LATERAL (SELECT * FROM aclexplode(nspacl) AS x) AS acl ON true
JOIN pg_catalog.pg_user users ON acl.grantee = users.usesysid
WHERE users.usename OPERATOR(pg_catalog.=) $1::pg_catalog.name
AND nspname OPERATOR(pg_catalog.=) $2::pg_catalog.name";
FROM pg_namespace
JOIN LATERAL (SELECT * FROM aclexplode(nspacl) AS x) acl ON true
JOIN pg_user users ON acl.grantee = users.usesysid
WHERE users.usename = $1
AND nspname = $2";
let rows = db_client
.query(select, &[role_name, schema_name])
.await
@@ -2546,9 +2538,8 @@ LIMIT 100",
.await
.with_context(|| format!("Failed to execute query: {query}"))?;
} else {
let query = format!(
"CREATE EXTENSION IF NOT EXISTS {ext_name} WITH SCHEMA public VERSION {quoted_version}"
);
let query =
format!("CREATE EXTENSION IF NOT EXISTS {ext_name} WITH VERSION {quoted_version}");
db_client
.simple_query(&query)
.await
@@ -2651,22 +2642,22 @@ LIMIT 100",
/// The operation will time out after a specified duration.
pub fn wait_timeout_while_pageserver_connstr_unchanged(&self, duration: Duration) {
let state = self.state.lock().unwrap();
let old_pageserver_conninfo = state
let old_pageserver_connstr = state
.pspec
.as_ref()
.expect("spec must be set")
.pageserver_conninfo
.pageserver_connstr
.clone();
let mut unchanged = true;
let _ = self
.state_changed
.wait_timeout_while(state, duration, |s| {
let pageserver_conninfo = &s
let pageserver_connstr = &s
.pspec
.as_ref()
.expect("spec must be set")
.pageserver_conninfo;
unchanged = pageserver_conninfo == &old_pageserver_conninfo;
.pageserver_connstr;
unchanged = pageserver_connstr == &old_pageserver_connstr;
unchanged
})
.unwrap();
@@ -2924,10 +2915,7 @@ mod tests {
match ParsedSpec::try_from(spec.clone()) {
Ok(_p) => panic!("Failed to detect duplicate entry"),
Err(e) => assert!(
e.to_string()
.starts_with("duplicate entry in safekeeper_connstrings:")
),
Err(e) => assert!(e.starts_with("duplicate entry in safekeeper_connstrings:")),
};
}
}

View File

@@ -78,7 +78,7 @@ impl ComputeNode {
const RETRIES: i32 = 20;
for i in 0..=RETRIES {
let row = client
.query_one("SELECT pg_catalog.pg_last_wal_replay_lsn()", &[])
.query_one("SELECT pg_last_wal_replay_lsn()", &[])
.await
.context("getting last replay lsn")?;
let lsn: u64 = row.get::<usize, postgres_types::PgLsn>(0).into();
@@ -103,7 +103,7 @@ impl ComputeNode {
.await
.context("setting safekeepers")?;
client
.query("SELECT pg_catalog.pg_reload_conf()", &[])
.query("SELECT pg_reload_conf()", &[])
.await
.context("reloading postgres config")?;
@@ -113,7 +113,7 @@ impl ComputeNode {
});
let row = client
.query_one("SELECT * FROM pg_catalog.pg_promote()", &[])
.query_one("SELECT * FROM pg_promote()", &[])
.await
.context("pg_promote")?;
if !row.get::<usize, bool>(0) {

View File

@@ -18,8 +18,6 @@ use crate::pg_helpers::{
};
use crate::tls::{self, SERVER_CRT, SERVER_KEY};
use utils::shard::{ShardIndex, ShardNumber};
/// Check that `line` is inside a text file and put it there if it is not.
/// Create file if it doesn't exist.
pub fn line_in_file(path: &Path, line: &str) -> Result<bool> {
@@ -65,102 +63,15 @@ pub fn write_postgres_conf(
writeln!(file, "{conf}")?;
}
// Stripe size GUC should be defined prior to connection string
if let Some(stripe_size) = spec.shard_stripe_size {
writeln!(file, "neon.stripe_size={stripe_size}")?;
}
// Add options for connecting to storage
writeln!(file, "# Neon storage settings")?;
writeln!(file)?;
if let Some(conninfo) = &spec.pageserver_connection_info {
// Stripe size GUC should be defined prior to connection string
if let Some(stripe_size) = conninfo.stripe_size {
writeln!(
file,
"# from compute spec's pageserver_conninfo.stripe_size field"
)?;
writeln!(file, "neon.stripe_size={stripe_size}")?;
}
let mut libpq_urls: Option<Vec<String>> = Some(Vec::new());
let mut grpc_urls: Option<Vec<String>> = Some(Vec::new());
let num_shards = if conninfo.shard_count.0 == 0 {
1 // unsharded, treat it as a single shard
} else {
conninfo.shard_count.0
};
for shard_number in 0..num_shards {
let shard_index = ShardIndex {
shard_number: ShardNumber(shard_number),
shard_count: conninfo.shard_count,
};
let info = conninfo.shards.get(&shard_index).ok_or_else(|| {
anyhow::anyhow!(
"shard {shard_index} missing from pageserver_connection_info shard map"
)
})?;
let first_pageserver = info
.pageservers
.first()
.expect("must have at least one pageserver");
// Add the libpq URL to the array, or if the URL is missing, reset the array
// forgetting any previous entries. All servers must have a libpq URL, or none
// at all.
if let Some(url) = &first_pageserver.libpq_url {
if let Some(ref mut urls) = libpq_urls {
urls.push(url.clone());
}
} else {
libpq_urls = None
}
// Similarly for gRPC URLs
if let Some(url) = &first_pageserver.grpc_url {
if let Some(ref mut urls) = grpc_urls {
urls.push(url.clone());
}
} else {
grpc_urls = None
}
}
if let Some(libpq_urls) = libpq_urls {
writeln!(
file,
"# derived from compute spec's pageserver_conninfo field"
)?;
writeln!(
file,
"neon.pageserver_connstring={}",
escape_conf_value(&libpq_urls.join(","))
)?;
writeln!(file, "neon.use_communicator_worker=false")?;
} else {
writeln!(file, "# no neon.pageserver_connstring")?;
}
if let Some(grpc_urls) = grpc_urls {
writeln!(
file,
"# derived from compute spec's pageserver_conninfo field"
)?;
writeln!(
file,
"neon.pageserver_grpc_urls={}",
escape_conf_value(&grpc_urls.join(","))
)?;
writeln!(file, "neon.use_communicator_worker=true")?;
} else {
writeln!(file, "# no neon.pageserver_grpc_urls")?;
}
} else {
// Stripe size GUC should be defined prior to connection string
if let Some(stripe_size) = spec.shard_stripe_size {
writeln!(file, "# from compute spec's shard_stripe_size field")?;
writeln!(file, "neon.stripe_size={stripe_size}")?;
}
if let Some(s) = &spec.pageserver_connstring {
writeln!(file, "# from compute spec's pageserver_connstring field")?;
writeln!(file, "neon.pageserver_connstring={}", escape_conf_value(s))?;
}
if let Some(s) = &spec.pageserver_connstring {
writeln!(file, "neon.pageserver_connstring={}", escape_conf_value(s))?;
}
if !spec.safekeeper_connstrings.is_empty() {
let mut neon_safekeepers_value = String::new();
tracing::info!(

View File

@@ -122,11 +122,8 @@ fn configurator_main_loop(compute: &Arc<ComputeNode>) {
// into the type system.
assert_eq!(state.status, ComputeStatus::RefreshConfiguration);
if state
.pspec
.as_ref()
.map(|ps| ps.pageserver_conninfo.clone())
== Some(pspec.pageserver_conninfo.clone())
if state.pspec.as_ref().map(|ps| ps.pageserver_connstr.clone())
== Some(pspec.pageserver_connstr.clone())
{
info!(
"Refresh configuration: Retrieved spec is the same as the current spec. Waiting for control plane to update the spec before attempting reconfiguration."

View File

@@ -19,7 +19,7 @@ async fn list_dbs(client: &mut Client) -> Result<Vec<String>, PostgresError> {
.query(
"SELECT datname FROM pg_catalog.pg_database
WHERE datallowconn
AND datconnlimit OPERATOR(pg_catalog.<>) (OPERATOR(pg_catalog.-) 2::pg_catalog.int4)
AND datconnlimit <> - 2
LIMIT 500",
&[],
)
@@ -67,7 +67,7 @@ pub async fn get_installed_extensions(
let extensions: Vec<(String, String, i32)> = client
.query(
"SELECT extname, extversion, extowner::pg_catalog.int4 FROM pg_catalog.pg_extension",
"SELECT extname, extversion, extowner::integer FROM pg_catalog.pg_extension",
&[],
)
.await?

View File

@@ -4,13 +4,14 @@ use std::thread;
use std::time::{Duration, SystemTime};
use anyhow::{Result, bail};
use compute_api::spec::{ComputeMode, PageserverConnectionInfo, PageserverProtocol};
use compute_api::spec::{ComputeMode, PageserverProtocol};
use itertools::Itertools as _;
use pageserver_page_api as page_api;
use postgres::{NoTls, SimpleQueryMessage};
use tracing::{info, warn};
use utils::id::{TenantId, TimelineId};
use utils::lsn::Lsn;
use utils::shard::TenantShardId;
use utils::shard::{ShardCount, ShardNumber, TenantShardId};
use crate::compute::ComputeNode;
@@ -77,16 +78,17 @@ fn acquire_lsn_lease_with_retry(
loop {
// Note: List of pageservers is dynamic, need to re-read configs before each attempt.
let (conninfo, auth) = {
let (connstrings, auth) = {
let state = compute.state.lock().unwrap();
let spec = state.pspec.as_ref().expect("spec must be set");
(
spec.pageserver_conninfo.clone(),
spec.pageserver_connstr.clone(),
spec.storage_auth_token.clone(),
)
};
let result = try_acquire_lsn_lease(conninfo, auth.as_deref(), tenant_id, timeline_id, lsn);
let result =
try_acquire_lsn_lease(&connstrings, auth.as_deref(), tenant_id, timeline_id, lsn);
match result {
Ok(Some(res)) => {
return Ok(res);
@@ -110,44 +112,35 @@ fn acquire_lsn_lease_with_retry(
/// Tries to acquire LSN leases on all Pageserver shards.
fn try_acquire_lsn_lease(
conninfo: PageserverConnectionInfo,
connstrings: &str,
auth: Option<&str>,
tenant_id: TenantId,
timeline_id: TimelineId,
lsn: Lsn,
) -> Result<Option<SystemTime>> {
let connstrings = connstrings.split(',').collect_vec();
let shard_count = connstrings.len();
let mut leases = Vec::new();
for (shard_index, shard) in conninfo.shards.into_iter() {
let tenant_shard_id = TenantShardId {
tenant_id,
shard_number: shard_index.shard_number,
shard_count: shard_index.shard_count,
for (shard_number, &connstring) in connstrings.iter().enumerate() {
let tenant_shard_id = match shard_count {
0 | 1 => TenantShardId::unsharded(tenant_id),
shard_count => TenantShardId {
tenant_id,
shard_number: ShardNumber(shard_number as u8),
shard_count: ShardCount::new(shard_count as u8),
},
};
// XXX: If there are more than pageserver for the one shard, do we need to get a
// leas on all of them? Currently, that's what we assume, but this is hypothetical
// as of this writing, as we never pass the info for more than one pageserver per
// shard.
for pageserver in shard.pageservers {
let lease = match conninfo.prefer_protocol {
PageserverProtocol::Grpc => acquire_lsn_lease_grpc(
&pageserver.grpc_url.unwrap(),
auth,
tenant_shard_id,
timeline_id,
lsn,
)?,
PageserverProtocol::Libpq => acquire_lsn_lease_libpq(
&pageserver.libpq_url.unwrap(),
auth,
tenant_shard_id,
timeline_id,
lsn,
)?,
};
leases.push(lease);
}
let lease = match PageserverProtocol::from_connstring(connstring)? {
PageserverProtocol::Libpq => {
acquire_lsn_lease_libpq(connstring, auth, tenant_shard_id, timeline_id, lsn)?
}
PageserverProtocol::Grpc => {
acquire_lsn_lease_grpc(connstring, auth, tenant_shard_id, timeline_id, lsn)?
}
};
leases.push(lease);
}
Ok(leases.into_iter().min().flatten())

View File

@@ -76,7 +76,7 @@ impl<'m> MigrationRunner<'m> {
self.client
.simple_query("CREATE SCHEMA IF NOT EXISTS neon_migration")
.await?;
self.client.simple_query("CREATE TABLE IF NOT EXISTS neon_migration.migration_id (key pg_catalog.int4 NOT NULL PRIMARY KEY, id pg_catalog.int8 NOT NULL DEFAULT 0)").await?;
self.client.simple_query("CREATE TABLE IF NOT EXISTS neon_migration.migration_id (key INT NOT NULL PRIMARY KEY, id bigint NOT NULL DEFAULT 0)").await?;
self.client
.simple_query(
"INSERT INTO neon_migration.migration_id VALUES (0, 0) ON CONFLICT DO NOTHING",

View File

@@ -15,17 +15,17 @@ DO $$
DECLARE
role_name text;
BEGIN
FOR role_name IN SELECT rolname FROM pg_catalog.pg_roles WHERE pg_catalog.pg_has_role(rolname, '{privileged_role_name}', 'member')
FOR role_name IN SELECT rolname FROM pg_roles WHERE pg_has_role(rolname, '{privileged_role_name}', 'member')
LOOP
RAISE NOTICE 'EXECUTING ALTER ROLE % INHERIT', pg_catalog.quote_ident(role_name);
EXECUTE pg_catalog.format('ALTER ROLE %I INHERIT;', role_name);
RAISE NOTICE 'EXECUTING ALTER ROLE % INHERIT', quote_ident(role_name);
EXECUTE 'ALTER ROLE ' || quote_ident(role_name) || ' INHERIT';
END LOOP;
FOR role_name IN SELECT rolname FROM pg_catalog.pg_roles
FOR role_name IN SELECT rolname FROM pg_roles
WHERE
NOT pg_catalog.pg_has_role(rolname, '{privileged_role_name}', 'member') AND NOT pg_catalog.starts_with(rolname, 'pg_')
NOT pg_has_role(rolname, '{privileged_role_name}', 'member') AND NOT starts_with(rolname, 'pg_')
LOOP
RAISE NOTICE 'EXECUTING ALTER ROLE % NOBYPASSRLS', pg_catalog.quote_ident(role_name);
EXECUTE pg_catalog.format('ALTER ROLE %I NOBYPASSRLS;', role_name);
RAISE NOTICE 'EXECUTING ALTER ROLE % NOBYPASSRLS', quote_ident(role_name);
EXECUTE 'ALTER ROLE ' || quote_ident(role_name) || ' NOBYPASSRLS';
END LOOP;
END $$;

View File

@@ -1,6 +1,6 @@
DO $$
BEGIN
IF (SELECT setting::pg_catalog.numeric >= 160000 FROM pg_catalog.pg_settings WHERE name = 'server_version_num') THEN
IF (SELECT setting::numeric >= 160000 FROM pg_settings WHERE name = 'server_version_num') THEN
EXECUTE 'GRANT pg_create_subscription TO {privileged_role_name}';
END IF;
END $$;

View File

@@ -5,9 +5,9 @@ DO $$
DECLARE
role_name TEXT;
BEGIN
FOR role_name IN SELECT rolname FROM pg_catalog.pg_roles WHERE rolreplication IS TRUE
FOR role_name IN SELECT rolname FROM pg_roles WHERE rolreplication IS TRUE
LOOP
RAISE NOTICE 'EXECUTING ALTER ROLE % NOREPLICATION', pg_catalog.quote_ident(role_name);
EXECUTE pg_catalog.format('ALTER ROLE %I NOREPLICATION;', role_name);
RAISE NOTICE 'EXECUTING ALTER ROLE % NOREPLICATION', quote_ident(role_name);
EXECUTE 'ALTER ROLE ' || quote_ident(role_name) || ' NOREPLICATION';
END LOOP;
END $$;

View File

@@ -1,6 +1,6 @@
DO $$
BEGIN
IF (SELECT setting::pg_catalog.numeric >= 160000 FROM pg_catalog.pg_settings WHERE name OPERATOR(pg_catalog.=) 'server_version_num'::pg_catalog.text) THEN
IF (SELECT setting::numeric >= 160000 FROM pg_settings WHERE name = 'server_version_num') THEN
EXECUTE 'GRANT EXECUTE ON FUNCTION pg_export_snapshot TO {privileged_role_name}';
EXECUTE 'GRANT EXECUTE ON FUNCTION pg_log_standby_snapshot TO {privileged_role_name}';
END IF;

View File

@@ -2,7 +2,7 @@ DO $$
DECLARE
bypassrls boolean;
BEGIN
SELECT rolbypassrls INTO bypassrls FROM pg_catalog.pg_roles WHERE rolname = 'neon_superuser';
SELECT rolbypassrls INTO bypassrls FROM pg_roles WHERE rolname = 'neon_superuser';
IF NOT bypassrls THEN
RAISE EXCEPTION 'neon_superuser cannot bypass RLS';
END IF;

View File

@@ -4,8 +4,8 @@ DECLARE
BEGIN
FOR role IN
SELECT rolname AS name, rolinherit AS inherit
FROM pg_catalog.pg_roles
WHERE pg_catalog.pg_has_role(rolname, 'neon_superuser', 'member')
FROM pg_roles
WHERE pg_has_role(rolname, 'neon_superuser', 'member')
LOOP
IF NOT role.inherit THEN
RAISE EXCEPTION '% cannot inherit', quote_ident(role.name);
@@ -14,12 +14,12 @@ BEGIN
FOR role IN
SELECT rolname AS name, rolbypassrls AS bypassrls
FROM pg_catalog.pg_roles
WHERE NOT pg_catalog.pg_has_role(rolname, 'neon_superuser', 'member')
AND NOT pg_catalog.starts_with(rolname, 'pg_')
FROM pg_roles
WHERE NOT pg_has_role(rolname, 'neon_superuser', 'member')
AND NOT starts_with(rolname, 'pg_')
LOOP
IF role.bypassrls THEN
RAISE EXCEPTION '% can bypass RLS', pg_catalog.quote_ident(role.name);
RAISE EXCEPTION '% can bypass RLS', quote_ident(role.name);
END IF;
END LOOP;
END $$;

View File

@@ -1,10 +1,10 @@
DO $$
BEGIN
IF (SELECT pg_catalog.current_setting('server_version_num')::pg_catalog.numeric < 160000) THEN
IF (SELECT current_setting('server_version_num')::numeric < 160000) THEN
RETURN;
END IF;
IF NOT (SELECT pg_catalog.pg_has_role('neon_superuser', 'pg_create_subscription', 'member')) THEN
IF NOT (SELECT pg_has_role('neon_superuser', 'pg_create_subscription', 'member')) THEN
RAISE EXCEPTION 'neon_superuser cannot execute pg_create_subscription';
END IF;
END $$;

View File

@@ -2,12 +2,12 @@ DO $$
DECLARE
monitor record;
BEGIN
SELECT pg_catalog.pg_has_role('neon_superuser', 'pg_monitor', 'member') AS member,
SELECT pg_has_role('neon_superuser', 'pg_monitor', 'member') AS member,
admin_option AS admin
INTO monitor
FROM pg_catalog.pg_auth_members
WHERE roleid = 'pg_monitor'::pg_catalog.regrole
AND member = 'neon_superuser'::pg_catalog.regrole;
FROM pg_auth_members
WHERE roleid = 'pg_monitor'::regrole
AND member = 'neon_superuser'::regrole;
IF monitor IS NULL THEN
RAISE EXCEPTION 'no entry in pg_auth_members for neon_superuser and pg_monitor';

View File

@@ -2,11 +2,11 @@ DO $$
DECLARE
can_execute boolean;
BEGIN
SELECT pg_catalog.bool_and(pg_catalog.has_function_privilege('neon_superuser', oid, 'execute'))
SELECT bool_and(has_function_privilege('neon_superuser', oid, 'execute'))
INTO can_execute
FROM pg_catalog.pg_proc
FROM pg_proc
WHERE proname IN ('pg_export_snapshot', 'pg_log_standby_snapshot')
AND pronamespace = 'pg_catalog'::pg_catalog.regnamespace;
AND pronamespace = 'pg_catalog'::regnamespace;
IF NOT can_execute THEN
RAISE EXCEPTION 'neon_superuser cannot execute both pg_export_snapshot and pg_log_standby_snapshot';
END IF;

View File

@@ -2,9 +2,9 @@ DO $$
DECLARE
can_execute boolean;
BEGIN
SELECT pg_catalog.has_function_privilege('neon_superuser', oid, 'execute')
SELECT has_function_privilege('neon_superuser', oid, 'execute')
INTO can_execute
FROM pg_catalog.pg_proc
FROM pg_proc
WHERE proname = 'pg_show_replication_origin_status'
AND pronamespace = 'pg_catalog'::regnamespace;
IF NOT can_execute THEN

View File

@@ -2,10 +2,10 @@ DO $$
DECLARE
signal_backend record;
BEGIN
SELECT pg_catalog.pg_has_role('neon_superuser', 'pg_signal_backend', 'member') AS member,
SELECT pg_has_role('neon_superuser', 'pg_signal_backend', 'member') AS member,
admin_option AS admin
INTO signal_backend
FROM pg_catalog.pg_auth_members
FROM pg_auth_members
WHERE roleid = 'pg_signal_backend'::regrole
AND member = 'neon_superuser'::regrole;

View File

@@ -407,9 +407,9 @@ fn get_database_stats(cli: &mut Client) -> anyhow::Result<(f64, i64)> {
// like `postgres_exporter` use it to query Postgres statistics.
// Use explicit 8 bytes type casts to match Rust types.
let stats = cli.query_one(
"SELECT pg_catalog.coalesce(pg_catalog.sum(active_time), 0.0)::pg_catalog.float8 AS total_active_time,
pg_catalog.coalesce(pg_catalog.sum(sessions), 0)::pg_catalog.bigint AS total_sessions
FROM pg_catalog.pg_stat_database
"SELECT coalesce(sum(active_time), 0.0)::float8 AS total_active_time,
coalesce(sum(sessions), 0)::bigint AS total_sessions
FROM pg_stat_database
WHERE datname NOT IN (
'postgres',
'template0',
@@ -445,11 +445,11 @@ fn get_backends_state_change(cli: &mut Client) -> anyhow::Result<Option<DateTime
let mut last_active: Option<DateTime<Utc>> = None;
// Get all running client backends except ourself, use RFC3339 DateTime format.
let backends = cli.query(
"SELECT state, pg_catalog.to_char(state_change, 'YYYY-MM-DD\"T\"HH24:MI:SS.US\"Z\"'::pg_catalog.text) AS state_change
"SELECT state, to_char(state_change, 'YYYY-MM-DD\"T\"HH24:MI:SS.US\"Z\"') AS state_change
FROM pg_stat_activity
WHERE backend_type OPERATOR(pg_catalog.=) 'client backend'::pg_catalog.text
AND pid OPERATOR(pg_catalog.!=) pg_catalog.pg_backend_pid()
AND usename OPERATOR(pg_catalog.!=) 'cloud_admin'::pg_catalog.name;", // XXX: find a better way to filter other monitors?
WHERE backend_type = 'client backend'
AND pid != pg_backend_pid()
AND usename != 'cloud_admin';", // XXX: find a better way to filter other monitors?
&[],
);

View File

@@ -299,9 +299,9 @@ pub async fn get_existing_dbs_async(
.query_raw::<str, &String, &[String; 0]>(
"SELECT
datname AS name,
(SELECT rolname FROM pg_catalog.pg_roles WHERE oid OPERATOR(pg_catalog.=) datdba) AS owner,
(SELECT rolname FROM pg_roles WHERE oid = datdba) AS owner,
NOT datallowconn AS restrict_conn,
datconnlimit OPERATOR(pg_catalog.=) (OPERATOR(pg_catalog.-) 2) AS invalid
datconnlimit = - 2 AS invalid
FROM
pg_catalog.pg_database;",
&[],

View File

@@ -82,7 +82,7 @@ impl ComputeNode {
info!("Checking if drop subscription operation was already performed for timeline_id: {}", timeline_id);
drop_subscriptions_done = match
client.query("select 1 from neon.drop_subscriptions_done where timeline_id OPERATOR(pg_catalog.=) $1", &[&timeline_id.to_string()]).await {
client.query("select 1 from neon.drop_subscriptions_done where timeline_id = $1", &[&timeline_id.to_string()]).await {
Ok(result) => !result.is_empty(),
Err(e) =>
{
@@ -679,12 +679,7 @@ async fn get_operations<'a>(
ApplySpecPhase::CreatePrivilegedRole => Ok(Box::new(once(Operation {
query: format!(
include_str!("sql/create_privileged_role.sql"),
privileged_role_name = params.privileged_role_name,
privileges = if params.lakebase_mode {
"CREATEDB CREATEROLE NOLOGIN BYPASSRLS"
} else {
"CREATEDB CREATEROLE NOLOGIN REPLICATION BYPASSRLS"
}
privileged_role_name = params.privileged_role_name
),
comment: None,
}))),
@@ -1142,9 +1137,7 @@ async fn get_operations<'a>(
if let Some(libs) = spec.cluster.settings.find("shared_preload_libraries") {
if libs.contains("pg_stat_statements") {
return Ok(Box::new(once(Operation {
query: String::from(
"CREATE EXTENSION IF NOT EXISTS pg_stat_statements WITH SCHEMA public",
),
query: String::from("CREATE EXTENSION IF NOT EXISTS pg_stat_statements"),
comment: Some(String::from("create system extensions")),
})));
}
@@ -1152,13 +1145,11 @@ async fn get_operations<'a>(
Ok(Box::new(empty()))
}
ApplySpecPhase::CreatePgauditExtension => Ok(Box::new(once(Operation {
query: String::from("CREATE EXTENSION IF NOT EXISTS pgaudit WITH SCHEMA public"),
query: String::from("CREATE EXTENSION IF NOT EXISTS pgaudit"),
comment: Some(String::from("create pgaudit extensions")),
}))),
ApplySpecPhase::CreatePgauditlogtofileExtension => Ok(Box::new(once(Operation {
query: String::from(
"CREATE EXTENSION IF NOT EXISTS pgauditlogtofile WITH SCHEMA public",
),
query: String::from("CREATE EXTENSION IF NOT EXISTS pgauditlogtofile"),
comment: Some(String::from("create pgauditlogtofile extensions")),
}))),
// Disable pgaudit logging for postgres database.
@@ -1182,7 +1173,7 @@ async fn get_operations<'a>(
},
Operation {
query: String::from(
"UPDATE pg_catalog.pg_extension SET extrelocatable = true WHERE extname OPERATOR(pg_catalog.=) 'neon'::pg_catalog.name AND extrelocatable OPERATOR(pg_catalog.=) false",
"UPDATE pg_extension SET extrelocatable = true WHERE extname = 'neon'",
),
comment: Some(String::from("compat/fix: make neon relocatable")),
},

View File

@@ -3,17 +3,16 @@ BEGIN
IF NOT EXISTS(
SELECT 1
FROM pg_catalog.pg_tables
WHERE tablename::pg_catalog.name OPERATOR(pg_catalog.=) 'health_check'::pg_catalog.name
AND schemaname::pg_catalog.name OPERATOR(pg_catalog.=) 'public'::pg_catalog.name
WHERE tablename = 'health_check'
)
THEN
CREATE TABLE public.health_check (
id pg_catalog.int4 primary key generated by default as identity,
updated_at pg_catalog.timestamptz default pg_catalog.now()
CREATE TABLE health_check (
id serial primary key,
updated_at timestamptz default now()
);
INSERT INTO public.health_check VALUES (1, pg_catalog.now())
INSERT INTO health_check VALUES (1, now())
ON CONFLICT (id) DO UPDATE
SET updated_at = pg_catalog.now();
SET updated_at = now();
END IF;
END
$$

View File

@@ -2,11 +2,10 @@ DO $$
DECLARE
query varchar;
BEGIN
FOR query IN
SELECT pg_catalog.format('ALTER FUNCTION %I(%s) OWNER TO {db_owner};', p.oid::regproc, pg_catalog.pg_get_function_identity_arguments(p.oid))
FROM pg_catalog.pg_proc p
WHERE p.pronamespace OPERATOR(pg_catalog.=) 'anon'::regnamespace::oid
LOOP
FOR query IN SELECT 'ALTER FUNCTION '||nsp.nspname||'.'||p.proname||'('||pg_get_function_identity_arguments(p.oid)||') OWNER TO {db_owner};'
FROM pg_proc p
JOIN pg_namespace nsp ON p.pronamespace = nsp.oid
WHERE nsp.nspname = 'anon' LOOP
EXECUTE query;
END LOOP;
END

View File

@@ -1,8 +1,8 @@
DO $$
BEGIN
IF NOT EXISTS (SELECT FROM pg_catalog.pg_roles WHERE rolname OPERATOR(pg_catalog.=) '{privileged_role_name}'::pg_catalog.name)
IF NOT EXISTS (SELECT FROM pg_catalog.pg_roles WHERE rolname = '{privileged_role_name}')
THEN
CREATE ROLE {privileged_role_name} {privileges} IN ROLE pg_read_all_data, pg_write_all_data;
CREATE ROLE {privileged_role_name} CREATEDB CREATEROLE NOLOGIN REPLICATION BYPASSRLS IN ROLE pg_read_all_data, pg_write_all_data;
END IF;
END
$$;

View File

@@ -4,14 +4,14 @@ $$
IF EXISTS(
SELECT nspname
FROM pg_catalog.pg_namespace
WHERE nspname OPERATOR(pg_catalog.=) 'public'
WHERE nspname = 'public'
) AND
pg_catalog.current_setting('server_version_num')::int OPERATOR(pg_catalog./) 10000 OPERATOR(pg_catalog.>=) 15
current_setting('server_version_num')::int / 10000 >= 15
THEN
IF EXISTS(
SELECT rolname
FROM pg_catalog.pg_roles
WHERE rolname OPERATOR(pg_catalog.=) 'web_access'
WHERE rolname = 'web_access'
)
THEN
GRANT CREATE ON SCHEMA public TO web_access;
@@ -20,7 +20,7 @@ $$
IF EXISTS(
SELECT nspname
FROM pg_catalog.pg_namespace
WHERE nspname OPERATOR(pg_catalog.=) 'public'
WHERE nspname = 'public'
)
THEN
ALTER DEFAULT PRIVILEGES IN SCHEMA public GRANT ALL ON TABLES TO neon_superuser WITH GRANT OPTION;

View File

@@ -2,17 +2,11 @@ DO ${outer_tag}$
DECLARE
subname TEXT;
BEGIN
LOCK TABLE pg_catalog.pg_subscription IN ACCESS EXCLUSIVE MODE;
FOR subname IN
SELECT pg_subscription.subname
FROM pg_catalog.pg_subscription
WHERE subdbid OPERATOR(pg_catalog.=) (
SELECT oid FROM pg_database WHERE datname OPERATOR(pg_catalog.=) {datname_str}::pg_catalog.name
)
LOOP
EXECUTE pg_catalog.format('ALTER SUBSCRIPTION %I DISABLE;', subname);
EXECUTE pg_catalog.format('ALTER SUBSCRIPTION %I SET (slot_name = NONE);', subname);
EXECUTE pg_catalog.format('DROP SUBSCRIPTION %I;', subname);
LOCK TABLE pg_subscription IN ACCESS EXCLUSIVE MODE;
FOR subname IN SELECT pg_subscription.subname FROM pg_subscription WHERE subdbid = (SELECT oid FROM pg_database WHERE datname = {datname_str}) LOOP
EXECUTE format('ALTER SUBSCRIPTION %I DISABLE;', subname);
EXECUTE format('ALTER SUBSCRIPTION %I SET (slot_name = NONE);', subname);
EXECUTE format('DROP SUBSCRIPTION %I;', subname);
END LOOP;
END;
${outer_tag}$;

View File

@@ -3,19 +3,19 @@ BEGIN
IF NOT EXISTS(
SELECT 1
FROM pg_catalog.pg_tables
WHERE tablename OPERATOR(pg_catalog.=) 'drop_subscriptions_done'::pg_catalog.name
AND schemaname OPERATOR(pg_catalog.=) 'neon'::pg_catalog.name
WHERE tablename = 'drop_subscriptions_done'
AND schemaname = 'neon'
)
THEN
CREATE TABLE neon.drop_subscriptions_done
(id pg_catalog.int4 primary key generated by default as identity, timeline_id pg_catalog.text);
(id serial primary key, timeline_id text);
END IF;
-- preserve the timeline_id of the last drop_subscriptions run
-- to ensure that the cleanup of a timeline is executed only once.
-- use upsert to avoid the table bloat in case of cascade branching (branch of a branch)
INSERT INTO neon.drop_subscriptions_done VALUES (1, pg_catalog.current_setting('neon.timeline_id'))
INSERT INTO neon.drop_subscriptions_done VALUES (1, current_setting('neon.timeline_id'))
ON CONFLICT (id) DO UPDATE
SET timeline_id = pg_catalog.current_setting('neon.timeline_id')::pg_catalog.text;
SET timeline_id = current_setting('neon.timeline_id');
END
$$

View File

@@ -15,15 +15,15 @@ BEGIN
WHERE schema_name IN ('public')
LOOP
FOR grantor IN EXECUTE
pg_catalog.format(
'SELECT DISTINCT rtg.grantor FROM information_schema.role_table_grants AS rtg WHERE grantee OPERATOR(pg_catalog.=) %s',
format(
'SELECT DISTINCT rtg.grantor FROM information_schema.role_table_grants AS rtg WHERE grantee = %s',
-- N.B. this has to be properly dollar-escaped with `pg_quote_dollar()`
quote_literal({role_name})
)
LOOP
EXECUTE pg_catalog.format('SET LOCAL ROLE %I', grantor);
EXECUTE format('SET LOCAL ROLE %I', grantor);
revoke_query := pg_catalog.format(
revoke_query := format(
'REVOKE ALL PRIVILEGES ON ALL TABLES IN SCHEMA %I FROM %I GRANTED BY %I',
schema,
-- N.B. this has to be properly dollar-escaped with `pg_quote_dollar()`

View File

@@ -5,17 +5,17 @@ DO ${outer_tag}$
IF EXISTS(
SELECT nspname
FROM pg_catalog.pg_namespace
WHERE nspname OPERATOR(pg_catalog.=) 'public'::pg_catalog.name
WHERE nspname = 'public'
)
THEN
SELECT nspowner::regrole::text
FROM pg_catalog.pg_namespace
WHERE nspname OPERATOR(pg_catalog.=) 'public'::pg_catalog.text
WHERE nspname = 'public'
INTO schema_owner;
IF schema_owner OPERATOR(pg_catalog.=) 'cloud_admin'::pg_catalog.text OR schema_owner OPERATOR(pg_catalog.=) 'zenith_admin'::pg_catalog.text
IF schema_owner = 'cloud_admin' OR schema_owner = 'zenith_admin'
THEN
EXECUTE pg_catalog.format('ALTER SCHEMA public OWNER TO %I', {db_owner});
EXECUTE format('ALTER SCHEMA public OWNER TO %I', {db_owner});
END IF;
END IF;
END

View File

@@ -3,10 +3,10 @@ DO ${outer_tag}$
IF EXISTS(
SELECT 1
FROM pg_catalog.pg_database
WHERE datname OPERATOR(pg_catalog.=) {datname}::pg_catalog.name
WHERE datname = {datname}
)
THEN
EXECUTE pg_catalog.format('ALTER DATABASE %I is_template false', {datname});
EXECUTE format('ALTER DATABASE %I is_template false', {datname});
END IF;
END
${outer_tag}$;

View File

@@ -19,9 +19,6 @@ use compute_api::requests::ComputeClaimsScope;
use compute_api::spec::{ComputeMode, PageserverProtocol};
use control_plane::broker::StorageBroker;
use control_plane::endpoint::{ComputeControlPlane, EndpointTerminateMode};
use control_plane::endpoint::{
local_pageserver_conf_to_conn_info, tenant_locate_response_to_conn_info,
};
use control_plane::endpoint_storage::{ENDPOINT_STORAGE_DEFAULT_ADDR, EndpointStorage};
use control_plane::local_env;
use control_plane::local_env::{
@@ -47,6 +44,7 @@ use pageserver_api::models::{
};
use pageserver_api::shard::{DEFAULT_STRIPE_SIZE, ShardCount, ShardStripeSize, TenantShardId};
use postgres_backend::AuthType;
use postgres_connection::parse_host_port;
use safekeeper_api::membership::{SafekeeperGeneration, SafekeeperId};
use safekeeper_api::{
DEFAULT_HTTP_LISTEN_PORT as DEFAULT_SAFEKEEPER_HTTP_PORT,
@@ -54,6 +52,7 @@ use safekeeper_api::{
};
use storage_broker::DEFAULT_LISTEN_ADDR as DEFAULT_BROKER_ADDR;
use tokio::task::JoinSet;
use url::Host;
use utils::auth::{Claims, Scope};
use utils::id::{NodeId, TenantId, TenantTimelineId, TimelineId};
use utils::lsn::Lsn;
@@ -1547,41 +1546,62 @@ async fn handle_endpoint(subcmd: &EndpointCmd, env: &local_env::LocalEnv) -> Res
)?;
}
let prefer_protocol = if endpoint.grpc {
PageserverProtocol::Grpc
} else {
PageserverProtocol::Libpq
};
let mut pageserver_conninfo = if let Some(ps_id) = pageserver_id {
let conf = env.get_pageserver_conf(ps_id).unwrap();
local_pageserver_conf_to_conn_info(conf)?
let (pageservers, stripe_size) = if let Some(pageserver_id) = pageserver_id {
let conf = env.get_pageserver_conf(pageserver_id).unwrap();
// Use gRPC if requested.
let pageserver = if endpoint.grpc {
let grpc_addr = conf.listen_grpc_addr.as_ref().expect("bad config");
let (host, port) = parse_host_port(grpc_addr)?;
let port = port.unwrap_or(DEFAULT_PAGESERVER_GRPC_PORT);
(PageserverProtocol::Grpc, host, port)
} else {
let (host, port) = parse_host_port(&conf.listen_pg_addr)?;
let port = port.unwrap_or(5432);
(PageserverProtocol::Libpq, host, port)
};
// If caller is telling us what pageserver to use, this is not a tenant which is
// fully managed by storage controller, therefore not sharded.
(vec![pageserver], DEFAULT_STRIPE_SIZE)
} else {
// Look up the currently attached location of the tenant, and its striping metadata,
// to pass these on to postgres.
let storage_controller = StorageController::from_env(env);
let locate_result = storage_controller.tenant_locate(endpoint.tenant_id).await?;
assert!(!locate_result.shards.is_empty());
// Initialize LSN leases for static computes.
if let ComputeMode::Static(lsn) = endpoint.mode {
futures::future::try_join_all(locate_result.shards.iter().map(
|shard| async move {
let pageservers = futures::future::try_join_all(
locate_result.shards.into_iter().map(|shard| async move {
if let ComputeMode::Static(lsn) = endpoint.mode {
// Initialize LSN leases for static computes.
let conf = env.get_pageserver_conf(shard.node_id).unwrap();
let pageserver = PageServerNode::from_env(env, conf);
pageserver
.http_client
.timeline_init_lsn_lease(shard.shard_id, endpoint.timeline_id, lsn)
.await
},
))
.await?;
}
.await?;
}
tenant_locate_response_to_conn_info(&locate_result)?
let pageserver = if endpoint.grpc {
(
PageserverProtocol::Grpc,
Host::parse(&shard.listen_grpc_addr.expect("no gRPC address"))?,
shard.listen_grpc_port.expect("no gRPC port"),
)
} else {
(
PageserverProtocol::Libpq,
Host::parse(&shard.listen_pg_addr)?,
shard.listen_pg_port,
)
};
anyhow::Ok(pageserver)
}),
)
.await?;
let stripe_size = locate_result.shard_params.stripe_size;
(pageservers, stripe_size)
};
pageserver_conninfo.prefer_protocol = prefer_protocol;
assert!(!pageservers.is_empty());
let ps_conf = env.get_pageserver_conf(DEFAULT_PAGESERVER_ID)?;
let auth_token = if matches!(ps_conf.pg_auth_type, AuthType::NeonJWT) {
@@ -1611,8 +1631,9 @@ async fn handle_endpoint(subcmd: &EndpointCmd, env: &local_env::LocalEnv) -> Res
endpoint_storage_addr,
safekeepers_generation,
safekeepers,
pageserver_conninfo,
pageservers,
remote_ext_base_url: remote_ext_base_url.clone(),
shard_stripe_size: stripe_size.0 as usize,
create_test_user: args.create_test_user,
start_timeout: args.start_timeout,
autoprewarm: args.autoprewarm,
@@ -1629,29 +1650,37 @@ async fn handle_endpoint(subcmd: &EndpointCmd, env: &local_env::LocalEnv) -> Res
.endpoints
.get(endpoint_id.as_str())
.with_context(|| format!("postgres endpoint {endpoint_id} is not found"))?;
let prefer_protocol = if endpoint.grpc {
PageserverProtocol::Grpc
} else {
PageserverProtocol::Libpq
};
let mut pageserver_conninfo = match args.pageserver_id {
let pageservers = match args.pageserver_id {
Some(pageserver_id) => {
let conf = env.get_pageserver_conf(pageserver_id)?;
local_pageserver_conf_to_conn_info(conf)?
let pageserver =
PageServerNode::from_env(env, env.get_pageserver_conf(pageserver_id)?);
vec![(
PageserverProtocol::Libpq,
pageserver.pg_connection_config.host().clone(),
pageserver.pg_connection_config.port(),
)]
}
None => {
let storage_controller = StorageController::from_env(env);
let locate_result =
storage_controller.tenant_locate(endpoint.tenant_id).await?;
tenant_locate_response_to_conn_info(&locate_result)?
storage_controller
.tenant_locate(endpoint.tenant_id)
.await?
.shards
.into_iter()
.map(|shard| {
(
PageserverProtocol::Libpq,
Host::parse(&shard.listen_pg_addr)
.expect("Storage controller reported malformed host"),
shard.listen_pg_port,
)
})
.collect::<Vec<_>>()
}
};
pageserver_conninfo.prefer_protocol = prefer_protocol;
endpoint
.update_pageservers_in_config(&pageserver_conninfo)
.await?;
endpoint.update_pageservers_in_config(pageservers).await?;
}
EndpointCmd::Reconfigure(args) => {
let endpoint_id = &args.endpoint_id;
@@ -1659,30 +1688,51 @@ async fn handle_endpoint(subcmd: &EndpointCmd, env: &local_env::LocalEnv) -> Res
.endpoints
.get(endpoint_id.as_str())
.with_context(|| format!("postgres endpoint {endpoint_id} is not found"))?;
let prefer_protocol = if endpoint.grpc {
PageserverProtocol::Grpc
} else {
PageserverProtocol::Libpq
};
let mut pageserver_conninfo = if let Some(ps_id) = args.endpoint_pageserver_id {
let pageservers = if let Some(ps_id) = args.endpoint_pageserver_id {
let conf = env.get_pageserver_conf(ps_id)?;
local_pageserver_conf_to_conn_info(conf)?
// Use gRPC if requested.
let pageserver = if endpoint.grpc {
let grpc_addr = conf.listen_grpc_addr.as_ref().expect("bad config");
let (host, port) = parse_host_port(grpc_addr)?;
let port = port.unwrap_or(DEFAULT_PAGESERVER_GRPC_PORT);
(PageserverProtocol::Grpc, host, port)
} else {
let (host, port) = parse_host_port(&conf.listen_pg_addr)?;
let port = port.unwrap_or(5432);
(PageserverProtocol::Libpq, host, port)
};
vec![pageserver]
} else {
// Look up the currently attached location of the tenant, and its striping metadata,
// to pass these on to postgres.
let storage_controller = StorageController::from_env(env);
let locate_result = storage_controller.tenant_locate(endpoint.tenant_id).await?;
tenant_locate_response_to_conn_info(&locate_result)?
storage_controller
.tenant_locate(endpoint.tenant_id)
.await?
.shards
.into_iter()
.map(|shard| {
// Use gRPC if requested.
if endpoint.grpc {
(
PageserverProtocol::Grpc,
Host::parse(&shard.listen_grpc_addr.expect("no gRPC address"))
.expect("bad hostname"),
shard.listen_grpc_port.expect("no gRPC port"),
)
} else {
(
PageserverProtocol::Libpq,
Host::parse(&shard.listen_pg_addr).expect("bad hostname"),
shard.listen_pg_port,
)
}
})
.collect::<Vec<_>>()
};
pageserver_conninfo.prefer_protocol = prefer_protocol;
// If --safekeepers argument is given, use only the listed
// safekeeper nodes; otherwise all from the env.
let safekeepers = parse_safekeepers(&args.safekeepers)?;
endpoint
.reconfigure(Some(&pageserver_conninfo), safekeepers, None)
.reconfigure(Some(pageservers), None, safekeepers, None)
.await?;
}
EndpointCmd::RefreshConfiguration(args) => {

View File

@@ -37,7 +37,7 @@
//! <other PostgreSQL files>
//! ```
//!
use std::collections::{BTreeMap, HashMap};
use std::collections::BTreeMap;
use std::fmt::Display;
use std::net::{IpAddr, Ipv4Addr, SocketAddr, TcpStream};
use std::path::PathBuf;
@@ -58,12 +58,8 @@ use compute_api::responses::{
};
use compute_api::spec::{
Cluster, ComputeAudit, ComputeFeature, ComputeMode, ComputeSpec, Database, PageserverProtocol,
PageserverShardInfo, PgIdent, RemoteExtSpec, Role,
PgIdent, RemoteExtSpec, Role,
};
// re-export these, because they're used in the reconfigure() function
pub use compute_api::spec::{PageserverConnectionInfo, PageserverShardConnectionInfo};
use jsonwebtoken::jwk::{
AlgorithmParameters, CommonParameters, EllipticCurve, Jwk, JwkSet, KeyAlgorithm, KeyOperations,
OctetKeyPairParameters, OctetKeyPairType, PublicKeyUse,
@@ -78,11 +74,9 @@ use sha2::{Digest, Sha256};
use spki::der::Decode;
use spki::{SubjectPublicKeyInfo, SubjectPublicKeyInfoRef};
use tracing::debug;
use url::Host;
use utils::id::{NodeId, TenantId, TimelineId};
use utils::shard::{ShardCount, ShardIndex, ShardNumber};
use pageserver_api::config::DEFAULT_GRPC_LISTEN_PORT as DEFAULT_PAGESERVER_GRPC_PORT;
use postgres_connection::parse_host_port;
use utils::shard::ShardStripeSize;
use crate::local_env::LocalEnv;
use crate::postgresql_conf::PostgresConf;
@@ -393,8 +387,9 @@ pub struct EndpointStartArgs {
pub endpoint_storage_addr: String,
pub safekeepers_generation: Option<SafekeeperGeneration>,
pub safekeepers: Vec<NodeId>,
pub pageserver_conninfo: PageserverConnectionInfo,
pub pageservers: Vec<(PageserverProtocol, Host, u16)>,
pub remote_ext_base_url: Option<String>,
pub shard_stripe_size: usize,
pub create_test_user: bool,
pub start_timeout: Duration,
pub autoprewarm: bool,
@@ -667,6 +662,14 @@ impl Endpoint {
}
}
fn build_pageserver_connstr(pageservers: &[(PageserverProtocol, Host, u16)]) -> String {
pageservers
.iter()
.map(|(scheme, host, port)| format!("{scheme}://no_user@{host}:{port}"))
.collect::<Vec<_>>()
.join(",")
}
/// Map safekeepers ids to the actual connection strings.
fn build_safekeepers_connstrs(&self, sk_ids: Vec<NodeId>) -> Result<Vec<String>> {
let mut safekeeper_connstrings = Vec::new();
@@ -712,6 +715,9 @@ impl Endpoint {
std::fs::remove_dir_all(self.pgdata())?;
}
let pageserver_connstring = Self::build_pageserver_connstr(&args.pageservers);
assert!(!pageserver_connstring.is_empty());
let safekeeper_connstrings = self.build_safekeepers_connstrs(args.safekeepers)?;
// check for file remote_extensions_spec.json
@@ -726,44 +732,6 @@ impl Endpoint {
remote_extensions = None;
};
// For the sake of backwards-compatibility, also fill in 'pageserver_connstring'
//
// XXX: I believe this is not really needed, except to make
// test_forward_compatibility happy.
//
// Use a closure so that we can conviniently return None in the middle of the
// loop.
let pageserver_connstring: Option<String> = (|| {
let num_shards = args.pageserver_conninfo.shard_count.count();
let mut connstrings = Vec::new();
for shard_no in 0..num_shards {
let shard_index = ShardIndex {
shard_count: args.pageserver_conninfo.shard_count,
shard_number: ShardNumber(shard_no),
};
let shard = args
.pageserver_conninfo
.shards
.get(&shard_index)
.ok_or_else(|| {
anyhow!(
"shard {} not found in pageserver_connection_info",
shard_index
)
})?;
let pageserver = shard
.pageservers
.first()
.ok_or(anyhow!("must have at least one pageserver"))?;
if let Some(libpq_url) = &pageserver.libpq_url {
connstrings.push(libpq_url.clone());
} else {
return Ok::<_, anyhow::Error>(None);
}
}
Ok(Some(connstrings.join(",")))
})()?;
// Create config file
let config = {
let mut spec = ComputeSpec {
@@ -808,14 +776,13 @@ impl Endpoint {
branch_id: None,
endpoint_id: Some(self.endpoint_id.clone()),
mode: self.mode,
pageserver_connection_info: Some(args.pageserver_conninfo.clone()),
pageserver_connstring,
pageserver_connstring: Some(pageserver_connstring),
safekeepers_generation: args.safekeepers_generation.map(|g| g.into_inner()),
safekeeper_connstrings,
storage_auth_token: args.auth_token.clone(),
remote_extensions,
pgbouncer_settings: None,
shard_stripe_size: args.pageserver_conninfo.stripe_size, // redundant with pageserver_connection_info.stripe_size
shard_stripe_size: Some(args.shard_stripe_size),
local_proxy_config: None,
reconfigure_concurrency: self.reconfigure_concurrency,
drop_subscriptions_before_start: self.drop_subscriptions_before_start,
@@ -999,7 +966,7 @@ impl Endpoint {
// Update the pageservers in the spec file of the endpoint. This is useful to test the spec refresh scenario.
pub async fn update_pageservers_in_config(
&self,
pageserver_conninfo: &PageserverConnectionInfo,
pageservers: Vec<(PageserverProtocol, Host, u16)>,
) -> Result<()> {
let config_path = self.endpoint_path().join("config.json");
let mut config: ComputeConfig = {
@@ -1007,8 +974,10 @@ impl Endpoint {
serde_json::from_reader(file)?
};
let pageserver_connstring = Self::build_pageserver_connstr(&pageservers);
assert!(!pageserver_connstring.is_empty());
let mut spec = config.spec.unwrap();
spec.pageserver_connection_info = Some(pageserver_conninfo.clone());
spec.pageserver_connstring = Some(pageserver_connstring);
config.spec = Some(spec);
let file = std::fs::File::create(&config_path)?;
@@ -1051,7 +1020,8 @@ impl Endpoint {
pub async fn reconfigure(
&self,
pageserver_conninfo: Option<&PageserverConnectionInfo>,
pageservers: Option<Vec<(PageserverProtocol, Host, u16)>>,
stripe_size: Option<ShardStripeSize>,
safekeepers: Option<Vec<NodeId>>,
safekeeper_generation: Option<SafekeeperGeneration>,
) -> Result<()> {
@@ -1066,15 +1036,15 @@ impl Endpoint {
let postgresql_conf = self.read_postgresql_conf()?;
spec.cluster.postgresql_conf = Some(postgresql_conf);
if let Some(pageserver_conninfo) = pageserver_conninfo {
// If pageservers are provided, we need to ensure that they are not empty.
// This is a requirement for the compute_ctl configuration.
anyhow::ensure!(
!pageserver_conninfo.shards.is_empty(),
"no pageservers provided"
);
spec.pageserver_connection_info = Some(pageserver_conninfo.clone());
spec.shard_stripe_size = pageserver_conninfo.stripe_size;
// If pageservers are not specified, don't change them.
if let Some(pageservers) = pageservers {
anyhow::ensure!(!pageservers.is_empty(), "no pageservers provided");
let pageserver_connstr = Self::build_pageserver_connstr(&pageservers);
spec.pageserver_connstring = Some(pageserver_connstr);
if stripe_size.is_some() {
spec.shard_stripe_size = stripe_size.map(|s| s.0 as usize);
}
}
// If safekeepers are not specified, don't change them.
@@ -1123,9 +1093,11 @@ impl Endpoint {
pub async fn reconfigure_pageservers(
&self,
pageservers: &PageserverConnectionInfo,
pageservers: Vec<(PageserverProtocol, Host, u16)>,
stripe_size: Option<ShardStripeSize>,
) -> Result<()> {
self.reconfigure(Some(pageservers), None, None).await
self.reconfigure(Some(pageservers), stripe_size, None, None)
.await
}
pub async fn reconfigure_safekeepers(
@@ -1133,7 +1105,7 @@ impl Endpoint {
safekeepers: Vec<NodeId>,
generation: SafekeeperGeneration,
) -> Result<()> {
self.reconfigure(None, Some(safekeepers), Some(generation))
self.reconfigure(None, None, Some(safekeepers), Some(generation))
.await
}
@@ -1216,84 +1188,3 @@ impl Endpoint {
)
}
}
/// If caller is telling us what pageserver to use, this is not a tenant which is
/// fully managed by storage controller, therefore not sharded.
pub fn local_pageserver_conf_to_conn_info(
conf: &crate::local_env::PageServerConf,
) -> Result<PageserverConnectionInfo> {
let libpq_url = {
let (host, port) = parse_host_port(&conf.listen_pg_addr)?;
let port = port.unwrap_or(5432);
Some(format!("postgres://no_user@{host}:{port}"))
};
let grpc_url = if let Some(grpc_addr) = &conf.listen_grpc_addr {
let (host, port) = parse_host_port(grpc_addr)?;
let port = port.unwrap_or(DEFAULT_PAGESERVER_GRPC_PORT);
Some(format!("grpc://no_user@{host}:{port}"))
} else {
None
};
let ps_conninfo = PageserverShardConnectionInfo {
id: Some(conf.id),
libpq_url,
grpc_url,
};
let shard_info = PageserverShardInfo {
pageservers: vec![ps_conninfo],
};
let shards: HashMap<_, _> = vec![(ShardIndex::unsharded(), shard_info)]
.into_iter()
.collect();
Ok(PageserverConnectionInfo {
shard_count: ShardCount::unsharded(),
stripe_size: None,
shards,
prefer_protocol: PageserverProtocol::default(),
})
}
pub fn tenant_locate_response_to_conn_info(
response: &pageserver_api::controller_api::TenantLocateResponse,
) -> Result<PageserverConnectionInfo> {
let mut shards = HashMap::new();
for shard in response.shards.iter() {
tracing::info!("parsing {}", shard.listen_pg_addr);
let libpq_url = {
let host = &shard.listen_pg_addr;
let port = shard.listen_pg_port;
Some(format!("postgres://no_user@{host}:{port}"))
};
let grpc_url = if let Some(grpc_addr) = &shard.listen_grpc_addr {
let host = grpc_addr;
let port = shard.listen_grpc_port.expect("no gRPC port");
Some(format!("grpc://no_user@{host}:{port}"))
} else {
None
};
let shard_info = PageserverShardInfo {
pageservers: vec![PageserverShardConnectionInfo {
id: Some(shard.node_id),
libpq_url,
grpc_url,
}],
};
shards.insert(shard.shard_id.to_index(), shard_info);
}
let stripe_size = if response.shard_params.count.is_unsharded() {
None
} else {
Some(response.shard_params.stripe_size)
};
Ok(PageserverConnectionInfo {
shard_count: response.shard_params.count,
stripe_size,
shards,
prefer_protocol: PageserverProtocol::default(),
})
}

View File

@@ -120,11 +120,6 @@
"value": "host=pageserver port=6400",
"vartype": "string"
},
{
"name": "neon.pageserver_grpc_urls",
"value": "grpc://pageserver:6401/",
"vartype": "string"
},
{
"name": "max_replication_write_lag",
"value": "500MB",

View File

@@ -1,7 +1,6 @@
broker_endpoint='http://storage_broker:50051'
pg_distrib_dir='/usr/local/'
listen_pg_addr='0.0.0.0:6400'
listen_grpc_addr='0.0.0.0:6401'
listen_http_addr='0.0.0.0:9898'
remote_storage={ endpoint='http://minio:9000', bucket_name='neon', bucket_region='eu-north-1', prefix_in_bucket='/pageserver' }
control_plane_api='http://0.0.0.0:6666' # No storage controller in docker compose, specify a junk address

View File

@@ -12,9 +12,8 @@ use regex::Regex;
use remote_storage::RemotePath;
use serde::{Deserialize, Serialize};
use url::Url;
use utils::id::{NodeId, TenantId, TimelineId};
use utils::id::{TenantId, TimelineId};
use utils::lsn::Lsn;
use utils::shard::{ShardCount, ShardIndex, ShardNumber, ShardStripeSize};
use crate::responses::TlsConfig;
@@ -106,27 +105,8 @@ pub struct ComputeSpec {
// updated to fill these fields, we can make these non optional.
pub tenant_id: Option<TenantId>,
pub timeline_id: Option<TimelineId>,
/// Pageserver information can be passed in three different ways:
/// 1. Here in `pageserver_connection_info`
/// 2. In the `pageserver_connstring` field.
/// 3. in `cluster.settings`.
///
/// The goal is to use method 1. everywhere. But for backwards-compatibility with old
/// versions of the control plane, `compute_ctl` will check 2. and 3. if the
/// `pageserver_connection_info` field is missing.
///
/// If both `pageserver_connection_info` and `pageserver_connstring`+`shard_stripe_size` are
/// given, they must contain the same information.
pub pageserver_connection_info: Option<PageserverConnectionInfo>,
pub pageserver_connstring: Option<String>,
/// Stripe size for pageserver sharding, in pages. This is set together with the legacy
/// `pageserver_connstring` field. When the modern `pageserver_connection_info` field is used,
/// the stripe size is stored in `pageserver_connection_info.stripe_size` instead.
pub shard_stripe_size: Option<ShardStripeSize>,
// More neon ids that we expose to the compute_ctl
// and to postgres as neon extension GUCs.
pub project_id: Option<String>,
@@ -159,6 +139,10 @@ pub struct ComputeSpec {
pub pgbouncer_settings: Option<IndexMap<String, String>>,
// Stripe size for pageserver sharding, in pages
#[serde(default)]
pub shard_stripe_size: Option<usize>,
/// Local Proxy configuration used for JWT authentication
#[serde(default)]
pub local_proxy_config: Option<LocalProxySpec>,
@@ -233,140 +217,6 @@ pub enum ComputeFeature {
UnknownFeature,
}
#[derive(Clone, Debug, Deserialize, Serialize, Eq, PartialEq)]
pub struct PageserverConnectionInfo {
/// NB: 0 for unsharded tenants, 1 for sharded tenants with 1 shard, following storage
pub shard_count: ShardCount,
/// INVARIANT: null if shard_count is 0, otherwise non-null and immutable
pub stripe_size: Option<ShardStripeSize>,
pub shards: HashMap<ShardIndex, PageserverShardInfo>,
/// If the compute supports both protocols, this indicates which one it should use. The compute
/// may use other available protocols too, if it doesn't support the preferred one. The URL's
/// for the protocol specified here must be present for all shards, i.e. do not mark a protocol
/// as preferred if it cannot actually be used with all the pageservers.
#[serde(default)]
pub prefer_protocol: PageserverProtocol,
}
/// Extract PageserverConnectionInfo from a comma-separated list of libpq connection strings.
///
/// This is used for backwards-compatibility, to parse the legacy
/// [ComputeSpec::pageserver_connstring] field, or the 'neon.pageserver_connstring' GUC. Nowadays,
/// the 'pageserver_connection_info' field should be used instead.
impl PageserverConnectionInfo {
pub fn from_connstr(
connstr: &str,
stripe_size: Option<ShardStripeSize>,
) -> Result<PageserverConnectionInfo, anyhow::Error> {
let shard_infos: Vec<_> = connstr
.split(',')
.map(|connstr| PageserverShardInfo {
pageservers: vec![PageserverShardConnectionInfo {
id: None,
libpq_url: Some(connstr.to_string()),
grpc_url: None,
}],
})
.collect();
match shard_infos.len() {
0 => anyhow::bail!("empty connection string"),
1 => {
// We assume that if there's only connection string, it means "unsharded",
// rather than a sharded system with just a single shard. The latter is
// possible in principle, but we never do it.
let shard_count = ShardCount::unsharded();
let only_shard = shard_infos.first().unwrap().clone();
let shards = vec![(ShardIndex::unsharded(), only_shard)];
Ok(PageserverConnectionInfo {
shard_count,
stripe_size: None,
shards: shards.into_iter().collect(),
prefer_protocol: PageserverProtocol::Libpq,
})
}
n => {
if stripe_size.is_none() {
anyhow::bail!("{n} shards but no stripe_size");
}
let shard_count = ShardCount(n.try_into()?);
let shards = shard_infos
.into_iter()
.enumerate()
.map(|(idx, shard_info)| {
(
ShardIndex {
shard_count,
shard_number: ShardNumber(
idx.try_into().expect("shard number fits in u8"),
),
},
shard_info,
)
})
.collect();
Ok(PageserverConnectionInfo {
shard_count,
stripe_size,
shards,
prefer_protocol: PageserverProtocol::Libpq,
})
}
}
}
/// Convenience routine to get the connection string for a shard.
pub fn shard_url(
&self,
shard_number: ShardNumber,
protocol: PageserverProtocol,
) -> anyhow::Result<&str> {
let shard_index = ShardIndex {
shard_number,
shard_count: self.shard_count,
};
let shard = self.shards.get(&shard_index).ok_or_else(|| {
anyhow::anyhow!("shard connection info missing for shard {}", shard_index)
})?;
// Just use the first pageserver in the list. That's good enough for this
// convenience routine; if you need more control, like round robin policy or
// failover support, roll your own. (As of this writing, we never have more than
// one pageserver per shard anyway, but that will change in the future.)
let pageserver = shard
.pageservers
.first()
.ok_or(anyhow::anyhow!("must have at least one pageserver"))?;
let result = match protocol {
PageserverProtocol::Grpc => pageserver
.grpc_url
.as_ref()
.ok_or(anyhow::anyhow!("no grpc_url for shard {shard_index}"))?,
PageserverProtocol::Libpq => pageserver
.libpq_url
.as_ref()
.ok_or(anyhow::anyhow!("no libpq_url for shard {shard_index}"))?,
};
Ok(result)
}
}
#[derive(Clone, Debug, Deserialize, Serialize, Eq, PartialEq)]
pub struct PageserverShardInfo {
pub pageservers: Vec<PageserverShardConnectionInfo>,
}
#[derive(Clone, Debug, Deserialize, Serialize, Eq, PartialEq)]
pub struct PageserverShardConnectionInfo {
pub id: Option<NodeId>,
pub libpq_url: Option<String>,
pub grpc_url: Option<String>,
}
#[derive(Clone, Debug, Default, Deserialize, Serialize)]
pub struct RemoteExtSpec {
pub public_extensions: Option<Vec<String>>,
@@ -484,12 +334,6 @@ impl ComputeMode {
}
}
impl Display for ComputeMode {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.write_str(self.to_type_str())
}
}
/// Log level for audit logging
#[derive(Clone, Debug, Default, Eq, PartialEq, Deserialize, Serialize)]
pub enum ComputeAudit {
@@ -626,15 +470,13 @@ pub struct JwksSettings {
pub jwt_audience: Option<String>,
}
/// Protocol used to connect to a Pageserver.
#[derive(Clone, Copy, Debug, Default, Deserialize, Serialize, PartialEq, Eq)]
/// Protocol used to connect to a Pageserver. Parsed from the connstring scheme.
#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)]
pub enum PageserverProtocol {
/// The original protocol based on libpq and COPY. Uses postgresql:// or postgres:// scheme.
#[default]
#[serde(rename = "libpq")]
Libpq,
/// A newer, gRPC-based protocol. Uses grpc:// scheme.
#[serde(rename = "grpc")]
Grpc,
}

View File

@@ -6,26 +6,15 @@ license.workspace = true
[dependencies]
thiserror.workspace = true
nix.workspace = true
nix.workspace=true
workspace_hack = { version = "0.1", path = "../../workspace_hack" }
libc.workspace = true
lock_api.workspace = true
rustc-hash.workspace = true
[dev-dependencies]
criterion = { workspace = true, features = ["html_reports"] }
rand = "0.9"
rand_distr = "0.5.1"
xxhash-rust = { version = "0.8.15", features = ["xxh3"] }
ahash.workspace = true
twox-hash = { version = "2.1.1" }
seahash = "4.1.0"
hashbrown = { git = "https://github.com/quantumish/hashbrown.git", rev = "6610e6d" }
[target.'cfg(target_os = "macos")'.dependencies]
tempfile = "3.14.0"
[[bench]]
name = "hmap_resize"
harness = false
[dev-dependencies]
rand.workspace = true
rand_distr = "0.5.1"

View File

@@ -1,330 +0,0 @@
use criterion::{BatchSize, BenchmarkId, Criterion, criterion_group, criterion_main};
use neon_shmem::hash::HashMapAccess;
use neon_shmem::hash::HashMapInit;
use neon_shmem::hash::entry::Entry;
use rand::distr::{Distribution, StandardUniform};
use rand::prelude::*;
use std::default::Default;
use std::hash::BuildHasher;
// Taken from bindings to C code
#[derive(Clone, Debug, Hash, Eq, PartialEq)]
#[repr(C)]
pub struct FileCacheKey {
pub _spc_id: u32,
pub _db_id: u32,
pub _rel_number: u32,
pub _fork_num: u32,
pub _block_num: u32,
}
impl Distribution<FileCacheKey> for StandardUniform {
// questionable, but doesn't need to be good randomness
fn sample<R: Rng + ?Sized>(&self, rng: &mut R) -> FileCacheKey {
FileCacheKey {
_spc_id: rng.random(),
_db_id: rng.random(),
_rel_number: rng.random(),
_fork_num: rng.random(),
_block_num: rng.random(),
}
}
}
#[derive(Clone, Debug)]
#[repr(C)]
pub struct FileCacheEntry {
pub _offset: u32,
pub _access_count: u32,
pub _prev: *mut FileCacheEntry,
pub _next: *mut FileCacheEntry,
pub _state: [u32; 8],
}
impl FileCacheEntry {
fn dummy() -> Self {
Self {
_offset: 0,
_access_count: 0,
_prev: std::ptr::null_mut(),
_next: std::ptr::null_mut(),
_state: [0; 8],
}
}
}
// Utilities for applying operations.
#[derive(Clone, Debug)]
struct TestOp<K, V>(K, Option<V>);
fn apply_op<K: Clone + std::hash::Hash + Eq, V, S: std::hash::BuildHasher>(
op: TestOp<K, V>,
map: &mut HashMapAccess<K, V, S>,
) {
let entry = map.entry(op.0);
match op.1 {
Some(new) => match entry {
Entry::Occupied(mut e) => Some(e.insert(new)),
Entry::Vacant(e) => {
_ = e.insert(new).unwrap();
None
}
},
None => match entry {
Entry::Occupied(e) => Some(e.remove()),
Entry::Vacant(_) => None,
},
};
}
// Hash utilities
struct SeaRandomState {
k1: u64,
k2: u64,
k3: u64,
k4: u64,
}
impl std::hash::BuildHasher for SeaRandomState {
type Hasher = seahash::SeaHasher;
fn build_hasher(&self) -> Self::Hasher {
seahash::SeaHasher::with_seeds(self.k1, self.k2, self.k3, self.k4)
}
}
impl SeaRandomState {
fn new() -> Self {
let mut rng = rand::rng();
Self {
k1: rng.random(),
k2: rng.random(),
k3: rng.random(),
k4: rng.random(),
}
}
}
fn small_benchs(c: &mut Criterion) {
let mut group = c.benchmark_group("Small maps");
group.sample_size(10);
group.bench_function("small_rehash", |b| {
let ideal_filled = 4_000_000;
let size = 5_000_000;
let mut writer = HashMapInit::new_resizeable(size, size * 2).attach_writer();
let mut rng = rand::rng();
while writer.get_num_buckets_in_use() < ideal_filled as usize {
let key: FileCacheKey = rng.random();
let val = FileCacheEntry::dummy();
apply_op(TestOp(key, Some(val)), &mut writer);
}
b.iter(|| writer.shuffle());
});
group.bench_function("small_rehash_xxhash", |b| {
let ideal_filled = 4_000_000;
let size = 5_000_000;
let mut writer = HashMapInit::new_resizeable(size, size * 2)
.with_hasher(twox_hash::xxhash64::RandomState::default())
.attach_writer();
let mut rng = rand::rng();
while writer.get_num_buckets_in_use() < ideal_filled as usize {
let key: FileCacheKey = rng.random();
let val = FileCacheEntry::dummy();
apply_op(TestOp(key, Some(val)), &mut writer);
}
b.iter(|| writer.shuffle());
});
group.bench_function("small_rehash_ahash", |b| {
let ideal_filled = 4_000_000;
let size = 5_000_000;
let mut writer = HashMapInit::new_resizeable(size, size * 2)
.with_hasher(ahash::RandomState::default())
.attach_writer();
let mut rng = rand::rng();
while writer.get_num_buckets_in_use() < ideal_filled as usize {
let key: FileCacheKey = rng.random();
let val = FileCacheEntry::dummy();
apply_op(TestOp(key, Some(val)), &mut writer);
}
b.iter(|| writer.shuffle());
});
group.bench_function("small_rehash_seahash", |b| {
let ideal_filled = 4_000_000;
let size = 5_000_000;
let mut writer = HashMapInit::new_resizeable(size, size * 2)
.with_hasher(SeaRandomState::new())
.attach_writer();
let mut rng = rand::rng();
while writer.get_num_buckets_in_use() < ideal_filled as usize {
let key: FileCacheKey = rng.random();
let val = FileCacheEntry::dummy();
apply_op(TestOp(key, Some(val)), &mut writer);
}
b.iter(|| writer.shuffle());
});
group.finish();
}
fn real_benchs(c: &mut Criterion) {
let mut group = c.benchmark_group("Realistic workloads");
group.sample_size(10);
group.bench_function("real_bulk_insert", |b| {
let size = 125_000_000;
let ideal_filled = 100_000_000;
let mut rng = rand::rng();
b.iter_batched(
|| HashMapInit::new_resizeable(size, size * 2).attach_writer(),
|writer| {
for _ in 0..ideal_filled {
let key: FileCacheKey = rng.random();
let val = FileCacheEntry::dummy();
let entry = writer.entry(key);
match entry {
Entry::Occupied(mut e) => {
std::hint::black_box(e.insert(val));
}
Entry::Vacant(e) => {
let _ = std::hint::black_box(e.insert(val).unwrap());
}
}
}
},
BatchSize::SmallInput,
)
});
group.bench_function("real_rehash", |b| {
let size = 125_000_000;
let ideal_filled = 100_000_000;
let mut writer = HashMapInit::new_resizeable(size, size).attach_writer();
let mut rng = rand::rng();
while writer.get_num_buckets_in_use() < ideal_filled {
let key: FileCacheKey = rng.random();
let val = FileCacheEntry::dummy();
apply_op(TestOp(key, Some(val)), &mut writer);
}
b.iter(|| writer.shuffle());
});
group.bench_function("real_rehash_hashbrown", |b| {
let size = 125_000_000;
let ideal_filled = 100_000_000;
let mut writer = hashbrown::raw::RawTable::new();
let mut rng = rand::rng();
let hasher = rustc_hash::FxBuildHasher;
unsafe {
writer
.resize(
size,
|(k, _)| hasher.hash_one(k),
hashbrown::raw::Fallibility::Infallible,
)
.unwrap();
}
while writer.len() < ideal_filled as usize {
let key: FileCacheKey = rng.random();
let val = FileCacheEntry::dummy();
writer.insert(hasher.hash_one(&key), (key, val), |(k, _)| {
hasher.hash_one(k)
});
}
b.iter(|| unsafe {
writer.table.rehash_in_place(
&|table, index| {
hasher.hash_one(
&table
.bucket::<(FileCacheKey, FileCacheEntry)>(index)
.as_ref()
.0,
)
},
std::mem::size_of::<(FileCacheKey, FileCacheEntry)>(),
if std::mem::needs_drop::<(FileCacheKey, FileCacheEntry)>() {
Some(|ptr| std::ptr::drop_in_place(ptr as *mut (FileCacheKey, FileCacheEntry)))
} else {
None
},
)
});
});
for elems in [2, 4, 8, 16, 32, 64, 96, 112] {
group.bench_with_input(
BenchmarkId::new("real_rehash_varied", elems),
&elems,
|b, &size| {
let ideal_filled = size * 1_000_000;
let size = 125_000_000;
let mut writer = HashMapInit::new_resizeable(size, size).attach_writer();
let mut rng = rand::rng();
while writer.get_num_buckets_in_use() < ideal_filled as usize {
let key: FileCacheKey = rng.random();
let val = FileCacheEntry::dummy();
apply_op(TestOp(key, Some(val)), &mut writer);
}
b.iter(|| writer.shuffle());
},
);
group.bench_with_input(
BenchmarkId::new("real_rehash_varied_hashbrown", elems),
&elems,
|b, &size| {
let ideal_filled = size * 1_000_000;
let size = 125_000_000;
let mut writer = hashbrown::raw::RawTable::new();
let mut rng = rand::rng();
let hasher = rustc_hash::FxBuildHasher;
unsafe {
writer
.resize(
size,
|(k, _)| hasher.hash_one(k),
hashbrown::raw::Fallibility::Infallible,
)
.unwrap();
}
while writer.len() < ideal_filled as usize {
let key: FileCacheKey = rng.random();
let val = FileCacheEntry::dummy();
writer.insert(hasher.hash_one(&key), (key, val), |(k, _)| {
hasher.hash_one(k)
});
}
b.iter(|| unsafe {
writer.table.rehash_in_place(
&|table, index| {
hasher.hash_one(
&table
.bucket::<(FileCacheKey, FileCacheEntry)>(index)
.as_ref()
.0,
)
},
std::mem::size_of::<(FileCacheKey, FileCacheEntry)>(),
if std::mem::needs_drop::<(FileCacheKey, FileCacheEntry)>() {
Some(|ptr| {
std::ptr::drop_in_place(ptr as *mut (FileCacheKey, FileCacheEntry))
})
} else {
None
},
)
});
},
);
}
group.finish();
}
criterion_group!(benches, small_benchs, real_benchs);
criterion_main!(benches);

View File

@@ -16,7 +16,6 @@
//!
//! Concurrency is managed very simply: the entire map is guarded by one shared-memory RwLock.
use std::fmt::Debug;
use std::hash::{BuildHasher, Hash};
use std::mem::MaybeUninit;
@@ -57,22 +56,6 @@ pub struct HashMapInit<'a, K, V, S = rustc_hash::FxBuildHasher> {
num_buckets: u32,
}
impl<'a, K, V, S> Debug for HashMapInit<'a, K, V, S>
where
K: Debug,
V: Debug,
{
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("HashMapInit")
.field("shmem_handle", &self.shmem_handle)
.field("shared_ptr", &self.shared_ptr)
.field("shared_size", &self.shared_size)
// .field("hasher", &self.hasher)
.field("num_buckets", &self.num_buckets)
.finish()
}
}
/// This is a per-process handle to a hash table that (possibly) lives in shared memory.
/// If a child process is launched with fork(), the child process should
/// get its own HashMapAccess by calling HashMapInit::attach_writer/reader().
@@ -88,20 +71,6 @@ pub struct HashMapAccess<'a, K, V, S = rustc_hash::FxBuildHasher> {
unsafe impl<K: Sync, V: Sync, S> Sync for HashMapAccess<'_, K, V, S> {}
unsafe impl<K: Send, V: Send, S> Send for HashMapAccess<'_, K, V, S> {}
impl<'a, K, V, S> Debug for HashMapAccess<'a, K, V, S>
where
K: Debug,
V: Debug,
{
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("HashMapAccess")
.field("shmem_handle", &self.shmem_handle)
.field("shared_ptr", &self.shared_ptr)
// .field("hasher", &self.hasher)
.finish()
}
}
impl<'a, K: Clone + Hash + Eq, V, S> HashMapInit<'a, K, V, S> {
/// Change the 'hasher' used by the hash table.
///
@@ -329,7 +298,7 @@ where
/// Get a reference to the entry containing a key.
///
/// NB: This takes a write lock as there's no way to distinguish whether the intention
/// NB: THis takes a write lock as there's no way to distinguish whether the intention
/// is to use the entry for reading or for writing in advance.
pub fn entry(&self, key: K) -> Entry<'a, '_, K, V> {
let hash = self.get_hash_value(&key);

View File

@@ -1,6 +1,5 @@
//! Simple hash table with chaining.
use std::fmt::Debug;
use std::hash::Hash;
use std::mem::MaybeUninit;
@@ -18,19 +17,6 @@ pub(crate) struct Bucket<K, V> {
pub(crate) inner: Option<(K, V)>,
}
impl<K, V> Debug for Bucket<K, V>
where
K: Debug,
V: Debug,
{
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("Bucket")
.field("next", &self.next)
.field("inner", &self.inner)
.finish()
}
}
/// Core hash table implementation.
pub(crate) struct CoreHashMap<'a, K, V> {
/// Dictionary used to map hashes to bucket indices.
@@ -45,22 +31,6 @@ pub(crate) struct CoreHashMap<'a, K, V> {
pub(crate) buckets_in_use: u32,
}
impl<'a, K, V> Debug for CoreHashMap<'a, K, V>
where
K: Debug,
V: Debug,
{
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("CoreHashMap")
.field("dictionary", &self.dictionary)
.field("buckets", &self.buckets)
.field("free_head", &self.free_head)
.field("alloc_limit", &self.alloc_limit)
.field("buckets_in_use", &self.buckets_in_use)
.finish()
}
}
/// Error for when there are no empty buckets left but one is needed.
#[derive(Debug, PartialEq)]
pub struct FullError;

View File

@@ -61,10 +61,6 @@ impl<K, V> OccupiedEntry<'_, '_, K, V> {
///
/// This may result in multiple bucket accesses if the entry was obtained by index as the
/// previous chain entry needs to be discovered in this case.
///
/// # Panics
/// Panics if the `prev_pos` field is equal to [`PrevPos::Unknown`]. In practice, this means
/// the entry was obtained via calling something like [`super::HashMapAccess::entry_at_bucket`].
pub fn remove(mut self) -> V {
// If this bucket was queried by index, go ahead and follow its chain from the start.
let prev = if let PrevPos::Unknown(hash) = self.prev_pos {

View File

@@ -21,7 +21,6 @@ use nix::unistd::ftruncate as nix_ftruncate;
/// the underlying file is resized. Do not access the area beyond the current size. Currently, that
/// will cause the file to be expanded, but we might use `mprotect()` etc. to enforce that in the
/// future.
#[derive(Debug)]
pub struct ShmemHandle {
/// memfd file descriptor
fd: OwnedFd,
@@ -36,7 +35,6 @@ pub struct ShmemHandle {
}
/// This is stored at the beginning in the shared memory area.
#[derive(Debug)]
struct SharedStruct {
max_size: usize,

View File

@@ -34,16 +34,13 @@ macro_rules! critical {
#[macro_export]
macro_rules! critical_timeline {
($tenant_shard_id:expr, $timeline_id:expr, $corruption_detected:expr, $($arg:tt)*) => {{
($tenant_shard_id:expr, $timeline_id:expr, $($arg:tt)*) => {{
if cfg!(debug_assertions) {
panic!($($arg)*);
}
// Increment both metrics
$crate::logging::TRACING_EVENT_COUNT_METRIC.inc_critical();
$crate::logging::HADRON_CRITICAL_STORAGE_EVENT_COUNT_METRIC.inc(&$tenant_shard_id.to_string(), &$timeline_id.to_string());
if let Some(c) = $corruption_detected.as_ref() {
c.store(true, std::sync::atomic::Ordering::Relaxed);
}
let backtrace = std::backtrace::Backtrace::capture();
tracing::error!("CRITICAL: [tenant_shard_id: {}, timeline_id: {}] {}\n{backtrace}",
$tenant_shard_id, $timeline_id, format!($($arg)*));

View File

@@ -310,11 +310,6 @@ impl AtomicLsn {
}
}
/// Consumes the atomic and returns the contained value.
pub const fn into_inner(self) -> Lsn {
Lsn(self.inner.into_inner())
}
/// Atomically retrieve the `Lsn` value from memory.
pub fn load(&self) -> Lsn {
Lsn(self.inner.load(Ordering::Acquire))

View File

@@ -32,9 +32,6 @@ pub struct PageserverFeedback {
pub replytime: SystemTime,
/// Used to track feedbacks from different shards. Always zero for unsharded tenants.
pub shard_number: u32,
/// If true, the pageserver has detected corruption and the safekeeper and postgres
/// should stop sending WAL.
pub corruption_detected: bool,
}
impl PageserverFeedback {
@@ -46,7 +43,6 @@ impl PageserverFeedback {
disk_consistent_lsn: Lsn::INVALID,
replytime: *PG_EPOCH,
shard_number: 0,
corruption_detected: false,
}
}
@@ -105,13 +101,6 @@ impl PageserverFeedback {
buf.put_u32(self.shard_number);
}
if self.corruption_detected {
nkeys += 1;
buf.put_slice(b"corruption_detected\0");
buf.put_i32(1);
buf.put_u8(1);
}
buf[buf_ptr] = nkeys;
}
@@ -158,11 +147,6 @@ impl PageserverFeedback {
assert_eq!(len, 4);
rf.shard_number = buf.get_u32();
}
b"corruption_detected" => {
let len = buf.get_i32();
assert_eq!(len, 1);
rf.corruption_detected = buf.get_u8() != 0;
}
_ => {
let len = buf.get_i32();
warn!(
@@ -222,26 +206,6 @@ mod tests {
assert_eq!(rf, rf_parsed);
}
// Test that databricks-specific fields added to the PageserverFeedback message are serialized
// and deserialized correctly, in addition to the existing fields from upstream.
#[test]
fn test_replication_feedback_databricks_fields() {
let mut rf = PageserverFeedback::empty();
rf.current_timeline_size = 12345678;
rf.last_received_lsn = Lsn(23456789);
rf.disk_consistent_lsn = Lsn(34567890);
rf.remote_consistent_lsn = Lsn(45678901);
rf.replytime = *PG_EPOCH + Duration::from_secs(100_000_000);
rf.shard_number = 1;
rf.corruption_detected = true;
let mut data = BytesMut::new();
rf.serialize(&mut data);
let rf_parsed = PageserverFeedback::parse(data.freeze());
assert_eq!(rf, rf_parsed);
}
#[test]
fn test_replication_feedback_unknown_key() {
let mut rf = PageserverFeedback::empty();

View File

@@ -59,10 +59,6 @@ impl ShardCount {
pub const MAX: Self = Self(u8::MAX);
pub const MIN: Self = Self(0);
pub fn unsharded() -> Self {
ShardCount(0)
}
/// The internal value of a ShardCount may be zero, which means "1 shard, but use
/// legacy format for TenantShardId that excludes the shard suffix", also known
/// as [`TenantShardId::unsharded`].

View File

@@ -341,34 +341,6 @@ extern "C-unwind" fn log_internal(
}
}
/* BEGIN_HADRON */
extern "C" fn reset_safekeeper_statuses_for_metrics(wp: *mut WalProposer, num_safekeepers: u32) {
unsafe {
let callback_data = (*(*wp).config).callback_data;
let api = callback_data as *mut Box<dyn ApiImpl>;
if api.is_null() {
return;
}
(*api).reset_safekeeper_statuses_for_metrics(&mut (*wp), num_safekeepers);
}
}
extern "C" fn update_safekeeper_status_for_metrics(
wp: *mut WalProposer,
sk_index: u32,
status: u8,
) {
unsafe {
let callback_data = (*(*wp).config).callback_data;
let api = callback_data as *mut Box<dyn ApiImpl>;
if api.is_null() {
return;
}
(*api).update_safekeeper_status_for_metrics(&mut (*wp), sk_index, status);
}
}
/* END_HADRON */
#[derive(Debug, PartialEq)]
pub enum Level {
Debug5,
@@ -442,10 +414,6 @@ pub(crate) fn create_api() -> walproposer_api {
finish_sync_safekeepers: Some(finish_sync_safekeepers),
process_safekeeper_feedback: Some(process_safekeeper_feedback),
log_internal: Some(log_internal),
/* BEGIN_HADRON */
reset_safekeeper_statuses_for_metrics: Some(reset_safekeeper_statuses_for_metrics),
update_safekeeper_status_for_metrics: Some(update_safekeeper_status_for_metrics),
/* END_HADRON */
}
}
@@ -458,7 +426,6 @@ pub fn empty_shmem() -> crate::bindings::WalproposerShmemState {
remote_consistent_lsn: 0,
replytime: 0,
shard_number: 0,
corruption_detected: false,
};
let empty_wal_rate_limiter = crate::bindings::WalRateLimiter {
@@ -483,8 +450,6 @@ pub fn empty_shmem() -> crate::bindings::WalproposerShmemState {
replica_promote: false,
min_ps_feedback: empty_feedback,
wal_rate_limiter: empty_wal_rate_limiter,
num_safekeepers: 0,
safekeeper_status: [0; 32],
}
}

View File

@@ -159,21 +159,6 @@ pub trait ApiImpl {
fn after_election(&self, _wp: &mut WalProposer) {
todo!()
}
/* BEGIN_HADRON */
fn reset_safekeeper_statuses_for_metrics(&self, _wp: &mut WalProposer, _num_safekeepers: u32) {
// Do nothing for testing purposes.
}
fn update_safekeeper_status_for_metrics(
&self,
_wp: &mut WalProposer,
_sk_index: u32,
_status: u8,
) {
// Do nothing for testing purposes.
}
/* END_HADRON */
}
#[derive(Debug)]

View File

@@ -54,7 +54,6 @@ pageserver_api.workspace = true
pageserver_client.workspace = true # for ResponseErrorMessageExt TOOD refactor that
pageserver_compaction.workspace = true
pageserver_page_api.workspace = true
peekable.workspace = true
pem.workspace = true
pin-project-lite.workspace = true
postgres_backend.workspace = true
@@ -67,7 +66,6 @@ postgres-types.workspace = true
posthog_client_lite.workspace = true
pprof.workspace = true
pq_proto.workspace = true
prost.workspace = true
rand.workspace = true
range-set-blaze = { version = "0.1.16", features = ["alloc"] }
regex.workspace = true

View File

@@ -230,14 +230,16 @@ impl PageserverClient {
) -> tonic::Result<page_api::GetPageResponse> {
// Fast path: request is for a single shard.
if let Some(shard_id) =
GetPageSplitter::for_single_shard(&req, shards.count, shards.stripe_size)?
GetPageSplitter::for_single_shard(&req, shards.count, shards.stripe_size)
.map_err(|err| tonic::Status::internal(err.to_string()))?
{
return Self::get_page_with_shard(req, shards.get(shard_id)?).await;
}
// Request spans multiple shards. Split it, dispatch concurrent per-shard requests, and
// reassemble the responses.
let mut splitter = GetPageSplitter::split(req, shards.count, shards.stripe_size)?;
let mut splitter = GetPageSplitter::split(req, shards.count, shards.stripe_size)
.map_err(|err| tonic::Status::internal(err.to_string()))?;
let mut shard_requests = FuturesUnordered::new();
for (shard_id, shard_req) in splitter.drain_requests() {
@@ -247,10 +249,14 @@ impl PageserverClient {
}
while let Some((shard_id, shard_response)) = shard_requests.next().await.transpose()? {
splitter.add_response(shard_id, shard_response)?;
splitter
.add_response(shard_id, shard_response)
.map_err(|err| tonic::Status::internal(err.to_string()))?;
}
Ok(splitter.collect_response()?)
splitter
.get_response()
.map_err(|err| tonic::Status::internal(err.to_string()))
}
/// Fetches pages on the given shard. Does not retry internally.

View File

@@ -3,4 +3,3 @@ mod pool;
mod retry;
pub use client::{PageserverClient, ShardSpec};
pub use pageserver_api::shard::ShardStripeSize; // used in ShardSpec

View File

@@ -24,4 +24,4 @@ mod split;
pub use client::Client;
pub use model::*;
pub use split::{GetPageSplitter, SplitError};
pub use split::GetPageSplitter;

View File

@@ -33,8 +33,6 @@ pub enum ProtocolError {
Invalid(&'static str, String),
#[error("required field '{0}' is missing")]
Missing(&'static str),
#[error("invalid combination of not_modified_lsn '{0}' and request_lsn '{1}'")]
InvalidLsns(Lsn, Lsn),
}
impl ProtocolError {
@@ -87,9 +85,9 @@ impl TryFrom<proto::ReadLsn> for ReadLsn {
return Err(ProtocolError::invalid("request_lsn", pb.request_lsn));
}
if pb.not_modified_since_lsn > pb.request_lsn {
return Err(ProtocolError::InvalidLsns(
Lsn(pb.not_modified_since_lsn),
Lsn(pb.request_lsn),
return Err(ProtocolError::invalid(
"not_modified_since_lsn",
pb.not_modified_since_lsn,
));
}
Ok(Self {

View File

@@ -1,5 +1,6 @@
use std::collections::HashMap;
use anyhow::anyhow;
use bytes::Bytes;
use crate::model::*;
@@ -26,19 +27,19 @@ impl GetPageSplitter {
req: &GetPageRequest,
count: ShardCount,
stripe_size: Option<ShardStripeSize>,
) -> Result<Option<ShardIndex>, SplitError> {
) -> anyhow::Result<Option<ShardIndex>> {
// Fast path: unsharded tenant.
if count.is_unsharded() {
return Ok(Some(ShardIndex::unsharded()));
}
let Some(stripe_size) = stripe_size else {
return Err("stripe size must be given for sharded tenants".into());
return Err(anyhow!("stripe size must be given for sharded tenants"));
};
// Find the first page's shard, for comparison.
let Some(&first_page) = req.block_numbers.first() else {
return Err("no block numbers in request".into());
return Err(anyhow!("no block numbers in request"));
};
let key = rel_block_to_key(req.rel, first_page);
let shard_number = key_to_shard_number(count, stripe_size, &key);
@@ -59,7 +60,7 @@ impl GetPageSplitter {
req: GetPageRequest,
count: ShardCount,
stripe_size: Option<ShardStripeSize>,
) -> Result<Self, SplitError> {
) -> anyhow::Result<Self> {
// The caller should make sure we don't split requests unnecessarily.
debug_assert!(
Self::for_single_shard(&req, count, stripe_size)?.is_none(),
@@ -67,10 +68,10 @@ impl GetPageSplitter {
);
if count.is_unsharded() {
return Err("unsharded tenant, no point in splitting request".into());
return Err(anyhow!("unsharded tenant, no point in splitting request"));
}
let Some(stripe_size) = stripe_size else {
return Err("stripe size must be given for sharded tenants".into());
return Err(anyhow!("stripe size must be given for sharded tenants"));
};
// Split the requests by shard index.
@@ -128,32 +129,35 @@ impl GetPageSplitter {
/// Adds a response from the given shard. The response must match the request ID and have an OK
/// status code. A response must not already exist for the given shard ID.
#[allow(clippy::result_large_err)]
pub fn add_response(
&mut self,
shard_id: ShardIndex,
response: GetPageResponse,
) -> Result<(), SplitError> {
) -> anyhow::Result<()> {
// The caller should already have converted status codes into tonic::Status.
if response.status_code != GetPageStatusCode::Ok {
return Err(SplitError(format!(
return Err(anyhow!(
"unexpected non-OK response for shard {shard_id}: {} {}",
response.status_code,
response.reason.unwrap_or_default()
)));
));
}
if response.request_id != self.response.request_id {
return Err(SplitError(format!(
return Err(anyhow!(
"response ID mismatch for shard {shard_id}: expected {}, got {}",
self.response.request_id, response.request_id
)));
self.response.request_id,
response.request_id
));
}
if response.request_id != self.response.request_id {
return Err(SplitError(format!(
return Err(anyhow!(
"response ID mismatch for shard {shard_id}: expected {}, got {}",
self.response.request_id, response.request_id
)));
self.response.request_id,
response.request_id
));
}
// Place the shard response pages into the assembled response, in request order.
@@ -165,27 +169,26 @@ impl GetPageSplitter {
}
let Some(slot) = self.response.pages.get_mut(i) else {
return Err(SplitError(format!(
"no block_shards slot {i} for shard {shard_id}"
)));
return Err(anyhow!("no block_shards slot {i} for shard {shard_id}"));
};
let Some(page) = pages.next() else {
return Err(SplitError(format!(
return Err(anyhow!(
"missing page {} in shard {shard_id} response",
slot.block_number
)));
));
};
if page.block_number != slot.block_number {
return Err(SplitError(format!(
return Err(anyhow!(
"shard {shard_id} returned wrong page at index {i}, expected {} got {}",
slot.block_number, page.block_number
)));
slot.block_number,
page.block_number
));
}
if !slot.image.is_empty() {
return Err(SplitError(format!(
return Err(anyhow!(
"shard {shard_id} returned duplicate page {} at index {i}",
slot.block_number
)));
));
}
*slot = page;
@@ -193,54 +196,32 @@ impl GetPageSplitter {
// Make sure we've consumed all pages from the shard response.
if let Some(extra_page) = pages.next() {
return Err(SplitError(format!(
return Err(anyhow!(
"shard {shard_id} returned extra page: {}",
extra_page.block_number
)));
));
}
Ok(())
}
/// Collects the final, assembled response.
pub fn collect_response(self) -> Result<GetPageResponse, SplitError> {
/// Fetches the final, assembled response.
#[allow(clippy::result_large_err)]
pub fn get_response(self) -> anyhow::Result<GetPageResponse> {
// Check that the response is complete.
for (i, page) in self.response.pages.iter().enumerate() {
if page.image.is_empty() {
return Err(SplitError(format!(
return Err(anyhow!(
"missing page {} for shard {}",
page.block_number,
self.block_shards
.get(i)
.map(|s| s.to_string())
.unwrap_or_else(|| "?".to_string())
)));
));
}
}
Ok(self.response)
}
}
/// A GetPageSplitter error.
#[derive(Debug, thiserror::Error)]
#[error("{0}")]
pub struct SplitError(String);
impl From<&str> for SplitError {
fn from(err: &str) -> Self {
SplitError(err.to_string())
}
}
impl From<String> for SplitError {
fn from(err: String) -> Self {
SplitError(err)
}
}
impl From<SplitError> for tonic::Status {
fn from(err: SplitError) -> Self {
tonic::Status::internal(err.0)
}
}

View File

@@ -25,9 +25,6 @@ tracing.workspace = true
tokio.workspace = true
tokio-stream.workspace = true
tokio-util.workspace = true
axum.workspace = true
http.workspace = true
metrics.workspace = true
tonic.workspace = true
url.workspace = true

View File

@@ -34,10 +34,6 @@ use crate::util::{request_stats, tokio_thread_local_stats};
/// GetPage@LatestLSN, uniformly distributed across the compute-accessible keyspace.
#[derive(clap::Parser)]
pub(crate) struct Args {
#[clap(long, default_value = "false")]
grpc: bool,
#[clap(long, default_value = "false")]
grpc_stream: bool,
#[clap(long, default_value = "http://localhost:9898")]
mgmt_api_endpoint: String,
/// Pageserver connection string. Supports postgresql:// and grpc:// protocols.
@@ -82,9 +78,6 @@ pub(crate) struct Args {
#[clap(long)]
set_io_mode: Option<pageserver_api::models::virtual_file::IoMode>,
#[clap(long)]
only_relnode: Option<u32>,
/// Queue depth generated in each client.
#[clap(long, default_value = "1")]
queue_depth: NonZeroUsize,
@@ -99,31 +92,10 @@ pub(crate) struct Args {
#[clap(long, default_value = "1")]
batch_size: NonZeroUsize,
#[clap(long)]
only_relnode: Option<u32>,
targets: Option<Vec<TenantTimelineId>>,
#[clap(long, default_value = "100")]
pool_max_consumers: NonZeroUsize,
#[clap(long, default_value = "5")]
pool_error_threshold: NonZeroUsize,
#[clap(long, default_value = "5000")]
pool_connect_timeout: NonZeroUsize,
#[clap(long, default_value = "1000")]
pool_connect_backoff: NonZeroUsize,
#[clap(long, default_value = "60000")]
pool_max_idle_duration: NonZeroUsize,
#[clap(long, default_value = "0")]
max_delay_ms: usize,
#[clap(long, default_value = "0")]
percent_drops: usize,
#[clap(long, default_value = "0")]
percent_hangs: usize,
}
/// State shared by all clients
@@ -180,6 +152,7 @@ pub(crate) fn main(args: Args) -> anyhow::Result<()> {
main_impl(args, thread_local_stats)
})
}
async fn main_impl(
args: Args,
all_thread_local_stats: AllThreadLocalStats<request_stats::Stats>,
@@ -344,7 +317,6 @@ async fn main_impl(
let rps_period = args
.per_client_rate
.map(|rps_limit| Duration::from_secs_f64(1.0 / (rps_limit as f64)));
let make_worker: &dyn Fn(WorkerId) -> Pin<Box<dyn Send + Future<Output = ()>>> = &|worker_id| {
let ss = shared_state.clone();
let cancel = cancel.clone();

View File

@@ -453,7 +453,6 @@ impl TimelineHandles {
handles: Default::default(),
}
}
async fn get(
&mut self,
tenant_id: TenantId,
@@ -3686,7 +3685,8 @@ impl GrpcPageServiceHandler {
// Fast path: the request fits in a single shard.
if let Some(shard_index) =
GetPageSplitter::for_single_shard(&req, shard_id.count, Some(shard_id.stripe_size))?
GetPageSplitter::for_single_shard(&req, shard_id.count, Some(shard_id.stripe_size))
.map_err(|err| tonic::Status::internal(err.to_string()))?
{
// We got the shard ID from the first page, so these must be equal.
assert_eq!(shard_index.shard_number, shard_id.number);
@@ -3697,7 +3697,8 @@ impl GrpcPageServiceHandler {
// The request spans multiple shards; split it and dispatch parallel requests. All pages
// were originally in the parent shard, and during a split all children are local, so we
// expect to find local shards for all pages.
let mut splitter = GetPageSplitter::split(req, shard_id.count, Some(shard_id.stripe_size))?;
let mut splitter = GetPageSplitter::split(req, shard_id.count, Some(shard_id.stripe_size))
.map_err(|err| tonic::Status::internal(err.to_string()))?;
let mut shard_requests = FuturesUnordered::new();
for (shard_index, shard_req) in splitter.drain_requests() {
@@ -3716,10 +3717,14 @@ impl GrpcPageServiceHandler {
}
while let Some((shard_index, shard_response)) = shard_requests.next().await.transpose()? {
splitter.add_response(shard_index, shard_response)?;
splitter
.add_response(shard_index, shard_response)
.map_err(|err| tonic::Status::internal(err.to_string()))?;
}
Ok(splitter.collect_response()?)
splitter
.get_response()
.map_err(|err| tonic::Status::internal(err.to_string()))
}
}

View File

@@ -397,11 +397,6 @@ pub struct Timeline {
/// If true, the last compaction failed.
compaction_failed: AtomicBool,
/// Begin Hadron: If true, the pageserver has likely detected data corruption in the timeline.
/// We need to feed this information back to the Safekeeper and postgres for them to take the
/// appropriate action.
corruption_detected: AtomicBool,
/// Notifies the tenant compaction loop that there is pending L0 compaction work.
l0_compaction_trigger: Arc<Notify>,
@@ -3315,7 +3310,6 @@ impl Timeline {
compaction_lock: tokio::sync::Mutex::default(),
compaction_failed: AtomicBool::default(),
corruption_detected: AtomicBool::default(),
l0_compaction_trigger: resources.l0_compaction_trigger,
gc_lock: tokio::sync::Mutex::default(),
@@ -6010,17 +6004,6 @@ impl Timeline {
)))
});
// Begin Hadron
//
fail_point!("create-image-layer-fail-simulated-corruption", |_| {
self.corruption_detected
.store(true, std::sync::atomic::Ordering::Relaxed);
Err(CreateImageLayersError::Other(anyhow::anyhow!(
"failpoint create-image-layer-fail-simulated-corruption"
)))
});
// End Hadron
let io_concurrency = IoConcurrency::spawn_from_conf(
self.conf.get_vectored_concurrent_io,
self.gate
@@ -7166,7 +7149,6 @@ impl Timeline {
critical_timeline!(
self.tenant_shard_id,
self.timeline_id,
Some(&self.corruption_detected),
"walredo failure during page reconstruction: {err:?}"
);
}

View File

@@ -1397,7 +1397,6 @@ impl Timeline {
critical_timeline!(
self.tenant_shard_id,
self.timeline_id,
Some(&self.corruption_detected),
"missing key during compaction: {err:?}"
);
}
@@ -1442,7 +1441,6 @@ impl Timeline {
critical_timeline!(
self.tenant_shard_id,
self.timeline_id,
Some(&self.corruption_detected),
"could not compact, repartitioning keyspace failed: {e:?}"
);
}

View File

@@ -365,7 +365,6 @@ pub(super) async fn handle_walreceiver_connection(
critical_timeline!(
timeline.tenant_shard_id,
timeline.timeline_id,
Some(&timeline.corruption_detected),
"{msg}"
);
return Err(WalReceiverError::Other(anyhow!(msg)));
@@ -383,7 +382,6 @@ pub(super) async fn handle_walreceiver_connection(
critical_timeline!(
timeline.tenant_shard_id,
timeline.timeline_id,
Some(&timeline.corruption_detected),
"{msg}"
);
return Err(WalReceiverError::Other(anyhow!(msg)));
@@ -457,7 +455,6 @@ pub(super) async fn handle_walreceiver_connection(
critical_timeline!(
timeline.tenant_shard_id,
timeline.timeline_id,
Some(&timeline.corruption_detected),
"{err:?}"
);
}
@@ -589,9 +586,6 @@ pub(super) async fn handle_walreceiver_connection(
remote_consistent_lsn,
replytime: ts,
shard_number: timeline.tenant_shard_id.shard_number.0 as u32,
corruption_detected: timeline
.corruption_detected
.load(std::sync::atomic::Ordering::Relaxed),
};
debug!("neon_status_update {status_update:?}");

View File

@@ -23,7 +23,6 @@
use std::backtrace::Backtrace;
use std::collections::HashMap;
use std::sync::atomic::AtomicBool;
use std::sync::{Arc, OnceLock};
use std::time::{Duration, Instant, SystemTime};
@@ -423,8 +422,6 @@ impl WalIngest {
critical_timeline!(
modification.tline.tenant_shard_id,
modification.tline.timeline_id,
// Hadron: No need to raise the corruption flag here; the caller of `ingest_record()` will do it.
None::<&AtomicBool>,
"clear_vm_bits for unknown VM relation {vm_rel}"
);
return Ok(());
@@ -434,8 +431,6 @@ impl WalIngest {
critical_timeline!(
modification.tline.tenant_shard_id,
modification.tline.timeline_id,
// Hadron: No need to raise the corruption flag here; the caller of `ingest_record()` will do it.
None::<&AtomicBool>,
"new_vm_blk {blknum} not in {vm_rel} of size {vm_size}"
);
new_vm_blk = None;
@@ -446,8 +441,6 @@ impl WalIngest {
critical_timeline!(
modification.tline.tenant_shard_id,
modification.tline.timeline_id,
// Hadron: No need to raise the corruption flag here; the caller of `ingest_record()` will do it.
None::<&AtomicBool>,
"old_vm_blk {blknum} not in {vm_rel} of size {vm_size}"
);
old_vm_blk = None;

View File

@@ -5,12 +5,10 @@ MODULE_big = neon
OBJS = \
$(WIN32RES) \
communicator.o \
communicator_new.o \
communicator_process.o \
extension_server.o \
file_cache.o \
hll.o \
lfc_prewarm.o \
libpagestore.o \
logical_replication_monitor.o \
neon.o \
@@ -69,7 +67,6 @@ WALPROP_OBJS = \
# libcommunicator.a is built by cargo from the Rust sources under communicator/
# subdirectory. `cargo build` also generates communicator_bindings.h.
communicator_new.o: communicator/communicator_bindings.h
communicator_process.o: communicator/communicator_bindings.h
file_cache.o: communicator/communicator_bindings.h

View File

@@ -17,31 +17,12 @@ rest_broker = []
[dependencies]
axum.workspace = true
bytes.workspace = true
clashmap.workspace = true
http.workspace = true
libc.workspace = true
nix.workspace = true
atomic_enum = "0.3.0"
measured.workspace = true
prometheus.workspace = true
prost.workspace = true
strum_macros.workspace = true
thiserror.workspace = true
tonic = { workspace = true, default-features = false, features=["codegen", "prost", "transport"] }
tokio = { workspace = true, features = ["macros", "net", "io-util", "rt", "rt-multi-thread"] }
tokio-pipe = { version = "0.2.12" }
tracing.workspace = true
tracing-subscriber.workspace = true
metrics.workspace = true
uring-common = { workspace = true, features = ["bytes"] }
pageserver_client_grpc.workspace = true
pageserver_api.workspace = true
pageserver_page_api.workspace = true
neon-shmem.workspace = true
measured.workspace = true
utils.workspace = true
workspace_hack = { version = "0.1", path = "../../../workspace_hack" }

View File

@@ -3,18 +3,9 @@
This package provides the so-called "compute-pageserver communicator",
or just "communicator" in short. The communicator is a separate
background worker process that runs in the PostgreSQL server. It's
part of the neon extension.
The commuicator handles the communication with the pageservers, and
also provides an HTTP endpoint for metrics over a local Unix Domain
socket (aka. the "communicator control socket"). On the PostgreSQL
side, the glue code in pgxn/neon/ uses the communicator to implement
the PostgreSQL Storage Manager (SMGR) interface.
## Design criteria
- Low latency
- Saturate a 10 Gbit / s network interface without becoming a bottleneck
part of the neon extension. Currently, it only provides an HTTP
endpoint for metrics, but in the future it will evolve to handle all
communications with the pageservers.
## Source code view
@@ -23,122 +14,10 @@ pgxn/neon/communicator_process.c
the glue that interacts with PostgreSQL code and the Rust
code in the communicator process.
pgxn/neon/communicator_new.c
Contains the backend code that interacts with the communicator
process.
pgxn/neon/communicator/src/backend_interface.rs
The entry point for calls from each backend.
pgxn/neon/communicator/src/init.rs
Initialization at server startup
pgxn/neon/communicator/src/worker_process/
Worker process main loop and glue code
At compilation time, pgxn/neon/communicator/ produces a static
library, libcommunicator.a. It is linked to the neon.so extension
library.
The real networking code, which is independent of PostgreSQL, is in
the pageserver/client_grpc crate.
## Process view
The communicator runs in a dedicated background worker process, the
"communicator process". The communicator uses a multi-threaded Tokio
runtime to execute the IO requests. So the communicator process has
multiple threads running. That's unusual for Postgres processes and
care must be taken to make that work.
### Backend <-> worker communication
Each backend has a number of I/O request slots in shared memory. The
slots are statically allocated for each backend, and must not be
accessed by other backends. The worker process reads requests from the
shared memory slots, and writes responses back to the slots.
Here's an example snapshot of the system, when two requests from two
different backends are in progress:
```
Backends Request slots Communicator process
--------- ------------- --------------------
Backend 1 1: Idle
2: Idle
3: Processing tokio task handling request 3
Backend 2 4: Completed
5: Processing tokio task handling request 5
6: Idle
... ...
```
To submit an IO request, the backend first picks one of its Idle
slots, writes the IO request in the slot, and updates it to
'Submitted' state. That transfers the ownership of the slot to the
worker process, until the worker process marks the request as
Completed. The worker process spawns a separate Tokio task for each
request.
To inform the worker process that a request slot has a pending IO
request, there's a pipe shared by the worker process and all backend
processes. The backend writes the index of the request slot to the
pipe after changing the slot's state to Submitted. This wakes up the
worker process.
(Note that the pipe is just used for wakeups, but the worker process
is free to pick up Submitted IO requests even without receiving the
wakeup. As of this writing, it doesn't do that, but it might be useful
in the future to reduce latency even further, for example.)
When the worker process has completed processing the request, it
writes the result back in the request slot. A GetPage request can also
contain a pointer to buffer in the shared buffer cache. In that case,
the worker process writes the resulting page contents directly to the
buffer, and just a result code in the request slot. It then updates
the 'state' field to Completed, which passes the owner ship back to
the originating backend. Finally, it signals the process Latch of the
originating backend, waking it up.
### Differences between PostgreSQL v16, v17 and v18
PostgreSQL v18 introduced the new AIO mechanism. The PostgreSQL AIO
mechanism uses a very similar mechanism as described in the previous
section, for the communication between AIO worker processes and
backends. With our communicator, the AIO worker processes are not
used, but we use the same PgAioHandle request slots as in upstream.
For Neon-specific IO requests like GetDbSize, a neon request slot is
used. But for the actual IO requests, the request slot merely contains
a pointer to the PgAioHandle slot. The worker process updates the
status of that, calls the IO callbacks upon completionetc, just like
the upstream AIO worker processes do.
## Sequence diagram
neon
PostgreSQL extension backend_interface.rs worker_process.rs processor tonic
| . . . .
| smgr_read() . . . .
+-------------> + . . .
. | . . .
. | rcommunicator_ . . .
. | get_page_at_lsn . . .
. +------------------> + . .
| . .
| write request to . . .
| slot . .
| . .
| . .
| submit_request() . .
+-----------------> + .
| | .
| | db_size_request . .
+---------------->.
. TODO
### Compute <-> pageserver protocol
The protocol between Compute and the pageserver is based on gRPC. See `protos/`.

View File

@@ -1,224 +0,0 @@
//! This module implements a request/response "slot" for submitting
//! requests from backends to the communicator process.
//!
//! NB: The "backend" side of this code runs in Postgres backend processes,
//! which means that it is not safe to use the 'tracing' crate for logging, nor
//! to launch threads or use tokio tasks!
use std::cell::UnsafeCell;
use std::sync::atomic::{AtomicI32, Ordering};
use crate::neon_request::{NeonIORequest, NeonIOResult};
use atomic_enum::atomic_enum;
/// One request/response slot. Each backend has its own set of slots that it
/// uses.
///
/// This is the moral equivalent of PgAioHandle for Postgres AIO requests
/// Like PgAioHandle, try to keep this small.
///
/// There is an array of these in shared memory. Therefore, this must be Sized.
///
/// ## Lifecycle of a request
///
/// A slot is always owned by either the backend process or the communicator
/// process, depending on the 'state'. Only the owning process is allowed to
/// read or modify the slot, except for reading the 'state' itself to check who
/// owns it.
///
/// A slot begins in the Idle state, where it is owned by the backend process.
/// To submit a request, the backend process fills the slot with the request
/// data, and changes it to the Submitted state. After changing the state, the
/// slot is owned by the communicator process, and the backend is not allowed
/// to access it until the communicator process marks it as Completed.
///
/// When the communicator process sees that the slot is in Submitted state, it
/// starts to process the request. After processing the request, it stores the
/// result in the slot, and changes the state to Completed. It is now owned by
/// the backend process again, which may now read the result, and reuse the
/// slot for a new request.
///
/// For correctness of the above protocol, we really only need two states:
/// "owned by backend" and "owned by communicator process". But to help with
/// debugging and better assertions, there are a few more states. When the
/// backend starts to fill in the request details in the slot, it first sets the
/// state from Idle to Filling, and when it's done with that, from Filling to
/// Submitted. In the Filling state, the slot is still owned by the
/// backend. Similarly, when the communicator process starts to process a
/// request, it sets it to Processing state first, but the slot is still owned
/// by the communicator process.
///
/// This struct doesn't handle waking up the communicator process when a request
/// has been submitted or when a response is ready. The 'owner_procno' is used
/// for waking up the backend on completion, but that happens elsewhere.
pub struct NeonIORequestSlot {
/// similar to PgAioHandleState
state: AtomicNeonIORequestSlotState,
/// The owning process's ProcNumber. The worker process uses this to set the
/// process's latch on completion.
///
/// (This could be calculated from num_neon_request_slots_per_backend and
/// the index of this slot in the overall 'neon_requst_slots array'. But we
/// prefer the communicator process to not know how the request slots are
/// divided between the backends.)
owner_procno: AtomicI32,
/// SAFETY: This is modified by submit_request(), after it has established
/// ownership of the slot by setting state from Idle to Filling
request: UnsafeCell<NeonIORequest>,
/// Valid when state is Completed
///
/// SAFETY: This is modified by RequestProcessingGuard::complete(). There
/// can be only one RequestProcessingGuard outstanding for a slot at a time,
/// because it is returned by start_processing_request() which checks the
/// state, so RequestProcessingGuard has exclusive access to the slot.
result: UnsafeCell<NeonIOResult>,
}
// The protocol described in the "Lifecycle of a request" section above ensures
// the safe access to the fields
unsafe impl Send for NeonIORequestSlot {}
unsafe impl Sync for NeonIORequestSlot {}
impl Default for NeonIORequestSlot {
fn default() -> NeonIORequestSlot {
NeonIORequestSlot {
owner_procno: AtomicI32::new(-1),
request: UnsafeCell::new(NeonIORequest::Empty),
result: UnsafeCell::new(NeonIOResult::Empty),
state: AtomicNeonIORequestSlotState::new(NeonIORequestSlotState::Idle),
}
}
}
#[atomic_enum]
#[derive(Eq, PartialEq)]
pub enum NeonIORequestSlotState {
Idle,
/// Backend is filling in the request
Filling,
/// Backend has submitted the request to the communicator, but the
/// communicator process has not yet started processing it.
Submitted,
/// Communicator is processing the request
Processing,
/// Communicator has completed the request, and the 'result' field is now
/// valid, but the backend has not read the result yet.
Completed,
}
impl NeonIORequestSlot {
/// Write a request to the slot, and mark it as Submitted.
///
/// Note: This does not wake up the worker process to actually process
/// the request. It's the caller's responsibility to do that.
pub fn submit_request(&self, request: &NeonIORequest, proc_number: i32) {
// Verify that the slot is in Idle state previously, and put it in
// Filling state.
//
// XXX: This step isn't strictly necessary. Assuming the caller didn't
// screw up and try to use a slot that's already in use, we could fill
// the slot and switch it directly from Idle to Submitted state.
if let Err(s) = self.state.compare_exchange(
NeonIORequestSlotState::Idle,
NeonIORequestSlotState::Filling,
Ordering::Relaxed,
Ordering::Relaxed,
) {
panic!("unexpected state in request slot: {s:?}");
}
// Fill in the request details
self.owner_procno.store(proc_number, Ordering::Relaxed);
unsafe { *self.request.get() = *request }
// This synchronizes-with store/swap in [`start_processing_request`].
// Note that this ensures that the previous non-atomic writes visible
// to other threads too.
self.state
.store(NeonIORequestSlotState::Submitted, Ordering::Release);
}
pub fn get_state(&self) -> NeonIORequestSlotState {
self.state.load(Ordering::Relaxed)
}
pub fn try_get_result(&self) -> Option<NeonIOResult> {
// This synchronizes-with the store/swap in [`RequestProcessingGuard::completed`]
let state = self.state.load(Ordering::Acquire);
if state == NeonIORequestSlotState::Completed {
let result = unsafe { *self.result.get() };
self.state
.store(NeonIORequestSlotState::Idle, Ordering::Relaxed);
Some(result)
} else {
None
}
}
/// Read the IO request from the slot indicated in the wakeup
pub fn start_processing_request<'a>(&'a self) -> Option<RequestProcessingGuard<'a>> {
// XXX: using atomic load rather than compare_exchange would be
// sufficient here, as long as the communicator process has _some_ means
// of tracking which requests it's already processing. That could be a
// flag somewhere in communicator's private memory, for example.
//
// This synchronizes-with the store in [`submit_request`].
if let Err(s) = self.state.compare_exchange(
NeonIORequestSlotState::Submitted,
NeonIORequestSlotState::Processing,
Ordering::Acquire,
Ordering::Relaxed,
) {
// FIXME surprising state. This is unexpected at the moment, but if we
// started to process requests more aggressively, without waiting for the
// read from the pipe, then this could happen
panic!("unexpected state in request slot: {s:?}");
}
Some(RequestProcessingGuard(self))
}
}
/// [`NeonIORequestSlot::start_processing_request`] returns this guard object to
/// indicate that the the caller now "owns" the slot, until it calls
/// [`RequestProcessingGuard::completed`].
///
/// TODO: implement Drop on this, to mark the request as Aborted or Errored
/// if [`RequestProcessingGuard::completed`] is not called.
pub struct RequestProcessingGuard<'a>(&'a NeonIORequestSlot);
unsafe impl<'a> Send for RequestProcessingGuard<'a> {}
unsafe impl<'a> Sync for RequestProcessingGuard<'a> {}
impl<'a> RequestProcessingGuard<'a> {
pub fn get_request(&self) -> &NeonIORequest {
unsafe { &*self.0.request.get() }
}
pub fn get_owner_procno(&self) -> i32 {
self.0.owner_procno.load(Ordering::Relaxed)
}
pub fn completed(self, result: NeonIOResult) {
// Store the result to the slot.
unsafe {
*self.0.result.get() = result;
};
// Mark the request as completed. After that, we no longer have
// ownership of the slot, and must not modify it.
let old_state = self
.0
.state
.swap(NeonIORequestSlotState::Completed, Ordering::Release);
assert!(old_state == NeonIORequestSlotState::Processing);
}
}

View File

@@ -1,296 +0,0 @@
//! This code runs in each backend process. That means that launching Rust threads, panicking
//! etc. is forbidden!
use std::os::fd::OwnedFd;
use crate::backend_comms::NeonIORequestSlot;
use crate::init::CommunicatorInitStruct;
use crate::integrated_cache::{BackendCacheReadOp, IntegratedCacheReadAccess};
use crate::neon_request::{CCachedGetPageVResult, CLsn, COid};
use crate::neon_request::{NeonIORequest, NeonIOResult};
use utils::lsn::Lsn;
pub struct CommunicatorBackendStruct<'t> {
my_proc_number: i32,
neon_request_slots: &'t [NeonIORequestSlot],
submission_pipe_write_fd: OwnedFd,
pending_cache_read_op: Option<BackendCacheReadOp<'t>>,
integrated_cache: &'t IntegratedCacheReadAccess<'t>,
}
#[unsafe(no_mangle)]
pub extern "C" fn rcommunicator_backend_init(
cis: Box<CommunicatorInitStruct>,
my_proc_number: i32,
) -> &'static mut CommunicatorBackendStruct<'static> {
if my_proc_number < 0 {
panic!("cannot attach to communicator shared memory with procnumber {my_proc_number}");
}
let integrated_cache = Box::leak(Box::new(cis.integrated_cache_init_struct.backend_init()));
let bs: &'static mut CommunicatorBackendStruct =
Box::leak(Box::new(CommunicatorBackendStruct {
my_proc_number,
neon_request_slots: cis.neon_request_slots,
submission_pipe_write_fd: cis.submission_pipe_write_fd,
pending_cache_read_op: None,
integrated_cache,
}));
bs
}
/// Start a request. You can poll for its completion and get the result by
/// calling bcomm_poll_dbsize_request_completion(). The communicator will wake
/// us up by setting our process latch, so to wait for the completion, wait on
/// the latch and call bcomm_poll_dbsize_request_completion() every time the
/// latch is set.
///
/// Safety: The C caller must ensure that the references are valid.
/// The requested slot must be free, or this panics.
#[unsafe(no_mangle)]
pub extern "C" fn bcomm_start_io_request(
bs: &'_ mut CommunicatorBackendStruct,
slot_idx: i32,
request: &NeonIORequest,
immediate_result_ptr: &mut NeonIOResult,
) -> i32 {
assert!(bs.pending_cache_read_op.is_none());
// Check if the request can be satisfied from the cache first
if let NeonIORequest::RelSize(req) = request {
if let Some(nblocks) = bs.integrated_cache.get_rel_size(&req.reltag()) {
*immediate_result_ptr = NeonIOResult::RelSize(nblocks);
return -1;
}
}
// Create neon request and submit it
bs.start_neon_io_request(slot_idx, request);
slot_idx
}
#[unsafe(no_mangle)]
pub extern "C" fn bcomm_start_get_page_v_request(
bs: &mut CommunicatorBackendStruct,
slot_idx: i32,
request: &NeonIORequest,
immediate_result_ptr: &mut CCachedGetPageVResult,
) -> i32 {
let NeonIORequest::GetPageV(get_pagev_request) = request else {
panic!("invalid request passed to bcomm_start_get_page_v_request()");
};
assert!(matches!(request, NeonIORequest::GetPageV(_)));
assert!(bs.pending_cache_read_op.is_none());
// Check if the request can be satisfied from the cache first
let mut all_cached = true;
let mut read_op = bs.integrated_cache.start_read_op();
for i in 0..get_pagev_request.nblocks {
if let Some(cache_block) = read_op.get_page(
&get_pagev_request.reltag(),
get_pagev_request.block_number + i as u32,
) {
immediate_result_ptr.cache_block_numbers[i as usize] = cache_block;
} else {
// not found in cache
all_cached = false;
break;
}
}
if all_cached {
bs.pending_cache_read_op = Some(read_op);
return -1;
}
// Create neon request and submit it
bs.start_neon_io_request(slot_idx, request);
slot_idx
}
/// Check if a request has completed. Returns:
///
/// -1 if the request is still being processed
/// 0 on success
#[unsafe(no_mangle)]
pub extern "C" fn bcomm_poll_request_completion(
bs: &mut CommunicatorBackendStruct,
request_slot_idx: u32,
result_p: &mut NeonIOResult,
) -> i32 {
match bs.neon_request_slots[request_slot_idx as usize].try_get_result() {
None => -1, // still processing
Some(result) => {
*result_p = result;
0
}
}
}
/// Check if a request has completed. Returns:
///
/// 'false' if the slot is Idle. The backend process has ownership.
/// 'true' if the slot is busy, and should be polled for result.
#[unsafe(no_mangle)]
pub extern "C" fn bcomm_get_request_slot_status(
bs: &mut CommunicatorBackendStruct,
request_slot_idx: u32,
) -> bool {
use crate::backend_comms::NeonIORequestSlotState;
match bs.neon_request_slots[request_slot_idx as usize].get_state() {
NeonIORequestSlotState::Idle => false,
NeonIORequestSlotState::Filling => {
// 'false' would be the right result here. However, this
// is a very transient state. The C code should never
// leave a slot in this state, so if it sees that,
// something's gone wrong and it's not clear what to do
// with it.
panic!("unexpected Filling state in request slot {request_slot_idx}");
}
NeonIORequestSlotState::Submitted => true,
NeonIORequestSlotState::Processing => true,
NeonIORequestSlotState::Completed => true,
}
}
// LFC functions
/// Finish a local file cache read
///
//
#[unsafe(no_mangle)]
pub extern "C" fn bcomm_finish_cache_read(bs: &mut CommunicatorBackendStruct) -> bool {
if let Some(op) = bs.pending_cache_read_op.take() {
op.finish()
} else {
panic!("bcomm_finish_cache_read() called with no cached read pending");
}
}
/// Check if LFC contains the given buffer, and update its last-written LSN if not.
///
/// This is used in WAL replay in read replica, to skip updating pages that are
/// not in cache.
#[unsafe(no_mangle)]
pub extern "C" fn bcomm_update_lw_lsn_for_block_if_not_cached(
bs: &mut CommunicatorBackendStruct,
spc_oid: COid,
db_oid: COid,
rel_number: u32,
fork_number: u8,
block_number: u32,
lsn: CLsn,
) -> bool {
bs.integrated_cache.update_lw_lsn_for_block_if_not_cached(
&pageserver_page_api::RelTag {
spcnode: spc_oid,
dbnode: db_oid,
relnode: rel_number,
forknum: fork_number,
},
block_number,
Lsn(lsn),
)
}
#[repr(C)]
#[derive(Clone, Debug)]
pub struct FileCacheIterator {
next_bucket: u64,
pub spc_oid: COid,
pub db_oid: COid,
pub rel_number: u32,
pub fork_number: u8,
pub block_number: u32,
}
/// Iterate over LFC contents
#[allow(clippy::missing_safety_doc)]
#[unsafe(no_mangle)]
pub unsafe extern "C" fn bcomm_cache_iterate_begin(
_bs: &mut CommunicatorBackendStruct,
iter: *mut FileCacheIterator,
) {
unsafe { (*iter).next_bucket = 0 };
}
#[allow(clippy::missing_safety_doc)]
#[unsafe(no_mangle)]
pub unsafe extern "C" fn bcomm_cache_iterate_next(
bs: &mut CommunicatorBackendStruct,
iter: *mut FileCacheIterator,
) -> bool {
use crate::integrated_cache::GetBucketResult;
loop {
let next_bucket = unsafe { (*iter).next_bucket } as usize;
match bs.integrated_cache.get_bucket(next_bucket) {
GetBucketResult::Occupied(rel, blk) => {
unsafe {
(*iter).spc_oid = rel.spcnode;
(*iter).db_oid = rel.dbnode;
(*iter).rel_number = rel.relnode;
(*iter).fork_number = rel.forknum;
(*iter).block_number = blk;
(*iter).next_bucket += 1;
}
break true;
}
GetBucketResult::Vacant => {
unsafe {
(*iter).next_bucket += 1;
}
continue;
}
GetBucketResult::OutOfBounds => {
break false;
}
}
}
}
#[allow(clippy::missing_safety_doc)]
#[unsafe(no_mangle)]
pub unsafe extern "C" fn bcomm_cache_get_num_pages_used(bs: &mut CommunicatorBackendStruct) -> u64 {
bs.integrated_cache.get_num_buckets_in_use() as u64
}
impl<'t> CommunicatorBackendStruct<'t> {
/// The slot must be free, or this panics.
pub(crate) fn start_neon_io_request(&mut self, request_slot_idx: i32, request: &NeonIORequest) {
let my_proc_number = self.my_proc_number;
self.neon_request_slots[request_slot_idx as usize].submit_request(request, my_proc_number);
// Tell the communicator about it
self.notify_about_request(request_slot_idx);
}
/// Send a wakeup to the communicator process
fn notify_about_request(self: &CommunicatorBackendStruct<'t>, request_slot_idx: i32) {
// wake up communicator by writing the idx to the submission pipe
//
// This can block, if the pipe is full. That should be very rare,
// because the communicator tries hard to drain the pipe to prevent
// that. Also, there's a natural upper bound on how many wakeups can be
// queued up: there is only a limited number of request slots for each
// backend.
//
// If it does block very briefly, that's not too serious.
let idxbuf = request_slot_idx.to_ne_bytes();
let _res = nix::unistd::write(&self.submission_pipe_write_fd, &idxbuf);
// FIXME: check result, return any errors
}
}

View File

@@ -1,162 +0,0 @@
//! Implement the "low-level" parts of the file cache.
//!
//! This module just deals with reading and writing the file, and keeping track
//! which blocks in the cache file are in use and which are free. The "high
//! level" parts of tracking which block in the cache file corresponds to which
//! relation block is handled in 'integrated_cache' instead.
//!
//! This module is only used to access the file from the communicator
//! process. The backend processes *also* read the file (and sometimes also
//! write it? ), but the backends use direct C library calls for that.
use std::fs::File;
use std::os::unix::fs::FileExt;
use std::path::Path;
use std::sync::Arc;
use std::sync::Mutex;
use crate::BLCKSZ;
use tokio::task::spawn_blocking;
pub type CacheBlock = u64;
pub const INVALID_CACHE_BLOCK: CacheBlock = u64::MAX;
#[derive(Debug)]
pub struct FileCache {
file: Arc<File>,
free_list: Mutex<FreeList>,
// metrics
max_blocks_gauge: metrics::IntGauge,
num_free_blocks_gauge: metrics::IntGauge,
}
// TODO: We keep track of all free blocks in this vec. That doesn't really scale.
// Idea: when free_blocks fills up with more than 1024 entries, write them all to
// one block on disk.
#[derive(Debug)]
struct FreeList {
next_free_block: CacheBlock,
max_blocks: u64,
free_blocks: Vec<CacheBlock>,
}
impl FileCache {
pub fn new(file_cache_path: &Path, mut initial_size: u64) -> Result<FileCache, std::io::Error> {
if initial_size < 100 {
tracing::warn!(
"min size for file cache is 100 blocks, {} requested",
initial_size
);
initial_size = 100;
}
let file = std::fs::OpenOptions::new()
.read(true)
.write(true)
.truncate(true)
.create(true)
.open(file_cache_path)?;
let max_blocks_gauge = metrics::IntGauge::new(
"file_cache_max_blocks",
"Local File Cache size in 8KiB blocks",
)
.unwrap();
let num_free_blocks_gauge = metrics::IntGauge::new(
"file_cache_num_free_blocks",
"Number of free 8KiB blocks in Local File Cache",
)
.unwrap();
tracing::info!("initialized file cache with {} blocks", initial_size);
Ok(FileCache {
file: Arc::new(file),
free_list: Mutex::new(FreeList {
next_free_block: 0,
max_blocks: initial_size,
free_blocks: Vec::new(),
}),
max_blocks_gauge,
num_free_blocks_gauge,
})
}
// File cache management
pub async fn read_block(
&self,
cache_block: CacheBlock,
mut dst: impl uring_common::buf::IoBufMut + Send + Sync,
) -> Result<(), std::io::Error> {
assert!(dst.bytes_total() == BLCKSZ);
let file = self.file.clone();
let dst_ref = unsafe { std::slice::from_raw_parts_mut(dst.stable_mut_ptr(), BLCKSZ) };
spawn_blocking(move || file.read_exact_at(dst_ref, cache_block * BLCKSZ as u64)).await??;
Ok(())
}
pub async fn write_block(
&self,
cache_block: CacheBlock,
src: impl uring_common::buf::IoBuf + Send + Sync,
) -> Result<(), std::io::Error> {
assert!(src.bytes_init() == BLCKSZ);
let file = self.file.clone();
let src_ref = unsafe { std::slice::from_raw_parts(src.stable_ptr(), BLCKSZ) };
spawn_blocking(move || file.write_all_at(src_ref, cache_block * BLCKSZ as u64)).await??;
Ok(())
}
pub fn alloc_block(&self) -> Option<CacheBlock> {
let mut free_list = self.free_list.lock().unwrap();
if let Some(x) = free_list.free_blocks.pop() {
return Some(x);
}
if free_list.next_free_block < free_list.max_blocks {
let result = free_list.next_free_block;
free_list.next_free_block += 1;
return Some(result);
}
None
}
pub fn dealloc_block(&self, cache_block: CacheBlock) {
let mut free_list = self.free_list.lock().unwrap();
free_list.free_blocks.push(cache_block);
}
}
impl metrics::core::Collector for FileCache {
fn desc(&self) -> Vec<&metrics::core::Desc> {
let mut descs = Vec::new();
descs.append(&mut self.max_blocks_gauge.desc());
descs.append(&mut self.num_free_blocks_gauge.desc());
descs
}
fn collect(&self) -> Vec<metrics::proto::MetricFamily> {
// Update the gauges with fresh values first
{
let free_list = self.free_list.lock().unwrap();
self.max_blocks_gauge.set(free_list.max_blocks as i64);
let total_free_blocks: i64 = free_list.free_blocks.len() as i64
+ (free_list.max_blocks as i64 - free_list.next_free_block as i64);
self.num_free_blocks_gauge.set(total_free_blocks);
}
let mut values = Vec::new();
values.append(&mut self.max_blocks_gauge.collect());
values.append(&mut self.num_free_blocks_gauge.collect());
values
}
}

View File

@@ -1,109 +0,0 @@
//! Global allocator, for tracking memory usage of the Rust parts
//!
//! Postgres is designed to handle allocation failure (ie. malloc() returning NULL) gracefully. It
//! rolls backs the transaction and gives the user an "ERROR: out of memory" error. Rust code
//! however panics if an allocation fails. We don't want that to ever happen, because an unhandled
//! panic leads to Postgres crash and restart. Our strategy is to pre-allocate a large enough chunk
//! of memory for use by the Rust code, so that the allocations never fail.
//!
//! To pick the size for the pre-allocated chunk, we have a metric to track the high watermark
//! memory usage of all the Rust allocations in total.
//!
//! TODO:
//!
//! - Currently we just export the metrics. Actual allocations are still just passed through to
//! the system allocator.
//! - Take padding etc. overhead into account
use std::alloc::{GlobalAlloc, Layout, System};
use std::sync::atomic::{AtomicU64, AtomicUsize, Ordering};
use metrics::IntGauge;
struct MyAllocator {
allocations: AtomicU64,
deallocations: AtomicU64,
allocated: AtomicUsize,
high: AtomicUsize,
}
unsafe impl GlobalAlloc for MyAllocator {
unsafe fn alloc(&self, layout: Layout) -> *mut u8 {
self.allocations.fetch_add(1, Ordering::Relaxed);
let mut allocated = self.allocated.fetch_add(layout.size(), Ordering::Relaxed);
allocated += layout.size();
self.high.fetch_max(allocated, Ordering::Relaxed);
unsafe { System.alloc(layout) }
}
unsafe fn dealloc(&self, ptr: *mut u8, layout: Layout) {
self.deallocations.fetch_add(1, Ordering::Relaxed);
self.allocated.fetch_sub(layout.size(), Ordering::Relaxed);
unsafe { System.dealloc(ptr, layout) }
}
}
#[global_allocator]
static GLOBAL: MyAllocator = MyAllocator {
allocations: AtomicU64::new(0),
deallocations: AtomicU64::new(0),
allocated: AtomicUsize::new(0),
high: AtomicUsize::new(0),
};
pub struct MyAllocatorCollector {
allocations: IntGauge,
deallocations: IntGauge,
allocated: IntGauge,
high: IntGauge,
}
impl MyAllocatorCollector {
pub fn new() -> MyAllocatorCollector {
MyAllocatorCollector {
allocations: IntGauge::new("allocations_total", "Number of allocations in Rust code")
.unwrap(),
deallocations: IntGauge::new(
"deallocations_total",
"Number of deallocations in Rust code",
)
.unwrap(),
allocated: IntGauge::new("allocated_total", "Bytes currently allocated").unwrap(),
high: IntGauge::new("allocated_high", "High watermark of allocated bytes").unwrap(),
}
}
}
impl metrics::core::Collector for MyAllocatorCollector {
fn desc(&self) -> Vec<&metrics::core::Desc> {
let mut descs = Vec::new();
descs.append(&mut self.allocations.desc());
descs.append(&mut self.deallocations.desc());
descs.append(&mut self.allocated.desc());
descs.append(&mut self.high.desc());
descs
}
fn collect(&self) -> Vec<metrics::proto::MetricFamily> {
let mut values = Vec::new();
// update the gauges
self.allocations
.set(GLOBAL.allocations.load(Ordering::Relaxed) as i64);
self.deallocations
.set(GLOBAL.allocations.load(Ordering::Relaxed) as i64);
self.allocated
.set(GLOBAL.allocated.load(Ordering::Relaxed) as i64);
self.high.set(GLOBAL.high.load(Ordering::Relaxed) as i64);
values.append(&mut self.allocations.collect());
values.append(&mut self.deallocations.collect());
values.append(&mut self.allocated.collect());
values.append(&mut self.high.collect());
values
}
}

View File

@@ -1,166 +0,0 @@
//! Initialization functions. These are executed in the postmaster process,
//! at different stages of server startup.
//!
//!
//! Communicator initialization steps:
//!
//! 1. At postmaster startup, before shared memory is allocated,
//! rcommunicator_shmem_size() is called to get the amount of
//! shared memory that this module needs.
//!
//! 2. Later, after the shared memory has been allocated,
//! rcommunicator_shmem_init() is called to initialize the shmem
//! area.
//!
//! Per process initialization:
//!
//! When a backend process starts up, it calls rcommunicator_backend_init().
//! In the communicator worker process, other functions are called, see
//! `worker_process` module.
use std::ffi::c_int;
use std::mem;
use std::mem::MaybeUninit;
use std::os::fd::OwnedFd;
use crate::backend_comms::NeonIORequestSlot;
use crate::integrated_cache::IntegratedCacheInitStruct;
/// This struct is created in the postmaster process, and inherited to
/// the communicator process and all backend processes through fork()
#[repr(C)]
pub struct CommunicatorInitStruct {
pub submission_pipe_read_fd: OwnedFd,
pub submission_pipe_write_fd: OwnedFd,
// Shared memory data structures
pub num_neon_request_slots: u32,
pub neon_request_slots: &'static [NeonIORequestSlot],
pub integrated_cache_init_struct: IntegratedCacheInitStruct<'static>,
}
impl std::fmt::Debug for CommunicatorInitStruct {
fn fmt(&self, fmt: &mut std::fmt::Formatter<'_>) -> Result<(), std::fmt::Error> {
fmt.debug_struct("CommunicatorInitStruct")
.field("submission_pipe_read_fd", &self.submission_pipe_read_fd)
.field("submission_pipe_write_fd", &self.submission_pipe_write_fd)
.field("num_neon_request_slots", &self.num_neon_request_slots)
.field("neon_request_slots length", &self.neon_request_slots.len())
.finish()
}
}
#[unsafe(no_mangle)]
pub extern "C" fn rcommunicator_shmem_size(num_neon_request_slots: u32) -> u64 {
let mut size = 0;
size += mem::size_of::<NeonIORequestSlot>() * num_neon_request_slots as usize;
// For integrated_cache's Allocator. TODO: make this adjustable
size += IntegratedCacheInitStruct::shmem_size();
size as u64
}
/// Initialize the shared memory segment. Returns a backend-private
/// struct, which will be inherited by backend processes through fork
#[unsafe(no_mangle)]
pub extern "C" fn rcommunicator_shmem_init(
submission_pipe_read_fd: c_int,
submission_pipe_write_fd: c_int,
num_neon_request_slots: u32,
shmem_area_ptr: *mut MaybeUninit<u8>,
shmem_area_len: u64,
initial_file_cache_size: u64,
max_file_cache_size: u64,
) -> &'static mut CommunicatorInitStruct {
let shmem_area: &'static mut [MaybeUninit<u8>] =
unsafe { std::slice::from_raw_parts_mut(shmem_area_ptr, shmem_area_len as usize) };
let (neon_request_slots, remaining_area) =
alloc_array_from_slice::<NeonIORequestSlot>(shmem_area, num_neon_request_slots as usize);
for slot in neon_request_slots.iter_mut() {
slot.write(NeonIORequestSlot::default());
}
// 'neon_request_slots' is initialized now. (MaybeUninit::slice_assume_init_mut() is nightly-only
// as of this writing.)
let neon_request_slots = unsafe {
std::mem::transmute::<&mut [MaybeUninit<NeonIORequestSlot>], &mut [NeonIORequestSlot]>(
neon_request_slots,
)
};
// Give the rest of the area to the integrated cache
let integrated_cache_init_struct = IntegratedCacheInitStruct::shmem_init(
remaining_area,
initial_file_cache_size,
max_file_cache_size,
);
let (submission_pipe_read_fd, submission_pipe_write_fd) = unsafe {
use std::os::fd::FromRawFd;
(
OwnedFd::from_raw_fd(submission_pipe_read_fd),
OwnedFd::from_raw_fd(submission_pipe_write_fd),
)
};
let cis: &'static mut CommunicatorInitStruct = Box::leak(Box::new(CommunicatorInitStruct {
submission_pipe_read_fd,
submission_pipe_write_fd,
num_neon_request_slots,
neon_request_slots,
integrated_cache_init_struct,
}));
cis
}
pub fn alloc_from_slice<T>(
area: &mut [MaybeUninit<u8>],
) -> (&mut MaybeUninit<T>, &mut [MaybeUninit<u8>]) {
let layout = std::alloc::Layout::new::<T>();
let area_start = area.as_mut_ptr();
// pad to satisfy alignment requirements
let padding = area_start.align_offset(layout.align());
if padding + layout.size() > area.len() {
panic!("out of memory");
}
let area = &mut area[padding..];
let (result_area, remain) = area.split_at_mut(layout.size());
let result_ptr: *mut MaybeUninit<T> = result_area.as_mut_ptr().cast();
let result = unsafe { result_ptr.as_mut().unwrap() };
(result, remain)
}
pub fn alloc_array_from_slice<T>(
area: &mut [MaybeUninit<u8>],
len: usize,
) -> (&mut [MaybeUninit<T>], &mut [MaybeUninit<u8>]) {
let layout = std::alloc::Layout::new::<T>();
let area_start = area.as_mut_ptr();
// pad to satisfy alignment requirements
let padding = area_start.align_offset(layout.align());
if padding + layout.size() * len > area.len() {
panic!("out of memory");
}
let area = &mut area[padding..];
let (result_area, remain) = area.split_at_mut(layout.size() * len);
let result_ptr: *mut MaybeUninit<T> = result_area.as_mut_ptr().cast();
let result = unsafe { std::slice::from_raw_parts_mut(result_ptr.as_mut().unwrap(), len) };
(result, remain)
}

File diff suppressed because it is too large Load Diff

View File

@@ -1,29 +1,5 @@
//! Three main parts:
//! - async tokio communicator core, which receives requests and processes them.
//! - Main loop and requests queues, which routes requests from backends to the core
//! - the per-backend glue code, which submits requests
mod backend_comms;
// mark this 'pub', because these functions are called from C code. Otherwise, the compiler
// complains about a bunch of structs and enum variants being unused, because it thinkgs
// the functions that use them are never called. There are some C-callable functions in
// other modules too, but marking this as pub is currently enough to silence the warnings
//
// TODO: perhaps collect *all* the extern "C" functions to one module?
pub mod backend_interface;
mod file_cache;
mod init;
mod integrated_cache;
mod neon_request;
mod worker_process;
mod global_allocator;
/// Name of the Unix Domain Socket that serves the metrics, and other APIs in the
/// future. This is within the Postgres data directory.
const NEON_COMMUNICATOR_SOCKET_NAME: &str = "neon-communicator.socket";
// FIXME: get this from postgres headers somehow
pub const BLCKSZ: usize = 8192;

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