Expand description
Resolves the clusterd compute controller address. The driver connects to an externally-managed clusterd (mzcompose or a manually-launched process); it never spawns one.
Local iteration without mzcompose:
- Start CockroachDB (see CLAUDE.md).
- Launch the driver first so it hosts PubSub on a fixed port (e.g. 0.0.0.0:6879).
- Launch clusterd manually (run
cargo run -p mz-clusterd -- --helpto confirm required flags and the exactTimelyConfigJSON shape fromsrc/cluster-client/src/client.rs):
PERSIST_PUBSUB_URL=http://127.0.0.1:6879 \
cargo run -p mz-clusterd -- \
--compute-controller-listen-addr 127.0.0.1:2101 \
--storage-controller-listen-addr 127.0.0.1:2100 \
--compute-timely-config 'TIMELY_CONFIG_JSON' \
--storage-timely-config 'TIMELY_CONFIG_JSON' \
--process 0 --scratch-directory /tmp/clusterd-scratch- Run the e2e test with CLUSTERD_COMPUTE_ADDR=127.0.0.1:2101 and the same PersistHost bind/port the clusterd PubSub URL points at.
Functionsยง
- compute_
addr - The clusterd compute controller address, from
CLUSTERD_COMPUTE_ADDR(default127.0.0.1:2101). - e2e_
enabled - Whether an end-to-end test should run: true only when a target address was
explicitly provided. Integration tests skip when unset to keep
cargo testgreen without a running clusterd.