1
Integration of performance changes
2
==================================
4
To deliver a version of bzr with all our planned changes will require
5
significant integration work. Minimally each change needs to integrate with
6
some aspect of the bzr version it's merged into, but in reality many of these
7
changes while conceptually independent will in fact have to integrate with the
8
other changes we have planned before can have a completed system.
10
Additionally changes that alter disk formats are inherently more tricky to
11
integrate because we will often need to alter apis throughout the code base to
12
expose the increased or reduced model of the preferred disk format.
14
`Performance Tasks`_, which is generated from the Graphviz "dot" file ``performance.dot``, graphs out the dependencies to let us make
15
accurate assessments of the changes needed in terms of code and API, hopefully
16
minimising the number of different integration steps we have to take, while
17
giving us a broad surface area for development. It's based on a summary in the
18
next section of this document of the planned changes with their expected
19
collaborators and dependencies. Where a command is listed, the expectation is
20
that all uses of that command - local, remote, dumb transport and smart
21
transport are being addressed together.
24
The following provides a summary of the planned changes and their expected
25
collaborators within the code base, along with an estimate of whether they are
26
likely to require changes to their collaborators to be considered 'finished'.
28
* Use case target APIs: Each of these is likely to alter the Tree interface.
29
Some few of them focus on Branch and will alter Branch and Repository
30
accordingly. As they are targeted APIs we can deep changes all the way down
31
the stack to the underlying representation to make it all fit well.
32
Presenting a top level API for many things will be possible now as long as
33
the exposed data is audited for things we plan to make optional, or remove:
34
Such things cannot be present in the final API. Writing these APIs now will
35
provide strong feedback to the design process for those things which are
36
considered optional or removable, so these APIs should be implemented
37
before removing or making optional existing data.
39
* Deprecating versioned files as a supported API: This collaborates with the
40
Repository API but can probably be done by adding a replacement API for
41
places where the versioned-file api is used. We may well want to keep a
42
concept of 'a file over time' or 'inventories over time', so the existing
43
repository model of exposing versioned file objects may be ok; what we need
44
to ensure we do is remove the places in the code base where you create or
45
remove or otherwise describe manipulation of the storage by knit rather than
46
talking at the level of file ids and revision ids. The current
47
versioned-file API would be a burden for implementors of a blob based
48
repository format, so the removal of callers, and deprecation of those parts
49
of the API should be done before creating a blob based repository format.
51
* Creating a revision validator: Revision validators may depend on storage
52
layer changes to inventories so while we can create a revision validator
53
API, we cannot create the final one until we have the inventory structural
56
* Annotation caching API: This API is a prerequisite for new repository
57
formats. If written after they are introduced we may find that the
58
repository is lacking in functionality, so the API should be implemented
61
* _iter_changes based merging: If the current _iter_changes_ API is
62
insufficient, we should know about that before designing the disk format for
63
generating fast _iter_changes_ output.
65
* Network-efficient revision graph API: This influences what questions we will
66
want to ask a local repository very quickly; as such it's a driver for the
67
new repository format and should be in place first if possible. Its probably
68
not sufficiently different to local operations to make this a hard ordering
71
* Working tree disk ordering: Knowing the expected order for disk operations
72
may influence the needed use case specific APIs, so having a solid
73
understanding of what is optimal - and why - and whether it is pessimal on
74
non linux platforms is rather important.
76
* Be able to version files greater than memory in size: This cannot be
77
achieved until all parts of the library which deal with user files are able
78
to provide access to files larger than memory. Many strategies can be
79
considered for this - such as temporary files on disk, memory mapping etc.
80
We should have enough of a design laid out that developers of repository and
81
tree logic are able to start exposing apis, and considering requirements
82
related to them, to let this happen.
84
* Per-file graph access API: This should be implemented on top of or as part
85
of the newer API for accessing data about a file over time. It can be a
86
separate step easily; but as it's in the same area of the library should not
89
* Repository stacking API: The key dependency/change required for this is that
90
repositories must individually be happy with having partial data - e.g. many
91
ghosts. However the way the API needs to be used should be driven from the
92
command layer in, because its unclear at the moment what will work best.
94
* Revision stream API: This API will become clear as we streamline commands.
95
On the data insertion side commit will want to generate new data. The
96
commands pull, bundle, merge, push, possibly uncommit will want to copy
97
existing data in a streaming fashion.
99
* New container format: Its hard to tell what the right way to structure the
100
layering is. Probably having smooth layering down to the point that code
101
wants to operate on the containers directly will make this more clear. As
102
bundles will become a read-only branch & repository, the smart server wants
103
streaming-containers, and we are planning a pack based repository, it
104
appears that we will have three different direct container users. However,
105
the bundle user may in fact be fake - because it really is a repository.
107
* Separation of annotation cache: Making the disk changes to achieve this
108
depends on the new API being created. Bundles probably want to be
109
annotation-free, so they are a form of implementation of this and will need
110
the on-demand annotation facility.
112
* Repository operation disk ordering: Dramatically changing the ordering of
113
disk operations requires a new repository format. We have most of the
114
analysis done to be able to specify the desired ordering, so it should be
115
possible to write such a format now based on the container logic, but
116
without any of the inventory representation or delta representation changes.
117
This would for instance involve pack combining ordering the existing diffs
120
* Inventory representation: This has a dependency on what data is
121
dropped from the core and what is kept. Without those changes being known we
122
can implement a new representation, but it won't be a final one. One of the
123
services the new inventory representation is expected to deliver is one of
124
validators for subtrees -- a means of comparing just subtrees of two
125
inventories without comparing all the data within that subtree.
127
* Delta storage optimisation: This has a strict dependency on a new repository
128
format. Optimisation takes many forms - we probably cannot complete the
129
desired optimisations under knits though we could use xdelta within a
132
* Greatest distance from origin cache: The potential users of this exist
133
today, it is likely able to be implemented immediately, but we are not sure
134
that its needed anymore, so it is being shelved.
136
* Removing derivable data: Its very hard to do this while the derived data is
137
exposed in API's but not used by commands. Implemented the targeted API's
138
for our core use cases should allow use to remove accidental use of derived
139
data, making only explicit uses of it visible, and isolating the impact of
140
removing it : allowing us to experiment sensibly. This covers both dropping
141
the per-file merge graph and the hash-based-names proposals.
143
.. _`Performance Tasks`: performance.png
added
removed
doc/developers/planned-change-integration.txt
