Merge pull request #424 from ComputeCanada/clean-cloudinit

Overall cleanup of cloudinit

Author: cmd-ntrf

aws/infrastructure.tf

@@ -33,7 +33,6 @@ module "configuration" {
   cloud_provider  = local.cloud_provider
   cloud_region    = local.cloud_region
   skip_upgrade    = var.skip_upgrade
-  puppetfile      = var.puppetfile
 }
 
 module "provision" {

azure/infrastructure.tf

@@ -33,7 +33,6 @@ module "configuration" {
   cloud_provider  = local.cloud_provider
   cloud_region    = local.cloud_region
   skip_upgrade    = var.skip_upgrade
-  puppetfile      = var.puppetfile
 }
 
 module "provision" {

common/configuration/main.tf

@@ -23,7 +23,6 @@ variable "guest_passwd" {}
 variable "public_keys" {}
 
 variable "skip_upgrade" {}
-variable "puppetfile" {}
 variable "bastion_tags" {}
 
 resource "tls_private_key" "ssh" {
@@ -84,6 +83,7 @@ locals {
         domain_name     = var.domain_name
         guest_passwd    = local.guest_passwd
         nb_users        = var.nb_users
+        bastion_tags    = var.bastion_tags
       }
     }
   })
@@ -112,7 +112,8 @@ locals {
         tf_ssh_public_key     = chomp(tls_private_key.ssh.public_key_openssh)
         terraform_facts       = local.terraform_facts
         skip_upgrade          = var.skip_upgrade
-        puppetfile            = var.puppetfile
+        module_path           = path.module
+        user_tf_required      = contains(values.tags, "puppet") || length(setintersection(values.tags, var.bastion_tags)) > 0
         hostkeys = {
           rsa = {
             private = chomp(tls_private_key.rsa[values.prefix].private_key_openssh)

common/configuration/puppet.yaml.tftpl

@@ -6,13 +6,13 @@ mounts:
  - [ ephemeral0, /mnt/ephemeral0, auto, "defaults,nofail" ]
 
 users:
-%{ if contains(tags, "puppet") || length(setintersection(tags, bastion_tags)) > 0 ~}
+%{ if user_tf_required ~}
   - name: tf
     system: true
     passwd: '*'
     no_user_group: true
     homedir: /tmp
-%{ if contains(tags, "puppet") }
+%{ if contains(tags, "puppet") ~}
     sudo: "ALL = NOPASSWD: /usr/sbin/update_etc_puppetlabs.sh *.zip"
 %{ endif ~}
 %{ endif ~}
@@ -21,10 +21,10 @@ users:
     passwd: '*'
     homedir: /${sudoer_username}
     sudo: ALL=(ALL) NOPASSWD:ALL
-%{ if cloud_provider != "incus" }
+%{ if cloud_provider != "incus" ~}
     # most container images do not have selinux pre-installed
     selinux_user: unconfined_u
-%{ endif }
+%{ endif ~}
     ssh_authorized_keys:
 %{ for key in ssh_authorized_keys ~}
       - ${key}
@@ -35,18 +35,11 @@ runcmd:
   - trap 'flag_cloud_init_failed' ERR
   - chmod 755 /etc # avoid issue with Rocky 9.4
   - test ! -d /${sudoer_username} && userdel -f -r ${sudoer_username} && cloud-init clean -r
-%{ if cloud_provider != "incus" }
-    # most container images do not have selinux pre-installed
+  - blkid -L ephemeral0 &>/dev/null || sed -i -E '\;^[[:space:]]*ephemeral0[[:space:]]+/mnt/ephemeral0[[:space:]];s;^;#;' /etc/fstab
+%{ if cloud_provider != "incus" ~}
+  # most container images do not have selinux pre-installed
   - restorecon -R /${sudoer_username}
-%{ endif }
-# Make sure puppet server can be reached by name early in the process if we need to debug.
-# For some provider, the ip address of the puppetserver is not known in advance, so we make
-# sure it is not empty before adding it to /etc/hosts
-%{ for host, ip in puppetservers ~}
-%{ if ip != "" ~}
-  - echo "${ip} ${host}" >> /etc/hosts
 %{ endif ~}
-%{ endfor ~}
   # Install package and configure kernel only if building from a "vanilla" linux image
   - |
     if ! test -f /etc/magic-castle-release; then
@@ -84,9 +77,16 @@ runcmd:
       systemctl disable kdump
       grubby --update-kernel=ALL --args="rd.driver.blacklist=nouveau nouveau.modeset=0 crashkernel=0M initcall_blacklist=algif_aead_init"
       grub2-mkconfig -o /boot/grub2/grub.cfg
+%{ endif ~}
+%{ if cloud_provider == "gcp" ~}
+      # Google Cloud user-data fact generates a warning because its size is greater than what is allowed (<4096 bytes).
+      # We have no use for it, so we remove startup-script, user-data and user-data-encoding when running in GCE.
+      sed -i "/gce_data\['instance'\] = instance_data/i \ \ \ \ \ \ \ \ \ \ instance_data['attributes'].delete('startup-script')" /opt/puppetlabs/puppet/lib/ruby/vendor_ruby/facter/resolvers/gce.rb
+      sed -i "/gce_data\['instance'\] = instance_data/i \ \ \ \ \ \ \ \ \ \ instance_data['attributes'].delete('user-data')" /opt/puppetlabs/puppet/lib/ruby/vendor_ruby/facter/resolvers/gce.rb
+      sed -i "/gce_data\['instance'\] = instance_data/i \ \ \ \ \ \ \ \ \ \ instance_data['attributes'].delete('user-data-encoding')" /opt/puppetlabs/puppet/lib/ruby/vendor_ruby/facter/resolvers/gce.rb
 %{ endif ~}
     fi
-%{ if contains(tags, "puppet") }
+%{ if contains(tags, "puppet") ~}
 # Install puppetserver
   - dnf -y install openvox-server-8.11.0
 # Configure puppet-agent to start after puppetserver when on puppetserver
@@ -127,20 +127,13 @@ runcmd:
   - ln -sf /etc/puppetlabs/facts/terraform_facts.yaml /etc/puppetlabs/code/environments/production/site/profile/facts.d
 # We use r10k solely to install the modules of the production puppet environment.
   - /opt/puppetlabs/puppet/bin/r10k puppetfile install --moduledir=/etc/puppetlabs/code/environments/production/modules --puppetfile=/etc/puppetlabs/code/environments/production/Puppetfile
-%{ if puppetfile != "" ~}
-  - /opt/puppetlabs/puppet/bin/r10k puppetfile install --moduledir=/etc/puppetlabs/code/modules --puppetfile=/etc/puppetlabs/code/Puppetfile
-%{ endif ~}
-# Backport stdlib patch to support Puppet 8 json parsing until we update to stdlib >= 8.5.0
-# Patch origin: https://github.com/puppetlabs/puppetlabs-stdlib/commit/53a8ccf869aa49de7a77c8a35d89dd5aac8fa52d.patch
-  - sed -i -e 's/PSON/Puppet::Util::Json/' -e "/module Puppet/i require 'puppet/util/json'" /etc/puppetlabs/code/environments/production/modules/stdlib/lib/puppet/parser/functions/parsejson.rb
-# Wait for Terraform to scp its YAML data
-  - while [ ! -e "/etc/puppetlabs/data/terraform_data.yaml" ]; do echo "$(date -I'seconds') Waiting for terraform to scp terraform_data.yaml"; sleep 5; done
-%{ if node_name != keys(puppetservers)[0] }
+  - test -f /etc/puppetlabs/code/Puppetfile && /opt/puppetlabs/puppet/bin/r10k puppetfile install --moduledir=/etc/puppetlabs/code/modules --puppetfile=/etc/puppetlabs/code/Puppetfile
+%{ if node_name != keys(puppetservers)[0] ~}
   - sed -e '/certificate-authority-service/ s/^/#/' -i /etc/puppetlabs/puppetserver/services.d/ca.cfg
   - sed -e '/certificate-authority-disabled-service/ s/^#//' -i /etc/puppetlabs/puppetserver/services.d/ca.cfg
-%{ endif }
+%{ endif ~}
   - chgrp puppet /etc/puppetlabs/puppet/csr_attributes.yaml
-%{ endif }
+%{ endif ~}
 %{ if length(puppetservers) > 0 ~}
   - /opt/puppetlabs/bin/puppet config set server ${keys(puppetservers)[0]}
 %{ endif ~}
@@ -151,41 +144,44 @@ runcmd:
   - systemctl enable puppet
 # Remove all ifcfg configuration files that have no corresponding network interface in ip link show.
   - for i in /etc/sysconfig/network-scripts/ifcfg-*; do if ! ip link show | grep -q "$${i##*-}:"; then rm -f $i; fi; done
-%{ if cloud_provider == "gcp" }
-# Google Cloud user-data fact generates a warning because its size is greater than what is allowed (<4096 bytes).
-# We have no use for it, so we remove startup-script, user-data and user-data-encoding when running in GCE.
-  - sed -i "/gce_data\['instance'\] = instance_data/i \ \ \ \ \ \ \ \ \ \ instance_data['attributes'].delete('startup-script')" /opt/puppetlabs/puppet/lib/ruby/vendor_ruby/facter/resolvers/gce.rb
-  - sed -i "/gce_data\['instance'\] = instance_data/i \ \ \ \ \ \ \ \ \ \ instance_data['attributes'].delete('user-data')" /opt/puppetlabs/puppet/lib/ruby/vendor_ruby/facter/resolvers/gce.rb
-  - sed -i "/gce_data\['instance'\] = instance_data/i \ \ \ \ \ \ \ \ \ \ instance_data['attributes'].delete('user-data-encoding')" /opt/puppetlabs/puppet/lib/ruby/vendor_ruby/facter/resolvers/gce.rb
-%{ endif }
-%{ if contains(tags, "puppet") }
+%{ if contains(tags, "puppet") ~}
+# Wait for Terraform to scp its YAML data
+  - while [ ! -e "/etc/puppetlabs/data/terraform_data.yaml" ]; do echo "$(date -I'seconds') Waiting for terraform to scp terraform_data.yaml"; sleep 5; done
 # Bootstrap services that are essential to the puppet repo
   - (cd /etc/puppetlabs/code/environments/production; test -e bootstrap.sh && ./bootstrap.sh)
-%{ endif }
+%{ endif ~}
   # If the current image has already been configured with Magic Castle Puppet environment,
   # we can start puppet and skip reboot, reducing the delay for bringing the node up.
   - test -f /etc/magic-castle-release && systemctl start puppet || true
   - test -f /run/cloud-init-failed && echo 'WARNING - some steps cloud-init runcmd failed, listed in /run/cloud-init-failed. Manual fixing and rebooting required. ' | tee /etc/motd || true
 
 write_files:
-%{ if contains(tags, "puppet") || length(setintersection(tags, bastion_tags)) > 0 ~}
   - content: |
+%{ for host, ip in puppetservers ~}
+%{ if ip != "" ~}
+      ${ip} ${host}
+%{ endif ~}
+%{ endfor ~}
+    path: /etc/hosts
+    append: true
+%{ if user_tf_required ~}
+  - path: /etc/ssh/sshd_config.d/50-authenticationmethods.conf
+    permissions: "0600"
+    content: |
       Match User tf
         AuthorizedKeysFile /etc/ssh/authorized_keys.%u
         AuthenticationMethods publickey
-    path: /etc/ssh/sshd_config.d/50-authenticationmethods.conf
-    permissions: "0600"
-%{ endif ~}
-%{ if contains(tags, "puppet") ~}
-  - content: restrict,pty ${tf_ssh_public_key}
-    path: /etc/ssh/authorized_keys.tf
+  - path: /etc/ssh/authorized_keys.tf
     permissions: "0644"
+    content: |
+%{ if contains(tags, "puppet") ~}
+      restrict,pty ${tf_ssh_public_key}
 %{ else ~}
-%{ if length(setintersection(tags, bastion_tags)) > 0 ~}
-  # If the ip addresses of the puppet servers are not known in advance, we cannot restrict the ssh connection to them.
-  - content: restrict,%{if length(compact(values(puppetservers))) == length(keys(puppetservers)) }%{ for host, ip in puppetservers }permitopen="${ip}:22"%{ endfor }%{ else }pty%{ endif },port-forwarding,command="/sbin/nologin" ${tf_ssh_public_key}
-    path: /etc/ssh/authorized_keys.tf
-    permissions: "0644"
+%{ if length(compact(values(puppetservers))) != length(keys(puppetservers)) ~}
+      restrict,pty,port-forwarding,command="/sbin/nologin" ${tf_ssh_public_key}
+%{ else ~}
+      restrict,%{ for host, ip in puppetservers }permitopen="${ip}:22",%{ endfor }port-forwarding,command="/sbin/nologin" ${tf_ssh_public_key}
+%{ endif ~}
 %{ endif ~}
 %{ endif ~}
   - content: |
@@ -207,29 +203,8 @@ write_files:
     path: /etc/puppetlabs/facter/facts.d/prefix.yaml
     permissions: "0644"
 %{ if contains(tags, "puppet") ~}
-  - content: |
-      #!/bin/bash
-      ZIP_FILE=$${1}
-      ZIP_DIR=$${ZIP_FILE%.zip}
-
-      # unzip is not necessarily installed when connecting, but python is.
-      /usr/libexec/platform-python -c "import zipfile; zipfile.ZipFile('$${ZIP_FILE}').extractall()"
-
-      chmod g-w,o-rwx $(find $${ZIP_DIR}/ -type f ! -path $${ZIP_DIR}/code/*)
-      chown -R root:52 $${ZIP_DIR}
-      mkdir -p -m 755 /etc/puppetlabs/
-      rsync -avh --no-t --exclude 'data' $${ZIP_DIR}/ /etc/puppetlabs/
-      rsync -avh --no-t --del $${ZIP_DIR}/data/ /etc/puppetlabs/data/
-      rm -rf $${ZIP_DIR}/
-
-      if [ -f /opt/puppetlabs/puppet/bin/r10k ] && [ /etc/puppetlabs/code/Puppetfile -nt /etc/puppetlabs/code/modules ]; then
-          /opt/puppetlabs/puppet/bin/r10k puppetfile install --moduledir=/etc/puppetlabs/code/modules --puppetfile=/etc/puppetlabs/code/Puppetfile
-          touch /etc/puppetlabs/code/modules
-      fi
-
-      if [ -f /usr/local/bin/consul ] && [ -f /usr/bin/jq ]; then
-          /usr/local/bin/consul event -token=$(jq -r .acl.tokens.agent /etc/consul/config.json) -name=puppet $(date +%s)
-      fi
+  - content: ${filebase64("${module_path}/update_etc_puppetlabs.sh")}
+    encoding: b64
     permissions: "0755"
     path: /usr/sbin/update_etc_puppetlabs.sh
   - content: |
@@ -245,12 +220,6 @@ write_files:
     permissions: "0640"
   - path: /var/log/autosign.log
     permissions: "0640"
-%{ if puppetfile != "" ~}
-  - path: /etc/puppetlabs/code/Puppetfile
-    content: |
-      ${indent(6, puppetfile)}
-    permissions: "0644"
-%{ endif ~}
 %{ endif ~}
   - content: |
       ---

common/configuration/update_etc_puppetlabs.sh

@@ -0,0 +1,22 @@
+#!/bin/bash
+ZIP_FILE=${1}
+ZIP_DIR=${ZIP_FILE%.zip}
+
+# unzip is not necessarily installed when connecting, but python is.
+/usr/libexec/platform-python -c "import zipfile; zipfile.ZipFile('${ZIP_FILE}').extractall()"
+
+chmod g-w,o-rwx $(find ${ZIP_DIR}/ -type f ! -path ${ZIP_DIR}/code/*)
+chown -R root:52 ${ZIP_DIR}
+mkdir -p -m 755 /etc/puppetlabs/
+rsync -avh --no-t --exclude 'data' ${ZIP_DIR}/ /etc/puppetlabs/
+rsync -avh --no-t --del ${ZIP_DIR}/data/ /etc/puppetlabs/data/
+rm -rf ${ZIP_DIR}/
+
+if [ -f /opt/puppetlabs/puppet/bin/r10k ] && [ /etc/puppetlabs/code/Puppetfile -nt /etc/puppetlabs/code/modules ]; then
+    /opt/puppetlabs/puppet/bin/r10k puppetfile install --moduledir=/etc/puppetlabs/code/modules --puppetfile=/etc/puppetlabs/code/Puppetfile
+    touch /etc/puppetlabs/code/modules
+fi
+
+if [ -f /usr/local/bin/consul ] && [ -f /usr/bin/jq ]; then
+    /usr/local/bin/consul event -token=$(jq -r .acl.tokens.agent /etc/consul/config.json) -name=puppet $(date +%s)
+fi

gcp/infrastructure.tf

@@ -32,7 +32,6 @@ module "configuration" {
   cloud_provider  = local.cloud_provider
   cloud_region    = local.cloud_region
   skip_upgrade    = var.skip_upgrade
-  puppetfile      = var.puppetfile
 }
 
 module "provision" {

incus/infrastructure.tf

@@ -28,7 +28,6 @@ module "configuration" {
   cloud_provider  = "incus"
   cloud_region    = "local"
   skip_upgrade    = var.skip_upgrade
-  puppetfile      = var.puppetfile
 }
 
 module "provision" {

openstack/infrastructure.tf

@@ -28,7 +28,6 @@ module "configuration" {
   cloud_provider  = local.cloud_provider
   cloud_region    = local.cloud_region
   skip_upgrade    = var.skip_upgrade
-  puppetfile      = var.puppetfile
 }
 
 module "provision" {