Add VM informant to vm-compute-node (#3324)

The general idea is that the VM informant binary is added to the
vm-compute-node images only. `compute_tools` then will run whatever's at
`/bin/vm-informant`, if the path exists.

.github/workflows/build_and_test.yml

@@ -595,6 +595,8 @@ jobs:
     defaults:
       run:
         shell: sh -eu {0}
+    env:
+      VM_INFORMANT_VERSION: 0.1.1
 
     steps:
       - name: Downloading latest vm-builder
@@ -606,9 +608,22 @@ jobs:
         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 ${{ env.VM_INFORMANT_VERSION }}
+        run: |
+          curl -L https://github.com/neondatabase/autoscaling/releases/download/${{ 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 $ID
+
       - name: Build vm image
         run: |
-          ./vm-builder -src=369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}} -dst=369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
+          # note: as of 2023-01-12, vm-builder requires a trailing ":latest" for local images
+          ./vm-builder -src=temp-vm-compute-node:latest -dst=369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
 
       - name: Pushing vm-compute-node image
         run: |

compute_tools/README.md

@@ -19,6 +19,10 @@ Also `compute_ctl` spawns two separate service threads:
 - `http-endpoint` runs a Hyper HTTP API server, which serves readiness and the
   last activity requests.
 
+If the `vm-informant` binary is present at `/bin/vm-informant`, it will also be started. For VM
+compute nodes, `vm-informant` communicates with the VM autoscaling system. It coordinates
+downscaling and (eventually) will request immediate upscaling under resource pressure.
+
 Usage example:
 ```sh
 compute_ctl -D /var/db/postgres/compute \

compute_tools/src/bin/compute_ctl.rs

@@ -18,6 +18,10 @@
 //! - `http-endpoint` runs a Hyper HTTP API server, which serves readiness and the
 //!   last activity requests.
 //!
+//! If the `vm-informant` binary is present at `/bin/vm-informant`, it will also be started. For VM
+//! compute nodes, `vm-informant` communicates with the VM autoscaling system. It coordinates
+//! downscaling and (eventually) will request immediate upscaling under resource pressure.
+//!
 //! Usage example:
 //! ```sh
 //! compute_ctl -D /var/db/postgres/compute \
@@ -40,6 +44,7 @@ use log::{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::*;
@@ -114,6 +119,8 @@ 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");
 
     // Run compute (Postgres) and hang waiting on it.
     match compute.prepare_and_run() {

compute_tools/src/informant.rs

@@ -0,0 +1,50 @@
+use log::{info, warn};
+use std::path::Path;
+use std::process;
+use std::thread;
+use std::time::Duration;
+
+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

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