From 1709fd35e764c82bcf18777bc47155c5feb1604c Mon Sep 17 00:00:00 2001
From: "David M. Johnson" <johnsond@flux.utah.edu>
Date: Thu, 21 Apr 2016 12:35:54 -0600
Subject: [PATCH] Fix up host-evacuate for delete-nodes; add disfunctional
 migrate option.

(I'm not sure why migrate isn't working... the nodes try to migrate but
don't complete the journey :).)
---
 setup-controller-delete-nodes.sh | 95 ++++++++++++++++++++++++++------
 1 file changed, 79 insertions(+), 16 deletions(-)

diff --git a/setup-controller-delete-nodes.sh b/setup-controller-delete-nodes.sh
index 79049c8..853f833 100755
--- a/setup-controller-delete-nodes.sh
+++ b/setup-controller-delete-nodes.sh
@@ -17,6 +17,12 @@ if [ "$HOSTNAME" != "$CONTROLLER" ]; then
     exit 0;
 fi
 
+DOMIGRATE=0
+if [ "$1" = '-m' ]; then
+    DOMIGRATE=1
+    shift
+fi
+
 OLDNODES="$@"
 
 if [ -z "$OLDNODES" ] ; then
@@ -24,6 +30,9 @@ if [ -z "$OLDNODES" ] ; then
     exit 1
 fi
 
+. $SETTINGS
+. $OURDIR/admin-openrc.sh
+
 #
 # For now, we just do the stupid thing and "evacuate" the VMs from the
 # old hypervisor to a new hypervisor.  Let Openstack pick for now...
@@ -31,18 +40,41 @@ fi
 # To do this, we disable the compute service, force it down, and
 # evacuate!
 #
+VALIDNODES=""
 for node in $OLDNODES ; do
     echo "*** Forcing compute service on $node down and disabling it ..."
     fqdn=`getfqdn $node`
-    id=`nova service-list | awk "/ $fqdn / { print \\$2 }"`
-    nova service-disable $fqdn nova-compute
-    # Hm, this only supported in some versions, so...
-    nova service-force-down $fqdn nova-compute
-    # ... do this too, to make sure the service doesn't come up
-    $SSH $fqdn service nova-compute stop
-    echo "update services set updated_at=NULL where id=$id" \
-	| mysql -u nova --password=${NOVA_DBPASS} nova
+    if [ -z "$fqdn" ]; then
+	echo "ERROR: could not get FQDN for node $node; skipping!"
+	continue
+    fi
+    #id=`nova service-list | awk -v IGNORECASE=1 "/ $fqdn / { print \\$2 }"`
+    id=`nova service-list | grep -i $fqdn | awk '// { print $2 }'`
+    if [ -z "$id" ]; then
+	echo "ERROR: could not get service id for nova-compute on node $node; skipping! ($id)"
+
+	nova service-list
 
+	continue
+    fi
+
+    VALIDNODES="$VALIDNODES $node"
+
+    if [ $DOMIGRATE -eq 0 ]; then
+	nova service-disable $fqdn nova-compute
+        # Hm, this only supported in some versions, so...
+	nova service-force-down $fqdn nova-compute
+        # REALLY force it down :)
+	echo "update services set forced_down=1 where id=$id" \
+	    | mysql -u nova --password=${NOVA_DBPASS} nova
+	
+        # ... do this too, to make sure the service doesn't come up
+        # no, don't do it, we can't host-evacuate without it
+        #$SSH $fqdn service nova-compute stop
+	
+	echo "update services set updated_at=NULL where id=$id" \
+	    | mysql -u nova --password=${NOVA_DBPASS} nova
+    fi
 done
 
 #
@@ -51,11 +83,33 @@ done
 # evacuated VMs).
 #
 fqdnlist=""
-for node in $OLDNODES ; do
+for node in $VALIDNODES ; do
     echo "*** Evacuating all instances from $node ..."
     fqdn=`getfqdn $node`
+    if [ -z "$fqdn" ]; then
+	echo "ERROR: could not get FQDN for node $node; skipping!"
+	continue
+    fi
 
-    nova host-evacuate $fqdn
+    if [ $DOMIGRATE -eq 1 ]; then
+	servers=`nova hypervisor-servers $fqdn | grep -i $fqdn | awk '// { print $2 }' | xargs`
+	for server in $servers ; do
+	    nova migrate --poll $server
+	    VM_OUTPUT=`nova show $server`
+	    VM_STATUS=`echo "$VM_OUTPUT" | grep status | awk '{print $4}'`
+	    while [ "$VM_STATUS" != "VERIFY_RESIZE" ]; do
+		echo -n "."
+		sleep 2
+		VM_OUTPUT=`nova show $server`
+		VM_STATUS=`echo "$VM_OUTPUT" | grep status | awk '{print $4}'`
+	    done
+	    nova resize-confirm $server
+	    echo "$server instance migrated and resized."
+	    echo;
+	done
+    else
+	nova host-evacuate $fqdn
+    fi
 
     # Create a list for the next step so we don't have to keep resolving
     # FQDNs
@@ -66,12 +120,12 @@ done
 # Ok, now we want to wait until all those nodes no longer have instances
 # on them.
 #
-sucess=0
-while [ $success -ne 1 ]; do
+success=0
+while [ ! $success -eq 1 ]; do
     success=1
     for fqdn in $fqdnlist ; do
 	sleep 8
-	count=`nova hypervisor-servers $fqdn | awk '/ instance-.* / { print $2 }' | wc -l`
+	count=`nova hypervisor-servers $fqdn | awk -v IGNORECASE=1 '/ instance-.* / { print $2 }' | wc -l`
 	if [ $count -gt 0 ]; then
 	    success=0
 	    echo "*** $fqdn still has $count instances"
@@ -79,12 +133,21 @@ while [ $success -ne 1 ]; do
     done
 done
 
-for node in $OLDNODES ; do
+for node in $VALIDNODES ; do
     echo "*** Deleting compute service on $node ..."
     fqdn=`getfqdn $node`
-    id=`nova service-list | awk "/ $fqdn / { print \\$2 }"`
+    if [ -z "$fqdn" ]; then
+	echo "ERROR: could not get FQDN for node $node; skipping!"
+	continue
+    fi
+    #id=`nova service-list | awk -v IGNORECASE=1 "/ $fqdn / { print \\$2 }"`
+    id=`nova service-list | grep -i $fqdn | awk '// { print $2 }'`
+    if [ -z "$id" ]; then
+	echo "ERROR: could not get service id for nova-compute on node $node; skipping!"
+	continue
+    fi
     nova service-delete $id
 done
 
-echo "*** Evacuated and deleted nodes $OLDNODES !"
+echo "*** Successfully evacuated and deleted nodes $OLDNODES !"
 exit 0
-- 
GitLab