~bzr-pqm/bzr/bzr.dev

3641.3.29 by John Arbash Meinel
Cleanup the copyright headers
1
# Copyright (C) 2008 Canonical Ltd
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
2
#
3
# This program is free software; you can redistribute it and/or modify
3641.3.29 by John Arbash Meinel
Cleanup the copyright headers
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
7
#
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11
# GNU General Public License for more details.
12
#
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
4183.7.1 by Sabin Iacob
update FSF mailing address
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
16
#
17
6379.6.7 by Jelmer Vernooij
Move importing from future until after doc string, otherwise the doc string will disappear.
18
"""ChunkWriter: write compressed data out with a fixed upper bound."""
19
6379.6.1 by Jelmer Vernooij
Import absolute_import in a few places.
20
from __future__ import absolute_import
21
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
22
import zlib
23
from zlib import Z_FINISH, Z_SYNC_FLUSH
24
25
26
class ChunkWriter(object):
27
    """ChunkWriter allows writing of compressed data with a fixed size.
28
29
    If less data is supplied than fills a chunk, the chunk is padded with
30
    NULL bytes. If more data is supplied, then the writer packs as much
31
    in as it can, but never splits any item it was given.
32
33
    The algorithm for packing is open to improvement! Current it is:
34
     - write the bytes given
35
     - if the total seen bytes so far exceeds the chunk size, flush.
3641.3.4 by John Arbash Meinel
Tweak some 'sum' lines.
36
37
    :cvar _max_repack: To fit the maximum number of entries into a node, we
38
        will sometimes start over and compress the whole list to get tighter
39
        packing. We get diminishing returns after a while, so this limits the
40
        number of times we will try.
3641.5.18 by John Arbash Meinel
Clean out the global state, good for prototyping and tuning, bad for production code.
41
        The default is to try to avoid recompressing entirely, but setting this
42
        to something like 20 will give maximum compression.
43
44
    :cvar _max_zsync: Another tunable nob. If _max_repack is set to 0, then you
45
        can limit the number of times we will try to pack more data into a
46
        node. This allows us to do a single compression pass, rather than
47
        trying until we overflow, and then recompressing again.
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
48
    """
3641.5.18 by John Arbash Meinel
Clean out the global state, good for prototyping and tuning, bad for production code.
49
    #    In testing, some values for bzr.dev::
50
    #        repack  time  MB   max   full
51
    #         1       7.5  4.6  1140  0
3777.5.1 by John Arbash Meinel
Add ChunkWriter.optimize(for_size=True)
52
    #         2       8.4  4.2  1036  1
3641.5.18 by John Arbash Meinel
Clean out the global state, good for prototyping and tuning, bad for production code.
53
    #         3       9.8  4.1  1012  278
54
    #         4      10.8  4.1  728   945
55
    #        20      11.1  4.1  0     1012
56
    #        repack = 0
3777.5.1 by John Arbash Meinel
Add ChunkWriter.optimize(for_size=True)
57
    #        zsync   time  MB    repack  stop_for_z
58
    #         0       5.0  24.7  0       6270
59
    #         1       4.3  13.2  0       3342
60
    #         2       4.9   9.6  0       2414
61
    #         5       4.8   6.2  0       1549
62
    #         6       4.8   5.8  1       1435
63
    #         7       4.8   5.5  19      1337
64
    #         8       4.4   5.3  81      1220
65
    #        10       5.3   5.0  260     967
66
    #        11       5.3   4.9  366     839
67
    #        12       5.1   4.8  454     731
68
    #        15       5.8   4.7  704     450
69
    #        20       5.8   4.6  1133    7
3641.5.18 by John Arbash Meinel
Clean out the global state, good for prototyping and tuning, bad for production code.
70
71
    #    In testing, some values for mysql-unpacked::
72
    #                next_bytes estim
3777.5.1 by John Arbash Meinel
Add ChunkWriter.optimize(for_size=True)
73
    #        repack  time  MB    full    stop_for_repack
74
    #         1            15.4  0       3913
75
    #         2      35.4  13.7  0       346
76
    #        20      46.7  13.4  3380    0
3641.5.18 by John Arbash Meinel
Clean out the global state, good for prototyping and tuning, bad for production code.
77
    #        repack=0
3777.5.1 by John Arbash Meinel
Add ChunkWriter.optimize(for_size=True)
78
    #        zsync                       stop_for_z
79
    #         0      29.5 116.5  0       29782
80
    #         1      27.8  60.2  0       15356
81
    #         2      27.8  42.4  0       10822
82
    #         5      26.8  25.5  0       6491
83
    #         6      27.3  23.2  13      5896
84
    #         7      27.5  21.6  29      5451
85
    #         8      27.1  20.3  52      5108
86
    #        10      29.4  18.6  195     4526
87
    #        11      29.2  18.0  421     4143
88
    #        12      28.0  17.5  702     3738
89
    #        15      28.9  16.5  1223    2969
90
    #        20      29.6  15.7  2182    1810
91
    #        30      31.4  15.4  3891    23
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
92
3777.5.1 by John Arbash Meinel
Add ChunkWriter.optimize(for_size=True)
93
    # Tuple of (num_repack_attempts, num_zsync_attempts)
94
    # num_zsync_attempts only has meaning if num_repack_attempts is 0.
95
    _repack_opts_for_speed = (0, 8)
96
    _repack_opts_for_size = (20, 0)
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
97
3777.5.2 by John Arbash Meinel
Change the name to ChunkWriter.set_optimize()
98
    def __init__(self, chunk_size, reserved=0, optimize_for_size=False):
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
99
        """Create a ChunkWriter to write chunk_size chunks.
100
101
        :param chunk_size: The total byte count to emit at the end of the
102
            chunk.
103
        :param reserved: How many bytes to allow for reserved data. reserved
3641.5.19 by John Arbash Meinel
Documentation cleanup pass.
104
            data space can only be written to via the write(..., reserved=True).
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
105
        """
106
        self.chunk_size = chunk_size
107
        self.compressor = zlib.compressobj()
108
        self.bytes_in = []
109
        self.bytes_list = []
3641.3.16 by John Arbash Meinel
Somewhat surprisingly, tracking bytes_out_len makes a
110
        self.bytes_out_len = 0
3641.5.2 by John Arbash Meinel
(broken, but hopeful) Change the compact logic.
111
        # bytes that have been seen, but not included in a flush to out yet
112
        self.unflushed_in_bytes = 0
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
113
        self.num_repack = 0
3641.5.12 by John Arbash Meinel
Play around with max_repack=0 and limiting work done based on
114
        self.num_zsync = 0
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
115
        self.unused_bytes = None
116
        self.reserved_size = reserved
3777.5.1 by John Arbash Meinel
Add ChunkWriter.optimize(for_size=True)
117
        # Default is to make building fast rather than compact
3777.5.2 by John Arbash Meinel
Change the name to ChunkWriter.set_optimize()
118
        self.set_optimize(for_size=optimize_for_size)
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
119
120
    def finish(self):
121
        """Finish the chunk.
122
123
        This returns the final compressed chunk, and either None, or the
124
        bytes that did not fit in the chunk.
3641.5.19 by John Arbash Meinel
Documentation cleanup pass.
125
126
        :return: (compressed_bytes, unused_bytes, num_nulls_needed)
5891.1.3 by Andrew Bennetts
Move docstring formatting fixes.
127
128
            * compressed_bytes: a list of bytes that were output from the
129
              compressor. If the compressed length was not exactly chunk_size,
130
              the final string will be a string of all null bytes to pad this
131
              to chunk_size
132
            * unused_bytes: None, or the last bytes that were added, which we
133
              could not fit.
134
            * num_nulls_needed: How many nulls are padded at the end
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
135
        """
136
        self.bytes_in = None # Free the data cached so far, we don't need it
3641.3.16 by John Arbash Meinel
Somewhat surprisingly, tracking bytes_out_len makes a
137
        out = self.compressor.flush(Z_FINISH)
138
        self.bytes_list.append(out)
139
        self.bytes_out_len += len(out)
3641.5.12 by John Arbash Meinel
Play around with max_repack=0 and limiting work done based on
140
3641.3.16 by John Arbash Meinel
Somewhat surprisingly, tracking bytes_out_len makes a
141
        if self.bytes_out_len > self.chunk_size:
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
142
            raise AssertionError('Somehow we ended up with too much'
143
                                 ' compressed data, %d > %d'
3641.3.16 by John Arbash Meinel
Somewhat surprisingly, tracking bytes_out_len makes a
144
                                 % (self.bytes_out_len, self.chunk_size))
3644.2.3 by John Arbash Meinel
Do a bit more work to get all the tests to pass.
145
        nulls_needed = self.chunk_size - self.bytes_out_len
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
146
        if nulls_needed:
147
            self.bytes_list.append("\x00" * nulls_needed)
148
        return self.bytes_list, self.unused_bytes, nulls_needed
149
3777.5.2 by John Arbash Meinel
Change the name to ChunkWriter.set_optimize()
150
    def set_optimize(self, for_size=True):
151
        """Change how we optimize our writes.
152
153
        :param for_size: If True, optimize for minimum space usage, otherwise
154
            optimize for fastest writing speed.
155
        :return: None
156
        """
157
        if for_size:
158
            opts = ChunkWriter._repack_opts_for_size
159
        else:
160
            opts = ChunkWriter._repack_opts_for_speed
161
        self._max_repack, self._max_zsync = opts
162
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
163
    def _recompress_all_bytes_in(self, extra_bytes=None):
3641.3.12 by John Arbash Meinel
Collect some info on the space/time tradeoff for _max_repack.
164
        """Recompress the current bytes_in, and optionally more.
165
3641.5.19 by John Arbash Meinel
Documentation cleanup pass.
166
        :param extra_bytes: Optional, if supplied we will add it with
3641.3.12 by John Arbash Meinel
Collect some info on the space/time tradeoff for _max_repack.
167
            Z_SYNC_FLUSH
3641.5.19 by John Arbash Meinel
Documentation cleanup pass.
168
        :return: (bytes_out, bytes_out_len, alt_compressed)
5891.1.3 by Andrew Bennetts
Move docstring formatting fixes.
169
170
            * bytes_out: is the compressed bytes returned from the compressor
171
            * bytes_out_len: the length of the compressed output
172
            * compressor: An object with everything packed in so far, and
173
              Z_SYNC_FLUSH called.
3641.3.12 by John Arbash Meinel
Collect some info on the space/time tradeoff for _max_repack.
174
        """
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
175
        compressor = zlib.compressobj()
176
        bytes_out = []
3641.3.5 by John Arbash Meinel
For iter_all and three_level tests adjust spill-at.
177
        append = bytes_out.append
178
        compress = compressor.compress
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
179
        for accepted_bytes in self.bytes_in:
3641.3.5 by John Arbash Meinel
For iter_all and three_level tests adjust spill-at.
180
            out = compress(accepted_bytes)
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
181
            if out:
3641.3.5 by John Arbash Meinel
For iter_all and three_level tests adjust spill-at.
182
                append(out)
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
183
        if extra_bytes:
3641.3.5 by John Arbash Meinel
For iter_all and three_level tests adjust spill-at.
184
            out = compress(extra_bytes)
3641.5.10 by John Arbash Meinel
Only Z_SYNC_FLUSH when we have extra bytes.
185
            out += compressor.flush(Z_SYNC_FLUSH)
186
            append(out)
3641.3.27 by John Arbash Meinel
Bringing reserved in as a keyword to write() also saves some time.
187
        bytes_out_len = sum(map(len, bytes_out))
188
        return bytes_out, bytes_out_len, compressor
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
189
3641.3.27 by John Arbash Meinel
Bringing reserved in as a keyword to write() also saves some time.
190
    def write(self, bytes, reserved=False):
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
191
        """Write some bytes to the chunk.
192
193
        If the bytes fit, False is returned. Otherwise True is returned
194
        and the bytes have not been added to the chunk.
3641.5.19 by John Arbash Meinel
Documentation cleanup pass.
195
196
        :param bytes: The bytes to include
197
        :param reserved: If True, we can use the space reserved in the
198
            constructor.
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
199
        """
3641.5.2 by John Arbash Meinel
(broken, but hopeful) Change the compact logic.
200
        if self.num_repack > self._max_repack and not reserved:
201
            self.unused_bytes = bytes
202
            return True
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
203
        if reserved:
204
            capacity = self.chunk_size
205
        else:
206
            capacity = self.chunk_size - self.reserved_size
3641.3.27 by John Arbash Meinel
Bringing reserved in as a keyword to write() also saves some time.
207
        comp = self.compressor
3641.5.19 by John Arbash Meinel
Documentation cleanup pass.
208
3641.5.2 by John Arbash Meinel
(broken, but hopeful) Change the compact logic.
209
        # Check to see if the currently unflushed bytes would fit with a bit of
210
        # room to spare, assuming no compression.
211
        next_unflushed = self.unflushed_in_bytes + len(bytes)
212
        remaining_capacity = capacity - self.bytes_out_len - 10
213
        if (next_unflushed < remaining_capacity):
3641.5.19 by John Arbash Meinel
Documentation cleanup pass.
214
            # looks like it will fit
3641.3.27 by John Arbash Meinel
Bringing reserved in as a keyword to write() also saves some time.
215
            out = comp.compress(bytes)
3641.3.11 by John Arbash Meinel
Start working on an alternate way to track compressed_chunk state.
216
            if out:
217
                self.bytes_list.append(out)
3641.3.16 by John Arbash Meinel
Somewhat surprisingly, tracking bytes_out_len makes a
218
                self.bytes_out_len += len(out)
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
219
            self.bytes_in.append(bytes)
3641.5.2 by John Arbash Meinel
(broken, but hopeful) Change the compact logic.
220
            self.unflushed_in_bytes += len(bytes)
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
221
        else:
222
            # This may or may not fit, try to add it with Z_SYNC_FLUSH
3641.5.5 by John Arbash Meinel
Document my attempt to use copy() as a look-ahead.
223
            # Note: It is tempting to do this as a look-ahead pass, and to
3641.5.19 by John Arbash Meinel
Documentation cleanup pass.
224
            #       'copy()' the compressor before flushing. However, it seems
225
            #       that Which means that it is the same thing as increasing
226
            #       repack, similar cost, same benefit. And this way we still
227
            #       have the 'repack' knob that can be adjusted, and not depend
228
            #       on a platform-specific 'copy()' function.
3641.5.12 by John Arbash Meinel
Play around with max_repack=0 and limiting work done based on
229
            self.num_zsync += 1
230
            if self._max_repack == 0 and self.num_zsync > self._max_zsync:
231
                self.num_repack += 1
3641.5.19 by John Arbash Meinel
Documentation cleanup pass.
232
                self.unused_bytes = bytes
3641.5.12 by John Arbash Meinel
Play around with max_repack=0 and limiting work done based on
233
                return True
3641.3.27 by John Arbash Meinel
Bringing reserved in as a keyword to write() also saves some time.
234
            out = comp.compress(bytes)
235
            out += comp.flush(Z_SYNC_FLUSH)
3641.5.2 by John Arbash Meinel
(broken, but hopeful) Change the compact logic.
236
            self.unflushed_in_bytes = 0
3641.3.15 by John Arbash Meinel
Now that we have real data, remove the copy() code.
237
            if out:
238
                self.bytes_list.append(out)
3641.3.16 by John Arbash Meinel
Somewhat surprisingly, tracking bytes_out_len makes a
239
                self.bytes_out_len += len(out)
3641.5.3 by John Arbash Meinel
If we repack earlier, it catches this case.
240
241
            # We are a bit extra conservative, because it seems that you *can*
242
            # get better compression with Z_SYNC_FLUSH than a full compress. It
243
            # is probably very rare, but we were able to trigger it.
3641.5.4 by John Arbash Meinel
Using a different safety margin for the first repack,
244
            if self.num_repack == 0:
245
                safety_margin = 100
246
            else:
247
                safety_margin = 10
248
            if self.bytes_out_len + safety_margin <= capacity:
3641.5.1 by John Arbash Meinel
Update the stats for the current code layout.
249
                # It fit, so mark it added
250
                self.bytes_in.append(bytes)
251
            else:
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
252
                # We are over budget, try to squeeze this in without any
253
                # Z_SYNC_FLUSH calls
254
                self.num_repack += 1
3641.5.1 by John Arbash Meinel
Update the stats for the current code layout.
255
                (bytes_out, this_len,
256
                 compressor) = self._recompress_all_bytes_in(bytes)
257
                if self.num_repack >= self._max_repack:
258
                    # When we get *to* _max_repack, bump over so that the
259
                    # earlier > _max_repack will be triggered.
260
                    self.num_repack += 1
3641.3.16 by John Arbash Meinel
Somewhat surprisingly, tracking bytes_out_len makes a
261
                if this_len + 10 > capacity:
3641.5.1 by John Arbash Meinel
Update the stats for the current code layout.
262
                    (bytes_out, this_len,
263
                     compressor) = self._recompress_all_bytes_in()
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
264
                    self.compressor = compressor
3641.5.2 by John Arbash Meinel
(broken, but hopeful) Change the compact logic.
265
                    # Force us to not allow more data
266
                    self.num_repack = self._max_repack + 1
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
267
                    self.bytes_list = bytes_out
3641.3.27 by John Arbash Meinel
Bringing reserved in as a keyword to write() also saves some time.
268
                    self.bytes_out_len = this_len
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
269
                    self.unused_bytes = bytes
270
                    return True
271
                else:
272
                    # This fits when we pack it tighter, so use the new packing
273
                    self.compressor = compressor
274
                    self.bytes_in.append(bytes)
275
                    self.bytes_list = bytes_out
3641.3.16 by John Arbash Meinel
Somewhat surprisingly, tracking bytes_out_len makes a
276
                    self.bytes_out_len = this_len
3641.3.1 by John Arbash Meinel
Bring in the btree_index and chunk_writer code and their tests.
277
        return False
278