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