Create script to auto-deploy with retries

deploy/README.md

@@ -4,14 +4,26 @@ Note that you have to be connected to the `vpn.qo.is`
 (or execute the deployment from a host that is in the `backplane.net.qo.is` overlay network)
 and that you need to have SSH root access to the target machines.
 
+## Deploy system categories
+
+We currently split out nixosConfigurations into these categories:
+
+- `system-ci`: Systems should be updated separately because they might break automated deployment processes.
+- `system-vm`: Virtual systems.
+- `system-physical`: Physical systems.
+
+You can roll updates with retries by category with:
+
+```bash
+auto-deploy system-vm
+auto-deploy system-physical
+```
+
 ## Deploy to selected target hosts
 
 ```bash
-nix run .#deploy-qois .#<hostname>.system .#<hostname2>.system
-```
+nix develop
 
-## Deploy with extended timeouts (sometimes required for slow APU devices)
-
-```bash
-nix run .#deploy-qois  .#calanda.system -- --confirm-timeout 600 --activation-timeout 600
+deploy --skip-checks .#cyprianspitz.system-physical
+deploy --skip-checks .#lindberg-build.system-vm
 ```

deploy/docs-ops/default.nix

@@ -12,5 +12,6 @@ in
     sshUser = "nginx-${domain}";
     path = deployPkgs.deploy-rs.lib.activate.noop self.packages.${system}.docs;
     profilePath = "/var/lib/nginx-${domain}/root";
+    remoteBuild = true;
   };
 }

deploy/system-ci/default.nix

@@ -0,0 +1,27 @@
+{
+  deployPkgs,
+  pkgs,
+  self,
+  ...
+}:
+let
+  inherit (pkgs.lib) pipe filterAttrs mapAttrs;
+in
+{
+  nodes = pipe self.nixosConfigurations [
+    (filterAttrs (_n: v: v.config.qois.git-ci-runner.enable))
+    (mapAttrs (
+      host: config: {
+        hostname = "${host}.backplane.net.qo.is";
+        profiles.system-ci = {
+          sshUser = "root";
+          user = "root";
+          activationTimeout = 300;
+          confirmTimeout = 60;
+          remoteBuild = true;
+          path = deployPkgs.deploy-rs.lib.activate.nixos config;
+        };
+      }
+    ))
+  ];
+}

deploy/system-physical/default.nix

@@ -0,0 +1,27 @@
+{
+  deployPkgs,
+  pkgs,
+  self,
+  ...
+}:
+let
+  inherit (pkgs.lib) pipe filterAttrs mapAttrs;
+in
+{
+  nodes = pipe self.nixosConfigurations [
+    (filterAttrs (_n: v: !v.config.services.qemuGuest.enable && !v.config.qois.git-ci-runner.enable))
+    (mapAttrs (
+      host: config: {
+        hostname = "${host}.backplane.net.qo.is";
+        profiles.system-physical = {
+          sshUser = "root";
+          user = "root";
+          activationTimeout = 600;
+          confirmTimeout = 120;
+          remoteBuild = true;
+          path = deployPkgs.deploy-rs.lib.activate.nixos config;
+        };
+      }
+    ))
+  ];
+}

deploy/system-vm/default.nix

@@ -0,0 +1,27 @@
+{
+  deployPkgs,
+  pkgs,
+  self,
+  ...
+}:
+let
+  inherit (pkgs.lib) pipe filterAttrs mapAttrs;
+in
+{
+  nodes = pipe self.nixosConfigurations [
+    (filterAttrs (_n: v: v.config.services.qemuGuest.enable && !v.config.qois.git-ci-runner.enable))
+    (mapAttrs (
+      host: config: {
+        hostname = "${host}.backplane.net.qo.is";
+        profiles.system-vm = {
+          sshUser = "root";
+          user = "root";
+          activationTimeout = 300;
+          confirmTimeout = 60;
+          remoteBuild = true;
+          path = deployPkgs.deploy-rs.lib.activate.nixos config;
+        };
+      }
+    ))
+  ];
+}

deploy/system/default.nix

@@ -1,20 +0,0 @@
-{
-  deployPkgs,
-  pkgs,
-  self,
-  system,
-  ...
-}:
-{
-  nodes = pkgs.lib.mapAttrs (host: config: {
-    hostname = "${host}.backplane.net.qo.is";
-    profiles.system = {
-      sshUser = "root";
-      user = "root";
-      activationTimeout = 420;
-      confirmTimeout = 120;
-
-      path = deployPkgs.deploy-rs.lib.activate.nixos config;
-    };
-  }) self.nixosConfigurations;
-}

dev-shells/default.nix

@@ -29,9 +29,9 @@ in
       pre-commit-check.enabledPackages
       ++ [ vscodium-with-extensions ]
       ++ (with self.packages.${system}; [
-        deploy-qois
         sops
         sops-rekey
+        auto-deploy
       ])
       ++ (with pkgs; [
         attic-client

packages/auto-deploy/default.nix

@@ -0,0 +1,16 @@
+{
+  deploy-rs,
+  gitMinimal,
+  writeShellApplication,
+  lib,
+  ...
+}:
+writeShellApplication {
+  name = "auto-deploy";
+  meta.description = "Deploy machines automatically.";
+  runtimeInputs = [
+    deploy-rs
+    gitMinimal
+  ];
+  text = lib.readFile ./script.bash;
+}

packages/auto-deploy/script.bash

@@ -0,0 +1,49 @@
+#!/usr/bin/env bash
+
+#### Environment
+FLAKE_ROOT="$(git rev-parse --show-toplevel)"
+
+export PROFILE="${1:-''}"
+if [ -z "${PROFILE}" ]; then
+  echo "🛑 Error: No deployment profile was specified as first parameter (e.g. \"${0} system-vm\")" 1>&2
+  exit 1
+fi
+
+HOSTS=$(nix eval --raw "${FLAKE_ROOT}"#deploy.nodes --apply "
+   nodes: let
+     inherit (builtins) attrNames filter concatStringsSep;
+     names = attrNames nodes;
+     profile = \"${PROFILE}\";
+     filteredNames = filter (name: nodes.\${name}.profiles ? \${profile}) names;
+   in concatStringsSep \"\\n\" filteredNames
+")
+if [ -z "$HOSTS" ]; then
+  echo "🛑 Error: No deployments matching the profile ${PROFILE} were found." 1>&2
+  exit 1
+fi
+
+KNOWN_HOSTS_FILE=$(nix eval --raw .#nixosConfigurations.lindberg.config.environment.etc."ssh/ssh_known_hosts".source)
+
+#### Helpers
+retry() {
+  local -r -i max_attempts="$1"
+  shift
+  local -i attempt_num=1
+  until "$@"; do
+    if ((attempt_num == max_attempts)); then
+      echo "⚠️ Warning: Attempt $attempt_num failed and there are no more attempts left!"
+      return 1
+    else
+      echo "⚠️ Attempt $attempt_num failed! Trying again in $attempt_num seconds..."
+      sleep $((attempt_num++))
+    fi
+  done
+}
+
+#### Execution
+for HOST in $HOSTS; do
+  retry 3 deploy \
+    --skip-checks \
+    --ssh-opts "-o UserKnownHostsFile=${KNOWN_HOSTS_FILE}" \
+    --targets "${FLAKE_ROOT}#\"${HOST}\".\"${PROFILE}\""
+done

packages/deploy-qois/default.nix

@@ -1,14 +0,0 @@
-{
-  deploy-rs,
-  flakeSelf,
-  writeShellApplication,
-  ...
-}:
-writeShellApplication {
-  name = "deploy-qois";
-  meta.description = "Deploy configuration to specificed targets.";
-  runtimeInputs = [ deploy-rs ];
-  text = ''
-    deploy --remote-build --skip-checks --interactive --targets "''${@:-${flakeSelf}}"
-  '';
-}

treefmt.nix

@@ -19,11 +19,13 @@
         "*.toml"
       ]
       ++ [
-        ".envrc"
-        "robots.txt"
         ".vscode/*"
         "nixos-modules/system/etc/*"
+        "private"
         "private/*"
+
+        ".envrc"
+        "robots.txt"
       ];
     formatter.jsonfmt.excludes = [ ".vscode/*.json" ];
   };

updates.md

@@ -22,13 +22,14 @@ Deploy updates:
 nix develop
 
 # Deploy vms
-deploy-qois .#lindberg-nextcloud .#lindberg-build
+auto-deploy system-vm
 
-# Deploy fast physical hosts
-deploy-qois .#lindberg
+# Deploy CI hosts
+auto-deploy system-ci
+
+# Deploy physical hosts
+auto-deploy system-physical
 
-# Deploy slow physical hosts (maybe do individually)
-deploy-qois --confirm-timeout 600 --activation-timeout 600 --targets .#stompert .#stompert
 
 ```