Running Dragon

Dragon can be run on either a single node (such as your laptop or other single compute resource) or on a cluster of many compute resources (multi-node system). In either case, launching a Dragon application is done in a similar way.

Ensure that the Dragon package has been installed into your Python environment. This is typically done within a virtual Python environment. See the Dragon installation instructions on performing this step.
Ensure that the Dragon module has been loaded. This adds the dragon command to your environment.

module use [/path to dragon]/modulefiles
module load dragon

Start your dragon application using the dragon launcher command, passing any relevant command line options.

dragon [dragon options] [program] [program options]

The dragon launcher’s full command help and basic usage appears below:

usage: dragon [-h] [-N NODE_COUNT] [--hostlist HOSTLIST | --hostfile HOSTFILE] [--network-prefix NETWORK_PREFIX] [--network-config NETWORK_CONFIG]
              [--wlm WORKLOAD_MANAGER] [-p PORT] [--transport TRANSPORT_AGENT] [-s | -m] [-l LOG_LEVEL] [--no-label] [--basic-label] [--verbose-label]
              [--version] [PROG] ...

Dragon Launcher Arguments and Options

positional arguments:
  PROG                  PROG specifies an executable program to be run on the primary compute node. In this case, the file may be either executable or not.
                        If PROG is not executable, then Python version 3 will be used to interpret it. All command-line arguments after PROG are passed to
                        the program as its arguments. The PROG and ARGS are optional.
  ARG                   Zero or more program arguments may be specified.

optional arguments:
  -h, --help            show this help message and exit
  -N NODE_COUNT, --nodes NODE_COUNT
                        NODE_COUNT specifies the number of nodes to use. NODE_COUNT must be less or equal to the number of available nodes within the WLM
                        allocation. A value of zero (0) indicates that all available nodes should be used (the default).
  --hostlist HOSTLIST   Specify backend hostnames as a comma-separated list, eg: `--hostlist host_1,host_2,host_3`. `--hostfile` or `--hostlist` is a
                        required argument for WLM SSH and is only used for SSH
  --hostfile HOSTFILE   Specify a list of hostnames to connect to via SSH launch. The file should be a newline character separated list of hostnames.
                        `--hostfile` or `--hostlist` is a required argument for WLM SSH and is only used for SSH
  --network-prefix NETWORK_PREFIX
                        NETWORK_PREFIX specifies the network prefix the dragon runtime will use to determine which IP addresses it should use to build
                        multinode connections from. By default the regular expression r'^(hsn|ipogif|ib)\d+$' is used -- the prefix for known HPE-Cray XC
                        and EX high speed networks. If uncertain which networks are available, the following will return them in pretty formatting: `dragon-
                        network-ifaddrs --ip --no-loopback --up --running | jq`. Prepending with `srun` may be necessary to get networks available on
                        backend compute nodes
  --network-config NETWORK_CONFIG
                        NETWORK_CONFIG specifies a YAML or JSON file generated via a call to the launcher's network config tool that successfully generated
                        a corresponding YAML or JSON file (eg: `dragon-network-config --output-to-yaml`) describing the available backend compute nodes
                        specified either by a workload manager (this is what the tool provides). Alternatively, one can be generated manually as is needed
                        in the case of ssh-only launch. An example with keywords and formatting can be found in the documentation
  --wlm WORKLOAD_MANAGER, -w WORKLOAD_MANAGER
                        Specify what workload manager is used. Currently supported WLMs are: slurm, pbs+pals, ssh
  -p PORT, --port PORT  PORT specifies the port to be used for multinode communication. By default, 7575 is used.
  --transport TRANSPORT_AGENT, -t TRANSPORT_AGENT
                        TRANSPORT_AGENT selects which transport agent will be used for backend node-to-node communication. By default, the TCP
                        transport agent (tcp) is selected. Currently supported agents are: hsta, tcp
  -s, --single-node-override
                        Override automatic launcher selection to force use of the single node launcher
  -m, --multi-node-override
                        Override automatic launcher selection to force use of the multi-node launcher
  -l LOG_LEVEL, --log-level LOG_LEVEL
                        The Dragon runtime enables the output of diagnostic log messages to multiple different output devices. Diagnotic log messages can be
                        seen on the Dragon stderr console, via a combined 'dragon_*.log' file, or via individual log files created by each of the Dragon
                        'actors' (Global Services, Local Services, etc). By default, the Dragon runtime disables all diagnostic log messaging. Passing one
                        of NONE, DEBUG, INFO, WARNING, ERROR, or CRITICAL to this option, the Dragon runtime will enable the specified log verbosity. When
                        enabling DEBUG level logging, the Dragon runtime will limit the stderr and combined dragon log file to INFO level messages. Each
                        actor's log file will contain the complete log history, including DEBUG messages. This is done to help limit the number of messages
                        sent between the Dragon frontend and the Dragon backend at scale. To override the default logging behavior and enable specific
                        logging to one or more Dragon output devices, the LOG_LEVEL option can be formatted as a keyword=value pair, where the KEYWORD is
                        one of the Dragon log output devices (stderr, dragon_file or actor_file), and the VALUE is one of NONE, DEBUG, INFO, WARNING, ERROR
                        or CRITICAL (eg `-l dragon_file=INFO -l actor_file=DEBUG`). Multiple -l|--log-level options may be passed to enable the logging
                        desired.
  --no-label
  --basic-label
  --verbose-label
  --version             show program's version number and exit

In the event that dragon exits abnormlly, use the helper script dragon-cleanup to clean up any zombie processes and reserved memory. The dragon-cleanup script is located in the [dragon install dir]/bin directory and added to the $PATH environment variable after loading the Dragon module.

Running Dragon on a Multi-Node System

To run in multinode mode, Dragon must know what resources are available for its use on the compute backend. When using a workload manager (WLM) such as Slurm or PBS+Pals, Dragon normally obtains the list of available backend compute resources automatically from the active WLM allocation. However, when Dragon is used on a generic cluster without a traditional WLM, Dragon has no way to automatically ascertain what backend compute resources are available. In these cases Dragon can be run using a generic SSH launch.

Dragon supports the following multinode configurations:

Running on a cluster or supercomputer that has been configured with a Work Load Manager (WLM), such has Slurm or PBS+Pals.
Running on a cluster without any Work Load Manager (WLM) using generic SSH launch.

Running Dragon with a Work Load Manager

To launch a Dragon program on several compute nodes, a Work Load Manager job allocation obtained via salloc or sbatch (Slurm) or qsub (PBS+Pals) is required, eg:

$ salloc --nodes=2
$ dragon p2p_lat.py --iterations 100 --lg_max_message_size 12 --dragon

In the event that Dragon is run outside of an active WLM allocation an exception is raised, and the program will not execute:

Listing 2 Dragon exception when no WLM allocation exists:

$ dragon p2p_lat.py --iterations 100 --lg_max_message_size 12 --dragon

RuntimeError: Executing in a Slurm environment, but with no job allocation.
Resubmit as part of an 'salloc' or 'sbatch' execution

To override this default behavior and execute a Dragon program on the same node as your shell, the --single-node-override / -s option is available.

The Dragon service runtime assumes all nodes in an allocation are to be used unless the --node-count option is used. This limits the user program to executing on a smaller subset of nodes, potentially useful for execution of scaling benchmarks. For example, if the user has a job allocation for 4 nodes, but only wants to use 2 for their Dragon program, they may do the following:

$ salloc --nodes=4
$ dragon --nodes 2 p2p_lat.py --iterations 100 --lg_max_message_size 12 --dragon

Running Dragon using generic SSH launch

To use SSH launch, the following configuration options must be provided on the dragon launcher command line:

Please see the FAQ for more information.

Select the SSH Workload Manager

The --wlm ssh / -w ssh option tells the dragon launcher to use generic SSH launch semantics.

Select the TCP Transport Agent

The --transport tcp / -t tcp option tells the dragon launcher to use the Dragon TCP transport agent when setting up the Dragon backend compute network. This is the default option with the open source Dragon package.

Provide available backend compute resources

The list of available backend compute resources can be provided to the dragon launcher in one of several ways

by providing a list of backend compute resources (either explicitly on the launcher command line or via a file) or

by providing a Dragon network configuration file

Note: Dragon requires that passwordless SSH is enabled for all backend compute resources.

Providing a Host List or Host File

Providing a list of hosts to the dragon launcher can be done either by listing them explicitly on the dragon command-line or by providing the dragon launcher the name of a newline seperated text file containing the list of host names.

To provide the available nodes explicitly on the dragon command line, specify the available backend hostnames as a comma-separated list, eg: --hostlist host_1,host_2,host_3.

Listing 3 Providing a list of hosts via the command line

(_env) root $ dragon -w ssh -t tcp --hostlist host_1,host_2,host_3 [PROG]

To provide the available nodes via a text file, create a newline separated text file with each backend node’s hostname on a separate line. Pass the name of the text file to the dragon launcher, eg: --hostfile hosts.txt.

Listing 4 Providing a list of hosts via a text file

(_env) root $ cat hosts.txt
host_1
host_2
host_3
(_env) root $ dragon -w ssh -t tcp --hostfile hosts.txt [PROG]

NOTE: You cannot use both --hostfile and --hostlist on the commandline at the same time.

When passing the list of available backend nodes in either of these ways, the dragon launcher needs to determine basic network configuration settings for each listed node before it can launch the Dragon user application. This is done by launching a utility application on each listed node to report the node’s IP and other relevant information. Running this utility application slightly delays the startup of Dragon. To prevent this delay, you can instead generate a Dragon network-config file as explained below.

Providing a Dragon Network-Config File

Dragon provides a utility application to gather and persist relevant network information from it’s backend compute resorces. This utility can be used to generate a persistent YAML or JSON configuration which, when passed to the dragon launcher, provides all required information about a set of backend compute nodes.

To generate a network configuration file for a given set of backend compute nodes, run the dragon-network-config tool as shown below:

Listing 5 Example of how to run the dragon-network-config tool

(_env) root $ dragon-network-config -w ssh --hostlist host1,host2,host3,host4 -j
(_env) root $  ls ssh.json
ssh.json

Once you have a network configuration file, the name of the configuration file can be passed to the dragon launcher to identify the available backend compute resources:

Listing 6 Providing a list of hosts via the command line

(_env) root $ dragon -w ssh -t tcp --network-config ssh.json [PROG]

NOTE: Changes to the backend compute node’s IP addresses or other relevant network settings will invalidate the saved network config file. If this happens, please re-run the dragon-network-config tool to collect updated information.

The dragon-network-config help is below:

Listing 7 Dragon Network Config (dragon-network-config) tool’s help and basic use

usage: dragon-network-config [-h] [-p PORT] [--network-prefix NETWORK_PREFIX] [--wlm WORKLOAD_MANAGER] [--log] [--output-to-yaml] [--output-to-json]
                             [--no-stdout] [--primary PRIMARY] [--hostlist HOSTLIST | --hostfile HOSTFILE]

Runs Dragon internal tool for generating network topology

optional arguments:
  -h, --help            show this help message and exit
  -p PORT, --port PORT  Infrastructure listening port (default: 6565)
  --network-prefix NETWORK_PREFIX
                        NETWORK_PREFIX specifies the network prefix the dragon runtime will use to determine which IP addresses it should use to build
                        multinode connections from. By default the regular expression r'^(hsn|ipogif|ib)\d+$' is used -- the prefix for known HPE-Cray XC
                        and EX high speed networks. If uncertain which networks are available, the following will return them in pretty formatting: `dragon-
                        network-ifaddrs --ip --no-loopback --up --running | jq`. Prepending with `srun` may be necessary to get networks available on
                        backend compute nodes
  --wlm WORKLOAD_MANAGER, -w WORKLOAD_MANAGER
                        Specify what workload manager is used. Currently supported WLMs are: slurm, pbs+pals, ssh
  --log, -l             Enable debug logging
  --output-to-yaml, -y  Output configuration to YAML file
  --output-to-json, -j  Output configuration to JSON file
  --no-stdout           Do not print the configuration to stdout
  --primary PRIMARY     Specify the hostname to be used for the primary compute node
  --hostlist HOSTLIST   Specify backend hostnames as a comma-separated list, eg: `--hostlist host_1,host_2,host_3`. `--hostfile` or `--hostlist` is a
                        required argument for WLM SSH and is only used for SSH
  --hostfile HOSTFILE   Specify a list of hostnames to connect to via SSH launch. The file should be a newline character separated list of hostnames.
                        `--hostfile` or `--hostlist` is a required argument for WLM SSH and is only used for SSH

# To create YAML and JSON files with a slurm WLM:
$ dragon-network-config --wlm slurm --output-to-yaml --output-to-json

Formatting of the network-config file appears below for both JSON and YAML:

Listing 8 Example of YAML formatted network configuration file

'0':
  h_uid: null
  host_id: 18446744071562724608
  ip_addrs:
  - 10.128.0.5:6565
  is_primary: true
  name: nid00004
  num_cpus: 0
  physical_mem: 0
  shep_cd: ''
  state: 4
'1':
  h_uid: null
  host_id: 18446744071562724864
  ip_addrs:
  - 10.128.0.6:6565
  is_primary: false
  name: nid00005
  num_cpus: 0
  physical_mem: 0
  shep_cd: ''
  state: 4

Listing 9 Example of JSON formatted network configuration file

{
  "0": {
        "state": 4,
        "h_uid": null,
        "name": "nid00004",
        "is_primary": true,
        "ip_addrs": [
            "10.128.0.5:6565"
        ],
        "host_id": 18446744071562724608,
        "num_cpus": 0,
        "physical_mem": 0,
        "shep_cd": ""
    },
    "1": {
        "state": 4,
        "h_uid": null,
        "name": "nid00005",
        "is_primary": false,
        "ip_addrs": [
            "10.128.0.6:6565"
        ],
        "host_id": 18446744071562724864,
        "num_cpus": 0,
        "physical_mem": 0,
        "shep_cd": ""
    }
}

Dragon’s Transport Agents

To facilitate cross node communications when running in a multi-node mode, Dragon provides a couple of different Transport Agents.

High Speed Transport Agent (HSTA)

The HSTA is new in Dragon 0.4. The HSTA transport is an RDMA based transport agent that combines MPI-like performance using Dragon Channels. There are no network ports to configure for HSTA, but it does depend on Cray-MPICH being installed on the system.

The HSTA transport agent is currently not available in the opensource version of Dragon. For inquiries about Dragon’s high speed RDMA-based transport, please contact HPE by emailing dragonhpc@hpe.com.

TCP-based Transport Agent

As of Dragon 0.5, the TCP-based transport agent is the default transport agent for the Dragon opensource package. The TCP transport agent utilizes standard TCP for inter-node communication through Dragon Channels.

When using a version of Dragon that includes the HSTA transport agent and you prefer to use the TCP transport agent, the --transport tcp option can be passed to the launcher (see: FAQ and Launcher options).

The TCP agent is configured to use port 7575 by default. If that port is blocked, it can be changed with the --port argument to dragon. If not specific, 7575 is used:, eg:

# Port 7575 used
$ dragon --nodes 2 p2p_lat.py --iterations 100 --lg_max_message_size 12 --dragon

# Port 7000 used
$ dragon --port 7000 --nodes 2 p2p_lat.py --iterations 100 --lg_max_message_size 12 --dragon

The TCP transport agent also favors known Cray high-speed interconnect networks by default. This is accomplished via regex specification of the network’s named prefix matchin ipogif (Aries) or hsn (Slingshot): r'^(hsn|ipogif)d+$'. To change, for example, to match only hsn networks, the --network-prefix argument could be used:

$ dragon --network-prefix hsn --nodes 2 p2p_lat.py --iterations 100 --lg_max_message_size 12 --dragon

KNOWN ISSUE: If a --network-prefix argument is given that doesn’t actually exist, the Dragon runtime will enter a hung state. This will be fixed in future releases. For now, a ctrl+z and kill will be necessary to recover.

Dragon Logging

The Dragon runtime has extensive internal logging for its services. For performance reasons, this is disabled by default. However for debugging, various levels of logging can be requested via --log-level. The specific levels match those in Python’s logging module. As some examples:

# No runtime logging:
$ dragon p2p_lat.py --iterations 100 --lg_max_message_size 12 --dragon

# log messages of ERROR and CRITICAL level will be output to both stderr and dragon log
# file in pwd. No logging will be output to runtime actor-specific files
$ dragon -l ERROR program.py : Only file. No logging will be output the the actor files.

# INFO, WARNING, ERROR and CRITICAL level will be output to both stderr and dragon log
# file. No logging will be output to runtime actor-specific files
$ dragon -l INFO program.py

# INFO, WARNING, ERROR and CRITICAL level will be output to both stderr and dragon log
# file. The runtime actor log files will contain all log messages, up to and including
# DEBUG level.
$ dragon -l DEBUG program.py

# INFO, WARNING, ERROR and CRITICAL level will only be output to stderr. No dragon log
# file will be created
$ dragon -l stderr=INFO program.py

# ERROR and CRITICAL level will only be output to stderr. Log messages of INFO, WARNING,
# ERROR and CRITICAL level will only be output to the dragon log file.
$ dragon -l stderr=ERROR -l dragon_file=INFO program.py