class ArvadosFileReaderBase(_FileLikeObjectBase):
def __init__(self, name, mode, num_retries=None):
super(ArvadosFileReaderBase, self).__init__(name, mode)
- self._binary = 'b' in mode
- if sys.version_info >= (3, 0) and not self._binary:
- raise NotImplementedError("text mode {!r} is not implemented".format(mode))
self._filepos = 0
self.num_retries = num_retries
self._readline_cache = (None, None)
DEFAULT_PUT_THREADS = 2
DEFAULT_GET_THREADS = 2
- def __init__(self, keep, copies=None, put_threads=None):
+ def __init__(self, keep, copies=None, put_threads=None, num_retries=None, storage_classes_func=None, get_threads=None):
"""keep: KeepClient object to use"""
self._keep = keep
self._bufferblocks = collections.OrderedDict()
self._prefetch_threads = None
self.lock = threading.Lock()
self.prefetch_enabled = True
- if put_threads:
- self.num_put_threads = put_threads
- else:
- self.num_put_threads = _BlockManager.DEFAULT_PUT_THREADS
- self.num_get_threads = _BlockManager.DEFAULT_GET_THREADS
+ self.num_put_threads = put_threads or _BlockManager.DEFAULT_PUT_THREADS
+ self.num_get_threads = get_threads or _BlockManager.DEFAULT_GET_THREADS
self.copies = copies
+ self.storage_classes = storage_classes_func or (lambda: [])
self._pending_write_size = 0
self.threads_lock = threading.Lock()
self.padding_block = None
+ self.num_retries = num_retries
@synchronized
def alloc_bufferblock(self, blockid=None, starting_capacity=2**14, owner=None):
return
if self.copies is None:
- loc = self._keep.put(bufferblock.buffer_view[0:bufferblock.write_pointer].tobytes())
+ loc = self._keep.put(bufferblock.buffer_view[0:bufferblock.write_pointer].tobytes(), num_retries=self.num_retries, classes=self.storage_classes())
else:
- loc = self._keep.put(bufferblock.buffer_view[0:bufferblock.write_pointer].tobytes(), copies=self.copies)
+ loc = self._keep.put(bufferblock.buffer_view[0:bufferblock.write_pointer].tobytes(), num_retries=self.num_retries, copies=self.copies, classes=self.storage_classes())
bufferblock.set_state(_BufferBlock.COMMITTED, loc)
except Exception as e:
bufferblock.set_state(_BufferBlock.ERROR, e)
# If we don't limit the Queue size, the upload queue can quickly
# grow to take up gigabytes of RAM if the writing process is
- # generating data more quickly than it can be send to the Keep
+ # generating data more quickly than it can be sent to the Keep
# servers.
#
# With two upload threads and a queue size of 2, this means up to 4
b = self._prefetch_queue.get()
if b is None:
return
+ if self._keep.has_cache_slot(b):
+ continue
+ _logger.debug("prefetching %s", b)
self._keep.get(b)
except Exception:
_logger.exception("Exception doing block prefetch")
if sync:
try:
if self.copies is None:
- loc = self._keep.put(block.buffer_view[0:block.write_pointer].tobytes())
+ loc = self._keep.put(block.buffer_view[0:block.write_pointer].tobytes(), num_retries=self.num_retries, classes=self.storage_classes())
else:
- loc = self._keep.put(block.buffer_view[0:block.write_pointer].tobytes(), copies=self.copies)
+ loc = self._keep.put(block.buffer_view[0:block.write_pointer].tobytes(), num_retries=self.num_retries, copies=self.copies, classes=self.storage_classes())
block.set_state(_BufferBlock.COMMITTED, loc)
except Exception as e:
block.set_state(_BufferBlock.ERROR, e)
self._delete_bufferblock(locator)
def _delete_bufferblock(self, locator):
- bb = self._bufferblocks[locator]
- bb.clear()
- del self._bufferblocks[locator]
+ if locator in self._bufferblocks:
+ bb = self._bufferblocks[locator]
+ bb.clear()
+ del self._bufferblocks[locator]
def get_block_contents(self, locator, num_retries, cache_only=False):
"""Fetch a block.
if not self.prefetch_enabled:
return
- if self._keep.get_from_cache(locator) is not None:
+ if self._keep.has_cache_slot(locator):
return
with self.lock:
return
self.start_get_threads()
+ # _logger.debug("pushing %s to prefetch", locator)
self._prefetch_queue.put(locator)
if size == 0 or offset >= self.size():
return b''
readsegs = locators_and_ranges(self._segments, offset, size)
- prefetch = locators_and_ranges(self._segments, offset + size, config.KEEP_BLOCK_SIZE, limit=32)
+ prefetch = locators_and_ranges(self._segments, offset + size, config.KEEP_BLOCK_SIZE * self.parent._my_block_manager().num_get_threads, limit=32)
locs = set()
data = []
def stream_name(self):
return self.arvadosfile.parent.stream_name()
+ def readinto(self, b):
+ data = self.read(len(b))
+ b[:len(data)] = data
+ return len(data)
+
@_FileLikeObjectBase._before_close
@retry_method
def read(self, size=None, num_retries=None):
if not self.closed:
self.arvadosfile.remove_writer(self, flush)
super(ArvadosFileWriter, self).close()
+
+
+class WrappableFile(object):
+ """An interface to an Arvados file that's compatible with io wrappers.
+
+ """
+ def __init__(self, f):
+ self.f = f
+ self.closed = False
+ def close(self):
+ self.closed = True
+ return self.f.close()
+ def flush(self):
+ return self.f.flush()
+ def read(self, *args, **kwargs):
+ return self.f.read(*args, **kwargs)
+ def readable(self):
+ return self.f.readable()
+ def readinto(self, *args, **kwargs):
+ return self.f.readinto(*args, **kwargs)
+ def seek(self, *args, **kwargs):
+ return self.f.seek(*args, **kwargs)
+ def seekable(self):
+ return self.f.seekable()
+ def tell(self):
+ return self.f.tell()
+ def writable(self):
+ return self.f.writable()
+ def write(self, *args, **kwargs):
+ return self.f.write(*args, **kwargs)
projects / arvados.git / blobdiff