lib/bup/hashsplit.py

   1 import math
   2 from bup import _helpers
   3 from bup.helpers import *
   4
   5 BLOB_MAX = 8192*4   # 8192 is the "typical" blob size for bupsplit
   6 BLOB_READ_SIZE = 1024*1024
   7 MAX_PER_TREE = 256
   8 progress_callback = None
   9 max_pack_size = 1000*1000*1000  # larger packs will slow down pruning
  10 max_pack_objects = 200*1000  # cache memory usage is about 83 bytes per object
  11 fanout = 16
  12
  13 # The purpose of this type of buffer is to avoid copying on peek(), get(),
  14 # and eat().  We do copy the buffer contents on put(), but that should
  15 # be ok if we always only put() large amounts of data at a time.
  16 class Buf:
  17     def __init__(self):
  18         self.data = ''
  19         self.start = 0
  20
  21     def put(self, s):
  22         if s:
  23             self.data = buffer(self.data, self.start) + s
  24             self.start = 0
  25
  26     def peek(self, count):
  27         return buffer(self.data, self.start, count)
  28
  29     def eat(self, count):
  30         self.start += count
  31
  32     def get(self, count):
  33         v = buffer(self.data, self.start, count)
  34         self.start += count
  35         return v
  36
  37     def used(self):
  38         return len(self.data) - self.start
  39
  40
  41 def readfile_iter(files, progress=None):
  42     for filenum,f in enumerate(files):
  43         ofs = 0
  44         b = ''
  45         while 1:
  46             if progress:
  47                 progress(filenum, len(b))
  48             fadvise_done(f, max(0, ofs - 1024*1024))
  49             b = f.read(BLOB_READ_SIZE)
  50             ofs += len(b)
  51             if not b:
  52                 fadvise_done(f, ofs)
  53                 break
  54             yield b
  55
  56
  57 def _splitbuf(buf, basebits, fanbits):
  58     while 1:
  59         b = buf.peek(buf.used())
  60         (ofs, bits) = _helpers.splitbuf(b)
  61         if ofs > BLOB_MAX:
  62             ofs = BLOB_MAX
  63         if ofs:
  64             buf.eat(ofs)
  65             level = (bits-basebits)//fanbits  # integer division
  66             yield buffer(b, 0, ofs), level
  67         else:
  68             break
  69     while buf.used() >= BLOB_MAX:
  70         # limit max blob size
  71         yield buf.get(BLOB_MAX), 0
  72
  73
  74 def _hashsplit_iter(files, progress):
  75     assert(BLOB_READ_SIZE > BLOB_MAX)
  76     basebits = _helpers.blobbits()
  77     fanbits = int(math.log(fanout or 128, 2))
  78     buf = Buf()
  79     for inblock in readfile_iter(files, progress):
  80         buf.put(inblock)
  81         for buf_and_level in _splitbuf(buf, basebits, fanbits):
  82             yield buf_and_level
  83     if buf.used():
  84         yield buf.get(buf.used()), 0
  85
  86
  87 def _hashsplit_iter_keep_boundaries(files, progress):
  88     for real_filenum,f in enumerate(files):
  89         if progress:
  90             def prog(filenum, nbytes):
  91                 # the inner _hashsplit_iter doesn't know the real file count,
  92                 # so we'll replace it here.
  93                 return progress(real_filenum, nbytes)
  94         else:
  95             prog = None
  96         for buf_and_level in _hashsplit_iter([f], progress=prog):
  97             yield buf_and_level
  98
  99
 100 def hashsplit_iter(files, keep_boundaries, progress):
 101     if keep_boundaries:
 102         return _hashsplit_iter_keep_boundaries(files, progress)
 103     else:
 104         return _hashsplit_iter(files, progress)
 105
 106
 107 total_split = 0
 108 def _split_to_blobs(w, files, keep_boundaries, progress):
 109     global total_split
 110     for (blob, level) in hashsplit_iter(files, keep_boundaries, progress):
 111         sha = w.new_blob(blob)
 112         total_split += len(blob)
 113         if w.outbytes >= max_pack_size or w.count >= max_pack_objects:
 114             w.breakpoint()
 115         if progress_callback:
 116             progress_callback(len(blob))
 117         yield (sha, len(blob), level)
 118
 119
 120 def _make_shalist(l):
 121     ofs = 0
 122     shalist = []
 123     for (mode, sha, size) in l:
 124         shalist.append((mode, '%016x' % ofs, sha))
 125         ofs += size
 126     total = ofs
 127     return (shalist, total)
 128
 129
 130 def _squish(w, stacks, n):
 131     i = 0
 132     while i<n or len(stacks[i]) > MAX_PER_TREE:
 133         while len(stacks) <= i+1:
 134             stacks.append([])
 135         if len(stacks[i]) == 1:
 136             stacks[i+1] += stacks[i]
 137         elif stacks[i]:
 138             (shalist, size) = _make_shalist(stacks[i])
 139             tree = w.new_tree(shalist)
 140             stacks[i+1].append(('40000', tree, size))
 141         stacks[i] = []
 142         i += 1
 143
 144
 145 def split_to_shalist(w, files, keep_boundaries, progress=None):
 146     sl = _split_to_blobs(w, files, keep_boundaries, progress)
 147     if not fanout:
 148         shal = []
 149         for (sha,size,level) in sl:
 150             shal.append(('100644', sha, size))
 151         return _make_shalist(shal)[0]
 152     else:
 153         stacks = [[]]
 154         for (sha,size,level) in sl:
 155             stacks[0].append(('100644', sha, size))
 156             if level:
 157                 _squish(w, stacks, level)
 158         #log('stacks: %r\n' % [len(i) for i in stacks])
 159         _squish(w, stacks, len(stacks)-1)
 160         #log('stacks: %r\n' % [len(i) for i in stacks])
 161         return _make_shalist(stacks[-1])[0]
 162
 163
 164 def split_to_blob_or_tree(w, files, keep_boundaries):
 165     shalist = list(split_to_shalist(w, files, keep_boundaries))
 166     if len(shalist) == 1:
 167         return (shalist[0][0], shalist[0][2])
 168     elif len(shalist) == 0:
 169         return ('100644', w.new_blob(''))
 170     else:
 171         return ('40000', w.new_tree(shalist))
 172
 173
 174 def open_noatime(name):
 175     fd = _helpers.open_noatime(name)
 176     try:
 177         return os.fdopen(fd, 'rb', 1024*1024)
 178     except:
 179         try:
 180             os.close(fd)
 181         except:
 182             pass
 183         raise
 184
 185
 186 def fadvise_done(f, ofs):
 187     assert(ofs >= 0)
 188     if ofs > 0 and hasattr(f, 'fileno'):
 189         _helpers.fadvise_done(f.fileno(), ofs)