def download_and_sort(self, params):
dlname = Task.path("REDUCE_DL", Task.id)
msg("Reduce will be downloaded to %s" % dlname)
out_fd = AtomicFile(dlname, "w")
for url in self.inputs:
fd, sze, url = connect_input(url, params)
for k, v in fun_reader(fd, sze, url):
if " " in k:
err("Spaces are not allowed in keys "\
"with external sort.")
if "\0" in v:
err("Zero bytes are not allowed in "\
"values with external sort. "\
"Consider using base64 encoding.")
out_fd.write("%s %s\0" % (k, v))
out_fd.close()
msg("Reduce input downloaded ok")
msg("Starting external sort")
sortname = Task.path("REDUCE_SORTED", Task.id)
ensure_path(os.path.dirname(sortname))
cmd = ["sort", "-n", "-k", "1,1", "-z",\
"-t", " ", "-o", sortname, dlname]
proc = subprocess.Popen(cmd)
ret = proc.wait()
if ret:
err("Sorting %s to %s failed (%d)" %\
(dlname, sortname, ret))
msg("External sort done: %s" % sortname)
return self.multi_file_iterator([sortname], params, reader =\
lambda fd, sze, url:\
re_reader("(?s)(.*?) (.*?)\000", fd, sze, url))
def disk_sort(worker, input, filename, sort_buffer_size='10%'):
from os.path import getsize
from disco.comm import open_local
from disco.util import format_size
from disco.fileutils import AtomicFile
worker.send('MSG', "Downloading {0}".format(filename))
out_fd = AtomicFile(filename)
for key, value in input:
if not isinstance(key, bytes):
raise ValueError("Keys must be bytes for external sort", key)
if b'\xff' in key or b'\x00' in key:
raise ValueError("Cannot sort key with 0xFF or 0x00 bytes", key)
else:
# value pickled using protocol 0 will always be printable ASCII
out_fd.write(key + b'\xff')
out_fd.write(pickle_dumps(value, 0) + b'\x00')
out_fd.close()
worker.send('MSG',
"Downloaded {0:s} OK".format(format_size(getsize(filename))))
worker.send('MSG', "Sorting {0}...".format(filename))
unix_sort(filename, sort_buffer_size=sort_buffer_size)
worker.send('MSG', ("Finished sorting"))
fd = open_local(filename)
for k, v in re_reader(b"(?s)(.*?)\xff(.*?)\x00", fd, len(fd), fd.url):
yield k, pickle_loads(v)
def task_output_stream(stream, partition, url, params):
"""
An :func:`output_stream` which returns a handle to a task output.
The handle ensures that if a task fails, partially written data is ignored.
"""
from disco.fileutils import AtomicFile
return AtomicFile(url)
def disk_sort(worker, input, filename, sort_buffer_size='10%'):
from os.path import getsize
from disco.comm import open_local
from disco.fileutils import AtomicFile
from disco.worker.task_io import re_reader
if worker:
worker.send('MSG', "Downloading {0}".format(filename))
out_fd = AtomicFile(filename)
for key, value in input:
if not isinstance(key, bytes):
raise ValueError("Keys must be bytes for external sort", key)
if b'\xff' in key or b'\x00' in key:
raise ValueError("Cannot sort key with 0xFF or 0x00 bytes", key)
else:
# value pickled using protocol 0 will always be printable ASCII
out_fd.write(key + b'\xff')
out_fd.write(encode(pickle_dumps(value, 0)) + b'\x00')
out_fd.close()
if worker:
worker.send('MSG', "Downloaded {0:s} OK".format(format_size(getsize(filename))))
worker.send('MSG', "Sorting {0}...".format(filename))
unix_sort(filename, sort_buffer_size=sort_buffer_size)
if worker:
worker.send('MSG', ("Finished sorting"))
fd = open_local(filename)
for k, v in sort_reader(fd, fd.url):
yield k, bytes_to_str(decode(str_to_bytes(pickle_loads(v))))
def reduce_output_stream(stream, partition, url, params):
"""
An :func:`output_stream` which returns a handle to a reduce output.
The handle ensures that if a task fails, partially written data is ignored.
"""
from disco.fileutils import AtomicFile
path, url = Task.reduce_output
Task.blobs.append(path)
return AtomicFile(path, 'w'), url
def sorted_entries(self):
dlname = self.path('reduce-in-%d.dl' % self.id)
Message("Downloading %s" % dlname)
out_fd = AtomicFile(dlname, 'w')
for key, value in self.entries:
if not isinstance(key, str):
raise ValueError("Keys must be strings for external sort")
if '\xff' in key or '\x00' in key:
raise ValueError("Cannot sort keys with 0xFF or 0x00 bytes")
else:
# value pickled using protocol 0 will always be printable ASCII
out_fd.write('%s\xff%s\x00' % (key, cPickle.dumps(value, 0)))
out_fd.close()
Message("Downloaded OK")
self.disk_sort(dlname)
fd, size, url = comm.open_local(dlname)
for k, v in func.re_reader("(?s)(.*?)\xff(.*?)\x00", fd, size, url):
yield k, cPickle.loads(v)
def map_output_stream(stream, partition, url, params):
"""
An :func:`output_stream` which returns a handle to a partition output.
The handle ensures that if a task fails, partially written data is ignored.
"""
from disco.fileutils import AtomicFile
if Task.ispartitioned:
path, url = Task.partition_output(partition)
else:
path, url = Task.map_output(partition)
Task.blobs.append(path)
return AtomicFile(path, 'w'), url
def disk_sort(worker, input, filename, sort_buffer_size='10%'):
from os.path import getsize
from disco.comm import open_local
from disco.util import format_size
from disco.fileutils import AtomicFile
worker.send('MSG', "Downloading %s" % filename)
out_fd = AtomicFile(filename)
for key, value in input:
if not isinstance(key, str):
raise ValueError("Keys must be strings for external sort", key)
if '\xff' in key or '\x00' in key:
raise ValueError("Cannot sort key with 0xFF or 0x00 bytes", key)
else:
# value pickled using protocol 0 will always be printable ASCII
out_fd.write('%s\xff%s\x00' % (key, cPickle.dumps(value, 0)))
out_fd.close()
worker.send('MSG', "Downloaded %s OK" % format_size(getsize(filename)))
worker.send('MSG', "Sorting %s..." % filename)
unix_sort(filename, sort_buffer_size=sort_buffer_size)
worker.send('MSG', ("Finished sorting"))
fd = open_local(filename)
for k, v in delimited_reader(fd,
len(fd),
fd.url,
delimiter='\xff',
line_terminator='\x00'):
yield k, cPickle.loads(v)
def disk_sort(worker, input, filename, sort_buffer_size='10%'):
from os.path import getsize
from disco.comm import open_local
from disco.util import format_size
from disco.fileutils import AtomicFile
worker.send('MSG', "Downloading %s" % filename)
out_fd = AtomicFile(filename)
for key, value in input:
if not isinstance(key, str):
raise ValueError("Keys must be strings for external sort", key)
if '\xff' in key or '\x00' in key:
raise ValueError("Cannot sort key with 0xFF or 0x00 bytes", key)
else:
# value pickled using protocol 0 will always be printable ASCII
out_fd.write('%s\xff%s\x00' % (key, cPickle.dumps(value, 0)))
out_fd.close()
worker.send('MSG', "Downloaded %s OK" % format_size(getsize(filename)))
worker.send('MSG', "Sorting %s..." % filename)
unix_sort(filename, sort_buffer_size=sort_buffer_size)
worker.send('MSG', ("Finished sorting"))
fd = open_local(filename)
for k, v in delimited_reader(fd, len(fd), fd.url, delimiter='\xff', line_terminator='\x00'):
yield k, cPickle.loads(v)
def download_and_sort(self):
dlname = self.task.path('REDUCE_DL', self.task.id)
Message("Reduce will be downloaded to %s" % dlname)
out_fd = AtomicFile(dlname, 'w')
for url in self.inputs:
reader, sze, url = self.task.connect_input(url)
for k, v in reader:
self.sort_writer(out_fd, k, v)
out_fd.close()
Message("Reduce input downloaded ok")
Message("Starting external sort")
sortname = self.task.path('REDUCE_SORTED', self.task.id)
ensure_path(os.path.dirname(sortname))
cmd = ['sort', '-n', '-k', '1,1', '-T', '.',
'-z', '-t', '\xff', '-o', sortname, dlname]
proc = subprocess.Popen(cmd)
ret = proc.wait()
if ret:
TaskFailed("Sorting %s to %s failed (%d)" % (dlname, sortname, ret))
Message("External sort done: %s" % sortname)
return self.multi_file_iterator(self.sort_reader, inputs=[sortname])
def concat_input(cls, task, output_label, replicas):
output = AtomicFile(task.output_path(output_label))
BUFFER_SIZE = 1024 * 1024
for reps in replicas:
# Use only the first replica for now, since a set of one
# is the most common case.
# TODO: handle falling back to alternative replicas.
inp = open_url(reps[0])
buf = inp.read(BUFFER_SIZE)
while (len(buf) > 0):
output.write(buf)
buf = inp.read(BUFFER_SIZE)
inp.close()
output.close()
return output.path, output.size()
def concat_input(cls, task, output_label, replicas):
output = AtomicFile(task.output_path(output_label))
BUFFER_SIZE = 1024*1024
for reps in replicas:
# Use only the first replica for now, since a set of one
# is the most common case.
# TODO: handle falling back to alternative replicas.
inp = open_url(reps[0])
buf = inp.read(BUFFER_SIZE)
while (len(buf) > 0):
output.write(buf)
buf = inp.read(BUFFER_SIZE)
inp.close()
output.close()
return output.path, output.size()
def sorted_entries(self):
dlname = self.path('reduce-in-%d.dl' % self.id)
Status("Downloading %s" % dlname)
out_fd = AtomicFile(dlname, 'w')
for key, value in self.entries:
if not isinstance(key, str):
raise ValueError("Keys must be strings for external sort", key)
if '\xff' in key or '\x00' in key:
raise ValueError("Cannot sort key with 0xFF or 0x00 bytes", key)
else:
# value pickled using protocol 0 will always be printable ASCII
out_fd.write('%s\xff%s\x00' % (key, cPickle.dumps(value, 0)))
out_fd.close()
Status("Downloaded OK")
self.disk_sort(dlname)
fd, size, url = comm.open_local(dlname)
for k, v in func.re_reader("(?s)(.*?)\xff(.*?)\x00", fd, size, url):
yield k, cPickle.loads(v)
def disk_sort(input, filename, sort_keys, binaries=(), sort_buffer_size='10%', desc=False):
import ujson
from disco.comm import open_local
from disco.fileutils import AtomicFile
import base64
# import sys
# sys.path.append('/Library/Python/2.7/site-packages/pycharm-debug.egg')
# import pydevd
# pydevd.settrace('localhost', port=12999, stdoutToServer=True, stderrToServer=True)
out_fd = AtomicFile(filename)
key_types = None
MPT = ()
# print "SORTKEY: %s" % repr(sort_keys)
for key, _ in input:
if isinstance(key, (str, unicode)):
raise ValueError("Keys must be sequences", key)
# determine if the key is numeric
if key_types is None:
key_types = []
for kt in key:
try:
float(kt)
key_types.append('n')
except:
key_types.append('')
#serialize the key - encoded either as NULL, json, or b64 - note that
for i, each_key in enumerate(key):
if each_key is None:
ukey = b'\x00'
elif i in binaries and key_types[i] != 'n':
ukey = base64.b64encode(each_key)
else:
ukey = ujson.dumps(each_key)
out_fd.write(ukey)
out_fd.write(b'\xff')
out_fd.write('\n')
out_fd.flush()
out_fd.close()
unix_sort(filename,
[(sk, key_types[sk]) for sk in sort_keys],
sort_buffer_size=sort_buffer_size,
desc=desc)
fd = open_local(filename)
for k in sort_reader(fd, fd.url):
# yield [ujson.loads(key) if key != b'\x00' else None for key in k], MPT
rval = []
for i, key in enumerate(k):
if key == b'\x00':
rkey = None
elif i in binaries:
rkey = base64.b64decode(key)
else:
rkey = ujson.loads(key)
rval.append(rkey)
yield rval, MPT
def disk_sort(input, filename, sort_keys, binaries=(), sort_buffer_size='10%',
desc=False):
import ujson
from disco.comm import open_local
from disco.fileutils import AtomicFile
import base64
# import sys
# sys.path.append('/Library/Python/2.7/site-packages/pycharm-debug.egg')
# import pydevd
# pydevd.settrace('localhost', port=12999, stdoutToServer=True, stderrToServer=True)
out_fd = AtomicFile(filename)
key_types = None
MPT = ()
# print "SORTKEY: %s" % repr(sort_keys)
for key, _ in input:
if isinstance(key, (str, unicode)):
raise ValueError("Keys must be sequences", key)
# determine if the key is numeric
if key_types is None:
key_types = []
for kt in key:
try:
float(kt)
key_types.append('n')
except:
key_types.append('')
#serialize the key - encoded either as NULL, json, or b64 - note that
for i, each_key in enumerate(key):
if each_key is None:
ukey = b'\x00'
elif i in binaries and key_types[i] != 'n':
ukey = base64.b64encode(each_key)
else:
ukey = ujson.dumps(each_key)
out_fd.write(ukey)
out_fd.write(b'\xff')
out_fd.write('\n')
out_fd.flush()
out_fd.close()
unix_sort(filename,
[(sk, key_types[sk]) for sk in sort_keys],
sort_buffer_size=sort_buffer_size,
desc=desc)
fd = open_local(filename)
for k in sort_reader(fd, fd.url):
# yield [ujson.loads(key) if key != b'\x00' else None for key in k], MPT
rval = []
for i, key in enumerate(k):
if key == b'\x00':
rkey = None
elif i in binaries:
rkey = base64.b64decode(key)
else:
rkey = ujson.loads(key)
rval.append(rkey)
yield rval, MPT
