Fix broken handling of signals

Our tooling both past (Upstart?) and present (Kubernetes) sends a
`SIGTERM` signal to notify a process that it should gracefully shut
down. At the moment, this library doesn't properly handle this - it does
rescue the error, but does so in the wrong place (so it's not actually
handled) and indiscriminately exits with a non-zero exit code.

- Ensure `SignalException`s raised in `Bunny::Queue#subscribe` (rather
  than the block passed to it) are rescued
- Exit cleanly when receiving a `SignalException` (the comment next to
  `exit(1)` used to say "ensure rabbitmq requeues outstanding messages",
  but that has nothing to do with the exit code, it happens by default
  if the message hasn't been acked by then)
- Add test for signal handling

Ideally in the future we might take this further and allow in-flight
processing to complete before shutting down.

lib/govuk_message_queue_consumer/consumer.rb

@@ -44,15 +44,15 @@ def run(subscribe_opts: {})
         @statsd_client.increment("#{@queue_name}.started")
         message_consumer.process(message)
         @statsd_client.increment("#{@queue_name}.#{message.status}")
-      rescue SignalException => e
-        @logger.error "SignalException in processor: \n\n #{e.class}: #{e.message}\n\n#{e.backtrace.join("\n")}"
-        exit(1) # Ensure rabbitmq requeues outstanding messages
       rescue StandardError => e
         @statsd_client.increment("#{@queue_name}.uncaught_exception")
         GovukError.notify(e) if defined?(GovukError)
         @logger.error "Uncaught exception in processor: \n\n #{e.class}: #{e.message}\n\n#{e.backtrace.join("\n")}"
         exit(1) # Ensure rabbitmq requeues outstanding messages
       end
+    rescue SignalException => e
+      @logger.error "SignalException in processor: \n\n #{e.class}: #{e.message}\n\n#{e.backtrace.join("\n")}"
+      exit
     end
 
   private

spec/govuk_message_queue_consumer/consumer_spec.rb

@@ -8,28 +8,23 @@
   let(:client_processor) { instance_double("Client::Processor") }
 
   describe "#run" do
-    it "doesn't create the queue" do
-      stubs = create_bunny_stubs
-      channel = stubs.channel
+    let(:stubs) { create_bunny_stubs }
+    let(:channel) { stubs.channel }
+    let(:queue) { stubs.queue }
 
+    it "doesn't create the queue" do
       expect(channel).to receive(:queue).with("some-queue", { no_declare: true })
 
       described_class.new(queue_name: "some-queue", processor: client_processor, rabbitmq_connection: stubs.connection, logger: logger).run
     end
 
     it "doesn't bind the queue" do
-      stubs = create_bunny_stubs
-      queue = stubs.queue
-
       expect(queue).not_to receive(:bind)
 
       described_class.new(queue_name: "some-queue", processor: client_processor, rabbitmq_connection: stubs.connection, logger: logger).run
     end
 
     it "calls the heartbeat processor when subscribing to messages" do
-      stubs = create_bunny_stubs
-      queue = stubs.queue
-
       expect(queue).to receive(:subscribe).and_yield(:delivery_info_object, :headers, "payload")
       heartbeat_processor_stub = instance_double(GovukMessageQueueConsumer::HeartbeatProcessor)
       allow(GovukMessageQueueConsumer::HeartbeatProcessor).to receive(:new).and_return(heartbeat_processor_stub)
@@ -38,5 +33,16 @@
 
       described_class.new(queue_name: "some-queue", processor: client_processor, rabbitmq_connection: stubs.connection, logger: logger).run
     end
+
+    context "when a SignalException is raised due to the process being terminated" do
+      before do
+        allow(logger).to receive(:error)
+        allow(queue).to receive(:subscribe).and_raise(SignalException, "SIGTERM")
+      end
+
+      it "gracefully exits" do
+        expect { described_class.new(queue_name: "some-queue", processor: client_processor, rabbitmq_connection: stubs.connection, logger: logger).run }.to raise_error(SystemExit)
+      end
+    end
   end
 end