import os
import zlib
import bz2
-from .ranges import *
+from ._ranges import locators_and_ranges, replace_range, Range
from arvados.retry import retry_method
import config
import hashlib
import Queue
import copy
import errno
-from .errors import KeepWriteError
+from .errors import KeepWriteError, AssertionError
+from .keep import KeepLocator
+from _normalize_stream import normalize_stream
def split(path):
"""Separate the stream name and file name in a /-separated stream path and
stream_name, file_name = '.', path
return stream_name, file_name
-class ArvadosFileBase(object):
+class _FileLikeObjectBase(object):
def __init__(self, name, mode):
self.name = name
self.mode = mode
self.closed = True
-class ArvadosFileReaderBase(ArvadosFileBase):
+class ArvadosFileReaderBase(_FileLikeObjectBase):
class _NameAttribute(str):
# The Python file API provides a plain .name attribute.
# Older SDK provided a name() method.
def decompressed_name(self):
return re.sub('\.(bz2|gz)$', '', self.name)
- @ArvadosFileBase._before_close
+ @_FileLikeObjectBase._before_close
def seek(self, pos, whence=os.SEEK_CUR):
if whence == os.SEEK_CUR:
pos += self._filepos
def tell(self):
return self._filepos
- @ArvadosFileBase._before_close
+ @_FileLikeObjectBase._before_close
@retry_method
def readall(self, size=2**20, num_retries=None):
while True:
break
yield data
- @ArvadosFileBase._before_close
+ @_FileLikeObjectBase._before_close
@retry_method
def readline(self, size=float('inf'), num_retries=None):
cache_pos, cache_data = self._readline_cache
self._readline_cache = (self.tell(), data[nextline_index:])
return data[:nextline_index]
- @ArvadosFileBase._before_close
+ @_FileLikeObjectBase._before_close
@retry_method
def decompress(self, decompress, size, num_retries=None):
for segment in self.readall(size, num_retries):
if data:
yield data
- @ArvadosFileBase._before_close
+ @_FileLikeObjectBase._before_close
@retry_method
def readall_decompressed(self, size=2**20, num_retries=None):
self.seek(0)
else:
return self.readall(size, num_retries=num_retries)
- @ArvadosFileBase._before_close
+ @_FileLikeObjectBase._before_close
@retry_method
def readlines(self, sizehint=float('inf'), num_retries=None):
data = []
n = self.segments[-1]
return n.range_start + n.range_size
- @ArvadosFileBase._before_close
+ @_FileLikeObjectBase._before_close
@retry_method
def read(self, size, num_retries=None):
"""Read up to 'size' bytes from the stream, starting at the current file position"""
self._filepos += len(data)
return data
- @ArvadosFileBase._before_close
+ @_FileLikeObjectBase._before_close
@retry_method
def readfrom(self, start, size, num_retries=None):
"""Read up to 'size' bytes from the stream, starting at 'start'"""
return ''.join(data)
def as_manifest(self):
- from stream import normalize_stream
segs = []
for r in self.segments:
segs.extend(self._stream.locators_and_ranges(r.locator, r.range_size))
return " ".join(normalize_stream(".", {self.name: segs})) + "\n"
-class BufferBlock(object):
+def synchronized(orig_func):
+ @functools.wraps(orig_func)
+ def synchronized_wrapper(self, *args, **kwargs):
+ with self.lock:
+ return orig_func(self, *args, **kwargs)
+ return synchronized_wrapper
+
+class _BufferBlock(object):
"""A BufferBlock is a stand-in for a Keep block that is in the process of being
written.
self.buffer_block = bytearray(starting_capacity)
self.buffer_view = memoryview(self.buffer_block)
self.write_pointer = 0
- self.state = BufferBlock.WRITABLE
+ self._state = _BufferBlock.WRITABLE
self._locator = None
self.owner = owner
+ self.lock = threading.Lock()
+ @synchronized
def append(self, data):
"""Append some data to the buffer.
buffer, doubling capacity as needed to accomdate all the data.
"""
- if self.state == BufferBlock.WRITABLE:
+ if self._state == _BufferBlock.WRITABLE:
while (self.write_pointer+len(data)) > len(self.buffer_block):
new_buffer_block = bytearray(len(self.buffer_block) * 2)
new_buffer_block[0:self.write_pointer] = self.buffer_block[0:self.write_pointer]
else:
raise AssertionError("Buffer block is not writable")
+ @synchronized
+ def set_state(self, nextstate, loc=None):
+ if ((self._state == _BufferBlock.WRITABLE and nextstate == _BufferBlock.PENDING) or
+ (self._state == _BufferBlock.PENDING and nextstate == _BufferBlock.COMMITTED)):
+ self._state = nextstate
+ if self._state == _BufferBlock.COMMITTED:
+ self._locator = loc
+ self.buffer_view = None
+ self.buffer_block = None
+ else:
+ raise AssertionError("Invalid state change from %s to %s" % (self.state, state))
+
+ @synchronized
+ def state(self):
+ return self._state
+
def size(self):
"""The amount of data written to the buffer."""
return self.write_pointer
+ @synchronized
def locator(self):
"""The Keep locator for this buffer's contents."""
if self._locator is None:
return self._locator
-def synchronized(orig_func):
- @functools.wraps(orig_func)
- def synchronized_wrapper(self, *args, **kwargs):
- with self.lock:
- return orig_func(self, *args, **kwargs)
- return synchronized_wrapper
-
class NoopLock(object):
def __enter__(self):
return self
return must_be_writable_wrapper
-class BlockManager(object):
+class _BlockManager(object):
"""BlockManager handles buffer blocks, background block uploads, and background
block prefetch for a Collection of ArvadosFiles.
"""
if blockid is None:
blockid = "bufferblock%i" % len(self._bufferblocks)
- bufferblock = BufferBlock(blockid, starting_capacity=starting_capacity, owner=owner)
+ bufferblock = _BufferBlock(blockid, starting_capacity=starting_capacity, owner=owner)
self._bufferblocks[bufferblock.blockid] = bufferblock
return bufferblock
@synchronized
- def dup_block(self, blockid, owner):
+ def dup_block(self, block, owner):
"""Create a new bufferblock in WRITABLE state, initialized with the content of
an existing bufferblock.
- :blockid:
- the block to copy. May be an existing buffer block id.
+ :block:
+ the buffer block to copy.
:owner:
ArvadosFile that owns the new block
"""
new_blockid = "bufferblock%i" % len(self._bufferblocks)
- block = self._bufferblocks[blockid]
- bufferblock = BufferBlock(new_blockid, len(block), owner)
- bufferblock.append(block)
+ with block.lock:
+ if block._state == _BufferBlock.COMMITTED:
+ raise AssertionError("Can only duplicate a writable or pending buffer block")
+
+ bufferblock = _BufferBlock(new_blockid, block.size(), owner)
+ bufferblock.append(block.buffer_view[0:block.size()])
self._bufferblocks[bufferblock.blockid] = bufferblock
return bufferblock
bufferblock = self._put_queue.get()
if bufferblock is None:
return
- bufferblock._locator = self._keep.put(bufferblock.buffer_view[0:bufferblock.write_pointer].tobytes())
- bufferblock.state = BufferBlock.COMMITTED
- bufferblock.buffer_view = None
- bufferblock.buffer_block = None
+ loc = self._keep.put(bufferblock.buffer_view[0:bufferblock.write_pointer].tobytes())
+ bufferblock.set_state(_BufferBlock.COMMITTED, loc)
+
except Exception as e:
print e
self._put_errors.put((bufferblock.locator(), e))
thread.start()
# Mark the block as PENDING so to disallow any more appends.
- block.state = BufferBlock.PENDING
+ block.set_state(_BufferBlock.PENDING)
self._put_queue.put(block)
- def get_block(self, locator, num_retries, cache_only=False):
+ @synchronized
+ def get_bufferblock(self, locator):
+ return self._bufferblocks.get(locator)
+
+ def get_block_contents(self, locator, num_retries, cache_only=False):
"""Fetch a block.
First checks to see if the locator is a BufferBlock and return that, if
with self.lock:
if locator in self._bufferblocks:
bufferblock = self._bufferblocks[locator]
- if bufferblock.state != BufferBlock.COMMITTED:
+ if bufferblock.state() != _BufferBlock.COMMITTED:
return bufferblock.buffer_view[0:bufferblock.write_pointer].tobytes()
else:
locator = bufferblock._locator
items = self._bufferblocks.items()
for k,v in items:
- if v.state == BufferBlock.WRITABLE:
+ if v.state() == _BufferBlock.WRITABLE:
self.commit_bufferblock(v)
with self.lock:
if self._put_queue is not None:
self._put_queue.join()
+
if not self._put_errors.empty():
err = []
try:
def clone(self, new_parent):
"""Make a copy of this file."""
cp = ArvadosFile(new_parent)
-
- map_loc = {}
- for r in self._segments:
- new_loc = r.locator
- if self.parent._my_block_manager().is_bufferblock(r.locator):
- if r.locator not in map_loc:
- map_loc[r.locator] = self.parent._my_block_manager().dup_block(r.locator, cp).blockid
- new_loc = map_loc[r.locator]
-
- cp._segments.append(Range(new_loc, r.range_start, r.range_size, r.segment_offset))
-
+ cp.replace_contents(self)
return cp
@must_be_writable
@synchronized
def replace_contents(self, other):
"""Replace segments of this file with segments from another `ArvadosFile` object."""
- self._segments = other.segments()
+
+ map_loc = {}
+ self._segments = []
+ for other_segment in other.segments():
+ new_loc = other_segment.locator
+ if other.parent._my_block_manager().is_bufferblock(other_segment.locator):
+ if other_segment.locator not in map_loc:
+ bufferblock = other.parent._my_block_manager().get_bufferblock(other_segment.locator)
+ if bufferblock.state() != _BufferBlock.WRITABLE:
+ map_loc[other_segment.locator] = bufferblock.locator()
+ else:
+ map_loc[other_segment.locator] = self.parent._my_block_manager().dup_block(bufferblock, self).blockid
+ new_loc = map_loc[other_segment.locator]
+
+ self._segments.append(Range(new_loc, other_segment.range_start, other_segment.range_size, other_segment.segment_offset))
+
self._modified = True
def __eq__(self, other):
if not isinstance(other, ArvadosFile):
return False
- s = other.segments()
+ othersegs = other.segments()
with self.lock:
- return self._segments == s
+ if len(self._segments) != len(othersegs):
+ return False
+ for i in xrange(0, len(othersegs)):
+ seg1 = self._segments[i]
+ seg2 = othersegs[i]
+ loc1 = seg1.locator
+ loc2 = seg2.locator
+
+ if self.parent._my_block_manager().is_bufferblock(loc1):
+ loc1 = self.parent._my_block_manager().get_bufferblock(loc1).locator()
+
+ if other.parent._my_block_manager().is_bufferblock(loc2):
+ loc2 = other.parent._my_block_manager().get_bufferblock(loc2).locator()
+
+ if (KeepLocator(loc1).stripped() != KeepLocator(loc2).stripped() or
+ seg1.range_start != seg2.range_start or
+ seg1.range_size != seg2.range_size or
+ seg1.segment_offset != seg2.segment_offset):
+ return False
+
+ return True
def __ne__(self, other):
return not self.__eq__(other)
data = []
for lr in readsegs:
- block = self.parent._my_block_manager().get_block(lr.locator, num_retries=num_retries, cache_only=bool(data))
+ block = self.parent._my_block_manager().get_block_contents(lr.locator, num_retries=num_retries, cache_only=bool(data))
if block:
data.append(block[lr.segment_offset:lr.segment_offset+lr.segment_size])
else:
self._modified = True
- if self._current_bblock is None or self._current_bblock.state != BufferBlock.WRITABLE:
+ if self._current_bblock is None or self._current_bblock.state() != _BufferBlock.WRITABLE:
self._current_bblock = self.parent._my_block_manager().alloc_bufferblock(owner=self)
if (self._current_bblock.size() + len(data)) > config.KEEP_BLOCK_SIZE:
else:
return 0
+
+ @synchronized
+ def manifest_text(self, stream_name=".", portable_locators=False, normalize=False):
+ buf = ""
+ item = self
+ filestream = []
+ for segment in item.segments:
+ loc = segment.locator
+ if loc.startswith("bufferblock"):
+ loc = item._bufferblocks[loc].calculate_locator()
+ if portable_locators:
+ loc = KeepLocator(loc).stripped()
+ filestream.append(LocatorAndRange(loc, locator_block_size(loc),
+ segment.segment_offset, segment.range_size))
+ stream[stream_name] = filestream
+ buf += ' '.join(normalize_stream(stream_name, stream))
+ buf += "\n"
+ return buf
+
+
class ArvadosFileReader(ArvadosFileReaderBase):
"""Wraps ArvadosFile in a file-like object supporting reading only.
def size(self):
return self.arvadosfile.size()
- @ArvadosFileBase._before_close
+ def stream_name(self):
+ return self.arvadosfile.parent.stream_name()
+
+ @_FileLikeObjectBase._before_close
@retry_method
def read(self, size, num_retries=None):
"""Read up to `size` bytes from the stream, starting at the current file position."""
self._filepos += len(data)
return data
- @ArvadosFileBase._before_close
+ @_FileLikeObjectBase._before_close
@retry_method
def readfrom(self, offset, size, num_retries=None):
"""Read up to `size` bytes from the stream, starting at the current file position."""
def __init__(self, arvadosfile, name, mode, num_retries=None):
super(ArvadosFileWriter, self).__init__(arvadosfile, name, mode, num_retries=num_retries)
- @ArvadosFileBase._before_close
+ @_FileLikeObjectBase._before_close
@retry_method
def write(self, data, num_retries=None):
if self.mode[0] == "a":
self.arvadosfile.writeto(self._filepos, data, num_retries)
self._filepos += len(data)
- @ArvadosFileBase._before_close
+ @_FileLikeObjectBase._before_close
@retry_method
def writelines(self, seq, num_retries=None):
for s in seq:
projects / arvados.git / blobdiff