Drop meaningless --port from resilience webless entrypoint

[jmchilton]

Note: this PR description was prepared and posted by Claude (AI assistant) on jmchilton's behalf, not authored by them personally.

test/resilience/entrypoint.sh launches Pulsar with a --port it cannot use:

exec pulsar --mode webless --config_dir "$runtime" --port 8913

webless mode binds no port — scripts/pulsar --help says so explicitly ("In webless mode Pulsar will not attempt to bind to a port") — and pulsar-main declares no --port argument.

Why this is a no-op today

scripts/pulsar consumes -h, -m|--mode and -c|--config, then forwards the rest. --port and --config_dir both survive into "$@", and the webless branch runs a bare pulsar-main that never references "$@". So the flag is discarded unread, and the command line pulsar-main actually sees is identical before and after this change. The config directory reaches Pulsar through the PULSAR_CONFIG_DIR export on the preceding line.

Nothing in the suite talks to Pulsar on 8913 either: the pulsar service in test/resilience/docker-compose.yml publishes no ports, and readiness is docker-compose logs plus the RabbitMQ management API (_amqp_setup_has_consumer), never HTTP. The flag looks copied from the gunicorn/chaussette invocations, where a port is meaningful.

Why it stops being a no-op

#480 makes the webless branch forward "$@" to pulsar-main, which is what docs/scripts/pulsar.rst has always documented:

See the documentation for the pulsar-main for the arguments that may be supplied to pulsar in this mode.

With that change in place, argparse rejects the flag and the container dies at startup:

Starting pulsar with command [pulsar-main --config_dir /tmp/pulsar-runtime --port 8913]
usage: pulsar-main [-h] [-c CONFIG_DIR] [--ini_path INI_PATH] ...
pulsar-main: error: unrecognized arguments: --port 8913

No AMQP consumers ever bind, and the first scenario test times out:

E   TimeoutError: Pulsar did not bind amqp consumers within 60.0s
ERROR test/resilience/scenarios/test_broker_outage.py::test_b1_outage_during_preprocessing[mode=amqp]

Because the job runs pytest test/resilience -v -x, that first error blocks every scenario after it. The 9 passed in that run is mock_galaxy/recorder_test.py, which never touches a container — the scenario suite is entirely blocked.

Verification

Checked out #480's branch locally with only this one-line change applied and brought up the stack (Docker 29.7.2). The pulsar container stays up and logs:

pulsar.messaging.bind_amqp INFO bind_manager_to_queue called for [amqp://guest:********@toxiproxy:5672//] and manager [_default_]

which is the exact bind_marker string harness/pulsar_control.py polls for. Pre-fix that line never appeared, because the process exited at argparse.

Landing this separately keeps the harness fix independent of the behaviour change in #480 — it is a pre-existing bug in the harness, and it costs #480 nothing to have it already merged.

🤖 Generated with Claude Code