75bf97b575
59e63bc minor 6233050 ok 14477f9 add namenode 75a6d34 add libhdfs opts e3c76bf minmum fix 8b3c435 chg 2035799 test code 7751b2b add debug 7690313 ok bd346b4 ok faba1dc add testload 6f7783e add testload e5f0340 ok 3ed9ec8 chg e552ac4 ask for more ram in am b2505e3 only stop nm when sucess bc696c9 add queue info f3e867e add option queue 5dc843c refactor fileio cd9c81b quick fix 1e23af2 add virtual destructor to iseekstream f165ffb fix hdfs 8cc6508 allow demo to pass in env fad4d69 ok 0fd6197 fix more 7423837 fix more d25de54 add temporal solution, run_yarn_prog.py e5a9e31 final attempt ed3bee8 add command back 0774000 add hdfs to resource 9b66e7e fix hadoop 6812f14 ok 08e1c16 change hadoop prefix back to hadoop home d6b6828 Update build.sh 146e069 bugfix: logical boundary for ring buffer 19cb685 ok 4cf3c13 Merge branch 'master' of ssh://github.com/tqchen/rabit 20daddb add tracker c57dad8 add ringbased passing and batch schedule 295d8a1 update 994cb02 add sge 014c866 OK git-subtree-dir: subtree/rabit git-subtree-split: 59e63bc1354c9ff516d72d9a6468f6c431627202
141 lines
6.6 KiB
Python
Executable File
141 lines
6.6 KiB
Python
Executable File
#!/usr/bin/env python
|
|
"""
|
|
This is a script to submit rabit job via Yarn
|
|
rabit will run as a Yarn application
|
|
"""
|
|
import argparse
|
|
import sys
|
|
import os
|
|
import time
|
|
import subprocess
|
|
import warnings
|
|
import rabit_tracker as tracker
|
|
|
|
WRAPPER_PATH = os.path.dirname(__file__) + '/../wrapper'
|
|
YARN_JAR_PATH = os.path.dirname(__file__) + '/../yarn/rabit-yarn.jar'
|
|
YARN_BOOT_PY = os.path.dirname(__file__) + '/../yarn/run_hdfs_prog.py'
|
|
|
|
if not os.path.exists(YARN_JAR_PATH):
|
|
warnings.warn("cannot find \"%s\", I will try to run build" % YARN_JAR_PATH)
|
|
cmd = 'cd %s;./build.sh' % (os.path.dirname(__file__) + '/../yarn/')
|
|
print cmd
|
|
subprocess.check_call(cmd, shell = True, env = os.environ)
|
|
assert os.path.exists(YARN_JAR_PATH), "failed to build rabit-yarn.jar, try it manually"
|
|
|
|
hadoop_binary = None
|
|
# code
|
|
hadoop_home = os.getenv('HADOOP_HOME')
|
|
|
|
if hadoop_home != None:
|
|
if hadoop_binary == None:
|
|
hadoop_binary = hadoop_home + '/bin/hadoop'
|
|
assert os.path.exists(hadoop_binary), "HADOOP_HOME does not contain the hadoop binary"
|
|
|
|
|
|
parser = argparse.ArgumentParser(description='Rabit script to submit rabit jobs to Yarn.')
|
|
parser.add_argument('-n', '--nworker', required=True, type=int,
|
|
help = 'number of worker proccess to be launched')
|
|
parser.add_argument('-hip', '--host_ip', default='auto', type=str,
|
|
help = 'host IP address if cannot be automatically guessed, specify the IP of submission machine')
|
|
parser.add_argument('-v', '--verbose', default=0, choices=[0, 1], type=int,
|
|
help = 'print more messages into the console')
|
|
parser.add_argument('-q', '--queue', default='default', type=str,
|
|
help = 'the queue we want to submit the job to')
|
|
parser.add_argument('-ac', '--auto_file_cache', default=1, choices=[0, 1], type=int,
|
|
help = 'whether automatically cache the files in the command to hadoop localfile, this is on by default')
|
|
parser.add_argument('-f', '--files', default = [], action='append',
|
|
help = 'the cached file list in mapreduce,'\
|
|
' the submission script will automatically cache all the files which appears in command'\
|
|
' This will also cause rewritten of all the file names in the command to current path,'\
|
|
' for example `../../kmeans ../kmeans.conf` will be rewritten to `./kmeans kmeans.conf`'\
|
|
' because the two files are cached to running folder.'\
|
|
' You may need this option to cache additional files.'\
|
|
' You can also use it to manually cache files when auto_file_cache is off')
|
|
parser.add_argument('--jobname', default='auto', help = 'customize jobname in tracker')
|
|
parser.add_argument('--tempdir', default='/tmp', help = 'temporary directory in HDFS that can be used to store intermediate results')
|
|
parser.add_argument('--vcores', default = 1, type=int,
|
|
help = 'number of vcpores to request in each mapper, set it if each rabit job is multi-threaded')
|
|
parser.add_argument('-mem', '--memory_mb', default=1024, type=int,
|
|
help = 'maximum memory used by the process. Guide: set it large (near mapred.cluster.max.map.memory.mb)'\
|
|
'if you are running multi-threading rabit,'\
|
|
'so that each node can occupy all the mapper slots in a machine for maximum performance')
|
|
parser.add_argument('--libhdfs-opts', default='-Xmx128m', type=str,
|
|
help = 'setting to be passed to libhdfs')
|
|
parser.add_argument('--name-node', default='default', type=str,
|
|
help = 'the namenode address of hdfs, libhdfs should connect to, normally leave it as default')
|
|
|
|
parser.add_argument('command', nargs='+',
|
|
help = 'command for rabit program')
|
|
args = parser.parse_args()
|
|
|
|
if args.jobname == 'auto':
|
|
args.jobname = ('Rabit[nworker=%d]:' % args.nworker) + args.command[0].split('/')[-1];
|
|
|
|
if hadoop_binary == None:
|
|
parser.add_argument('-hb', '--hadoop_binary', required = True,
|
|
help="path to hadoop binary file")
|
|
else:
|
|
parser.add_argument('-hb', '--hadoop_binary', default = hadoop_binary,
|
|
help="path to hadoop binary file")
|
|
|
|
args = parser.parse_args()
|
|
|
|
if args.jobname == 'auto':
|
|
args.jobname = ('Rabit[nworker=%d]:' % args.nworker) + args.command[0].split('/')[-1];
|
|
|
|
# detech hadoop version
|
|
(out, err) = subprocess.Popen('%s version' % args.hadoop_binary, shell = True, stdout=subprocess.PIPE).communicate()
|
|
out = out.split('\n')[0].split()
|
|
assert out[0] == 'Hadoop', 'cannot parse hadoop version string'
|
|
hadoop_version = out[1].split('.')
|
|
|
|
(classpath, err) = subprocess.Popen('%s classpath --glob' % args.hadoop_binary, shell = True, stdout=subprocess.PIPE).communicate()
|
|
|
|
if hadoop_version < 2:
|
|
print 'Current Hadoop Version is %s, rabit_yarn will need Yarn(Hadoop 2.0)' % out[1]
|
|
|
|
def submit_yarn(nworker, worker_args, worker_env):
|
|
fset = set([YARN_JAR_PATH, YARN_BOOT_PY])
|
|
if args.auto_file_cache != 0:
|
|
for i in range(len(args.command)):
|
|
f = args.command[i]
|
|
if os.path.exists(f):
|
|
fset.add(f)
|
|
if i == 0:
|
|
args.command[i] = './' + args.command[i].split('/')[-1]
|
|
else:
|
|
args.command[i] = './' + args.command[i].split('/')[-1]
|
|
if args.command[0].endswith('.py'):
|
|
flst = [WRAPPER_PATH + '/rabit.py',
|
|
WRAPPER_PATH + '/librabit_wrapper.so',
|
|
WRAPPER_PATH + '/librabit_wrapper_mock.so']
|
|
for f in flst:
|
|
if os.path.exists(f):
|
|
fset.add(f)
|
|
|
|
cmd = 'java -cp `%s classpath`:%s org.apache.hadoop.yarn.rabit.Client ' % (args.hadoop_binary, YARN_JAR_PATH)
|
|
env = os.environ.copy()
|
|
for k, v in worker_env.items():
|
|
env[k] = str(v)
|
|
env['rabit_cpu_vcores'] = str(args.vcores)
|
|
env['rabit_memory_mb'] = str(args.memory_mb)
|
|
env['rabit_world_size'] = str(args.nworker)
|
|
env['rabit_hdfs_opts'] = str(args.libhdfs_opts)
|
|
env['rabit_hdfs_namenode'] = str(args.name_node)
|
|
|
|
if args.files != None:
|
|
for flst in args.files:
|
|
for f in flst.split('#'):
|
|
fset.add(f)
|
|
for f in fset:
|
|
cmd += ' -file %s' % f
|
|
cmd += ' -jobname %s ' % args.jobname
|
|
cmd += ' -tempdir %s ' % args.tempdir
|
|
cmd += ' -queue %s ' % args.queue
|
|
cmd += (' '.join(['./run_hdfs_prog.py'] + args.command + worker_args))
|
|
if args.verbose != 0:
|
|
print cmd
|
|
subprocess.check_call(cmd, shell = True, env = env)
|
|
|
|
tracker.submit(args.nworker, [], fun_submit = submit_yarn, verbose = args.verbose, hostIP = args.host_ip)
|