X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/ab3b9556865e694a4c33ed623cf2e55fc410f10f..04f5f75659e963f02441a566afcc6d22683d0485:/sdk/python/arvados/stream.py diff --git a/sdk/python/arvados/stream.py b/sdk/python/arvados/stream.py index f5975b9a77..a7e3be38b7 100644 --- a/sdk/python/arvados/stream.py +++ b/sdk/python/arvados/stream.py @@ -1,214 +1,69 @@ -import gflags -import httplib -import httplib2 +import collections +import hashlib import os -import pprint -import sys -import types -import subprocess -import json -import UserDict import re -import hashlib -import string -import bz2 -import zlib -import fcntl -import time import threading -import collections +import functools +import copy +from .ranges import * +from .arvfile import ArvadosFileBase, StreamFileReader from arvados.retry import retry_method from keep import * import config import errors -LOCATOR = 0 -BLOCKSIZE = 1 -OFFSET = 2 -SEGMENTSIZE = 3 +def locator_block_size(loc): + s = re.match(r'[0-9a-f]{32}\+(\d+)(\+\S+)*', loc) + return long(s.group(1)) -def locators_and_ranges(data_locators, range_start, range_size, debug=False): +def normalize_stream(s, stream): ''' - Get blocks that are covered by the range - data_locators: list of [locator, block_size, block_start], assumes that blocks are in order and contigous - range_start: start of range - range_size: size of range - returns list of [block locator, blocksize, segment offset, segment size] that satisfies the range + s is the stream name + stream is a dict mapping each filename to a list in the form [block locator, block size, segment offset (from beginning of block), segment size] + returns the stream as a list of tokens ''' - if range_size == 0: - return [] - resp = [] - range_start = long(range_start) - range_size = long(range_size) - range_end = range_start + range_size - block_start = 0L - - # range_start/block_start is the inclusive lower bound - # range_end/block_end is the exclusive upper bound - - hi = len(data_locators) - lo = 0 - i = int((hi + lo) / 2) - block_size = data_locators[i][BLOCKSIZE] - block_start = data_locators[i][OFFSET] - block_end = block_start + block_size - if debug: print '---' - - # perform a binary search for the first block - # assumes that all of the blocks are contigious, so range_start is guaranteed - # to either fall into the range of a block or be outside the block range entirely - while not (range_start >= block_start and range_start < block_end): - if lo == i: - # must be out of range, fail - return [] - if range_start > block_start: - lo = i - else: - hi = i - i = int((hi + lo) / 2) - if debug: print lo, i, hi - block_size = data_locators[i][BLOCKSIZE] - block_start = data_locators[i][OFFSET] - block_end = block_start + block_size - - while i < len(data_locators): - locator, block_size, block_start = data_locators[i] - block_end = block_start + block_size - if debug: - print locator, "range_start", range_start, "block_start", block_start, "range_end", range_end, "block_end", block_end - if range_end <= block_start: - # range ends before this block starts, so don't look at any more locators - break - - #if range_start >= block_end: - # range starts after this block ends, so go to next block - # we should always start at the first block due to the binary above, so this test is redundant - #next - - if range_start >= block_start and range_end <= block_end: - # range starts and ends in this block - resp.append([locator, block_size, range_start - block_start, range_size]) - elif range_start >= block_start and range_end > block_end: - # range starts in this block - resp.append([locator, block_size, range_start - block_start, block_end - range_start]) - elif range_start < block_start and range_end > block_end: - # range starts in a previous block and extends to further blocks - resp.append([locator, block_size, 0L, block_size]) - elif range_start < block_start and range_end <= block_end: - # range starts in a previous block and ends in this block - resp.append([locator, block_size, 0L, range_end - block_start]) - block_start = block_end - i += 1 - return resp - - -class StreamFileReader(object): - def __init__(self, stream, segments, name): - self._stream = stream - self.segments = segments - self._name = name - self._filepos = 0L - self.num_retries = stream.num_retries - - def name(self): - return self._name - - def decompressed_name(self): - return re.sub('\.(bz2|gz)$', '', self._name) - - def stream_name(self): - return self._stream.name() - - def seek(self, pos): - self._filepos = min(max(pos, 0L), self.size()) - - def tell(self): - return self._filepos - - def size(self): - n = self.segments[-1] - return n[OFFSET] + n[BLOCKSIZE] - - @retry_method - def read(self, size, num_retries=None): - """Read up to 'size' bytes from the stream, starting at the current file position""" - if size == 0: - return '' - - data = '' - available_chunks = locators_and_ranges(self.segments, self._filepos, size) - if available_chunks: - locator, blocksize, segmentoffset, segmentsize = available_chunks[0] - data = self._stream.readfrom(locator+segmentoffset, segmentsize, - num_retries=num_retries) - - self._filepos += len(data) - return data - - @retry_method - def readfrom(self, start, size, num_retries=None): - """Read up to 'size' bytes from the stream, starting at 'start'""" - if size == 0: - return '' - - data = [] - for locator, blocksize, segmentoffset, segmentsize in locators_and_ranges(self.segments, start, size): - data.append(self._stream.readfrom(locator+segmentoffset, segmentsize, - num_retries=num_retries)) - return ''.join(data) - - @retry_method - def readall(self, size=2**20, num_retries=None): - while True: - data = self.read(size, num_retries=num_retries) - if data == '': - break - yield data + stream_tokens = [s] + sortedfiles = list(stream.keys()) + sortedfiles.sort() + + blocks = {} + streamoffset = 0L + # Go through each file and add each referenced block exactly once. + for f in sortedfiles: + for b in stream[f]: + if b.locator not in blocks: + stream_tokens.append(b.locator) + blocks[b.locator] = streamoffset + streamoffset += locator_block_size(b.locator) + + # Add the empty block if the stream is otherwise empty. + if len(stream_tokens) == 1: + stream_tokens.append(config.EMPTY_BLOCK_LOCATOR) + + for f in sortedfiles: + # Add in file segments + current_span = None + fout = f.replace(' ', '\\040') + for segment in stream[f]: + # Collapse adjacent segments + streamoffset = blocks[segment.locator] + segment.segment_offset + if current_span is None: + current_span = [streamoffset, streamoffset + segment.segment_size] + else: + if streamoffset == current_span[1]: + current_span[1] += segment.segment_size + else: + stream_tokens.append("{0}:{1}:{2}".format(current_span[0], current_span[1] - current_span[0], fout)) + current_span = [streamoffset, streamoffset + segment.segment_size] - @retry_method - def decompress(self, decompress, size, num_retries=None): - for segment in self.readall(size, num_retries): - data = decompress(segment) - if data and data != '': - yield data + if current_span is not None: + stream_tokens.append("{0}:{1}:{2}".format(current_span[0], current_span[1] - current_span[0], fout)) - @retry_method - def readall_decompressed(self, size=2**20, num_retries=None): - self.seek(0) - if re.search('\.bz2$', self._name): - dc = bz2.BZ2Decompressor() - return self.decompress(dc.decompress, size, - num_retries=num_retries) - elif re.search('\.gz$', self._name): - dc = zlib.decompressobj(16+zlib.MAX_WBITS) - return self.decompress(lambda segment: dc.decompress(dc.unconsumed_tail + segment), - size, num_retries=num_retries) - else: - return self.readall(size, num_retries=num_retries) + if not stream[f]: + stream_tokens.append("0:0:{0}".format(fout)) - @retry_method - def readlines(self, decompress=True, num_retries=None): - read_func = self.readall_decompressed if decompress else self.readall - data = '' - for newdata in read_func(num_retries=num_retries): - data += newdata - sol = 0 - while True: - eol = string.find(data, "\n", sol) - if eol < 0: - break - yield data[sol:eol+1] - sol = eol+1 - data = data[sol:] - if data != '': - yield data - - def as_manifest(self): - manifest_text = ['.'] - manifest_text.extend([d[LOCATOR] for d in self._stream._data_locators]) - manifest_text.extend(["{}:{}:{}".format(seg[LOCATOR], seg[BLOCKSIZE], self.name().replace(' ', '\\040')) for seg in self.segments]) - return arvados.CollectionReader(' '.join(manifest_text) + '\n').manifest_text(normalize=True) + return stream_tokens class StreamReader(object): @@ -232,7 +87,7 @@ class StreamReader(object): s = re.match(r'^[0-9a-f]{32}\+(\d+)(\+\S+)*$', tok) if s: blocksize = long(s.group(1)) - self._data_locators.append([tok, blocksize, streamoffset]) + self._data_locators.append(Range(tok, streamoffset, blocksize)) streamoffset += blocksize continue @@ -242,10 +97,10 @@ class StreamReader(object): size = long(s.group(2)) name = s.group(3).replace('\\040', ' ') if name not in self._files: - self._files[name] = StreamFileReader(self, [[pos, size, 0]], name) + self._files[name] = StreamFileReader(self, [Range(pos, 0, size)], name) else: - n = self._files[name] - n.segments.append([pos, size, n.size()]) + filereader = self._files[name] + filereader.segments.append(Range(pos, filereader.size(), size)) continue raise errors.SyntaxError("Invalid manifest format") @@ -259,34 +114,172 @@ class StreamReader(object): def all_files(self): return self._files.values() - def size(self): + def _size(self): n = self._data_locators[-1] - return n[OFFSET] + n[BLOCKSIZE] + return n.range_start + n.range_size + + def size(self): + return self._size() def locators_and_ranges(self, range_start, range_size): return locators_and_ranges(self._data_locators, range_start, range_size) + @retry_method + def _keepget(self, locator, num_retries=None): + return self._keep.get(locator, num_retries=num_retries) + @retry_method def readfrom(self, start, size, num_retries=None): + return self._readfrom(start, size, num_retries=num_retries) + + @retry_method + def _readfrom(self, start, size, num_retries=None): """Read up to 'size' bytes from the stream, starting at 'start'""" if size == 0: return '' if self._keep is None: self._keep = KeepClient(num_retries=self.num_retries) data = [] - for locator, blocksize, segmentoffset, segmentsize in locators_and_ranges(self._data_locators, start, size): - data.append(self._keep.get(locator, num_retries=num_retries)[segmentoffset:segmentoffset+segmentsize]) + for lr in locators_and_ranges(self._data_locators, start, size): + data.append(self._keepget(lr.locator, num_retries=num_retries)[lr.segment_offset:lr.segment_offset+lr.segment_size]) return ''.join(data) def manifest_text(self, strip=False): manifest_text = [self.name().replace(' ', '\\040')] if strip: for d in self._data_locators: - m = re.match(r'^[0-9a-f]{32}\+\d+', d[LOCATOR]) + m = re.match(r'^[0-9a-f]{32}\+\d+', d.locator) manifest_text.append(m.group(0)) else: - manifest_text.extend([d[LOCATOR] for d in self._data_locators]) - manifest_text.extend([' '.join(["{}:{}:{}".format(seg[LOCATOR], seg[BLOCKSIZE], f.name().replace(' ', '\\040')) + manifest_text.extend([d.locator for d in self._data_locators]) + manifest_text.extend([' '.join(["{}:{}:{}".format(seg.locator, seg.range_size, f.name.replace(' ', '\\040')) for seg in f.segments]) for f in self._files.values()]) return ' '.join(manifest_text) + '\n' + + + + +# class StreamWriter(StreamReader): +# def __init__(self, tokens, keep=None, debug=False, _empty=False, +# num_retries=0): +# super(StreamWriter, self).__init__(tokens, keep, debug, _empty, num_retries) + +# if len(self._files) != 1: +# raise AssertionError("StreamWriter can only have one file at a time") +# sr = self._files.popitem()[1] +# self._files[sr.name] = StreamFileWriter(self, sr.segments, sr.name) + +# self.mutex = threading.Lock() +# self.current_bblock = None +# self.bufferblocks = {} + +# # wrap superclass methods in mutex +# def _proxy_method(name): +# method = getattr(StreamReader, name) +# @functools.wraps(method, ('__name__', '__doc__')) +# def wrapper(self, *args, **kwargs): +# with self.mutex: +# return method(self, *args, **kwargs) +# return wrapper + +# for _method_name in ['files', 'all_files', 'size', 'locators_and_ranges', 'readfrom', 'manifest_text']: +# locals()[_method_name] = _proxy_method(_method_name) + +# @retry_method +# def _keepget(self, locator, num_retries=None): +# if locator in self.bufferblocks: +# bb = self.bufferblocks[locator] +# return str(bb.buffer_block[0:bb.write_pointer]) +# else: +# return self._keep.get(locator, num_retries=num_retries) + +# def _init_bufferblock(self): +# last = self._data_locators[-1] +# streamoffset = last.range_start + last.range_size +# if last.range_size == 0: +# del self._data_locators[-1] +# self.current_bblock = BufferBlock("bufferblock%i" % len(self.bufferblocks), streamoffset) +# self.bufferblocks[self.current_bblock.locator] = self.current_bblock +# self._data_locators.append(self.current_bblock.locator_list_entry) + +# def _repack_writes(self): +# '''Test if the buffer block has more data than is referenced by actual segments +# (this happens when a buffered write over-writes a file range written in +# a previous buffered write). Re-pack the buffer block for efficiency +# and to avoid leaking information. +# ''' +# segs = self._files.values()[0].segments + +# bufferblock_segs = [] +# i = 0 +# tmp_segs = copy.copy(segs) +# while i < len(tmp_segs): +# # Go through each segment and identify segments that include the buffer block +# s = tmp_segs[i] +# if s[LOCATOR] < self.current_bblock.locator_list_entry.range_start and (s[LOCATOR] + s.range_size) > self.current_bblock.locator_list_entry.range_start: +# # The segment straddles the previous block and the current buffer block. Split the segment. +# b1 = self.current_bblock.locator_list_entry.range_start - s[LOCATOR] +# b2 = (s[LOCATOR] + s.range_size) - self.current_bblock.locator_list_entry.range_start +# bb_seg = [self.current_bblock.locator_list_entry.range_start, b2, s.range_start+b1] +# tmp_segs[i] = [s[LOCATOR], b1, s.range_start] +# tmp_segs.insert(i+1, bb_seg) +# bufferblock_segs.append(bb_seg) +# i += 1 +# elif s[LOCATOR] >= self.current_bblock.locator_list_entry.range_start: +# # The segment's data is in the buffer block. +# bufferblock_segs.append(s) +# i += 1 + +# # Now sum up the segments to get the total bytes +# # of the file referencing into the buffer block. +# write_total = sum([s.range_size for s in bufferblock_segs]) + +# if write_total < self.current_bblock.locator_list_entry.range_size: +# # There is more data in the buffer block than is actually accounted for by segments, so +# # re-pack into a new buffer by copying over to a new buffer block. +# new_bb = BufferBlock(self.current_bblock.locator, +# self.current_bblock.locator_list_entry.range_start, +# starting_size=write_total) +# for t in bufferblock_segs: +# t_start = t[LOCATOR] - self.current_bblock.locator_list_entry.range_start +# t_end = t_start + t.range_size +# t[0] = self.current_bblock.locator_list_entry.range_start + new_bb.write_pointer +# new_bb.append(self.current_bblock.buffer_block[t_start:t_end]) + +# self.current_bblock = new_bb +# self.bufferblocks[self.current_bblock.locator] = self.current_bblock +# self._data_locators[-1] = self.current_bblock.locator_list_entry +# self._files.values()[0].segments = tmp_segs + +# def _commit(self): +# # commit buffer block + +# # TODO: do 'put' in the background? +# pdh = self._keep.put(self.current_bblock.buffer_block[0:self.current_bblock.write_pointer]) +# self._data_locators[-1][0] = pdh +# self.current_bblock = None + +# def commit(self): +# with self.mutex: +# self._repack_writes() +# self._commit() + +# def _append(self, data): +# if len(data) > config.KEEP_BLOCK_SIZE: +# raise ArgumentError("Please append data chunks smaller than config.KEEP_BLOCK_SIZE") + +# if self.current_bblock is None: +# self._init_bufferblock() + +# if (self.current_bblock.write_pointer + len(data)) > config.KEEP_BLOCK_SIZE: +# self._repack_writes() +# if (self.current_bblock.write_pointer + len(data)) > config.KEEP_BLOCK_SIZE: +# self._commit() +# self._init_bufferblock() + +# self.current_bblock.append(data) + +# def append(self, data): +# with self.mutex: +# self._append(data)