Merge pull request #4314 from openshift-cherrypick-robot/cherry-pick-4182-to-release-4.13

[release-4.13] OCPBUGS-32208: Run resolv-prepender entirely async

Author: openshift-merge-bot[bot]

templates/common/on-prem/files/NetworkManager-resolv-prepender.yaml

@@ -10,18 +10,22 @@ contents:
 
     function resolv_prepender {
       mkdir -p /run/resolv-prepender
-      echo "DHCP6_FQDN_FQDN=$DHCP6_FQDN_FQDN" > /run/resolv-prepender/env
-      echo "IP4_DOMAINS=$IP4_DOMAINS" >> /run/resolv-prepender/env
-      echo "IP6_DOMAINS=$IP6_DOMAINS" >> /run/resolv-prepender/env
-      systemctl start on-prem-resolv-prepender
-      # Wait for the service to complete so we don't mark the network up too soon
-      while systemctl is-active on-prem-resolv-prepender
-      do
-        sleep 1
-      done
+      echo "IP4_DOMAINS=$IP4_DOMAINS" > /run/resolv-prepender/env.new
+      echo "IP6_DOMAINS=$IP6_DOMAINS" >> /run/resolv-prepender/env.new
+      # If we changed the environment, we should restart the service to pick up the
+      # new values. However, if the image hasn't been pulled successfully yet we can't
+      # restart the service or we may interrupt the pull and end up with a corrupt image.
+      # We're better off with incorrect search domains for a while than wedging the
+      # system with a bad image.
+      if ! diff -q /run/resolv-prepender/env /run/resolv-prepender/env.new && /usr/bin/podman image exists "{{ .Images.baremetalRuntimeCfgImage }}"; then
+        >&2 echo "NM resolv-prepender: Environment variable(s) changed. Restarting service."
+        systemctl is-active on-prem-resolv-prepender && systemctl kill on-prem-resolv-prepender
+      fi
+      mv -f /run/resolv-prepender/env.new /run/resolv-prepender/env
+      systemctl start --no-block on-prem-resolv-prepender
     }
 
-    export DHCP6_FQDN_FQDN IP4_DOMAINS IP6_DOMAINS
+    export IP4_DOMAINS IP6_DOMAINS
     export -f resolv_prepender
     # Given an overall Network Manager dispatcher timeout of 90 seconds, and multiple events which
     # may occur within this time period, we must enforce a time limit for each event. As some
@@ -34,6 +38,16 @@ contents:
             >&2 echo "NM resolv-prepender: Timeout occurred"
             exit 1
         fi
+        # If $DHCP6_FQDN_FQDN is not empty and is not localhost.localdomain and static hostname was not already set
+        if [[ -n "$DHCP6_FQDN_FQDN" && "$DHCP6_FQDN_FQDN" != "localhost.localdomain" && "$DHCP6_FQDN_FQDN" =~ "." ]] ; then
+           STATIC_HOSTNAME="$(test ! -e /etc/hostname && echo -n || cat /etc/hostname | xargs)"
+
+           if [[ -z "$STATIC_HOSTNAME" || "$STATIC_HOSTNAME" == "localhost.localdomain" ]] ; then
+              # run with systemd-run to avoid selinux problems
+              systemd-run --property=Type=oneshot --unit resolve-prepender-hostnamectl -Pq \
+                  hostnamectl set-hostname --static --transient $DHCP6_FQDN_FQDN
+           fi
+        fi
       ;;
       *)
       ;;

templates/common/on-prem/files/resolv-prepender.yaml

@@ -23,27 +23,16 @@ contents:
         # Ref.: https://github.com/containers/common/blob/e028741ef77fdfa3ae261b9d23cdd50253d586c4/libimage/copier.go#L27-L30
 
         >&2 echo "NM resolv-prepender: Checking if baremetal runtime cfg image already exists"
-        if ! /usr/bin/podman image exists {{ .Images.baremetalRuntimeCfgImage }}; then
+        if ! /usr/bin/podman image exists "{{ .Images.baremetalRuntimeCfgImage }}"; then
           >&2 echo "NM resolv-prepender: Starting download of baremetal runtime cfg image"
-          while ! /usr/bin/podman pull --authfile /var/lib/kubelet/config.json {{ .Images.baremetalRuntimeCfgImage }}; do sleep 1; done
+          while ! /usr/bin/podman pull --authfile /var/lib/kubelet/config.json "{{ .Images.baremetalRuntimeCfgImage }}"; do sleep 1; done
           >&2 echo "NM resolv-prepender: Download of baremetal runtime cfg image completed"
         else
           >&2 echo "NM resolv-prepender: Image exists, no need to download"
         fi
     }
 
     function resolv_prepender {
-        # If $DHCP6_FQDN_FQDN is not empty and is not localhost.localdomain and static hostname was not already set
-        if [[ -n "$DHCP6_FQDN_FQDN" && "$DHCP6_FQDN_FQDN" != "localhost.localdomain" && "$DHCP6_FQDN_FQDN" =~ "." ]] ; then
-           STATIC_HOSTNAME="$(test ! -e /etc/hostname && echo -n || cat /etc/hostname | xargs)"
-
-           if [[ -z "$STATIC_HOSTNAME" || "$STATIC_HOSTNAME" == "localhost.localdomain" ]] ; then
-              # run with systemd-run to avoid selinux problems
-              systemd-run --property=Type=oneshot --unit resolve-prepender-hostnamectl -Pq \
-                  hostnamectl set-hostname --static --transient $DHCP6_FQDN_FQDN
-           fi
-        fi
-
             # In DHCP connections, the resolv.conf content may be late, thus we wait for nameservers
             while ! grep nameserver /var/run/NetworkManager/resolv.conf; do
                 >&2 echo  "NM resolv-prepender: NM resolv.conf still empty of nameserver"