Files
neon/compute/vm-image-spec-bookworm.yaml
Heikki Linnakangas 0abff59e97 compute: Allow postgres user to power off the VM (#10710)
I plan to use this when launching a fast_import job in a VM. There's
currently no good way for an executable running in a NeonVM to exit
gracefully and have the VM shut down. The inittab we use always respawns
the payload command. The idea is that the control plane can use
"fast_import ... && poweroff" as the command, so that when fast_import
completes successfully, the VM is terminated, and the k8s Pod and
VirtualMachine object are marked as completed successfully.

I'm working on bigger changes to how we launch VMs, and will try to come
up with a nicer system for that, but in the meanwhile, this quick hack
allows us to proceed with using VMs for one-off jobs like fast_import.
2025-02-07 16:03:01 +00:00

133 lines
5.1 KiB
YAML

# 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'
# restrict permissions on /neonvm/bin/resize-swap, because we grant access to compute_ctl for
# running it as root.
- name: chmod-resize-swap
user: root
sysvInitAction: sysinit
shell: 'chmod 711 /neonvm/bin/resize-swap'
- name: chmod-set-disk-quota
user: root
sysvInitAction: sysinit
shell: 'chmod 711 /neonvm/bin/set-disk-quota'
- name: pgbouncer
user: postgres
sysvInitAction: respawn
shell: '/usr/local/bin/pgbouncer /etc/pgbouncer.ini 2>&1 > /dev/virtio-ports/tech.neon.log.0'
- name: local_proxy
user: postgres
sysvInitAction: respawn
shell: '/usr/local/bin/local_proxy --config-path /etc/local_proxy/config.json --pid-path /etc/local_proxy/pid --http 0.0.0.0:10432'
- name: postgres-exporter
user: nobody
sysvInitAction: respawn
shell: 'DATA_SOURCE_NAME="user=cloud_admin sslmode=disable dbname=postgres application_name=postgres-exporter" /bin/postgres_exporter --config.file=/etc/postgres_exporter.yml'
- name: pgbouncer-exporter
user: postgres
sysvInitAction: respawn
shell: '/bin/pgbouncer_exporter --pgBouncer.connectionString="postgres:///pgbouncer?host=/tmp&port=6432&dbname=pgbouncer&user=pgbouncer"'
- name: sql-exporter
user: nobody
sysvInitAction: respawn
shell: '/bin/sql_exporter -config.file=/etc/sql_exporter.yml -web.listen-address=:9399'
- name: sql-exporter-autoscaling
user: nobody
sysvInitAction: respawn
shell: '/bin/sql_exporter -config.file=/etc/sql_exporter_autoscaling.yml -web.listen-address=:9499'
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: compute_ctl-sudoers
content: |
# Allow postgres user (which is what compute_ctl runs as) to run /neonvm/bin/resize-swap
# and /neonvm/bin/set-disk-quota as root without requiring entering a password (NOPASSWD),
# regardless of hostname (ALL)
#
# Also allow it to shut down the VM. The fast_import job does that when it's finished.
postgres ALL=(root) NOPASSWD: /neonvm/bin/resize-swap, /neonvm/bin/set-disk-quota, /neonvm/bin/poweroff
- filename: cgconfig.conf
content: |
# Configuration for cgroups in VM compute nodes
group neon-postgres {
perm {
admin {
uid = postgres;
}
task {
gid = users;
}
}
memory {}
}
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.
#
# At time of migration to bookworm (2024-10-09), debian has a version of libcgroup/cgroup-tools 2.0.2,
# and it _probably_ can be used as-is. However, we'll build it ourselves to minimise the changeset
# for debian version migration.
#
FROM debian:bookworm-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
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 \
)
# Allow postgres user (compute_ctl) to run swap resizer.
# Need to install sudo in order to allow this.
#
# Also, remove the 'read' permission from group/other on /neonvm/bin/resize-swap, just to be safe.
RUN set -e \
&& apt update \
&& apt install --no-install-recommends -y \
sudo \
&& rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
COPY compute_ctl-sudoers /etc/sudoers.d/compute_ctl-sudoers
COPY cgconfig.conf /etc/cgconfig.conf
RUN set -e \
&& chmod 0644 /etc/cgconfig.conf
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/