Try to enable a custom postgres_exporter query.

.github/workflows/build_and_test.yml

@@ -852,7 +852,7 @@ jobs:
       run:
         shell: sh -eu {0}
     env:
-      VM_BUILDER_VERSION: v0.18.5
+      VM_BUILDER_VERSION: v0.19.0
 
     steps:
       - name: Checkout
@@ -874,8 +874,7 @@ jobs:
       - name: Build vm image
         run: |
           ./vm-builder \
-            -enable-file-cache \
-            -cgroup-uid=postgres \
+            -spec=vm-image-spec.yaml \
             -src=369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}} \
             -dst=369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
 

Dockerfile.compute-node

@@ -714,6 +714,23 @@ RUN wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.3.tar.gz -
     cargo pgrx install --release && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/ulid.control
 
+#########################################################################################
+#
+# Layer "pg-wait-sampling-pg-build"
+# compile pg_wait_sampling extension
+#
+#########################################################################################
+FROM build-deps AS pg-wait-sampling-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+ENV PATH "/usr/local/pgsql/bin/:$PATH"
+RUN wget https://github.com/postgrespro/pg_wait_sampling/archive/refs/tags/v1.1.5.tar.gz -O pg_wait_sampling.tar.gz && \
+    echo 'a03da6a413f5652ce470a3635ed6ebba528c74cb26aa4cfced8aff8a8441f81ec6dd657ff62cd6ce96a4e6ce02cad9f2519ae9525367ece60497aa20faafde5c  pg_wait_sampling.tar.gz' | sha512sum -c && \
+    mkdir pg_wait_sampling-src && cd pg_wait_sampling-src && tar xvzf ../pg_wait_sampling.tar.gz --strip-components=1 -C . && \
+    make USE_PGXS=1 -j $(getconf _NPROCESSORS_ONLN) && \
+    make USE_PGXS=1 -j $(getconf _NPROCESSORS_ONLN) install && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pg_wait_sampling.control
+
 #########################################################################################
 #
 # Layer "neon-pg-ext-build"
@@ -750,6 +767,7 @@ COPY --from=rdkit-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-uuidv7-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-roaringbitmap-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-embedding-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=pg-wait-sampling-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY pgxn/ pgxn/
 
 RUN make -j $(getconf _NPROCESSORS_ONLN) \

proxy/src/serverless/conn_pool.rs

@@ -210,12 +210,7 @@ impl GlobalConnPool {
                 client.session.send(session_id)?;
                 latency_timer.pool_hit();
                 latency_timer.success();
-                return Ok(Client {
-                    conn_id: client.conn_id,
-                    inner: Some(client),
-                    span: Span::current(),
-                    pool,
-                });
+                return Ok(Client::new(client, pool).await);
             }
         } else {
             let conn_id = uuid::Uuid::new_v4();
@@ -263,15 +258,11 @@ impl GlobalConnPool {
             _ => {}
         }
 
-        new_client.map(|inner| Client {
-            conn_id: inner.conn_id,
-            inner: Some(inner),
-            span: Span::current(),
-            pool,
-        })
+        // new_client.map(|inner| Client::new(inner, pool).await)
+        Ok(Client::new(new_client?, pool).await)
     }
 
-    fn put(&self, conn_info: &ConnInfo, client: ClientInner) -> anyhow::Result<()> {
+    fn put(&self, conn_info: &ConnInfo, client: ClientInner, pid: i32) -> anyhow::Result<()> {
         let conn_id = client.conn_id;
 
         // We want to hold this open while we return. This ensures that the pool can't close
@@ -315,9 +306,9 @@ impl GlobalConnPool {
 
         // do logging outside of the mutex
         if returned {
-            info!(%conn_id, "pool: returning connection '{conn_info}' back to the pool, total_conns={total_conns}, for this (db, user)={per_db_size}");
+            info!(%conn_id, "pool: returning connection '{conn_info}' back to the pool, total_conns={total_conns}, for this (db, user)={per_db_size}, pid={pid}");
         } else {
-            info!(%conn_id, "pool: throwing away connection '{conn_info}' because pool is full, total_conns={total_conns}");
+            info!(%conn_id, "pool: throwing away connection '{conn_info}' because pool is full, total_conns={total_conns}, pid={pid}");
         }
 
         Ok(())
@@ -528,6 +519,22 @@ struct ClientInner {
     conn_id: uuid::Uuid,
 }
 
+impl ClientInner {
+    pub async fn get_pid(&mut self) -> anyhow::Result<i32> {
+        let rows = self.inner.query("select pg_backend_pid();", &[]).await?;
+        if rows.len() != 1 {
+            Err(anyhow::anyhow!(
+                "expected 1 row from pg_backend_pid(), got {}",
+                rows.len()
+            ))
+        } else {
+            let pid = rows[0].get(0);
+            info!(%pid, "got pid");
+            Ok(pid)
+        }
+    }
+}
+
 impl Client {
     pub fn metrics(&self) -> Arc<MetricCounter> {
         USAGE_METRICS.register(self.inner.as_ref().unwrap().ids.clone())
@@ -539,6 +546,7 @@ pub struct Client {
     span: Span,
     inner: Option<ClientInner>,
     pool: Option<(ConnInfo, Arc<GlobalConnPool>)>,
+    pid: i32,
 }
 
 pub struct Discard<'a> {
@@ -547,12 +555,25 @@ pub struct Discard<'a> {
 }
 
 impl Client {
+    pub(self) async fn new(
+        mut inner: ClientInner,
+        pool: Option<(ConnInfo, Arc<GlobalConnPool>)>,
+    ) -> Self {
+        Self {
+            conn_id: inner.conn_id,
+            pid: inner.get_pid().await.unwrap_or(-1),
+            inner: Some(inner),
+            span: Span::current(),
+            pool,
+        }
+    }
     pub fn inner(&mut self) -> (&mut tokio_postgres::Client, Discard<'_>) {
         let Self {
             inner,
             pool,
             conn_id,
             span: _,
+            pid: _,
         } = self;
         (
             &mut inner
@@ -609,10 +630,11 @@ impl Drop for Client {
             .expect("client inner should not be removed");
         if let Some((conn_info, conn_pool)) = self.pool.take() {
             let current_span = self.span.clone();
+            let pid = self.pid;
             // return connection to the pool
             tokio::task::spawn_blocking(move || {
                 let _span = current_span.enter();
-                let _ = conn_pool.put(&conn_info, client);
+                let _ = conn_pool.put(&conn_info, client, pid);
             });
         }
     }

vm-image-spec.yaml

@@ -0,0 +1,143 @@
+# Supplemental file for neondatabase/autoscaling's vm-builder, for producing the VM compute image.
+---
+commands:
+  - name: cgconfigparser
+    user: root
+    sysvInitAction: sysinit
+    shell: "cgconfigparser -l /etc/cgconfig.conf -s 1664"
+  - name: pgbouncer
+    user: nobody
+    sysvInitAction: respawn
+    shell: "/usr/local/bin/pgbouncer /etc/pgbouncer.ini"
+  - name: postgres-exporter
+    user: nobody
+    sysvInitAction: respawn
+    shell: 'DATA_SOURCE_NAME="user=cloud_admin sslmode=disable dbname=postgres" /bin/postgres_exporter --extend.query-path /etc/postgres_exporter_queries.yml'
+shutdownHook: |
+  su -p postgres --session-command '/usr/local/bin/pg_ctl stop -D /var/db/postgres/compute/pgdata -m fast --wait -t 10'
+files:
+  - filename: pgbouncer.ini
+    content: |
+      [databases]
+      *=host=localhost port=5432 auth_user=cloud_admin
+      [pgbouncer]
+      listen_port=6432
+      listen_addr=0.0.0.0
+      auth_type=scram-sha-256
+      auth_user=cloud_admin
+      auth_dbname=postgres
+      client_tls_sslmode=disable
+      server_tls_sslmode=disable
+      pool_mode=transaction
+      max_client_conn=10000
+      default_pool_size=16
+      max_prepared_statements=0
+  - filename: cgconfig.conf
+    content: |
+      # Configuration for cgroups in VM compute nodes
+      group neon-postgres {
+          perm {
+              admin {
+                  uid = postgres;
+              }
+              task {
+                  gid = users;
+              }
+          }
+          memory {}
+      }
+  - filename: postgres_exporter_queries.yml
+    content: |
+      postgres_exporter_pg_database_size:
+        query: "SELECT pg_database.datname, pg_database_size(pg_database.datname) as bytes, 42 as fourtytwo FROM pg_database"
+        cache_seconds: 30
+        metrics:
+          - datname:
+              usage: "LABEL"
+              description: "Name of the database"
+          - bytes:
+              usage: "GAUGE"
+              description: "Disk space used by the database"
+          - fourtytwo:
+              usage: "GAUGE"
+              description: "fourtytwo"
+build: |
+  # Build cgroup-tools
+  #
+  # At time of writing (2023-03-14), debian bullseye has a version of cgroup-tools (technically
+  # libcgroup) that doesn't support cgroup v2 (version 0.41-11). Unfortunately, the vm-monitor
+  # requires cgroup v2, so we'll build cgroup-tools ourselves.
+  FROM debian:bullseye-slim as libcgroup-builder
+  ENV LIBCGROUP_VERSION v2.0.3
+
+  RUN set -exu \
+      && apt update \
+      && apt install --no-install-recommends -y \
+          git \
+          ca-certificates \
+          automake \
+          cmake \
+          make \
+          gcc \
+          byacc \
+          flex \
+          libtool \
+          libpam0g-dev \
+      && git clone --depth 1 -b $LIBCGROUP_VERSION https://github.com/libcgroup/libcgroup \
+      && INSTALL_DIR="/libcgroup-install" \
+      && mkdir -p "$INSTALL_DIR/bin" "$INSTALL_DIR/include" \
+      && cd libcgroup \
+      # extracted from bootstrap.sh, with modified flags:
+      && (test -d m4 || mkdir m4) \
+      && autoreconf -fi \
+      && rm -rf autom4te.cache \
+      && CFLAGS="-O3" ./configure --prefix="$INSTALL_DIR" --sysconfdir=/etc --localstatedir=/var --enable-opaque-hierarchy="name=systemd" \
+      # actually build the thing...
+      && make install
+
+  FROM quay.io/prometheuscommunity/postgres-exporter:v0.12.0 AS postgres-exporter
+
+  # Build pgbouncer
+  #
+  FROM debian:bullseye-slim AS pgbouncer
+  RUN set -e \
+      && apt-get update \
+      && apt-get install -y \
+          curl \
+          build-essential \
+          pkg-config \
+          libevent-dev \
+          libssl-dev
+
+  ENV PGBOUNCER_VERSION 1.21.0
+  ENV PGBOUNCER_GITPATH 1_21_0
+  RUN set -e \
+      && curl -sfSL https://github.com/pgbouncer/pgbouncer/releases/download/pgbouncer_${PGBOUNCER_GITPATH}/pgbouncer-${PGBOUNCER_VERSION}.tar.gz -o pgbouncer-${PGBOUNCER_VERSION}.tar.gz \
+      && tar xzvf pgbouncer-${PGBOUNCER_VERSION}.tar.gz \
+      && cd pgbouncer-${PGBOUNCER_VERSION} \
+      && LDFLAGS=-static ./configure --prefix=/usr/local/pgbouncer --without-openssl \
+      && make -j $(nproc) \
+      && make install
+merge: |
+  # tweak nofile limits
+  RUN set -e \
+      && echo 'fs.file-max = 1048576' >>/etc/sysctl.conf \
+      && test ! -e /etc/security || ( \
+         echo '*    - nofile 1048576' >>/etc/security/limits.conf \
+      && echo 'root - nofile 1048576' >>/etc/security/limits.conf \
+         )
+
+  COPY cgconfig.conf /etc/cgconfig.conf
+  COPY pgbouncer.ini /etc/pgbouncer.ini
+  COPY postgres_exporter_queries.yml /etc/postgres_exporter_queries.yml
+  RUN set -e \
+      && chown postgres:postgres /etc/pgbouncer.ini \
+      && chmod 0644 /etc/pgbouncer.ini \
+      && chmod 0644 /etc/cgconfig.conf \
+      && chmod 0644 /etc/postgres_exporter_queries.yml
+
+  COPY --from=libcgroup-builder /libcgroup-install/bin/*  /usr/bin/
+  COPY --from=libcgroup-builder /libcgroup-install/lib/*  /usr/lib/
+  COPY --from=libcgroup-builder /libcgroup-install/sbin/* /usr/sbin/
+  COPY --from=postgres-exporter /bin/postgres_exporter /bin/postgres_exporter
+  COPY --from=pgbouncer         /usr/local/pgbouncer/bin/pgbouncer /usr/local/bin/pgbouncer