sdk/python/arvados/arvfile.py

   1 import functools
   2 import os
   3 import zlib
   4 import bz2
   5 import config
   6 import hashlib
   7 import threading
   8 import Queue
   9 import copy
  10 import errno
  11 import re
  12
  13 from .errors import KeepWriteError, AssertionError, ArgumentError
  14 from .keep import KeepLocator
  15 from ._normalize_stream import normalize_stream
  16 from ._ranges import locators_and_ranges, replace_range, Range
  17 from .retry import retry_method
  18
  19 def split(path):
  20     """split(path) -> streamname, filename
  21
  22     Separate the stream name and file name in a /-separated stream path and
  23     return a tuple (stream_name, file_name).  If no stream name is available,
  24     assume '.'.
  25
  26     """
  27     try:
  28         stream_name, file_name = path.rsplit('/', 1)
  29     except ValueError:  # No / in string
  30         stream_name, file_name = '.', path
  31     return stream_name, file_name
  32
  33 class _FileLikeObjectBase(object):
  34     def __init__(self, name, mode):
  35         self.name = name
  36         self.mode = mode
  37         self.closed = False
  38
  39     @staticmethod
  40     def _before_close(orig_func):
  41         @functools.wraps(orig_func)
  42         def before_close_wrapper(self, *args, **kwargs):
  43             if self.closed:
  44                 raise ValueError("I/O operation on closed stream file")
  45             return orig_func(self, *args, **kwargs)
  46         return before_close_wrapper
  47
  48     def __enter__(self):
  49         return self
  50
  51     def __exit__(self, exc_type, exc_value, traceback):
  52         try:
  53             self.close()
  54         except Exception:
  55             if exc_type is None:
  56                 raise
  57
  58     def close(self):
  59         self.closed = True
  60
  61
  62 class ArvadosFileReaderBase(_FileLikeObjectBase):
  63     def __init__(self, name, mode, num_retries=None):
  64         super(ArvadosFileReaderBase, self).__init__(name, mode)
  65         self._filepos = 0L
  66         self.num_retries = num_retries
  67         self._readline_cache = (None, None)
  68
  69     def __iter__(self):
  70         while True:
  71             data = self.readline()
  72             if not data:
  73                 break
  74             yield data
  75
  76     def decompressed_name(self):
  77         return re.sub('\.(bz2|gz)$', '', self.name)
  78
  79     @_FileLikeObjectBase._before_close
  80     def seek(self, pos, whence=os.SEEK_SET):
  81         if whence == os.SEEK_CUR:
  82             pos += self._filepos
  83         elif whence == os.SEEK_END:
  84             pos += self.size()
  85         self._filepos = min(max(pos, 0L), self.size())
  86
  87     def tell(self):
  88         return self._filepos
  89
  90     @_FileLikeObjectBase._before_close
  91     @retry_method
  92     def readall(self, size=2**20, num_retries=None):
  93         while True:
  94             data = self.read(size, num_retries=num_retries)
  95             if data == '':
  96                 break
  97             yield data
  98
  99     @_FileLikeObjectBase._before_close
 100     @retry_method
 101     def readline(self, size=float('inf'), num_retries=None):
 102         cache_pos, cache_data = self._readline_cache
 103         if self.tell() == cache_pos:
 104             data = [cache_data]
 105         else:
 106             data = ['']
 107         data_size = len(data[-1])
 108         while (data_size < size) and ('\n' not in data[-1]):
 109             next_read = self.read(2 ** 20, num_retries=num_retries)
 110             if not next_read:
 111                 break
 112             data.append(next_read)
 113             data_size += len(next_read)
 114         data = ''.join(data)
 115         try:
 116             nextline_index = data.index('\n') + 1
 117         except ValueError:
 118             nextline_index = len(data)
 119         nextline_index = min(nextline_index, size)
 120         self._readline_cache = (self.tell(), data[nextline_index:])
 121         return data[:nextline_index]
 122
 123     @_FileLikeObjectBase._before_close
 124     @retry_method
 125     def decompress(self, decompress, size, num_retries=None):
 126         for segment in self.readall(size, num_retries):
 127             data = decompress(segment)
 128             if data:
 129                 yield data
 130
 131     @_FileLikeObjectBase._before_close
 132     @retry_method
 133     def readall_decompressed(self, size=2**20, num_retries=None):
 134         self.seek(0)
 135         if self.name.endswith('.bz2'):
 136             dc = bz2.BZ2Decompressor()
 137             return self.decompress(dc.decompress, size,
 138                                    num_retries=num_retries)
 139         elif self.name.endswith('.gz'):
 140             dc = zlib.decompressobj(16+zlib.MAX_WBITS)
 141             return self.decompress(lambda segment: dc.decompress(dc.unconsumed_tail + segment),
 142                                    size, num_retries=num_retries)
 143         else:
 144             return self.readall(size, num_retries=num_retries)
 145
 146     @_FileLikeObjectBase._before_close
 147     @retry_method
 148     def readlines(self, sizehint=float('inf'), num_retries=None):
 149         data = []
 150         data_size = 0
 151         for s in self.readall(num_retries=num_retries):
 152             data.append(s)
 153             data_size += len(s)
 154             if data_size >= sizehint:
 155                 break
 156         return ''.join(data).splitlines(True)
 157
 158     def size(self):
 159         raise NotImplementedError()
 160
 161     def read(self, size, num_retries=None):
 162         raise NotImplementedError()
 163
 164     def readfrom(self, start, size, num_retries=None):
 165         raise NotImplementedError()
 166
 167
 168 class StreamFileReader(ArvadosFileReaderBase):
 169     class _NameAttribute(str):
 170         # The Python file API provides a plain .name attribute.
 171         # Older SDK provided a name() method.
 172         # This class provides both, for maximum compatibility.
 173         def __call__(self):
 174             return self
 175
 176     def __init__(self, stream, segments, name):
 177         super(StreamFileReader, self).__init__(self._NameAttribute(name), 'rb', num_retries=stream.num_retries)
 178         self._stream = stream
 179         self.segments = segments
 180
 181     def stream_name(self):
 182         return self._stream.name()
 183
 184     def size(self):
 185         n = self.segments[-1]
 186         return n.range_start + n.range_size
 187
 188     @_FileLikeObjectBase._before_close
 189     @retry_method
 190     def read(self, size, num_retries=None):
 191         """Read up to 'size' bytes from the stream, starting at the current file position"""
 192         if size == 0:
 193             return ''
 194
 195         data = ''
 196         available_chunks = locators_and_ranges(self.segments, self._filepos, size)
 197         if available_chunks:
 198             lr = available_chunks[0]
 199             data = self._stream.readfrom(lr.locator+lr.segment_offset,
 200                                           lr.segment_size,
 201                                           num_retries=num_retries)
 202
 203         self._filepos += len(data)
 204         return data
 205
 206     @_FileLikeObjectBase._before_close
 207     @retry_method
 208     def readfrom(self, start, size, num_retries=None):
 209         """Read up to 'size' bytes from the stream, starting at 'start'"""
 210         if size == 0:
 211             return ''
 212
 213         data = []
 214         for lr in locators_and_ranges(self.segments, start, size):
 215             data.append(self._stream.readfrom(lr.locator+lr.segment_offset, lr.segment_size,
 216                                               num_retries=num_retries))
 217         return ''.join(data)
 218
 219     def as_manifest(self):
 220         segs = []
 221         for r in self.segments:
 222             segs.extend(self._stream.locators_and_ranges(r.locator, r.range_size))
 223         return " ".join(normalize_stream(".", {self.name: segs})) + "\n"
 224
 225
 226 def synchronized(orig_func):
 227     @functools.wraps(orig_func)
 228     def synchronized_wrapper(self, *args, **kwargs):
 229         with self.lock:
 230             return orig_func(self, *args, **kwargs)
 231     return synchronized_wrapper
 232
 233 class _BufferBlock(object):
 234     """A stand-in for a Keep block that is in the process of being written.
 235
 236     Writers can append to it, get the size, and compute the Keep locator.
 237     There are three valid states:
 238
 239     WRITABLE
 240       Can append to block.
 241
 242     PENDING
 243       Block is in the process of being uploaded to Keep, append is an error.
 244
 245     COMMITTED
 246       The block has been written to Keep, its internal buffer has been
 247       released, fetching the block will fetch it via keep client (since we
 248       discarded the internal copy), and identifiers referring to the BufferBlock
 249       can be replaced with the block locator.
 250
 251     """
 252
 253     WRITABLE = 0
 254     PENDING = 1
 255     COMMITTED = 2
 256
 257     def __init__(self, blockid, starting_capacity, owner):
 258         """
 259         :blockid:
 260           the identifier for this block
 261
 262         :starting_capacity:
 263           the initial buffer capacity
 264
 265         :owner:
 266           ArvadosFile that owns this block
 267
 268         """
 269         self.blockid = blockid
 270         self.buffer_block = bytearray(starting_capacity)
 271         self.buffer_view = memoryview(self.buffer_block)
 272         self.write_pointer = 0
 273         self._state = _BufferBlock.WRITABLE
 274         self._locator = None
 275         self.owner = owner
 276         self.lock = threading.Lock()
 277
 278     @synchronized
 279     def append(self, data):
 280         """Append some data to the buffer.
 281
 282         Only valid if the block is in WRITABLE state.  Implements an expanding
 283         buffer, doubling capacity as needed to accomdate all the data.
 284
 285         """
 286         if self._state == _BufferBlock.WRITABLE:
 287             while (self.write_pointer+len(data)) > len(self.buffer_block):
 288                 new_buffer_block = bytearray(len(self.buffer_block) * 2)
 289                 new_buffer_block[0:self.write_pointer] = self.buffer_block[0:self.write_pointer]
 290                 self.buffer_block = new_buffer_block
 291                 self.buffer_view = memoryview(self.buffer_block)
 292             self.buffer_view[self.write_pointer:self.write_pointer+len(data)] = data
 293             self.write_pointer += len(data)
 294             self._locator = None
 295         else:
 296             raise AssertionError("Buffer block is not writable")
 297
 298     @synchronized
 299     def set_state(self, nextstate, loc=None):
 300         if ((self._state == _BufferBlock.WRITABLE and nextstate == _BufferBlock.PENDING) or
 301             (self._state == _BufferBlock.PENDING and nextstate == _BufferBlock.COMMITTED)):
 302             self._state = nextstate
 303             if self._state == _BufferBlock.COMMITTED:
 304                 self._locator = loc
 305                 self.buffer_view = None
 306                 self.buffer_block = None
 307         else:
 308             raise AssertionError("Invalid state change from %s to %s" % (self.state, nextstate))
 309
 310     @synchronized
 311     def state(self):
 312         return self._state
 313
 314     def size(self):
 315         """The amount of data written to the buffer."""
 316         return self.write_pointer
 317
 318     @synchronized
 319     def locator(self):
 320         """The Keep locator for this buffer's contents."""
 321         if self._locator is None:
 322             self._locator = "%s+%i" % (hashlib.md5(self.buffer_view[0:self.write_pointer]).hexdigest(), self.size())
 323         return self._locator
 324
 325     @synchronized
 326     def clone(self, new_blockid, owner):
 327         if self._state == _BufferBlock.COMMITTED:
 328             raise AssertionError("Can only duplicate a writable or pending buffer block")
 329         bufferblock = _BufferBlock(new_blockid, self.size(), owner)
 330         bufferblock.append(self.buffer_view[0:self.size()])
 331         return bufferblock
 332
 333
 334 class NoopLock(object):
 335     def __enter__(self):
 336         return self
 337
 338     def __exit__(self, exc_type, exc_value, traceback):
 339         pass
 340
 341     def acquire(self, blocking=False):
 342         pass
 343
 344     def release(self):
 345         pass
 346
 347
 348 def must_be_writable(orig_func):
 349     @functools.wraps(orig_func)
 350     def must_be_writable_wrapper(self, *args, **kwargs):
 351         if not self.writable():
 352             raise IOError((errno.EROFS, "Collection must be writable."))
 353         return orig_func(self, *args, **kwargs)
 354     return must_be_writable_wrapper
 355
 356
 357 class _BlockManager(object):
 358     """BlockManager handles buffer blocks.
 359
 360     Also handles background block uploads, and background block prefetch for a
 361     Collection of ArvadosFiles.
 362
 363     """
 364     def __init__(self, keep):
 365         """keep: KeepClient object to use"""
 366         self._keep = keep
 367         self._bufferblocks = {}
 368         self._put_queue = None
 369         self._put_errors = None
 370         self._put_threads = None
 371         self._prefetch_queue = None
 372         self._prefetch_threads = None
 373         self.lock = threading.Lock()
 374         self.prefetch_enabled = True
 375         self.num_put_threads = 2
 376         self.num_get_threads = 2
 377
 378     @synchronized
 379     def alloc_bufferblock(self, blockid=None, starting_capacity=2**14, owner=None):
 380         """Allocate a new, empty bufferblock in WRITABLE state and return it.
 381
 382         :blockid:
 383           optional block identifier, otherwise one will be automatically assigned
 384
 385         :starting_capacity:
 386           optional capacity, otherwise will use default capacity
 387
 388         :owner:
 389           ArvadosFile that owns this block
 390
 391         """
 392         if blockid is None:
 393             blockid = "bufferblock%i" % len(self._bufferblocks)
 394         bufferblock = _BufferBlock(blockid, starting_capacity=starting_capacity, owner=owner)
 395         self._bufferblocks[bufferblock.blockid] = bufferblock
 396         return bufferblock
 397
 398     @synchronized
 399     def dup_block(self, block, owner):
 400         """Create a new bufferblock initialized with the content of an existing bufferblock.
 401
 402         :block:
 403           the buffer block to copy.
 404
 405         :owner:
 406           ArvadosFile that owns the new block
 407
 408         """
 409         new_blockid = "bufferblock%i" % len(self._bufferblocks)
 410         bufferblock = block.clone(new_blockid, owner)
 411         self._bufferblocks[bufferblock.blockid] = bufferblock
 412         return bufferblock
 413
 414     @synchronized
 415     def is_bufferblock(self, locator):
 416         return locator in self._bufferblocks
 417
 418     @synchronized
 419     def stop_threads(self):
 420         """Shut down and wait for background upload and download threads to finish."""
 421
 422         if self._put_threads is not None:
 423             for t in self._put_threads:
 424                 self._put_queue.put(None)
 425             for t in self._put_threads:
 426                 t.join()
 427         self._put_threads = None
 428         self._put_queue = None
 429         self._put_errors = None
 430
 431         if self._prefetch_threads is not None:
 432             for t in self._prefetch_threads:
 433                 self._prefetch_queue.put(None)
 434             for t in self._prefetch_threads:
 435                 t.join()
 436         self._prefetch_threads = None
 437         self._prefetch_queue = None
 438
 439     def commit_bufferblock(self, block):
 440         """Initiate a background upload of a bufferblock.
 441
 442         This will block if the upload queue is at capacity, otherwise it will
 443         return immediately.
 444
 445         """
 446
 447         def commit_bufferblock_worker(self):
 448             """Background uploader thread."""
 449
 450             while True:
 451                 try:
 452                     bufferblock = self._put_queue.get()
 453                     if bufferblock is None:
 454                         return
 455                     loc = self._keep.put(bufferblock.buffer_view[0:bufferblock.write_pointer].tobytes())
 456                     bufferblock.set_state(_BufferBlock.COMMITTED, loc)
 457
 458                 except Exception as e:
 459                     self._put_errors.put((bufferblock.locator(), e))
 460                 finally:
 461                     if self._put_queue is not None:
 462                         self._put_queue.task_done()
 463
 464         with self.lock:
 465             if self._put_threads is None:
 466                 # Start uploader threads.
 467
 468                 # If we don't limit the Queue size, the upload queue can quickly
 469                 # grow to take up gigabytes of RAM if the writing process is
 470                 # generating data more quickly than it can be send to the Keep
 471                 # servers.
 472                 #
 473                 # With two upload threads and a queue size of 2, this means up to 4
 474                 # blocks pending.  If they are full 64 MiB blocks, that means up to
 475                 # 256 MiB of internal buffering, which is the same size as the
 476                 # default download block cache in KeepClient.
 477                 self._put_queue = Queue.Queue(maxsize=2)
 478                 self._put_errors = Queue.Queue()
 479
 480                 self._put_threads = []
 481                 for i in xrange(0, self.num_put_threads):
 482                     thread = threading.Thread(target=commit_bufferblock_worker, args=(self,))
 483                     self._put_threads.append(thread)
 484                     thread.daemon = True
 485                     thread.start()
 486
 487         if block.state() == _BufferBlock.WRITABLE:
 488             # Mark the block as PENDING so to disallow any more appends.
 489             block.set_state(_BufferBlock.PENDING)
 490             self._put_queue.put(block)
 491
 492     @synchronized
 493     def get_bufferblock(self, locator):
 494         return self._bufferblocks.get(locator)
 495
 496     def get_block_contents(self, locator, num_retries, cache_only=False):
 497         """Fetch a block.
 498
 499         First checks to see if the locator is a BufferBlock and return that, if
 500         not, passes the request through to KeepClient.get().
 501
 502         """
 503         with self.lock:
 504             if locator in self._bufferblocks:
 505                 bufferblock = self._bufferblocks[locator]
 506                 if bufferblock.state() != _BufferBlock.COMMITTED:
 507                     return bufferblock.buffer_view[0:bufferblock.write_pointer].tobytes()
 508                 else:
 509                     locator = bufferblock._locator
 510         if cache_only:
 511             return self._keep.get_from_cache(locator)
 512         else:
 513             return self._keep.get(locator, num_retries=num_retries)
 514
 515     def commit_all(self):
 516         """Commit all outstanding buffer blocks.
 517
 518         Unlike commit_bufferblock(), this is a synchronous call, and will not
 519         return until all buffer blocks are uploaded.  Raises
 520         KeepWriteError() if any blocks failed to upload.
 521
 522         """
 523         with self.lock:
 524             items = self._bufferblocks.items()
 525
 526         for k,v in items:
 527             v.owner.flush()
 528
 529         with self.lock:
 530             if self._put_queue is not None:
 531                 self._put_queue.join()
 532
 533                 if not self._put_errors.empty():
 534                     err = []
 535                     try:
 536                         while True:
 537                             err.append(self._put_errors.get(False))
 538                     except Queue.Empty:
 539                         pass
 540                     raise KeepWriteError("Error writing some blocks", err, label="block")
 541
 542     def block_prefetch(self, locator):
 543         """Initiate a background download of a block.
 544
 545         This assumes that the underlying KeepClient implements a block cache,
 546         so repeated requests for the same block will not result in repeated
 547         downloads (unless the block is evicted from the cache.)  This method
 548         does not block.
 549
 550         """
 551
 552         if not self.prefetch_enabled:
 553             return
 554
 555         def block_prefetch_worker(self):
 556             """The background downloader thread."""
 557             while True:
 558                 try:
 559                     b = self._prefetch_queue.get()
 560                     if b is None:
 561                         return
 562                     self._keep.get(b)
 563                 except Exception:
 564                     pass
 565
 566         with self.lock:
 567             if locator in self._bufferblocks:
 568                 return
 569             if self._prefetch_threads is None:
 570                 self._prefetch_queue = Queue.Queue()
 571                 self._prefetch_threads = []
 572                 for i in xrange(0, self.num_get_threads):
 573                     thread = threading.Thread(target=block_prefetch_worker, args=(self,))
 574                     self._prefetch_threads.append(thread)
 575                     thread.daemon = True
 576                     thread.start()
 577         self._prefetch_queue.put(locator)
 578
 579
 580 class ArvadosFile(object):
 581     """Represent a file in a Collection.
 582
 583     ArvadosFile manages the underlying representation of a file in Keep as a
 584     sequence of segments spanning a set of blocks, and implements random
 585     read/write access.
 586
 587     This object may be accessed from multiple threads.
 588
 589     """
 590
 591     def __init__(self, parent, stream=[], segments=[]):
 592         """
 593         ArvadosFile constructor.
 594
 595         :stream:
 596           a list of Range objects representing a block stream
 597
 598         :segments:
 599           a list of Range objects representing segments
 600         """
 601         self.parent = parent
 602         self._modified = True
 603         self._segments = []
 604         self.lock = parent.root_collection().lock
 605         for s in segments:
 606             self._add_segment(stream, s.locator, s.range_size)
 607         self._current_bblock = None
 608
 609     def writable(self):
 610         return self.parent.writable()
 611
 612     @synchronized
 613     def segments(self):
 614         return copy.copy(self._segments)
 615
 616     @synchronized
 617     def clone(self, new_parent):
 618         """Make a copy of this file."""
 619         cp = ArvadosFile(new_parent)
 620         cp.replace_contents(self)
 621         return cp
 622
 623     @must_be_writable
 624     @synchronized
 625     def replace_contents(self, other):
 626         """Replace segments of this file with segments from another `ArvadosFile` object."""
 627
 628         map_loc = {}
 629         self._segments = []
 630         for other_segment in other.segments():
 631             new_loc = other_segment.locator
 632             if other.parent._my_block_manager().is_bufferblock(other_segment.locator):
 633                 if other_segment.locator not in map_loc:
 634                     bufferblock = other.parent._my_block_manager().get_bufferblock(other_segment.locator)
 635                     if bufferblock.state() != _BufferBlock.WRITABLE:
 636                         map_loc[other_segment.locator] = bufferblock.locator()
 637                     else:
 638                         map_loc[other_segment.locator] = self.parent._my_block_manager().dup_block(bufferblock, self).blockid
 639                 new_loc = map_loc[other_segment.locator]
 640
 641             self._segments.append(Range(new_loc, other_segment.range_start, other_segment.range_size, other_segment.segment_offset))
 642
 643         self._modified = True
 644
 645     def __eq__(self, other):
 646         if other is self:
 647             return True
 648         if not isinstance(other, ArvadosFile):
 649             return False
 650
 651         othersegs = other.segments()
 652         with self.lock:
 653             if len(self._segments) != len(othersegs):
 654                 return False
 655             for i in xrange(0, len(othersegs)):
 656                 seg1 = self._segments[i]
 657                 seg2 = othersegs[i]
 658                 loc1 = seg1.locator
 659                 loc2 = seg2.locator
 660
 661                 if self.parent._my_block_manager().is_bufferblock(loc1):
 662                     loc1 = self.parent._my_block_manager().get_bufferblock(loc1).locator()
 663
 664                 if other.parent._my_block_manager().is_bufferblock(loc2):
 665                     loc2 = other.parent._my_block_manager().get_bufferblock(loc2).locator()
 666
 667                 if (KeepLocator(loc1).stripped() != KeepLocator(loc2).stripped() or
 668                     seg1.range_start != seg2.range_start or
 669                     seg1.range_size != seg2.range_size or
 670                     seg1.segment_offset != seg2.segment_offset):
 671                     return False
 672
 673         return True
 674
 675     def __ne__(self, other):
 676         return not self.__eq__(other)
 677
 678     @synchronized
 679     def set_unmodified(self):
 680         """Clear the modified flag"""
 681         self._modified = False
 682
 683     @synchronized
 684     def modified(self):
 685         """Test the modified flag"""
 686         return self._modified
 687
 688     @must_be_writable
 689     @synchronized
 690     def truncate(self, size):
 691         """Shrink the size of the file.
 692
 693         If `size` is less than the size of the file, the file contents after
 694         `size` will be discarded.  If `size` is greater than the current size
 695         of the file, an IOError will be raised.
 696
 697         """
 698         if size < self.size():
 699             new_segs = []
 700             for r in self._segments:
 701                 range_end = r.range_start+r.range_size
 702                 if r.range_start >= size:
 703                     # segment is past the trucate size, all done
 704                     break
 705                 elif size < range_end:
 706                     nr = Range(r.locator, r.range_start, size - r.range_start, 0)
 707                     nr.segment_offset = r.segment_offset
 708                     new_segs.append(nr)
 709                     break
 710                 else:
 711                     new_segs.append(r)
 712
 713             self._segments = new_segs
 714             self._modified = True
 715         elif size > self.size():
 716             raise IOError("truncate() does not support extending the file size")
 717
 718
 719     def readfrom(self, offset, size, num_retries, exact=False):
 720         """Read up to `size` bytes from the file starting at `offset`.
 721
 722         :exact:
 723          If False (default), return less data than requested if the read
 724          crosses a block boundary and the next block isn't cached.  If True,
 725          only return less data than requested when hitting EOF.
 726         """
 727
 728         with self.lock:
 729             if size == 0 or offset >= self.size():
 730                 return ''
 731             prefetch = locators_and_ranges(self._segments, offset, size + config.KEEP_BLOCK_SIZE)
 732             readsegs = locators_and_ranges(self._segments, offset, size)
 733
 734         for lr in prefetch:
 735             self.parent._my_block_manager().block_prefetch(lr.locator)
 736
 737         data = []
 738         for lr in readsegs:
 739             block = self.parent._my_block_manager().get_block_contents(lr.locator, num_retries=num_retries, cache_only=(bool(data) and not exact))
 740             if block:
 741                 data.append(block[lr.segment_offset:lr.segment_offset+lr.segment_size])
 742             else:
 743                 break
 744         return ''.join(data)
 745
 746     def _repack_writes(self, num_retries):
 747         """Test if the buffer block has more data than actual segments.
 748
 749         This happens when a buffered write over-writes a file range written in
 750         a previous buffered write.  Re-pack the buffer block for efficiency
 751         and to avoid leaking information.
 752
 753         """
 754         segs = self._segments
 755
 756         # Sum up the segments to get the total bytes of the file referencing
 757         # into the buffer block.
 758         bufferblock_segs = [s for s in segs if s.locator == self._current_bblock.blockid]
 759         write_total = sum([s.range_size for s in bufferblock_segs])
 760
 761         if write_total < self._current_bblock.size():
 762             # There is more data in the buffer block than is actually accounted for by segments, so
 763             # re-pack into a new buffer by copying over to a new buffer block.
 764             contents = self.parent._my_block_manager().get_block_contents(self._current_bblock.blockid, num_retries)
 765             new_bb = self.parent._my_block_manager().alloc_bufferblock(self._current_bblock.blockid, starting_capacity=write_total, owner=self)
 766             for t in bufferblock_segs:
 767                 new_bb.append(contents[t.segment_offset:t.segment_offset+t.range_size])
 768                 t.segment_offset = new_bb.size() - t.range_size
 769
 770             self._current_bblock = new_bb
 771
 772     @must_be_writable
 773     @synchronized
 774     def writeto(self, offset, data, num_retries):
 775         """Write `data` to the file starting at `offset`.
 776
 777         This will update existing bytes and/or extend the size of the file as
 778         necessary.
 779
 780         """
 781         if len(data) == 0:
 782             return
 783
 784         if offset > self.size():
 785             raise ArgumentError("Offset is past the end of the file")
 786
 787         if len(data) > config.KEEP_BLOCK_SIZE:
 788             # Chunk it up into smaller writes
 789             n = 0
 790             dataview = memoryview(data)
 791             while n < len(data):
 792                 self.writeto(offset+n, dataview[n:n + config.KEEP_BLOCK_SIZE].tobytes(), num_retries)
 793                 n += config.KEEP_BLOCK_SIZE
 794             return
 795
 796         self._modified = True
 797
 798         if self._current_bblock is None or self._current_bblock.state() != _BufferBlock.WRITABLE:
 799             self._current_bblock = self.parent._my_block_manager().alloc_bufferblock(owner=self)
 800
 801         if (self._current_bblock.size() + len(data)) > config.KEEP_BLOCK_SIZE:
 802             self._repack_writes(num_retries)
 803             if (self._current_bblock.size() + len(data)) > config.KEEP_BLOCK_SIZE:
 804                 self.parent._my_block_manager().commit_bufferblock(self._current_bblock)
 805                 self._current_bblock = self.parent._my_block_manager().alloc_bufferblock(owner=self)
 806
 807         self._current_bblock.append(data)
 808
 809         replace_range(self._segments, offset, len(data), self._current_bblock.blockid, self._current_bblock.write_pointer - len(data))
 810
 811     @synchronized
 812     def flush(self, num_retries=0):
 813         if self._current_bblock:
 814             self._repack_writes(num_retries)
 815             self.parent._my_block_manager().commit_bufferblock(self._current_bblock)
 816
 817     @must_be_writable
 818     @synchronized
 819     def add_segment(self, blocks, pos, size):
 820         """Add a segment to the end of the file.
 821
 822         `pos` and `offset` reference a section of the stream described by
 823         `blocks` (a list of Range objects)
 824
 825         """
 826         self._add_segment(blocks, pos, size)
 827
 828     def _add_segment(self, blocks, pos, size):
 829         """Internal implementation of add_segment."""
 830         self._modified = True
 831         for lr in locators_and_ranges(blocks, pos, size):
 832             last = self._segments[-1] if self._segments else Range(0, 0, 0, 0)
 833             r = Range(lr.locator, last.range_start+last.range_size, lr.segment_size, lr.segment_offset)
 834             self._segments.append(r)
 835
 836     @synchronized
 837     def size(self):
 838         """Get the file size."""
 839         if self._segments:
 840             n = self._segments[-1]
 841             return n.range_start + n.range_size
 842         else:
 843             return 0
 844
 845     @synchronized
 846     def manifest_text(self, stream_name=".", portable_locators=False, normalize=False):
 847         buf = ""
 848         filestream = []
 849         for segment in self.segments:
 850             loc = segment.locator
 851             if loc.startswith("bufferblock"):
 852                 loc = self._bufferblocks[loc].calculate_locator()
 853             if portable_locators:
 854                 loc = KeepLocator(loc).stripped()
 855             filestream.append(LocatorAndRange(loc, locator_block_size(loc),
 856                                  segment.segment_offset, segment.range_size))
 857         buf += ' '.join(normalize_stream(stream_name, {stream_name: filestream}))
 858         buf += "\n"
 859         return buf
 860
 861
 862 class ArvadosFileReader(ArvadosFileReaderBase):
 863     """Wraps ArvadosFile in a file-like object supporting reading only.
 864
 865     Be aware that this class is NOT thread safe as there is no locking around
 866     updating file pointer.
 867
 868     """
 869
 870     def __init__(self, arvadosfile, name, mode="r", num_retries=None):
 871         super(ArvadosFileReader, self).__init__(name, mode, num_retries=num_retries)
 872         self.arvadosfile = arvadosfile
 873
 874     def size(self):
 875         return self.arvadosfile.size()
 876
 877     def stream_name(self):
 878         return self.arvadosfile.parent.stream_name()
 879
 880     @_FileLikeObjectBase._before_close
 881     @retry_method
 882     def read(self, size, num_retries=None):
 883         """Read up to `size` bytes from the stream, starting at the current file position."""
 884         data = self.arvadosfile.readfrom(self._filepos, size, num_retries, exact=True)
 885         self._filepos += len(data)
 886         return data
 887
 888     @_FileLikeObjectBase._before_close
 889     @retry_method
 890     def readfrom(self, offset, size, num_retries=None):
 891         """Read up to `size` bytes from the stream, starting at the current file position."""
 892         return self.arvadosfile.readfrom(offset, size, num_retries)
 893
 894     def flush(self):
 895         pass
 896
 897
 898 class ArvadosFileWriter(ArvadosFileReader):
 899     """Wraps ArvadosFile in a file-like object supporting both reading and writing.
 900
 901     Be aware that this class is NOT thread safe as there is no locking around
 902     updating file pointer.
 903
 904     """
 905
 906     def __init__(self, arvadosfile, name, mode, num_retries=None):
 907         super(ArvadosFileWriter, self).__init__(arvadosfile, name, mode, num_retries=num_retries)
 908
 909     @_FileLikeObjectBase._before_close
 910     @retry_method
 911     def write(self, data, num_retries=None):
 912         if self.mode[0] == "a":
 913             self.arvadosfile.writeto(self.size(), data, num_retries)
 914         else:
 915             self.arvadosfile.writeto(self._filepos, data, num_retries)
 916             self._filepos += len(data)
 917
 918     @_FileLikeObjectBase._before_close
 919     @retry_method
 920     def writelines(self, seq, num_retries=None):
 921         for s in seq:
 922             self.write(s, num_retries)
 923
 924     @_FileLikeObjectBase._before_close
 925     def truncate(self, size=None):
 926         if size is None:
 927             size = self._filepos
 928         self.arvadosfile.truncate(size)
 929         if self._filepos > self.size():
 930             self._filepos = self.size()
 931
 932     @_FileLikeObjectBase._before_close
 933     def flush(self):
 934         self.arvadosfile.flush()
 935
 936     def close(self):
 937         if not self.closed:
 938             self.flush()
 939             super(ArvadosFileWriter, self).close()