fuzzer: introduce fuzzer runner (#28)

The fuzzer runs as a separate service on NayDuck worker machines and
run fuzzing tests while the worker is idle.  When worker starts
executing a test it signals this to the fuzzer which then stops and
waits for the worker to finish.

The comments around the top of the file are there mostly to help
toggle between ‘run on GCP’ and ‘run on local machine’.

The API this exposes is:
- on port 7055, an HTTP server that can in particular pause/resume
  fuzzing upon hitting `/pause` and `/resume`;
- on port 5507, Prometheus metrics; and
- when a new artefact is found (~= a new crash is found), the
  `fuzz_artifacts_found` metric gets bumped by 1

automation/setup-host.sh

@@ -99,13 +99,16 @@ else
 	fi
 fi
 
-service=$type
-if [ "$type" = frontend ]; then
-	service=ui
-fi
-cp -nvt /etc/systemd/system/ -- \
-   "/home/nayduck/nayduck/systemd/nayduck-$service.service"
-systemctl enable "nayduck-$service"
+case $type in
+frontend) services=ui              ;;
+worker)   services='worker fuzzer' ;;
+*)        services=$type
+esac
+for service in $services; do
+    cp -nvt /etc/systemd/system/ -- \
+		"/home/nayduck/nayduck/systemd/nayduck-$service.service"
+    systemctl enable "nayduck-$service"
+done
 
 rm -- "$basedir/setup-host.sh"
 

automation/update.sh

@@ -49,7 +49,7 @@ if $only_frontend; then
 	exit 0
 fi
 
-for service in nayduck-ui nayduck-builder nayduck-worker; do
+for service in nayduck-ui nayduck-builder nayduck-worker nayduck-fuzzer; do
 	if has_service "$service"; then
 		sudo systemctl restart "$service"
 	fi

check.sh

@@ -7,6 +7,7 @@ parallel 'sh -xc {}' ::: \
 	'exec pylint -j0 */*.py' \
 	'exec python -m yapf -pir .' \
 	'exec mypy -m backend.backend' \
+	'exec mypy -m fuzzers.main' \
 	'exec mypy -m workers.builder' \
 	'exec mypy -m workers.worker' \
 	'exec shellcheck */*.sh'