Improve Swarm support (#333)

* Query for labeled services as well

* Try scaling down services

* Scale services back up

* Use progress tool from Docker CLI

* In test, label both services

* Clean up error and log messages

* Document scale-up/down approach in docs

* Downgrade Docker CLI to match client

* Document services stats

* Do not rely on PreviousSpec for storing desired replica count

* Log warnings from Docker when updating services

* Check whether container and service labels collide

* Document script behavior on label collision

* Add additional check if all containers have been removed

* Scale services concurrently

* Move docker interaction code into own file

* Factor out code for service updating

* Time out after five minutes of not reaching desired container count

* Inline handling of in-swarm container level restart

* Timer is more suitable for timeout race

* Timeout when scaling down services should be configurable

* Choose better filename

* Reflect changes in naming

* Rename and deprecate BACKUP_STOP_CONTAINER_LABEL

* Improve logging

* Further simplify logging

cmd/backup/main.go

@@ -47,12 +47,12 @@ func main() {
 	}()
 
 	s.must(s.withLabeledCommands(lifecyclePhaseArchive, func() error {
-		restartContainers, err := s.stopContainers()
+		restartContainersAndServices, err := s.stopContainersAndServices()
 		// The mechanism for restarting containers is not using hooks as it
 		// should happen as soon as possible (i.e. before uploading backups or
 		// similar).
 		defer func() {
-			s.must(restartContainers())
+			s.must(restartContainersAndServices())
 		}()
 		if err != nil {
 			return err