# System Imports system_variables = import_module("github.com/logos-co/wakurtosis/src/system_variables.star") # Module Imports waku = import_module(system_variables.WAKU_MODULE) prometheus = import_module(system_variables.PROMETHEUS_MODULE) grafana = import_module(system_variables.GRAFANA_MODULE) args_parser = import_module(system_variables.ARGUMENT_PARSER_MODULE) wsl = import_module(system_variables.WSL_MODULE) def run(args): # Load global config file config_file = args_parser.load_config_args(args) config_json = read_file(src=config_file) config = json.decode(config_json) print(config) same_toml_configuration = config['same_toml_configuration'] # Load network topology waku_topology_json = read_file(src=system_variables.TOPOLOGIES_LOCATION + config['topology_file']) waku_topology = json.decode(waku_topology_json) # Set up nodes waku_services = waku.instantiate_waku_nodes(waku_topology, same_toml_configuration) # Set up prometheus + graphana prometheus_service = prometheus.set_up_prometheus(waku_services) grafana_service = grafana.set_up_graphana(prometheus_service) waku.interconnect_waku_nodes(waku_topology, waku_services) # waku.send_test_messages(waku_topology, system_variables.NUMBER_TEST_MESSAGES, # system_variables.DELAY_BETWEEN_TEST_MESSAGE) # waku.get_waku_peers(waku_topology.keys()[1]) # Setup WSL & Start the Simulation wsl_service = wsl.set_up_wsl(waku_services, config['simulation_time'], config['message_rate'], config['min_packet_size'], config['max_packet_size'], config['inter_msg_type'], config['dist_type'], config['emitters_fraction'])