From e8243b8ecd5f5da4352481486cc73973e7a9b64e Mon Sep 17 00:00:00 2001
From: Dirk Mueller <dmueller@suse.com>
Date: Tue, 18 Jul 2017 10:36:00 +0200
Subject: [PATCH] crowbar: do a graceful shutdown/reboot when invoked via
 crowbar (bsc#1047636)

Currently there is a hard reboot executed which is interrupting
a currently pending chef-client run. Normally that shouldn't be
a problem but in the HA cluster case it can cause significant harm.
---
 .../templates/default/crowbar_join.suse.sh.erb  | 17 +++++++++++++++++
 crowbar_framework/app/models/node.rb            |  2 +-
 2 files changed, 18 insertions(+), 1 deletion(-)

diff --git a/chef/cookbooks/provisioner/templates/default/crowbar_join.suse.sh.erb b/chef/cookbooks/provisioner/templates/default/crowbar_join.suse.sh.erb
index c96fdad08a..a474c89ccf 100644
--- a/chef/cookbooks/provisioner/templates/default/crowbar_join.suse.sh.erb
+++ b/chef/cookbooks/provisioner/templates/default/crowbar_join.suse.sh.erb
@@ -124,6 +124,18 @@ wait_for_admin_server() {
     return 0
 }
 
+wait_for_chef_run_finish() {
+
+    local tries_left=300
+    while [[ $tries_left > 0 ]] ; do
+        if [ ! -r /var/log/chef/client.log ] || grep -q "Chef Run complete" /var/log/chef/client.log; then
+            break
+        fi
+        tries_left=$(($tries_left-1))
+        sleep 1
+    done
+}
+
 sync_time() {
     if [ -z "$VALID_NTP_SERVERS" ]; then
         echo_verbose "Skipping time synchronization..."
@@ -508,6 +520,11 @@ elif [ "$MODE" == "stop" ]; then
     final_state="shutdown"
     [ "$RUNLEVEL" == "6" ] && final_state="reboot"
 
+    # Be graceful to a running chef-client run
+    wait_for_chef_run_finish
+
+    systemctl stop chef-client.service
+
     echo_debug "Stopping with state=$final_state (runlevel $RUNLEVEL)"
 
     HOSTNAME=$(hostname -f)
diff --git a/crowbar_framework/app/models/node.rb b/crowbar_framework/app/models/node.rb
index eac54693c9..93e00b0571 100644
--- a/crowbar_framework/app/models/node.rb
+++ b/crowbar_framework/app/models/node.rb
@@ -1058,7 +1058,7 @@ def set_state(state)
       if @node[:platform_family] == "windows"
         net_rpc_cmd(:power_cycle)
       else
-        ssh_cmd("/sbin/reboot")
+        ssh_cmd("/usr/sbin/crowbar_join --stop || /sbin/reboot")
       end
     end
     result