Run compute_ctl in a cgroup in VMs (#3577)

.dockerignore

@@ -21,3 +21,4 @@
 !workspace_hack/
 !neon_local/
 !scripts/ninstall.sh
+!vm-cgconfig.conf

.github/workflows/build_and_test.yml

@@ -611,34 +611,31 @@ jobs:
       run:
         shell: sh -eu {0}
     env:
-      VM_INFORMANT_VERSION: 0.1.1
+      VM_BUILDER_VERSION: v0.4.6
 
     steps:
-      - name: Downloading latest vm-builder
+      - name: Checkout
+        uses: actions/checkout@v1
+        with:
+          fetch-depth: 0
+
+      - name: Downloading vm-builder
         run: |
-          curl -L https://github.com/neondatabase/neonvm/releases/latest/download/vm-builder -o vm-builder
+          curl -L https://github.com/neondatabase/neonvm/releases/download/$VM_BUILDER_VERSION/vm-builder -o vm-builder
           chmod +x vm-builder
 
       - name: Pulling compute-node image
         run: |
           docker pull 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
 
-      - name: Downloading VM informant version ${{ env.VM_INFORMANT_VERSION }}
+      - name: Building VM compute-node rootfs
         run: |
-          curl -fL https://github.com/neondatabase/autoscaling/releases/download/${{ env.VM_INFORMANT_VERSION }}/vm-informant -o vm-informant
-          chmod +x vm-informant
-
-      - name: Adding VM informant to compute-node image
-        run: |
-          ID=$(docker create 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}})
-          docker cp vm-informant $ID:/bin/vm-informant
-          docker commit $ID temp-vm-compute-node
-          docker rm -f $ID
+          docker build -t temp-vm-compute-node --build-arg SRC_IMAGE=369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}} -f Dockerfile.vm-compute-node .
 
       - name: Build vm image
         run: |
           # note: as of 2023-01-12, vm-builder requires a trailing ":latest" for local images
-          ./vm-builder -src=temp-vm-compute-node:latest -dst=369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
+          ./vm-builder -use-inittab -src=temp-vm-compute-node:latest -dst=369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
 
       - name: Pushing vm-compute-node image
         run: |

Dockerfile.vm-compute-node

@@ -0,0 +1,32 @@
+# Note: this file *mostly* just builds on Dockerfile.compute-node
+
+ARG SRC_IMAGE
+ARG VM_INFORMANT_VERSION=v0.1.6
+
+# Pull VM informant and set up inittab
+FROM neondatabase/vm-informant:$VM_INFORMANT_VERSION as informant
+
+RUN set -e \
+	&& rm -f /etc/inittab \
+	&& touch /etc/inittab
+
+ADD vm-cgconfig.conf /etc/cgconfig.conf
+RUN set -e \
+	&& echo "::sysinit:cgconfigparser -l /etc/cgconfig.conf -s 1664" >> /etc/inittab \
+	&& echo "::respawn:su vm-informant -c '/usr/local/bin/vm-informant --auto-restart --cgroup=neon-postgres'" >> /etc/inittab
+
+# Combine, starting from non-VM compute node image.
+FROM $SRC_IMAGE as base
+
+# Temporarily set user back to root so we can run apt update and adduser
+USER root
+RUN apt update && \
+	apt install --no-install-recommends -y \
+        cgroup-tools
+RUN adduser vm-informant --disabled-password --no-create-home
+USER postgres
+
+COPY --from=informant /etc/inittab /etc/inittab
+COPY --from=informant /usr/bin/vm-informant /usr/local/bin/vm-informant
+
+ENTRYPOINT ["/usr/sbin/cgexec", "-g", "*:neon-postgres", "/usr/local/bin/compute_ctl"]

compute_tools/src/bin/compute_ctl.rs

@@ -44,7 +44,6 @@ use tracing::{error, info};
 
 use compute_tools::compute::{ComputeMetrics, ComputeNode, ComputeState, ComputeStatus};
 use compute_tools::http::api::launch_http_server;
-use compute_tools::informant::spawn_vm_informant_if_present;
 use compute_tools::logger::*;
 use compute_tools::monitor::launch_monitor;
 use compute_tools::params::*;
@@ -141,8 +140,6 @@ fn main() -> Result<()> {
     // requests, while configuration is still in progress.
     let _http_handle = launch_http_server(&compute).expect("cannot launch http endpoint thread");
     let _monitor_handle = launch_monitor(&compute).expect("cannot launch compute monitor thread");
-    // Also spawn the thread responsible for handling the VM informant -- if it's present
-    let _vm_informant_handle = spawn_vm_informant_if_present().expect("cannot launch VM informant");
 
     // Start Postgres
     let mut delay_exit = false;

compute_tools/src/informant.rs

@@ -1,50 +0,0 @@
-use std::path::Path;
-use std::process;
-use std::thread;
-use std::time::Duration;
-use tracing::{info, warn};
-
-use anyhow::{Context, Result};
-
-const VM_INFORMANT_PATH: &str = "/bin/vm-informant";
-const RESTART_INFORMANT_AFTER_MILLIS: u64 = 5000;
-
-/// Launch a thread to start the VM informant if it's present (and restart, on failure)
-pub fn spawn_vm_informant_if_present() -> Result<Option<thread::JoinHandle<()>>> {
-    let exists = Path::new(VM_INFORMANT_PATH)
-        .try_exists()
-        .context("could not check if path exists")?;
-
-    if !exists {
-        return Ok(None);
-    }
-
-    Ok(Some(
-        thread::Builder::new()
-            .name("run-vm-informant".into())
-            .spawn(move || run_informant())?,
-    ))
-}
-
-fn run_informant() -> ! {
-    let restart_wait = Duration::from_millis(RESTART_INFORMANT_AFTER_MILLIS);
-
-    info!("starting VM informant");
-
-    loop {
-        let mut cmd = process::Command::new(VM_INFORMANT_PATH);
-        // Block on subprocess:
-        let result = cmd.status();
-
-        match result {
-            Err(e) => warn!("failed to run VM informant at {VM_INFORMANT_PATH:?}: {e}"),
-            Ok(status) if !status.success() => {
-                warn!("{VM_INFORMANT_PATH} exited with code {status:?}, retrying")
-            }
-            Ok(_) => info!("{VM_INFORMANT_PATH} ended gracefully (unexpectedly). Retrying"),
-        }
-
-        // Wait before retrying
-        thread::sleep(restart_wait);
-    }
-}

compute_tools/src/lib.rs

@@ -8,7 +8,6 @@ pub mod http;
 #[macro_use]
 pub mod logger;
 pub mod compute;
-pub mod informant;
 pub mod monitor;
 pub mod params;
 pub mod pg_helpers;

vm-cgconfig.conf

@@ -0,0 +1,12 @@
+# Configuration for cgroups in VM compute nodes
+group neon-postgres {
+    perm {
+        admin {
+            uid = vm-informant;
+        }
+        task {
+            gid = users;
+        }
+    }
+    memory {}
+}