1
==========================
2
KnitPack repository format
3
==========================
7
Using KnitPack repositories
8
===========================
13
KnitPack is a new repository format for Bazaar, which is expected to be
14
faster both locally and over the network, is usually more compact, and
15
will work with more FTP servers.
17
Our benchmarking results to date have been very promising. We fully expect
18
to make a pack-based format the default in the near future. We would
19
therefore like as many people as possible using KnitPack repositories,
20
benchmarking the results and telling us where improvements are still needed.
25
A small percentage of existing repositories may have some inconsistent
26
data within them. It's is a good idea to check the integrity of your
27
repositories before migrating them to knitpack format. To do this, run::
31
If that reports a problem, run this command::
35
Note that this can take many hours for repositories with deep history
36
so be sure to set aside some time for this if it is required.
38
Creating a new knitpack branch
39
------------------------------
41
If you're starting a project from scratch, it's easy to make it a
42
``knitpack`` one. Here's how::
47
bzr commit -m "initial import"
49
In other words, use the normal sequence of commands but add the
50
``--pack-0.92`` option to the ``init`` command.
52
**Note:** In bzr 0.92, this format was called ``knitpack-experimental``.
54
Creating a new knitpack repository
55
----------------------------------
57
If you're starting a project from scratch and wish to use a shared repository
58
for branches, you can make it a ``knitpack`` repository like this::
61
bzr init-repo --pack-0.92 .
65
bzr commit -m "initial import"
67
In other words, use the normal sequence of commands but add the
68
``--pack-0.92`` option to the ``init-repo`` command.
70
Upgrading an existing branch or repository to knitpack format
71
-------------------------------------------------------------
73
If you have an existing branch and wish to migrate it to
74
a ``knitpack`` format, use the ``upgrade`` command like this::
76
bzr upgrade --pack-0.92 path-to-my-branch
78
If you are using a shared repository, run::
80
bzr upgrade --pack-0.92 ROOT_OF_REPOSITORY
82
to upgrade the history database. Note that this will not
83
alter the branch format of each branch, so
84
you will need to also upgrade each branch individually
85
if you are upgrading from an old (e.g. < 0.17) bzr.
86
More modern bzr's will already have the branch format at
87
our latest branch format which adds support for tags.
89
Starting a new knitpack branch from one in an older format
90
----------------------------------------------------------
92
This can be done in one of several ways:
94
1. Create a new branch and pull into it
95
2. Create a standalone branch and upgrade its format
96
3. Create a knitpack shared repository and branch into it
98
Here are the commands for using the ``pull`` approach::
100
bzr init --pack-0.92 my-new-branch
102
bzr pull my-source-branch
104
Here are the commands for using the ``upgrade`` approach::
106
bzr branch my-source-branch my-new-branch
108
bzr upgrade --pack-0.92 .
110
Here are the commands for the shared repository approach::
113
bzr init-repo --pack-0.92 .
114
bzr branch my-source-branch my-new-branch
117
As a reminder, any of the above approaches can fail if the source branch
118
has inconsistent data within it and hasn't been reconciled yet. Please
119
be sure to check that before reporting problems.
121
Testing packs for bzr-svn users
122
-------------------------------
124
If you are using ``bzr-svn`` or are testing the prototype subtree support,
125
you can still use and assist in testing KnitPacks. The commands to use
126
are identical to the ones given above except that the name of the format
127
to use is ``knitpack-subtree-experimental``.
129
WARNING: Note that the subtree formats, ``dirstate-subtree`` and
130
``knitpack-subtree-experimental``, are **not** production strength yet and
131
may cause unexpected problems. They are required for the bzr-svn
132
plug-in but should otherwise only be used by people happy to live on the
133
bleeding edge. If you are using bzr-svn, you're on the bleeding edge anyway.
139
If you need any help or encounter any problems, please contact the developers
140
via the usual ways, i.e. chat to us on IRC or send a message to our mailing
141
list. See http://wiki.bazaar.canonical.com/BzrSupport for contact details.
147
Bazaar 0.92 adds a new format (experimental at first) implemented in
148
``bzrlib.repofmt.pack_repo.py``.
150
This format provides a knit-like interface which is quite compatible
151
with knit format repositories: you can get a VersionedFile for a
152
particular file-id, or for revisions, or for the inventory, even though
153
these do not correspond to single files on disk.
155
The on-disk format is that the repository directory contains these
156
files and subdirectories:
158
==================== =============================================
159
packs/ completed readonly packs
160
indices/ indices for completed packs
161
upload/ temporary files for packs currently being
163
obsolete_packs/ packs that have been repacked and are no
164
longer normally needed
165
pack-names index of all live packs
167
==================== =============================================
169
Note that for consistency we always write "indices" not "indexes".
171
This is implemented on top of pack files, which are written once from
172
start to end, then left alone. A pack consists of a body file, plus
173
several index files. There are four index files for each pack, which
174
have the same basename and an extension indicating the purpose of the
177
======== ========== ======================== ==========================
178
extn Purpose Key References
179
======== ========== ======================== ==========================
180
``.tix`` File texts ``file_id, revision_id`` per-file parents,
183
``.six`` Signatures ``revision_id,`` -
184
``.rix`` Revisions ``revision_id,`` revision parents
185
``.iix`` Inventory ``revision_id,`` revision parents,
187
======== ========== ======================== ==========================
189
Indices are accessed through the ``bzrlib.index.GraphIndex`` class.
190
Indices are stored as sorted files on disk. Each line is one record,
194
* a value string - for all these indices, this is an ascii decimal pair
195
of "offset length" giving the position of the referenced data within
197
* a list of zero or more reference lists
199
The reference lists let a graph be stored within the index. Each
200
reference list entry points to another entry in the same index. The
201
references are represented as a byte offset for the target within the
204
When a compression base is given, it indicates that the body of the text
205
or inventory is a forward delta from the referenced revision. The
206
compression base list must have length 0 or 1.
208
Like packs, indexes are written only once and then unmodified. A
209
GraphIndex builder is a mutable in-memory graph that can be sorted,
210
cross-referenced and written out when the write group completes.
212
There can also be index entries with a value of 'a' for absent. These
213
records exist just to be pointed to in a graph. This is used, for
214
example, to give the revision-parent pointer when the parent revision is
217
The data content for each record is a knit data chunk. The knits are
218
always unannotated - the annotations must be generated when needed.
219
(We'd like to cache/memoize the annotations.) The data hunks can be
220
moved between packs without needing to recompress them.
222
It is not possible to regenerate an index from the body file, because it
223
contains information stored in the knit index that's not in the body.
224
(In particular, the per-file graph is only stored in the index.)
225
We would like to change this in a future format.
227
The lock is a regular LockDir lock. The lock is only held for a much
228
reduced scope, while updating the pack-names file. The bulk of the
229
insertion can be done without the repository locked. This is an
230
implementation detail; the repository user should still call
231
``repository.lock_write`` at the regular time but be aware this does not
232
correspond to a physical mutex.
234
Read locks control caching but do not affect writers.
236
The newly-added repository write group concept is very important to
237
KnitPack repositories. When ``start_write_group`` is called, a new
238
temporary pack is created and all modifications to the repository will
239
go into it until either ``commit_write_group`` or ``abort_write_group``
240
is called, at which time it is either finished and moved into place or
241
discarded respectively. Write groups cannot be nested, only one can be
242
underway at a time on a Repository instance and they must occur within a
245
Normally the data for each revision will be entirely within a single
246
pack but this is not required.
248
When a pack is finished, it gets a final name based on the md5 of all
249
the data written into the pack body file.
251
The ``pack-names`` file gives the list of all finished non-obsolete
252
packs. (This should always be the same as the list of files in the
253
``packs/`` directory, but the file is needed for readonly http clients
254
that can't easily list directories, and it includes other information.)
255
The constraint on the ``pack-names`` list is that every file mentioned
256
must exist in the ``packs/`` directory.
258
In rare cases, when a writer is interrupted, about-to-be-removed packs
259
may still be present in the directory but removed from the list.
261
As well as the list of names, the pack-names file also contains the
262
size, in bytes, of each of the four indices. This is used to bootstrap
263
bisection search within the indices.
265
In normal use, one pack will be created for each commit to a repository.
266
This would build up to an inefficient number of files over time, so a
267
``repack`` operation is available to recombine them, by producing larger
268
files containing data on multiple revisions. This can be done manually
269
by running ``bzr pack``, and it also may happen automatically when a
270
write group is committed.
272
The repacking strategy used at the moment tries to balance not doing too
273
much work during commit with not having too many small files left in the
274
repository. The algorithm is roughly this: the total number of
275
revisions in the repository is expressed as a decimal number, e.g.
276
"532". Then we'll repack until we have five packs containing a hundred
277
revisions each, three packs containing ten revisions each, and two packs
278
with single revisions. This means that each revision will normally
279
initially be created in a single-revision pack, then moved to a
280
ten-revision pack, then to a 100-pack, and so on.
282
As with other repositories, in normal use data is only inserted.
283
However, in some circumstances we may want to garbage-collect or prune
284
existing data, or reconcile indexes.
287
vim: tw=72 ft=rst expandtab
added
removed
doc/developers/packrepo.txt
