Migrate stuff: allow for different protocol (RDMA, woo), wait more accurately for when VM is ready for receive, wait for migrate to finish properly to work around qemu bug

2 files changed, 25 insertions, 12 deletions

diff --git a/kvm-wrapper.conf.default b/kvm-wrapper.conf.default
index 60a2be2..a426398 100644
--- a/kvm-wrapper.conf.default
+++ b/kvm-wrapper.conf.default
@@ -33,6 +33,8 @@ SCREEN_EXTRA_OPTS=""
 #SCREEN_NAME_PREFIX="$RUN_DIR/"
 #SCREEN_EXTRA_OPTS="-z -r winch" # change escape key: "-e ^]"
 
+KVM_MIGRATE_TRANSPORT="tcp" #or "x-rdma"
+
 STARTUP_LIST="$ROOTDIR/startup/startup-list"
 
 KVM_IMG_FORMAT="qcow2"
diff --git a/kvm-wrapper.sh b/kvm-wrapper.sh
index 51ba179..dc2cfc1 100755
--- a/kvm-wrapper.sh
+++ b/kvm-wrapper.sh
@@ -861,25 +861,33 @@ function kvm_migrate_vm ()
 	PORT=$((RANDOM%1000+4000))
 
 	# Launch new instance (on pre-configured node)
-	"$SCRIPT_PATH" receive-migrate "$VM_NAME" "$PORT"
+	if ! "$SCRIPT_PATH" receive-migrate "$VM_NAME" "$PORT"; then
+		desc_revert_setting "KVM_CLUSTER_NODE"
+		fail_exit "Could not receive migrate on remote host"
+	fi
+
+	sleep 0.5
 
-	monitor_send_cmd "migrate_set_speed 1024m"
+	# default speed is 32MiB/s, apparently. Set it higher.
+	monitor_send_cmd "migrate_set_speed 100g"
 #	monitor_send_cmd "migrate \"exec: ssh `get_cluster_host $REMOTE_NODE` socat - unix:$RUN_DIR/migrate-$REMOTE_NODE.sock\""
-	monitor_send_cmd "migrate tcp:`get_cluster_host $REMOTE_NODE`:$PORT"
+	monitor_send_cmd "migrate ${KVM_MIGRATE_TRANSPORT:-tcp}:`get_cluster_host $REMOTE_NODE`:$PORT"
+	while ! echo "info migrate" | socat - unix:"$MONITOR_FILE" | grep -q "Migration status: completed"; do
+		sleep 1
+	done
 	monitor_send_cmd "quit"
 }
 
 function kvm_receive_migrate_vm ()
 {
-        local PORT="$2"
+	local PORT="$2"
 
-        eval KVM_SCREEN="yes" $SCREEN_START_DETACHED "$SCREEN_SESSION_NAME" $SCREEN_EXTRA_OPTS "$SCRIPT_PATH" receive-migrate-here "$VM_NAME" "$PORT"
+	eval KVM_SCREEN="yes" $SCREEN_START_DETACHED "$SCREEN_SESSION_NAME" $SCREEN_EXTRA_OPTS "$SCRIPT_PATH" receive-migrate-here "$VM_NAME" "$PORT"
 
-        # Wait for the receiving qemu is ready.
-        #while ! test_exist $RUN_DIR/migrate-$VM_NAME.sock; do
-        while ! netstat -nplt | grep -q ":$PORT "; do
-                sleep 1;
-        done
+	# Wait for the receiving qemu is ready.
+	while ! monitor_send_cmd "info status" >&/dev/null; do
+		sleep 0.1
+	done
 }
 
 function kvm_receive_migrate_here_vm ()
@@ -887,7 +895,7 @@ function kvm_receive_migrate_here_vm ()
 	local PORT="$2"
 
 #	KVM_ADDITIONNAL_PARAMS+=" -incoming unix:$RUN_DIR/migrate-$VM_NAME.sock"
-	KVM_ADDITIONNAL_PARAMS+=" -incoming tcp:`get_cluster_host $(hostname -s)`:$PORT"
+	KVM_ADDITIONNAL_PARAMS+=" -incoming ${KVM_MIGRATE_TRANSPORT:-tcp}:`get_cluster_host $(hostname -s)`:$PORT"
 	FORCE="yes"
 	kvm_start_vm "$VM_NAME"
 }
@@ -897,8 +905,11 @@ kvm_save_state_vm ()
 	! test_exist "$PID_FILE" && fail_exit "Error: $VM_NAME doesn't seem to be running."
 	! test_socket_rw "$MONITOR_FILE" && fail_exit "Error: could not open monitor socket $MONITOR_FILE."
 	monitor_send_cmd "stop"
-	monitor_send_cmd "migrate_set_speed 4095m"
+	monitor_send_cmd "migrate_set_speed 100g"
 	monitor_send_cmd "migrate \"exec:gzip -c > $CACHE_DIR/$VM_NAME-state.gz\""
+	while ! echo "info migrate" | socat - unix:"$MONITOR_FILE" | grep -q "Migration status: completed"; do
+		sleep 1
+	done
 	monitor_send_cmd "quit"
 }