commit-graph: allow cross-alternate chains
[git/git.git] / Documentation / technical / commit-graph.txt
1 Git Commit Graph Design Notes
2 =============================
4 Git walks the commit graph for many reasons, including:
6 1. Listing and filtering commit history.
7 2. Computing merge bases.
9 These operations can become slow as the commit count grows. The merge
10 base calculation shows up in many user-facing commands, such as 'merge-base'
11 or 'status' and can take minutes to compute depending on history shape.
13 There are two main costs here:
15 1. Decompressing and parsing commits.
16 2. Walking the entire graph to satisfy topological order constraints.
18 The commit-graph file is a supplemental data structure that accelerates
19 commit graph walks. If a user downgrades or disables the 'core.commitGraph'
20 config setting, then the existing ODB is sufficient. The file is stored
21 as "commit-graph" either in the .git/objects/info directory or in the info
22 directory of an alternate.
24 The commit-graph file stores the commit graph structure along with some
25 extra metadata to speed up graph walks. By listing commit OIDs in lexi-
26 cographic order, we can identify an integer position for each commit and
27 refer to the parents of a commit using those integer positions. We use
28 binary search to find initial commits and then use the integer positions
29 for fast lookups during the walk.
31 A consumer may load the following info for a commit from the graph:
33 1. The commit OID.
34 2. The list of parents, along with their integer position.
35 3. The commit date.
36 4. The root tree OID.
37 5. The generation number (see definition below).
39 Values 1-4 satisfy the requirements of parse_commit_gently().
41 Define the "generation number" of a commit recursively as follows:
43 * A commit with no parents (a root commit) has generation number one.
45 * A commit with at least one parent has generation number one more than
46 the largest generation number among its parents.
48 Equivalently, the generation number of a commit A is one more than the
49 length of a longest path from A to a root commit. The recursive definition
50 is easier to use for computation and observing the following property:
52 If A and B are commits with generation numbers N and M, respectively,
53 and N <= M, then A cannot reach B. That is, we know without searching
54 that B is not an ancestor of A because it is further from a root commit
55 than A.
57 Conversely, when checking if A is an ancestor of B, then we only need
58 to walk commits until all commits on the walk boundary have generation
59 number at most N. If we walk commits using a priority queue seeded by
60 generation numbers, then we always expand the boundary commit with highest
61 generation number and can easily detect the stopping condition.
63 This property can be used to significantly reduce the time it takes to
64 walk commits and determine topological relationships. Without generation
65 numbers, the general heuristic is the following:
67 If A and B are commits with commit time X and Y, respectively, and
68 X < Y, then A _probably_ cannot reach B.
70 This heuristic is currently used whenever the computation is allowed to
71 violate topological relationships due to clock skew (such as "git log"
72 with default order), but is not used when the topological order is
73 required (such as merge base calculations, "git log --graph").
75 In practice, we expect some commits to be created recently and not stored
76 in the commit graph. We can treat these commits as having "infinite"
77 generation number and walk until reaching commits with known generation
78 number.
80 We use the macro GENERATION_NUMBER_INFINITY = 0xFFFFFFFF to mark commits not
81 in the commit-graph file. If a commit-graph file was written by a version
82 of Git that did not compute generation numbers, then those commits will
83 have generation number represented by the macro GENERATION_NUMBER_ZERO = 0.
85 Since the commit-graph file is closed under reachability, we can guarantee
86 the following weaker condition on all commits:
88 If A and B are commits with generation numbers N amd M, respectively,
89 and N < M, then A cannot reach B.
91 Note how the strict inequality differs from the inequality when we have
92 fully-computed generation numbers. Using strict inequality may result in
93 walking a few extra commits, but the simplicity in dealing with commits
94 with generation number *_INFINITY or *_ZERO is valuable.
96 We use the macro GENERATION_NUMBER_MAX = 0x3FFFFFFF to for commits whose
97 generation numbers are computed to be at least this value. We limit at
98 this value since it is the largest value that can be stored in the
99 commit-graph file using the 30 bits available to generation numbers. This
100 presents another case where a commit can have generation number equal to
101 that of a parent.
103 Design Details
104 --------------
106 - The commit-graph file is stored in a file named 'commit-graph' in the
107 .git/objects/info directory. This could be stored in the info directory
108 of an alternate.
110 - The core.commitGraph config setting must be on to consume graph files.
112 - The file format includes parameters for the object ID hash function,
113 so a future change of hash algorithm does not require a change in format.
115 - Commit grafts and replace objects can change the shape of the commit
116 history. The latter can also be enabled/disabled on the fly using
117 `--no-replace-objects`. This leads to difficultly storing both possible
118 interpretations of a commit id, especially when computing generation
119 numbers. The commit-graph will not be read or written when
120 replace-objects or grafts are present.
122 - Shallow clones create grafts of commits by dropping their parents. This
123 leads the commit-graph to think those commits have generation number 1.
124 If and when those commits are made unshallow, those generation numbers
125 become invalid. Since shallow clones are intended to restrict the commit
126 history to a very small set of commits, the commit-graph feature is less
127 helpful for these clones, anyway. The commit-graph will not be read or
128 written when shallow commits are present.
130 Commit Graphs Chains
131 --------------------
133 Typically, repos grow with near-constant velocity (commits per day). Over time,
134 the number of commits added by a fetch operation is much smaller than the
135 number of commits in the full history. By creating a "chain" of commit-graphs,
136 we enable fast writes of new commit data without rewriting the entire commit
137 history -- at least, most of the time.
139 ## File Layout
141 A commit-graph chain uses multiple files, and we use a fixed naming convention
142 to organize these files. Each commit-graph file has a name
143 `$OBJDIR/info/commit-graphs/graph-{hash}.graph` where `{hash}` is the hex-
144 valued hash stored in the footer of that file (which is a hash of the file's
145 contents before that hash). For a chain of commit-graph files, a plain-text
146 file at `$OBJDIR/info/commit-graphs/commit-graph-chain` contains the
147 hashes for the files in order from "lowest" to "highest".
149 For example, if the `commit-graph-chain` file contains the lines
151 ```
152 {hash0}
153 {hash1}
154 {hash2}
155 ```
157 then the commit-graph chain looks like the following diagram:
159 +-----------------------+
160 | graph-{hash2}.graph |
161 +-----------------------+
162 |
163 +-----------------------+
164 | |
165 | graph-{hash1}.graph |
166 | |
167 +-----------------------+
168 |
169 +-----------------------+
170 | |
171 | |
172 | |
173 | graph-{hash0}.graph |
174 | |
175 | |
176 | |
177 +-----------------------+
179 Let X0 be the number of commits in `graph-{hash0}.graph`, X1 be the number of
180 commits in `graph-{hash1}.graph`, and X2 be the number of commits in
181 `graph-{hash2}.graph`. If a commit appears in position i in `graph-{hash2}.graph`,
182 then we interpret this as being the commit in position (X0 + X1 + i), and that
183 will be used as its "graph position". The commits in `graph-{hash2}.graph` use these
184 positions to refer to their parents, which may be in `graph-{hash1}.graph` or
185 `graph-{hash0}.graph`. We can navigate to an arbitrary commit in position j by checking
186 its containment in the intervals [0, X0), [X0, X0 + X1), [X0 + X1, X0 + X1 +
187 X2).
189 Each commit-graph file (except the base, `graph-{hash0}.graph`) contains data
190 specifying the hashes of all files in the lower layers. In the above example,
191 `graph-{hash1}.graph` contains `{hash0}` while `graph-{hash2}.graph` contains
192 `{hash0}` and `{hash1}`.
194 ## Merging commit-graph files
196 If we only added a new commit-graph file on every write, we would run into a
197 linear search problem through many commit-graph files. Instead, we use a merge
198 strategy to decide when the stack should collapse some number of levels.
200 The diagram below shows such a collapse. As a set of new commits are added, it
201 is determined by the merge strategy that the files should collapse to
202 `graph-{hash1}`. Thus, the new commits, the commits in `graph-{hash2}` and
203 the commits in `graph-{hash1}` should be combined into a new `graph-{hash3}`
204 file.
206 +---------------------+
207 | |
208 | (new commits) |
209 | |
210 +---------------------+
211 | |
212 +-----------------------+ +---------------------+
213 | graph-{hash2} |->| |
214 +-----------------------+ +---------------------+
215 | | |
216 +-----------------------+ +---------------------+
217 | | | |
218 | graph-{hash1} |->| |
219 | | | |
220 +-----------------------+ +---------------------+
221 | tmp_graphXXX
222 +-----------------------+
223 | |
224 | |
225 | |
226 | graph-{hash0} |
227 | |
228 | |
229 | |
230 +-----------------------+
232 During this process, the commits to write are combined, sorted and we write the
233 contents to a temporary file, all while holding a `commit-graph-chain.lock`
234 lock-file. When the file is flushed, we rename it to `graph-{hash3}`
235 according to the computed `{hash3}`. Finally, we write the new chain data to
236 `commit-graph-chain.lock`:
238 ```
239 {hash3}
240 {hash0}
241 ```
243 We then close the lock-file.
245 ## Merge Strategy
247 When writing a set of commits that do not exist in the commit-graph stack of
248 height N, we default to creating a new file at level N + 1. We then decide to
249 merge with the Nth level if one of two conditions hold:
251 1. The expected file size for level N + 1 is at least half the file size for
252 level N.
254 2. Level N + 1 contains more than 64,0000 commits.
256 This decision cascades down the levels: when we merge a level we create a new
257 set of commits that then compares to the next level.
259 The first condition bounds the number of levels to be logarithmic in the total
260 number of commits. The second condition bounds the total number of commits in
261 a `graph-{hashN}` file and not in the `commit-graph` file, preventing
262 significant performance issues when the stack merges and another process only
263 partially reads the previous stack.
265 The merge strategy values (2 for the size multiple, 64,000 for the maximum
266 number of commits) could be extracted into config settings for full
267 flexibility.
269 ## Chains across multiple object directories
271 In a repo with alternates, we look for the `commit-graph-chain` file starting
272 in the local object directory and then in each alternate. The first file that
273 exists defines our chain. As we look for the `graph-{hash}` files for
274 each `{hash}` in the chain file, we follow the same pattern for the host
275 directories.
277 This allows commit-graphs to be split across multiple forks in a fork network.
278 The typical case is a large "base" repo with many smaller forks.
280 As the base repo advances, it will likely update and merge its commit-graph
281 chain more frequently than the forks. If a fork updates their commit-graph after
282 the base repo, then it should "reparent" the commit-graph chain onto the new
283 chain in the base repo. When reading each `graph-{hash}` file, we track
284 the object directory containing it. During a write of a new commit-graph file,
285 we check for any changes in the source object directory and read the
286 `commit-graph-chain` file for that source and create a new file based on those
287 files. During this "reparent" operation, we necessarily need to collapse all
288 levels in the fork, as all of the files are invalid against the new base file.
290 It is crucial to be careful when cleaning up "unreferenced" `graph-{hash}.graph`
291 files in this scenario. It falls to the user to define the proper settings for
292 their custom environment:
294 1. When merging levels in the base repo, the unreferenced files may still be
295 referenced by chains from fork repos.
297 2. The expiry time should be set to a length of time such that every fork has
298 time to recompute their commit-graph chain to "reparent" onto the new base
299 file(s).
301 3. If the commit-graph chain is updated in the base, the fork will not have
302 access to the new chain until its chain is updated to reference those files.
303 (This may change in the future [5].)
305 Related Links
306 -------------
307 [0]
308 Chromium work item for: Serialized Commit Graph
310 [1]
311 An abandoned patch that introduced generation numbers.
313 [2]
314 Discussion about generation numbers on commits and how they interact
315 with fsck.
317 [3]
318 More discussion about generation numbers and not storing them inside
319 commit objects. A valuable quote:
321 "I think we should be moving more in the direction of keeping
322 repo-local caches for optimizations. Reachability bitmaps have been
323 a big performance win. I think we should be doing the same with our
324 properties of commits. Not just generation numbers, but making it
325 cheap to access the graph structure without zlib-inflating whole
326 commit objects (i.e., packv4 or something like the "metapacks" I
327 proposed a few years ago)."
329 [4]
330 A patch to remove the ahead-behind calculation from 'status'.
332 [5]
333 A discussion of a "two-dimensional graph position" that can allow reading
334 multiple commit-graph chains at the same time.