Deployment of Burza in multinode mode

There are two different sets of settings and actions in the multinode mode:

Head node

Env

• Add RAY_NODE_NAME=head to the env to tell the head node about its role
• Add SECONDARY_NODE_NAMES=replica1,replica2 to tell the head node about the other nodes that it has to wait for

Ray cluster

• before starting 'burza run-tests' run 'ray start --head' to start the Ray head node. Ray is installed in Poetry, so you can run it with 'poetry run ray start --head'.
• when all burza runs are finished, run 'ray stop'

Secondary nodes

Env

• Add RAY_NODE_NAME=replicaN to the env
• Change TEST_RUNNER to 'freeloader_test_runner'
• Disable all the report generators (REPORT_GENERATORS=), because all the report generation will happen on the head node
• Remove perf_events from DATA_POINT_GENERATORS, it doesn't support multinode mode right now

Ray cluster

Run "ray start --address='<head_ip>:6379'" before starting burza on the secondary nodes