options:¶

-h, --help

show this help message and exit

-d, --debug

Enable logging at DEBUG level

-a ADDRESSES, --addresses ADDRESSES

Comma separated list of addresses at which the interchange could be reached

--cert_dir CERT_DIR

Path to certificate directory.

-l LOGDIR, --logdir LOGDIR

Process worker pool log directory

-u UID, --uid UID

Unique identifier string for Manager

-b BLOCK_ID, --block_id BLOCK_ID

Block identifier for Manager

-c CORES_PER_WORKER, --cores_per_worker CORES_PER_WORKER

Number of cores assigned to each worker process. Default=1.0

-m MEM_PER_WORKER, --mem_per_worker MEM_PER_WORKER

GB of memory assigned to each worker process. Default=0, no assignment

-t TASK_PORT, --task_port TASK_PORT

REQUIRED: Task port for receiving tasks from the interchange

--max_workers_per_node MAX_WORKERS_PER_NODE

Caps the maximum workers that can be launched, default:infinity

-p PREFETCH_CAPACITY, --prefetch_capacity PREFETCH_CAPACITY

Number of tasks that can be prefetched to the manager. Default is 0.

--hb_period HB_PERIOD

Heartbeat period in seconds. Uses manager default unless set

--hb_threshold HB_THRESHOLD

Heartbeat threshold in seconds. Uses manager default unless set

--drain_period DRAIN_PERIOD

Drain this pool after specified number of seconds. By default, does not drain.

--address_probe_timeout ADDRESS_PROBE_TIMEOUT

Timeout to probe for viable address to interchange. Default: 30s

--poll POLL

Poll period used in milliseconds

-r RESULT_PORT, --result_port RESULT_PORT

REQUIRED: Result port for posting results to the interchange

--cpu-affinity CPU_AFFINITY

Whether/how workers should control CPU affinity.

--available-accelerators [AVAILABLE_ACCELERATORS ...]

Names of available accelerators, if not given assumed to be zero accelerators available

--enable_mpi_mode

Enable MPI mode

--mpi-launcher {srun,aprun,mpiexec}

MPI launcher to use iff enable_mpi_mode=true