Runner - support 'pattern' in 'mpi' mode to run tasks in parallel (#430)

* add mpi-parallels mode

* update according to comments

* fix and update doc

* update

* merge into 'mpi' mode

* udpate according to comments

* fix testcases

* fix ansible

* regard pattern as field

* udpate

* fix flake8 version

* add flake8 range

* remove map-by from host config

* udpate comments

docs/superbench-config.mdx

@@ -401,6 +401,7 @@ Some attributes may only be suitable for specific mode.
 | `env`      |    ✓    |          ✓          |   ✓   |
 | `mca`      |    ✘    |          ✘          |   ✓   |
 | `parallel` |    ✓    |          ✘          |   ✘   |
+| `pattern`  |    ✘    |          ✘          |   ✓   |
 
 * accepted values: `local | torch.distributed | mpi`
 * default value: `local`
@@ -454,3 +455,12 @@ Whether run benchmarks in parallel (all ranks at the same time) or in sequence (
 Only available for `local` mode.
 
 * default value: `yes`
+
+### `pattern`
+
+Pattern variables to run benchmarks with nodes in specified traffic pattern combination, in a flatten key-value dictionary.
+Only available for `mpi` mode.
+
+Available variables in formatted string includes:
++ `name`
+    * accepted values: `all-nodes`

setup.py

@@ -188,7 +188,7 @@ def run(self):
             'test': [
                 'flake8-docstrings>=1.5.0',
                 'flake8-quotes>=3.2.0',
-                'flake8>=3.8.4',
+                'flake8>=3.8.4, <6.0.0',
                 'mypy>=0.800',
                 'pydocstyle>=5.1.1',
                 'pytest-cov>=2.11.1',

superbench/common/utils/__init__.py

@@ -9,6 +9,7 @@
 from superbench.common.utils.lazy_import import LazyImport
 from superbench.common.utils.process import run_command
 from superbench.common.utils.topo_aware import gen_topo_aware_config
+from superbench.common.utils.gen_traffic_pattern_config import gen_tarffic_pattern_host_group
 
 device_manager = LazyImport('superbench.common.utils.device_manager')
 
@@ -24,4 +25,5 @@
     'rotate_dir',
     'run_command',
     'gen_topo_aware_config',
+    'gen_tarffic_pattern_host_group',
 ]

superbench/common/utils/gen_traffic_pattern_config.py

+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+"""Utilities for traffic pattern config."""
+from superbench.common.utils import logger
+
+
+def gen_all_nodes_config(n):
+    """Generate all nodes config.
+
+    Args:
+        n (int): the number of participants.
+
+    Returns:
+        config (list): the generated config list, each item in the list is a str like "0,1,2,3".
+    """
+    config = []
+    if n <= 0:
+        logger.warning('n is not positive')
+        return config
+    config = [','.join(map(str, range(n)))]
+    return config
+
+
+def __convert_config_to_host_group(config, host_list):
+    """Convert config format to host node.
+
+    Args:
+        host_list (list): the list of hostnames read from hostfile.
+        config (list): the traffic pattern config.
+
+    Returns:
+        host_groups (list): the host groups converted from traffic pattern config.
+    """
+    host_groups = []
+    for item in config:
+        groups = item.strip().strip(';').split(';')
+        host_group = []
+        for group in groups:
+            hosts = []
+            for index in group.split(','):
+                hosts.append(host_list[int(index)])
+            host_group.append(hosts)
+        host_groups.append(host_group)
+    return host_groups
+
+
+def gen_tarffic_pattern_host_group(host_list, pattern):
+    """Generate host group from specified traffic pattern.
+
+    Args:
+        host_list (list): the list of hostnames read from hostfile.
+        pattern (DictConfig): the mpi pattern dict.
+
+    Returns:
+        host_group (list): the host group generated from traffic pattern.
+    """
+    config = []
+    n = len(host_list)
+    if pattern.name == 'all-nodes':
+        config = gen_all_nodes_config(n)
+    else:
+        logger.error('Unsupported traffic pattern: {}'.format(pattern.name))
+    host_group = __convert_config_to_host_group(config, host_list)
+    return host_group

superbench/runner/playbooks/check_env.yaml

@@ -56,9 +56,16 @@
         - '{{ workspace }}/sb.env'
         - /tmp/sb.env
       become: yes
-    - name: Updating Hostfile
+    - name: Updating Hostfile to Remote
       copy:
         content: "{{ sb_nodes | join('\n') }}\n"
         dest: '{{ workspace }}/hostfile'
         mode: 0644
       become: yes
+    - name: Generating Hostfile to Local
+      delegate_to: localhost
+      run_once: true
+      copy:
+        content: "{{ sb_nodes | join('\n') }}\n"
+        dest: '{{ output_dir }}/hostfile'
+        mode: 0644

superbench/runner/runner.py

@@ -14,7 +14,7 @@
 from joblib import Parallel, delayed
 from omegaconf import ListConfig, OmegaConf
 
-from superbench.common.utils import SuperBenchLogger, logger
+from superbench.common.utils import SuperBenchLogger, logger, gen_tarffic_pattern_host_group
 from superbench.runner.ansible import AnsibleClient
 from superbench.benchmarks import ReduceType, Reducer
 from superbench.monitor import MonitorRecord
@@ -109,6 +109,7 @@ def __get_mode_command(self, benchmark_name, mode, timeout=None):
             benchmark_name (str): Benchmark name.
             mode (DictConfig): Runner mode.
             timeout (int): The timeout value in seconds.
+            host_list (list): The specified Host node list.
 
         Return:
             str: Runner command.
@@ -143,12 +144,13 @@ def __get_mode_command(self, benchmark_name, mode, timeout=None):
                 'mpirun '    # use default OpenMPI in image
                 '-tag-output '    # tag mpi output with [jobid,rank]<stdout/stderr> prefix
                 '-allow-run-as-root '    # allow mpirun to run when executed by root user
-                '{host_list} '    # use prepared hostfile and launch {proc_num} processes on each node
+                '{host_list} '    # use prepared hostfile or specify nodes and launch {proc_num} processes on each node
                 '-bind-to numa '    # bind processes to numa
                 '{mca_list} {env_list} {command}'
             ).format(
-                host_list=f'-host localhost:{mode.proc_num}'
-                if mode.node_num == 1 else f'-hostfile hostfile -map-by ppr:{mode.proc_num}:node',
+                host_list=f'-host localhost:{mode.proc_num}' if mode.node_num == 1 else
+                f'-hostfile hostfile -map-by ppr:{mode.proc_num}:node' if mode.host_list is None else '-host ' +
+                ','.join(f'{host}:{mode.proc_num}' for host in mode.host_list),
                 mca_list=' '.join(f'-mca {k} {v}' for k, v in mode.mca.items()),
                 env_list=' '.join(
                     f'-x {k}={str(v).format(proc_rank=mode.proc_rank, proc_num=mode.proc_num)}'
@@ -444,7 +446,21 @@ def run(self):
                     )
                     ansible_rc = sum(rc_list)
                 elif mode.name == 'torch.distributed' or mode.name == 'mpi':
-                    ansible_rc = self._run_proc(benchmark_name, mode, {'proc_rank': 0})
+                    if not mode.pattern:
+                        ansible_rc = self._run_proc(benchmark_name, mode, {'proc_rank': 0})
+                    else:
+                        with open(self._output_path / 'hostfile', 'r') as f:
+                            host_list = f.read().splitlines()
+                        pattern_hostx = gen_tarffic_pattern_host_group(host_list, mode.pattern)
+                        for host_groups in pattern_hostx:
+                            para_rc_list = Parallel(n_jobs=len(host_groups))(
+                                delayed(self._run_proc)
+                                (benchmark_name, mode, vars={
+                                    'proc_rank': 0,
+                                    'host_list': host_group,
+                                }) for host_group in host_groups
+                            )
+                            ansible_rc = ansible_rc + sum(para_rc_list)
                 else:
                     logger.warning('Unknown mode %s.', mode.name)
                 if ansible_rc != 0:

tests/common/test_gen_tarffic_pattern_config.py

+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+"""Tests for traffic pattern config generation module."""
+import argparse
+import unittest
+
+from superbench.common.utils import gen_tarffic_pattern_host_group
+
+
+class GenConfigTest(unittest.TestCase):
+    """Test the utils for generating config."""
+    def test_gen_tarffic_pattern_host_group(self):
+        """Test the function of generating traffic pattern config from specified mode."""
+        # test under 8 nodes
+        hostx = ['node0', 'node1', 'node2', 'node3', 'node4', 'node5', 'node6', 'node7']
+        parser = argparse.ArgumentParser(
+            add_help=False,
+            usage=argparse.SUPPRESS,
+            allow_abbrev=False,
+        )
+        parser.add_argument(
+            '--name',
+            type=str,
+            default='all-nodes',
+            required=False,
+        )
+        pattern, _ = parser.parse_known_args()
+        expected_host_group = [[['node0', 'node1', 'node2', 'node3', 'node4', 'node5', 'node6', 'node7']]]
+        self.assertEqual(gen_tarffic_pattern_host_group(hostx, pattern), expected_host_group)

tests/runner/test_runner.py

@@ -196,12 +196,39 @@ def test_get_mode_command(self):
                     f'sb exec --output-dir {self.sb_output_dir} -c sb.config.yaml -C superbench.enable=foo'
                 ),
             },
+            {
+                'benchmark_name':
+                'foo',
+                'mode': {
+                    'name': 'mpi',
+                    'proc_num': 8,
+                    'proc_rank': 1,
+                    'mca': {},
+                    'pattern': {
+                        'name': 'all-nodes',
+                    },
+                    'env': {
+                        'PATH': None,
+                        'LD_LIBRARY_PATH': None,
+                    },
+                },
+                'expected_command': (
+                    'mpirun -tag-output -allow-run-as-root -host node0:8,node1:8 -bind-to numa '
+                    ' -x PATH -x LD_LIBRARY_PATH '
+                    f'sb exec --output-dir {self.sb_output_dir} -c sb.config.yaml -C superbench.enable=foo'
+                ),
+            },
         ]
+
         for test_case in test_cases:
             with self.subTest(msg='Testing with case', test_case=test_case):
+                mode = OmegaConf.create(test_case['mode'])
+                if 'pattern' in test_case['mode']:
+                    mode.update({'host_list': ['node0', 'node1']})
                 self.assertEqual(
                     self.runner._SuperBenchRunner__get_mode_command(
-                        test_case['benchmark_name'], OmegaConf.create(test_case['mode'])
+                        test_case['benchmark_name'],
+                        mode,
                     ), test_case['expected_command']
                 )
 
@@ -210,9 +237,14 @@ def test_get_mode_command(self):
                 index = test_case['expected_command'].find('sb exec')
                 expected_command = test_case['expected_command'][:index] + timeout_str + test_case['expected_command'][
                     index:]
+                mode = OmegaConf.create(test_case['mode'])
+                if 'pattern' in test_case['mode']:
+                    mode.update({'host_list': ['node0', 'node1']})
                 self.assertEqual(
                     self.runner._SuperBenchRunner__get_mode_command(
-                        test_case['benchmark_name'], OmegaConf.create(test_case['mode']), test_case['timeout']
+                        test_case['benchmark_name'],
+                        mode,
+                        test_case['timeout'],
                     ), expected_command
                 )