/
multithread.py
549 lines (474 loc) · 21.7 KB
/
multithread.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
# -*- coding: utf-8 -*-
# coding=utf-8
# --------------------------------------------------------------------------
# Copyright (c) Microsoft Corporation. All rights reserved.
# Licensed under the MIT License. See License.txt in the project root for
# license information.
# --------------------------------------------------------------------------
"""
High performance multi-threaded module to up/download
Calls method in `core` with thread pool executor to ensure the network
is used to its maximum throughput.
Only implements upload and download of (massive) files and directory trees.
"""
from contextlib import closing
import glob
import logging
import os
import pickle
import time
from io import open
from .core import AzureDLPath, _fetch_range
from .exceptions import FileExistsError
from .transfer import ADLTransferClient
from .utils import commonprefix, datadir, read_block, tokenize
logger = logging.getLogger(__name__)
def save(instance, filename, keep=True):
if os.path.exists(filename):
all_downloads = load(filename)
else:
all_downloads = {}
if not instance.client._fstates.contains_all('finished') and keep:
all_downloads[instance._name] = instance
else:
all_downloads.pop(instance._name, None)
try:
# persist failure should not halt things
with open(filename, 'wb') as f:
pickle.dump(all_downloads, f)
except IOError:
logger.debug("Persist failed: %s" % filename)
def load(filename):
try:
return pickle.load(open(filename, 'rb'))
except:
return {}
class ADLDownloader(object):
""" Download remote file(s) using chunks and threads
Launches multiple threads for efficient downloading, with `chunksize`
assigned to each. The remote path can be a single file, a directory
of files or a glob pattern.
Parameters
----------
adlfs: ADL filesystem instance
rpath: str
remote path/globstring to use to find remote files. Recursive glob
patterns using `**` are not supported.
lpath: str
local path. If downloading a single file, will write to this specific
file, unless it is an existing directory, in which case a file is
created within it. If downloading multiple files, this is the root
directory to write within. Will create directories as required.
nthreads: int [None]
Number of threads to use. If None, uses the number of cores.
chunksize: int [2**28]
Number of bytes for a chunk. Large files are split into chunks. Files
smaller than this number will always be transferred in a single thread.
buffersize: int [2**22]
Number of bytes for internal buffer. This block cannot be bigger than
a chunk and cannot be smaller than a block.
blocksize: int [2**22]
Number of bytes for a block. Within each chunk, we write a smaller
block for each API call. This block cannot be bigger than a chunk.
client: ADLTransferClient [None]
Set an instance of ADLTransferClient when finer-grained control over
transfer parameters is needed. Ignores `nthreads` and `chunksize` set
by constructor.
run: bool [True]
Whether to begin executing immediately.
overwrite: bool [False]
Whether to forcibly overwrite existing files/directories. If False and
local path is a directory, will quit regardless if any files would be
overwritten or not. If True, only matching filenames are actually
overwritten.
See Also
--------
azure.datalake.store.transfer.ADLTransferClient
"""
def __init__(self, adlfs, rpath, lpath, nthreads=None, chunksize=2**28,
buffersize=2**22, blocksize=2**22, client=None, run=True,
overwrite=False, verbose=False):
# validate that the src exists and the current user has access to it
# this only validates access to the top level folder. If there are files
# or folders underneath it that the user does not have access to the download
# will fail on those files. We clean the path in case there are wildcards.
# In this case, we will always invalidate the cache for this check to
# do our best to ensure that the path exists as close to run time of the transfer as possible.
# Due to the nature of a distributed filesystem, the path could be deleted later during execution,
# at which point the transfer's behavior may be non-deterministic, but it will indicate an error.
if not adlfs.exists(AzureDLPath(rpath).globless_prefix, invalidate_cache=True):
raise FileNotFoundError('Data Lake item at path: {} either does not exist or the current user does not have permission to access it.'.format(rpath))
if client:
self.client = client
else:
self.client = ADLTransferClient(
adlfs,
transfer=get_chunk,
nthreads=nthreads,
chunksize=chunksize,
buffersize=buffersize,
blocksize=blocksize,
chunked=False,
verbose=verbose,
parent=self)
self._name = tokenize(adlfs, rpath, lpath, chunksize, blocksize)
self.rpath = rpath
self.lpath = lpath
self._overwrite = overwrite
existing_files = self._setup()
if existing_files:
raise FileExistsError('Overwrite was not specified and the following files exist, blocking the transfer operation. Please specify overwrite to overwrite these files during transfer: {}'.format(','.join(existing_files)))
if run:
self.run()
def save(self, keep=True):
""" Persist this download
Saves a copy of this transfer process in its current state to disk.
This is done automatically for a running transfer, so that as a chunk
is completed, this is reflected. Thus, if a transfer is interrupted,
e.g., by user action, the transfer can be restarted at another time.
All chunks that were not already completed will be restarted at that
time.
See methods ``load`` to retrieved saved transfers and ``run`` to
resume a stopped transfer.
Parameters
----------
keep: bool (True)
If True, transfer will be saved if some chunks remain to be
completed; the transfer will be sure to be removed otherwise.
"""
save(self, os.path.join(datadir, 'downloads'), keep)
@staticmethod
def load():
""" Load list of persisted transfers from disk, for possible resumption.
Returns
-------
A dictionary of download instances. The hashes are auto-
generated unique. The state of the chunks completed, errored, etc.,
can be seen in the status attribute. Instances can be resumed with
``run()``.
"""
return load(os.path.join(datadir, 'downloads'))
@staticmethod
def clear_saved():
""" Remove references to all persisted downloads.
"""
if os.path.exists(os.path.join(datadir, 'downloads')):
os.remove(os.path.join(datadir, 'downloads'))
@property
def hash(self):
return self._name
def _setup(self):
""" Create set of parameters to loop over
"""
if "*" not in self.rpath:
rfiles = self.client._adlfs.walk(self.rpath, details=True, invalidate_cache=True)
else:
rfiles = self.client._adlfs.glob(self.rpath, details=True, invalidate_cache=True)
if len(rfiles) > 1:
prefix = commonprefix([f['name'] for f in rfiles])
file_pairs = [(os.path.join(self.lpath, os.path.relpath(f['name'], prefix)), f)
for f in rfiles]
elif len(rfiles) == 1:
if os.path.exists(self.lpath) and os.path.isdir(self.lpath):
file_pairs = [(os.path.join(self.lpath, os.path.basename(rfiles[0]['name'])),
rfiles[0])]
else:
file_pairs = [(self.lpath, rfiles[0])]
else:
raise ValueError('No files to download')
# this property is used for internal validation
# and should not be referenced directly by public callers
self._file_pairs = file_pairs
existing_files = []
for lfile, rfile in file_pairs:
if not self._overwrite and os.path.exists(lfile):
existing_files.append(lfile)
else:
self.client.submit(rfile['name'], lfile, rfile['length'])
return existing_files
def run(self, nthreads=None, monitor=True):
""" Populate transfer queue and execute downloads
Parameters
----------
nthreads: int [None]
Override default nthreads, if given
monitor: bool [True]
To watch and wait (block) until completion.
"""
def touch(self, src, dst):
root = os.path.dirname(dst)
if not os.path.exists(root) and root:
# don't attempt to create current directory
logger.debug('Creating directory %s', root)
os.makedirs(root)
logger.debug('Creating empty file %s', dst)
with open(dst, 'wb'):
pass
self.client.run(nthreads, monitor, before_start=touch)
def active(self):
""" Return whether the downloader is active """
return self.client.active
def successful(self):
"""
Return whether the downloader completed successfully.
It will raise AssertionError if the downloader is active.
"""
return self.client.successful
def __str__(self):
return "<ADL Download: %s -> %s (%s)>" % (self.rpath, self.lpath,
self.client.status)
__repr__ = __str__
def get_chunk(adlfs, src, dst, offset, size, buffersize, blocksize,
shutdown_event=None, retries=10, delay=0.01, backoff=3):
""" Download a piece of a remote file and write locally
Internal function used by `download`.
"""
err = None
total_bytes_downloaded = 0
for i in range(retries):
try:
nbytes = 0
with closing(_fetch_range(adlfs.azure, src, start=offset,
end=offset+size, stream=True)) as response:
with open(dst, 'rb+') as fout:
fout.seek(offset)
for chunk in response.iter_content(chunk_size=blocksize):
if shutdown_event and shutdown_event.is_set():
return total_bytes_downloaded, None
if chunk:
nwritten = fout.write(chunk)
if nwritten:
nbytes += nwritten
logger.debug('Downloaded %s bytes to %s, byte offset %s', nbytes, dst, offset)
# There are certain cases where we will be throttled and recieve less than the expected amount of data.
# In these cases, instead of failing right away, instead indicate a retry is occuring and update offset and
# size to attempt another read to get the rest of the data. We will only do this if the amount of bytes read
# is less than size, because if somehow we recieved too much data we are not clear on how to proceed.
if nbytes < size:
errMsg = 'Did not recieve total bytes requested from server. This can be due to server side throttling and will be retried. Data Expected: {}. Data Received: {}.'.format(size, nbytes)
size -= nbytes
offset += nbytes
total_bytes_downloaded += nbytes
raise IOError(errMsg)
elif nbytes > size:
raise IOError('Received more bytes than expected from the server. Expected: {}. Received: {}.'.format(size, nbytes))
else:
total_bytes_downloaded += nbytes
return total_bytes_downloaded, None
except Exception as e:
err = e
logger.debug('Exception %s on ADL download on attempt: %s, retrying in %s seconds',
repr(err), i, delay, exc_info=True)
time.sleep(delay)
delay *= backoff
exception = RuntimeError('Max number of ADL retries exceeded: exception ' + repr(err))
logger.error('Download failed %s; %s', dst, repr(exception))
return total_bytes_downloaded, exception
class ADLUploader(object):
""" Upload local file(s) using chunks and threads
Launches multiple threads for efficient uploading, with `chunksize`
assigned to each. The path can be a single file, a directory
of files or a glob pattern.
Parameters
----------
adlfs: ADL filesystem instance
rpath: str
remote path to upload to; if multiple files, this is the dircetory
root to write within
lpath: str
local path. Can be single file, directory (in which case, upload
recursively) or glob pattern. Recursive glob patterns using `**` are
not supported.
nthreads: int [None]
Number of threads to use. If None, uses the number of cores.
chunksize: int [2**28]
Number of bytes for a chunk. Large files are split into chunks. Files
smaller than this number will always be transferred in a single thread.
buffersize: int [2**22]
Number of bytes for internal buffer. This block cannot be bigger than
a chunk and cannot be smaller than a block.
blocksize: int [2**22]
Number of bytes for a block. Within each chunk, we write a smaller
block for each API call. This block cannot be bigger than a chunk.
client: ADLTransferClient [None]
Set an instance of ADLTransferClient when finer-grained control over
transfer parameters is needed. Ignores `nthreads` and `chunksize`
set by constructor.
run: bool [True]
Whether to begin executing immediately.
overwrite: bool [False]
Whether to forcibly overwrite existing files/directories. If False and
remote path is a directory, will quit regardless if any files would be
overwritten or not. If True, only matching filenames are actually
overwritten.
See Also
--------
azure.datalake.store.transfer.ADLTransferClient
"""
def __init__(self, adlfs, rpath, lpath, nthreads=None, chunksize=2**28,
buffersize=2**22, blocksize=2**22, client=None, run=True,
overwrite=False, verbose=False):
if client:
self.client = client
else:
self.client = ADLTransferClient(
adlfs,
transfer=put_chunk,
merge=merge_chunks,
nthreads=nthreads,
chunksize=chunksize,
buffersize=buffersize,
blocksize=blocksize,
delimiter=None, # TODO: see utils.cs for what is required to support delimiters.
parent=self,
verbose=verbose,
unique_temporary=True)
self._name = tokenize(adlfs, rpath, lpath, chunksize, blocksize)
self.rpath = AzureDLPath(rpath)
self.lpath = lpath
self._overwrite = overwrite
existing_files = self._setup()
if existing_files:
raise FileExistsError('Overwrite was not specified and the following files exist, blocking the transfer operation. Please specify overwrite to overwrite these files during transfer: {}'.format(','.join(existing_files)))
if run:
self.run()
def save(self, keep=True):
""" Persist this upload
Saves a copy of this transfer process in its current state to disk.
This is done automatically for a running transfer, so that as a chunk
is completed, this is reflected. Thus, if a transfer is interrupted,
e.g., by user action, the transfer can be restarted at another time.
All chunks that were not already completed will be restarted at that
time.
See methods ``load`` to retrieved saved transfers and ``run`` to
resume a stopped transfer.
Parameters
----------
keep: bool (True)
If True, transfer will be saved if some chunks remain to be
completed; the transfer will be sure to be removed otherwise.
"""
save(self, os.path.join(datadir, 'uploads'), keep)
@staticmethod
def load():
""" Load list of persisted transfers from disk, for possible resumption.
Returns
-------
A dictionary of upload instances. The hashes are auto-
generated unique. The state of the chunks completed, errored, etc.,
can be seen in the status attribute. Instances can be resumed with
``run()``.
"""
return load(os.path.join(datadir, 'uploads'))
@staticmethod
def clear_saved():
""" Remove references to all persisted uploads.
"""
if os.path.exists(os.path.join(datadir, 'uploads')):
os.remove(os.path.join(datadir, 'uploads'))
@property
def hash(self):
return self._name
def _setup(self):
""" Create set of parameters to loop over
"""
if "*" not in self.lpath:
out = os.walk(self.lpath)
lfiles = sum(([os.path.join(dir, f) for f in fnames] for
(dir, _, fnames) in out), [])
if (not lfiles and os.path.exists(self.lpath) and
not os.path.isdir(self.lpath)):
lfiles = [self.lpath]
else:
lfiles = glob.glob(self.lpath)
if len(lfiles) > 1:
prefix = commonprefix(lfiles)
file_pairs = [(f, self.rpath / AzureDLPath(f).relative_to(prefix)) for f in lfiles]
elif lfiles:
if self.client._adlfs.exists(self.rpath, invalidate_cache=True) and \
self.client._adlfs.info(self.rpath, invalidate_cache=False)['type'] == "DIRECTORY":
file_pairs = [(lfiles[0], self.rpath / AzureDLPath(lfiles[0]).name)]
else:
file_pairs = [(lfiles[0], self.rpath)]
else:
raise ValueError('No files to upload')
# this property is used for internal validation
# and should not be referenced directly by public callers
self._file_pairs = file_pairs
existing_files = []
for lfile, rfile in file_pairs:
if not self._overwrite and self.client._adlfs.exists(rfile, invalidate_cache=False):
existing_files.append(rfile.as_posix())
else:
fsize = os.stat(lfile).st_size
self.client.submit(lfile, rfile, fsize)
return existing_files
def run(self, nthreads=None, monitor=True):
""" Populate transfer queue and execute downloads
Parameters
----------
nthreads: int [None]
Override default nthreads, if given
monitor: bool [True]
To watch and wait (block) until completion.
"""
self.client.run(nthreads, monitor)
def active(self):
""" Return whether the uploader is active """
return self.client.active
def successful(self):
"""
Return whether the uploader completed successfully.
It will raise AssertionError if the uploader is active.
"""
return self.client.successful
def __str__(self):
return "<ADL Upload: %s -> %s (%s)>" % (self.lpath, self.rpath,
self.client.status)
__repr__ = __str__
def put_chunk(adlfs, src, dst, offset, size, buffersize, blocksize, delimiter=None,
shutdown_event=None):
""" Upload a piece of a local file
Internal function used by `upload`.
"""
nbytes = 0
try:
with adlfs.open(dst, 'wb', blocksize=buffersize, delimiter=delimiter) as fout:
end = offset + size
miniblock = min(size, blocksize)
# For empty files there is no need to take the IO hit.
if size != 0:
with open(src, 'rb') as fin:
for o in range(offset, end, miniblock):
if shutdown_event and shutdown_event.is_set():
return nbytes, None
data = read_block(fin, o, miniblock, delimiter)
nbytes += fout.write(data)
except Exception as e:
exception = repr(e)
logger.error('Upload failed %s; %s', src, exception)
return nbytes, exception
logger.debug('Uploaded from %s, byte offset %s', src, offset)
return nbytes, None
def merge_chunks(adlfs, outfile, files, shutdown_event=None, overwrite=False):
try:
# note that it is assumed that only temp files from this run are in the segment folder created.
# so this call is optimized to instantly delete the temp folder on concat.
# if somehow the target file was created between the beginning of upload
# and concat, we will remove it if the user specified overwrite.
# here we must get the most up to date information from the service,
# instead of relying on the local cache to ensure that we know if
# the merge target already exists.
if adlfs.exists(outfile, invalidate_cache=True):
if overwrite:
adlfs.remove(outfile, True)
else:
raise FileExistsError(outfile)
adlfs.concat(outfile, files, delete_source=True)
except Exception as e:
exception = repr(e)
logger.error('Merged failed %s; %s', outfile, exception)
return exception
logger.debug('Merged %s', outfile)
adlfs.invalidate_cache(outfile)
return None