def gen_broadcast_path(work_dirs, uuid):
work_dir = decide_dir(work_dirs)
broadcast_dir = os.path.join(work_dir, 'broadcast')
mkdir_p(broadcast_dir)
uuid_path = '%s_%d' % (uuid, os.getpid())
broadcast_path = os.path.join(broadcast_dir, uuid_path)
return broadcast_path
def __init__(self, tracker, path):
try:
mkdir_p(path)
except:
pass
self.tracker = tracker
self.root = path
def __init__(self, tracker, path):
try:
mkdir_p(path)
except:
pass
self.tracker = tracker
self.root = path
def start(self):
if self.started:
return
self.started = True
logger.debug("start env in %s", os.getpid())
for d in self.workdir:
util.mkdir_p(d)
if 'TRACKER_ADDR' not in self.environ:
from dpark.tracker import TrackerServer
trackerServer = self.trackerServer = TrackerServer()
self.trackerServer.start()
self.register('TRACKER_ADDR', trackerServer.addr)
from dpark.tracker import TrackerClient
addr = self.get('TRACKER_ADDR')
self.trackerClient = TrackerClient(addr)
from dpark.cache import CacheTracker
self.cacheTracker = CacheTracker()
from dpark.shuffle import MapOutputTracker
self.mapOutputTracker = MapOutputTracker()
from dpark.shuffle import ParallelShuffleFetcher
self.shuffleFetcher = ParallelShuffleFetcher(2)
from dpark.broadcast import start_guide_manager, GUIDE_ADDR
if GUIDE_ADDR not in self.environ:
start_guide_manager()
logger.debug("env started")
def _get_path(self):
dirs = env.get('WORKDIR')
if not dirs:
raise RuntimeError('No available workdir')
path = os.path.join(dirs[0], 'mutable_dict')
if os.path.exists(path):
return path
st = os.statvfs(dirs[0])
ratio = st.f_bfree * 1.0 / st.f_blocks
if ratio >= 0.66:
mkdir_p(path)
return path
for d in dirs[1:]:
p = os.path.join(d, 'mutable_dict')
try:
os.makedirs(p)
os.symlink(p, path)
except OSError as e:
pass
return path
raise RuntimeError('Cannot find suitable workdir')
def getOutputFile(cls, shuffle_id, input_id, output_id, datasize=0):
"""
datasize < 0: disk first
datasize > 0: memfirst
datasize = 0: read only, use link
"""
shuffleDir = env.get('WORKDIR')
path = os.path.join(shuffleDir[0], str(shuffle_id), str(input_id))
mkdir_p(path)
p = os.path.join(path, str(output_id))
if datasize != 0 and len(shuffleDir) > 1:
use_disk = datasize < 0
if datasize > 0:
st = os.statvfs(path)
free = st.f_bfree * st.f_bsize
ratio = st.f_bfree * 1.0 / st.f_blocks
use_disk = free < max(datasize, 1 << 30) or ratio < 0.66
if use_disk:
d2 = os.path.join(random.choice(shuffleDir[1:]),
str(shuffle_id), str(input_id))
mkdir_p(d2)
p2 = os.path.join(d2, str(output_id))
if os.path.exists(p):
os.remove(p)
os.symlink(p2, p)
if os.path.islink(p2):
os.unlink(p2) # p == p2
return p2
return p
def registered(self, driver, executorInfo, frameworkInfo, agent_info):
try:
global Script
(
Script, cwd, python_path, osenv, self.parallel,
out_logger, err_logger, logLevel, args
) = marshal.loads(decode_data(executorInfo.data))
self.init_args = args
sys.path = python_path
os.environ.update(osenv)
setproctitle('[Executor]' + Script)
prefix = '[%s] ' % socket.gethostname()
fmt = '%(asctime)-15s [%(levelname)s] [%(name)-9s] %(message)s'
logging.basicConfig(format=fmt, level=logLevel)
r1 = self.stdout_redirect = Redirect(1, out_logger, prefix)
sys.stdout = r1.pipe_wfile
r2 = self.stderr_redirect = Redirect(2, err_logger, prefix)
sys.stderr = r2.pipe_wfile
if os.path.exists(cwd):
try:
os.chdir(cwd)
except Exception as e:
logger.warning('change cwd to %s failed: %s', cwd, e)
else:
logger.warning('cwd (%s) not exists', cwd)
self.workdir = args['WORKDIR']
main_workdir = self.workdir[0]
root = os.path.dirname(main_workdir)
if not os.path.exists(root):
os.mkdir(root)
os.chmod(root, 0o777) # because umask
mkdir_p(main_workdir)
self._try_flock(main_workdir)
args['SERVER_URI'] = startWebServer(main_workdir)
if 'MESOS_SLAVE_PID' in os.environ: # make unit test happy
setup_cleaner_process(self.workdir)
spawn(self.check_memory, driver)
spawn(self.replier, driver)
logger.debug('executor started at %s', agent_info.hostname)
except Exception as e:
import traceback
msg = traceback.format_exc()
logger.error('init executor failed: %s', msg)
raise
def start(self, isMaster, environ={}):
if self.started:
return
logger.debug("start env in %s: %s %s", os.getpid(), isMaster, environ)
self.isMaster = isMaster
if isMaster:
roots = conf.DPARK_WORK_DIR
if isinstance(roots, str):
roots = roots.split(',')
name = '%s-%s-%d' % (time.strftime("%Y%m%d-%H%M%S"),
socket.gethostname(), os.getpid())
self.workdir = [os.path.join(root, name) for root in roots]
try:
for d in self.workdir:
util.mkdir_p(d)
except OSError as e:
if environ.get('is_local', False):
raise e
self.environ['SERVER_URI'] = 'file://' + self.workdir[0]
self.environ['WORKDIR'] = self.workdir
self.environ['COMPRESS'] = util.COMPRESS
else:
self.environ.update(environ)
if self.environ['COMPRESS'] != util.COMPRESS:
raise Exception("no %s available" % self.environ['COMPRESS'])
self.ctx = zmq.Context()
from dpark.tracker import TrackerServer, TrackerClient
if isMaster:
self.trackerServer = TrackerServer()
self.trackerServer.start()
addr = self.trackerServer.addr
env.register('TrackerAddr', addr)
else:
addr = env.get('TrackerAddr')
self.trackerClient = TrackerClient(addr)
from dpark.cache import CacheTracker
self.cacheTracker = CacheTracker()
from dpark.shuffle import LocalFileShuffle, MapOutputTracker
LocalFileShuffle.initialize(isMaster)
self.mapOutputTracker = MapOutputTracker()
from dpark.shuffle import ParallelShuffleFetcher
self.shuffleFetcher = ParallelShuffleFetcher(2)
from dpark.broadcast import start_manager
start_manager(isMaster)
self.started = True
logger.debug("env started")
def write(self, path):
mkdir_p(path)
fd, file_name = tempfile.mkstemp(dir=path)
output_file = os.path.join(path, 'metadata')
try:
with os.fdopen(fd, 'wb+') as f:
f.write(pickle.dumps(self, -1))
os.rename(file_name, output_file)
finally:
try:
os.remove(file_name)
except OSError:
pass
def write(self, path):
mkdir_p(path)
fd, file_name = tempfile.mkstemp(dir=path)
output_file = os.path.join(path, 'metadata')
try:
with os.fdopen(fd, 'wb+') as f:
f.write(pickle.dumps(self, -1))
os.rename(file_name, output_file)
finally:
try:
os.remove(file_name)
except OSError:
pass
def getOutputFile(cls, shuffleId, inputId, outputId, datasize=0):
path = os.path.join(cls.shuffleDir[0], str(shuffleId), str(inputId))
mkdir_p(path)
p = os.path.join(path, str(outputId))
if datasize > 0 and len(cls.shuffleDir) > 1:
st = os.statvfs(path)
free = st.f_bfree * st.f_bsize
ratio = st.f_bfree * 1.0 / st.f_blocks
if free < max(datasize, 1<<30) or ratio < 0.66:
d2 = os.path.join(random.choice(cls.shuffleDir[1:]), str(shuffleId), str(inputId))
mkdir_p(d2)
p2 = os.path.join(d2, str(outputId))
os.symlink(p2, p)
if os.path.islink(p2):
os.unlink(p2) # p == p2
return p2
return p
def getOutputFile(cls, shuffle_id, input_id, output_id, datasize=0):
shuffleDir = env.get('WORKDIR')
path = os.path.join(shuffleDir[0], str(shuffle_id), str(input_id))
mkdir_p(path)
p = os.path.join(path, str(output_id))
if datasize > 0 and len(shuffleDir) > 1:
# datasize > 0 means its writing
st = os.statvfs(path)
free = st.f_bfree * st.f_bsize
ratio = st.f_bfree * 1.0 / st.f_blocks
if free < max(datasize, 1 << 30) or ratio < 0.66:
d2 = os.path.join(random.choice(shuffleDir[1:]),
str(shuffle_id), str(input_id))
mkdir_p(d2)
p2 = os.path.join(d2, str(output_id))
if os.path.exists(p):
os.remove(p)
os.symlink(p2, p)
if os.path.islink(p2):
os.unlink(p2) # p == p2
return p2
return p
def _get_path(self):
dirs = env.get('WORKDIR')
if not dirs:
raise RuntimeError('No available workdir')
path = os.path.join(dirs[0], 'mutable_dict')
if os.path.exists(path):
return path
st = os.statvfs(dirs[0])
ratio = st.f_bfree * 1.0 / st.f_blocks
if ratio >= 0.66:
mkdir_p(path)
return path
for d in dirs[1:]:
p = os.path.join(d, 'mutable_dict')
try:
os.makedirs(p)
os.symlink(p, path)
except OSError, e:
pass
return path
def getExecutorInfo(self, framework_id):
info = mesos_pb2.ExecutorInfo()
if hasattr(info, 'framework_id'):
info.framework_id.value = framework_id
if self.use_self_as_exec:
info.command.value = os.path.abspath(sys.argv[0])
info.executor_id.value = sys.argv[0]
else:
info.command.value = '%s %s' % (
sys.executable,
os.path.abspath(
os.path.join(os.path.dirname(__file__), 'executor.py')))
info.executor_id.value = "default"
v = info.command.environment.variables.add()
v.name = 'UID'
v.value = str(os.getuid())
v = info.command.environment.variables.add()
v.name = 'GID'
v.value = str(os.getgid())
if self.options.image and hasattr(info, 'container'):
info.container.type = mesos_pb2.ContainerInfo.DOCKER
info.container.docker.image = self.options.image
for path in ['/etc/passwd', '/etc/group']:
v = info.container.volumes.add()
v.host_path = v.container_path = path
v.mode = mesos_pb2.Volume.RO
for path in conf.MOOSEFS_MOUNT_POINTS:
v = info.container.volumes.add()
v.host_path = v.container_path = path
v.mode = mesos_pb2.Volume.RW
for path in conf.DPARK_WORK_DIR.split(','):
v = info.container.volumes.add()
v.host_path = v.container_path = path
v.mode = mesos_pb2.Volume.RW
if self.options.volumes:
for volume in self.options.volumes.split(','):
fields = volume.split(':')
if len(fields) == 3:
host_path, container_path, mode = fields
mode = mesos_pb2.Volume.RO if mode.lower(
) == 'ro' else mesos_pb2.Volume.RW
elif len(fields) == 2:
host_path, container_path = fields
mode = mesos_pb2.Volume.RW
elif len(fields) == 1:
container_path, = fields
host_path = ''
mode = mesos_pb2.Volume.RW
else:
raise Exception("cannot parse volume %s", volume)
mkdir_p(host_path)
v = info.container.volumes.add()
v.container_path = container_path
v.mode = mode
if host_path:
v.host_path = host_path
mem = info.resources.add()
mem.name = 'mem'
mem.type = 0 #mesos_pb2.Value.SCALAR
mem.scalar.value = EXECUTOR_MEMORY
Script = os.path.realpath(sys.argv[0])
if hasattr(info, 'name'):
info.name = Script
info.data = marshal.dumps(
(Script, os.getcwd(), sys.path, dict(os.environ),
self.task_per_node, self.out_logger, self.err_logger,
self.logLevel, env.environ))
return info
def getExecutorInfo(self, framework_id):
info = Dict()
info.framework_id.value = framework_id
if self.use_self_as_exec:
info.command.value = os.path.abspath(sys.argv[0])
info.executor_id.value = sys.argv[0]
else:
info.command.value = '%s %s' % (
sys.executable,
os.path.abspath(
os.path.join(os.path.dirname(__file__), 'executor.py')))
info.executor_id.value = 'default'
info.command.environment.variables = variables = []
v = Dict()
variables.append(v)
v.name = 'UID'
v.value = str(os.getuid())
v = Dict()
variables.append(v)
v.name = 'GID'
v.value = str(os.getgid())
if self.options.image:
info.container.type = 'DOCKER'
info.container.docker.image = self.options.image
info.container.volumes = volumes = []
for path in ['/etc/passwd', '/etc/group']:
v = Dict()
volumes.append(v)
v.host_path = v.container_path = path
v.mode = 'RO'
for path in conf.MOOSEFS_MOUNT_POINTS:
v = Dict()
volumes.append(v)
v.host_path = v.container_path = path
v.mode = 'RW'
for path in conf.DPARK_WORK_DIR.split(','):
v = Dict()
volumes.append(v)
v.host_path = v.container_path = path
v.mode = 'RW'
if self.options.volumes:
for volume in self.options.volumes.split(','):
fields = volume.split(':')
if len(fields) == 3:
host_path, container_path, mode = fields
mode = mode.upper()
assert mode in ('RO', 'RW')
elif len(fields) == 2:
host_path, container_path = fields
mode = 'RW'
elif len(fields) == 1:
container_path, = fields
host_path = ''
mode = 'RW'
else:
raise Exception('cannot parse volume %s', volume)
mkdir_p(host_path)
v = Dict()
volumes.append(v)
v.container_path = container_path
v.mode = mode
if host_path:
v.host_path = host_path
info.resources = resources = []
mem = Dict()
resources.append(mem)
mem.name = 'mem'
mem.type = 'SCALAR'
mem.scalar.value = EXECUTOR_MEMORY
cpus = Dict()
resources.append(cpus)
cpus.name = 'cpus'
cpus.type = 'SCALAR'
cpus.scalar.value = EXECUTOR_CPUS
Script = os.path.realpath(sys.argv[0])
info.name = Script
info.data = encode_data(
marshal.dumps((Script, os.getcwd(), sys.path, dict(os.environ),
self.task_per_node, self.out_logger,
self.err_logger, self.logLevel, env.environ)))
return info
try:
os.chdir(cwd)
except Exception, e:
logger.warning("change cwd to %s failed: %s", cwd, e)
else:
logger.warning("cwd (%s) not exists", cwd)
self.workdir = args['WORKDIR']
main_workdir = self.workdir[0]
root = os.path.dirname(main_workdir)
if not os.path.exists(root):
os.mkdir(root)
os.chmod(root, 0777) # because umask
mkdir_p(main_workdir)
self._try_flock(main_workdir)
args['SERVER_URI'] = startWebServer(main_workdir)
if 'MESOS_SLAVE_PID' in os.environ: # make unit test happy
setup_cleaner_process(self.workdir)
spawn(self.check_memory, driver)
logger.debug("executor started at %s", slaveInfo.hostname)
except Exception, e:
import traceback
msg = traceback.format_exc()
logger.error("init executor failed: %s", msg)
raise
def get_tmp(cls):
dirs = env.get('WORKDIR')
d = random.choice(dirs[1:]) if dirs[1:] else dirs[0]
mkdir_p(d)
return os.path.join(d, 'shuffle-%s.tmp' % uuid.uuid4().hex)
def registered(self, driver, executorInfo, frameworkInfo, agent_info):
try:
global Script
(Script, cwd, python_path, osenv, self.parallel, out_logger,
err_logger, logLevel, use_color,
args) = marshal.loads(decode_data(executorInfo.data))
sys.path = python_path
os.environ.update(osenv)
setproctitle('[Executor]' + Script)
prefix = formatter_message(
'{MAGENTA}[%s]{RESET} ' % socket.gethostname().ljust(10),
use_color)
init_dpark_logger(logLevel, use_color=use_color)
logging.root.setLevel(logLevel)
r1 = self.stdout_redirect = Redirect(1, out_logger, prefix)
sys.stdout = r1.pipe_wfile
r2 = self.stderr_redirect = Redirect(2, err_logger, prefix)
sys.stderr = r2.pipe_wfile
if os.path.exists(cwd):
try:
os.chdir(cwd)
except Exception as e:
logger.warning('change cwd to %s failed: %s', cwd, e)
else:
logger.warning('cwd (%s) not exists', cwd)
self.workdir = args['WORKDIR']
main_workdir = self.workdir[0]
root = os.path.dirname(main_workdir)
if not os.path.exists(root):
os.mkdir(root)
os.chmod(root, 0o777) # because umask
mkdir_p(main_workdir)
self._try_flock(main_workdir)
args['SERVER_URI'] = startWebServer(main_workdir)
if 'MESOS_SLAVE_PID' in os.environ: # make unit test happy
setup_cleaner_process(self.workdir)
spawn(self.check_memory, driver)
spawn(self.replier, driver)
env.environ.update(args)
from dpark.broadcast import start_download_manager
start_download_manager()
logger.debug('executor started at %s', agent_info.hostname)
except Exception as e:
import traceback
msg = traceback.format_exc()
logger.error('init executor failed: %s', msg)
raise
def getExecutorInfo(self, framework_id):
info = mesos_pb2.ExecutorInfo()
if hasattr(info, 'framework_id'):
info.framework_id.value = framework_id
if self.use_self_as_exec:
info.command.value = os.path.abspath(sys.argv[0])
info.executor_id.value = sys.argv[0]
else:
info.command.value = '%s %s' % (
sys.executable,
os.path.abspath(os.path.join(os.path.dirname(__file__), 'executor.py'))
)
info.executor_id.value = "default"
v = info.command.environment.variables.add()
v.name = 'UID'
v.value = str(os.getuid())
v = info.command.environment.variables.add()
v.name = 'GID'
v.value = str(os.getgid())
if self.options.image and hasattr(info, 'container'):
info.container.type = mesos_pb2.ContainerInfo.DOCKER
info.container.docker.image = self.options.image
for path in ['/etc/passwd', '/etc/group']:
v = info.container.volumes.add()
v.host_path = v.container_path = path
v.mode = mesos_pb2.Volume.RO
for path in conf.MOOSEFS_MOUNT_POINTS:
v = info.container.volumes.add()
v.host_path = v.container_path = path
v.mode = mesos_pb2.Volume.RW
for path in conf.DPARK_WORK_DIR.split(','):
v = info.container.volumes.add()
v.host_path = v.container_path = path
v.mode = mesos_pb2.Volume.RW
if self.options.volumes:
for volume in self.options.volumes.split(','):
fields = volume.split(':')
if len(fields) == 3:
host_path, container_path, mode = fields
mode = mesos_pb2.Volume.RO if mode.lower() == 'ro' else mesos_pb2.Volume.RW
elif len(fields) == 2:
host_path, container_path = fields
mode = mesos_pb2.Volume.RW
elif len(fields) == 1:
container_path, = fields
host_path = ''
mode = mesos_pb2.Volume.RW
else:
raise Exception("cannot parse volume %s", volume)
mkdir_p(host_path)
v = info.container.volumes.add()
v.container_path = container_path
v.mode = mode
if host_path:
v.host_path = host_path
mem = info.resources.add()
mem.name = 'mem'
mem.type = mesos_pb2.Value.SCALAR
mem.scalar.value = EXECUTOR_MEMORY
cpus = info.resources.add()
cpus.name = 'cpus'
cpus.type = mesos_pb2.Value.SCALAR
cpus.scalar.value = EXECUTOR_CPUS
Script = os.path.realpath(sys.argv[0])
if hasattr(info, 'name'):
info.name = Script
info.data = marshal.dumps((Script, os.getcwd(), sys.path, dict(os.environ),
self.task_per_node, self.out_logger, self.err_logger, self.logLevel, env.environ))
return info
def __init__(self, tracker, path):
mkdir_p(path)
self.tracker = tracker
self.root = path
try:
os.chdir(cwd)
except Exception, e:
logger.warning("change cwd to %s failed: %s", cwd, e)
else:
logger.warning("cwd (%s) not exists", cwd)
self.workdir = args['WORKDIR']
main_workdir = self.workdir[0]
root = os.path.dirname(main_workdir)
if not os.path.exists(root):
os.mkdir(root)
os.chmod(root, 0777) # because umask
mkdir_p(main_workdir)
self._try_flock(main_workdir)
args['SERVER_URI'] = startWebServer(main_workdir)
if 'MESOS_SLAVE_PID' in os.environ: # make unit test happy
setup_cleaner_process(self.workdir)
spawn(self.check_memory, driver)
logger.debug("executor started at %s", slaveInfo.hostname)
except Exception, e:
import traceback
msg = traceback.format_exc()
logger.error("init executor failed: %s", msg)
raise
def getExecutorInfo(self, framework_id):
info = Dict()
info.framework_id.value = framework_id
if self.use_self_as_exec:
info.command.value = os.path.abspath(sys.argv[0])
info.executor_id.value = sys.argv[0]
else:
info.command.value = '%s %s' % (
sys.executable,
os.path.abspath(
os.path.join(
os.path.dirname(__file__),
'executor.py'))
)
info.executor_id.value = 'default'
info.command.environment.variables = variables = []
v = Dict()
variables.append(v)
v.name = 'UID'
v.value = str(os.getuid())
v = Dict()
variables.append(v)
v.name = 'GID'
v.value = str(os.getgid())
if self.options.image:
info.container.type = 'DOCKER'
info.container.docker.image = self.options.image
info.container.volumes = volumes = []
for path in ['/etc/passwd', '/etc/group']:
v = Dict()
volumes.append(v)
v.host_path = v.container_path = path
v.mode = 'RO'
for path in conf.MOOSEFS_MOUNT_POINTS:
v = Dict()
volumes.append(v)
v.host_path = v.container_path = path
v.mode = 'RW'
for path in conf.DPARK_WORK_DIR.split(','):
v = Dict()
volumes.append(v)
v.host_path = v.container_path = path
v.mode = 'RW'
if self.options.volumes:
for volume in self.options.volumes.split(','):
fields = volume.split(':')
if len(fields) == 3:
host_path, container_path, mode = fields
mode = mode.upper()
assert mode in ('RO', 'RW')
elif len(fields) == 2:
host_path, container_path = fields
mode = 'RW'
elif len(fields) == 1:
container_path, = fields
host_path = ''
mode = 'RW'
else:
raise Exception('cannot parse volume %s', volume)
mkdir_p(host_path)
v = Dict()
volumes.append(v)
v.container_path = container_path
v.mode = mode
if host_path:
v.host_path = host_path
info.resources = resources = []
mem = Dict()
resources.append(mem)
mem.name = 'mem'
mem.type = 'SCALAR'
mem.scalar.value = EXECUTOR_MEMORY
cpus = Dict()
resources.append(cpus)
cpus.name = 'cpus'
cpus.type = 'SCALAR'
cpus.scalar.value = EXECUTOR_CPUS
Script = os.path.realpath(sys.argv[0])
info.name = Script
info.data = encode_data(marshal.dumps(
(
Script, os.getcwd(), sys.path, dict(os.environ),
self.task_per_node, self.out_logger, self.err_logger,
self.logLevel, env.environ
)
))
return info
