4
This use case covers pulling in or pushing out some number of revisions which
5
is typically a small fraction of the number already present in the target
6
repository. Pushing and pulling are defined as branch level operations for ease
7
of interaction with VCS systems that have no repository abstraction (such as
8
bzr-svn or GNU Arch) but within bzrlib's core they are currently the
9
responsibility of the Repository object.
11
Functional Requirements
12
-----------------------
14
A push or pull operation must:
15
* Copy all the data to reconstruct the selected revisions in the target
16
branch. This is the goal of push and pull after all.
17
* Reject corrupt data. As bzr has no innate mechanism for discarding corrupted
18
data, corrupted data should not be incorporated accidentally.
20
Factors which should add work for push/pull
21
-------------------------------------------
23
* Baseline overhead: The time to connect to both branches.
24
* Actual new data in the revisions being pulled (drives the amount of data to
25
move around, includes the commit messages etc)
26
* Number of revisions in the two repositories (scaling affects the
27
determination of what revisions to move around).
32
1. New data is identified in the source repository.
33
2. That data is read from the source repository.
34
3. The same data is verified and written to the target repository in such a
35
manner that its not visible to readers until its ready for use.
37
New data identification
38
~~~~~~~~~~~~~~~~~~~~~~~
40
We have a single top level data object: revisions. Everything else is
41
subordinate to revisions, so determining the revisions to propagate should be
42
all thats needed. This depends on revisions with partial data - such as those
43
with no signature - being flagged in some efficient manner.
45
We could do this in two manners: determine revisions to sync and signatures to sync in two passes, or change the 'value' of a revision implicitly when the signature is different. E.g. by using merkle hash trees with the signature data a separate component the signatures will naturally be identified to sync.
47
We want to only exchange data proportional to the number of new revisions and
48
signatures in the system though. One way to achieve this for revisions is to
49
walk the graph out from the desired tips until the surface area intersection is
50
found. For signatures a set difference seems to be needed as there is no DAG of signatures: the presence of one has no implications on the presence of another, so a full pass over the set of signatures would be required to confirm no new signatures are needed (let alone replaced signatures).
52
IFF we can determine 'new revisions' and 'new signatures' without full graph access then we can scale acceptable for push and pull.
54
Ghosts are revisions which are not present in a particular repository. Filling ghosts refers to removing ghosts in the target repository when the ghost is present in the source repository. Filling ghosts can be either an explicit or implicit action. The common case is no ghosts.
56
Set synchronisation approaches
57
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
59
A set synchronisation approach is one which synchronises two sets without
60
regard for innate structure. This can be very efficient but requires adding a
61
new node to be processed with every commit. Caching of the results of the
62
various set based syncs I've seen is possible but because the data structures
63
look different depending on the tip revision being synced up to the cache needs
64
to be very complex. I recommend not using such an approach for the common case
65
pull because of the failure to scale. We can use such an approach for
66
synchronisation of new signatures and ghosts, which should be an explicit
69
DAG synchronisation approaches
70
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
72
A DAG based approach to synchronistion is one that uses the DAG structure to
73
determine the difference in present nodes. It can as a result operate from the
74
tip of the DAG backwards. A dag based approach should allow incremental access
75
to data and not require a full-graph scan for incremental operations.
80
We should read roughly as much of the revision level graph as is needed from
81
each repository to determine the node difference. If requested we should
82
perform a detailed scan to pick up ghost revisions and revisions which have had
83
signatures added. This should not be the default as it requires full history
86
Expected file IO and access pattern:
88
* Common case: repo with many branches of one project, to the same.
90
1. Source and Target branch tips read.
91
2. Find the tip of each branch in their repo (will require reading some of
92
the revision graph but is typically near the end of the graph).
93
3. Read and parse increasing amounts of the revision graph until one is
94
found to be a subset of the other, or a complete list of revisions to be
95
transmitted is created.
99
1. Repositories with many projects or branches which are very old may
100
require reading a lot of unrelated graph data.
102
1. Initial push/pull scenarios should not require reading an entire graph.
109
2. Determine one sided graph difference. To avoid obtaining a full graph over
110
the wire this needs to be done without reference to the full graph, and
111
with some logarthmic scaling algorithm. There are several already available
114
With ghost and new-signature detection:
116
* File IO access pattern will read the entire graph on the 'target' side - if
117
no ghosts are present then stop, otherwise seek the new revisions on the
118
source side with the regular algorithm and also explicitly search for the
119
ghost points from the target; plus a set difference search is needed on
122
* Semantic level can probably be tuned, but as its also complex I suggest
123
deferring analysis for optimal behaviour of this use case.
129
When transferring information about a revision the graph of data for the
130
revision is walked: revision -> inventory, revision -> matching signature,
131
inventory -> file ids:revision pairs.
136
As we're reading already committed data, as long as nothing is mutating data on
137
disk reading should be race free. We will:
139
- read each revision object
140
- read the matching inventory delta
141
- attempt to read a signature object
142
- parse the inventory delta
143
- read the fileid:revisionid compressed chunk for each line in the inventory
146
Theres no point validating that the data read is valid, as transmission through to the client writing the data might invalidate it; we need to validate before we write.
151
Given that we have established the revisions needed, a single API call should
152
suffice to obtain all data; the API should present the data in such an order
153
that it can be validated as it arrives and thus not require large scale
154
buffering on disk. Specifically each item of data should be validatable (e.g.
155
for some file data we want the fileid:revisionid:validationhash + content).
158
Data Verification and writing
159
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
161
New data written to a repository should be completed intact when it is made
162
visible. This suggests that either all the data for a revision must be made
163
atomically visible (e.g. by renaming a single file) or the leaf nodes of the
164
reference graph must become visible first.
166
Data is referred to via the following graph:
168
revision -> signature
169
revision -> inventory
170
inventory -> fileid:revisionid
171
fileid:revisionid -> fileid:revisionid
173
Data is verifiable via a different ordering:
174
signature -> revision -> inventory -> fileid:revisionid texts.
176
We dont gpg verify each revision today; this analysis only speaks to hash
177
verification of contents.
179
To validate a revision we need to validate the data it refers to. But to
180
validate the contents of a revision we need the new texts in the inventory for
181
the revision - to check a fileid:revisionid we need to know the expected sha1
182
of the full text and thus also need to read the delta chain to construct the
183
text as we accept it to determine if its valid. Providing separate validators
184
for the chosen representation would address this.
185
e.g: For an inventory entry FILEID:REVISIONID we store the validator of the
186
full text :SHA1:. If we also stored the validator of the chosen disk
187
representation (:DELTASHA1:) we could validate the transmitted representation
188
without expanding the delta in the common case. If that failed we could expand
189
the delta chain and try against the full text validator, and finally fail. As
190
different delta generators might generate different deltas, :DELTASHA1: should
191
not become part of the revision validator, only the inventory disk encoding. In
192
a related manner a transmission format that allowed cheap validation of content
193
without applying locally stored deltas would be advantageous because no local
194
reads would be incurred to validate new content. For instance, always sending a
195
full text for any file, possibly with a delta-chain when transmitting multiple
196
revisionids of the file, would allow this. (git pack-files have this property).
201
A single-file local format would allow safe atomic addition of data while
202
allowing optimisal transmission order of data. Failing this the validation of
203
data should be tuned to not require reading local texts during data addition
204
even in the presence of delta chains. We should have transmission-validators
205
separate from content validators that allow validation of the delta-transmitted
211
* Every new file text requires transmission and local serialisation.
212
* Every commit requires transmission and storage of a revision, signature and inventory.
214
Thus 4000 commits to a 50000 path tree of 10 files on averages requires (with
215
knits) between 26 writes (2*(3+10)) and 80006 (2*(4000*10 + 3)) writes. In all
216
cases there are 4000 * 13 distinct objects to record.
218
Grouping data by fileid, content and metadata, gives the figures above.
221
* File per full identifier (fileid:revisionid:meta|content): 104000
222
* Delta-chain per object: object id count * constant overhead per object id
224
* Collation/pack file: 1
226
Performance for these depends heavily on implementation:
227
- Using full ids we could name by validator or by id, giving best performance
228
that depends on either receiving data in validator order or in id order.
229
- using delta-chain per object we get least seek overhead and syscall overhead
230
if we recieve in topological order within the object id, and object ids in
232
- Using a collation/pack file we can stream it into place and validate as we go,
233
giving near ideal performance.
238
The api for writing new data recieved over the network will need to be geared
239
to the transmission and local storage method. What we need is for the
240
transmission method to reasonably closely match the desired write ordering
241
locally. This suggests that once we decide on the best local storage means we
242
should design the api.
245
take N commits from A to B, if B is local then merge changes into the tree.
246
copy ebough data to recreate snapshots
247
avoid ending up wth corrupt/bad data
254
look at graph of revisions for ~N comits to deretmine eligibility for
255
if preserve mainline is on, check LH only
257
identify objects to send that are not on the client repo
258
- revision - may be proportional to the graph
259
- inventory - proportional to work
260
- texts - proportional to work
265
* send data proportional to the new information
268
#. validate the sha1 of the full text of each transmitted text.
269
#. validate the sha1:name mapping in each newly referenced inventory item.
270
#. validate the sha1 of the XML of each inventory against the revision.
271
**this is proportional to tree size and must be fixed**
273
#. write the data to the local repo.
274
The API should output the file texts needed by the merge as by product of the transmission
278
Combine the output from the transmission step with additional 'new work data' for anything already in the local repository that is new in this tree.
279
should write new files and stat existing files proportional to the count of the new work and the size of the full texts.