Documentation/technical/partial-clone.txt

   1 Partial Clone Design Notes
   2 ==========================
   3
   4 The "Partial Clone" feature is a performance optimization for Git that
   5 allows Git to function without having a complete copy of the repository.
   6 The goal of this work is to allow Git better handle extremely large
   7 repositories.
   8
   9 During clone and fetch operations, Git downloads the complete contents
  10 and history of the repository.  This includes all commits, trees, and
  11 blobs for the complete life of the repository.  For extremely large
  12 repositories, clones can take hours (or days) and consume 100+GiB of disk
  13 space.
  14
  15 Often in these repositories there are many blobs and trees that the user
  16 does not need such as:
  17
  18   1. files outside of the user's work area in the tree.  For example, in
  19      a repository with 500K directories and 3.5M files in every commit,
  20      we can avoid downloading many objects if the user only needs a
  21      narrow "cone" of the source tree.
  22
  23   2. large binary assets.  For example, in a repository where large build
  24      artifacts are checked into the tree, we can avoid downloading all
  25      previous versions of these non-mergeable binary assets and only
  26      download versions that are actually referenced.
  27
  28 Partial clone allows us to avoid downloading such unneeded objects *in
  29 advance* during clone and fetch operations and thereby reduce download
  30 times and disk usage.  Missing objects can later be "demand fetched"
  31 if/when needed.
  32
  33 A remote that can later provide the missing objects is called a
  34 promisor remote, as it promises to send the objects when
  35 requested. Initially Git supported only one promisor remote, the origin
  36 remote from which the user cloned and that was configured in the
  37 "extensions.partialClone" config option. Later support for more than
  38 one promisor remote has been implemented.
  39
  40 Use of partial clone requires that the user be online and the origin
  41 remote or other promisor remotes be available for on-demand fetching
  42 of missing objects.  This may or may not be problematic for the user.
  43 For example, if the user can stay within the pre-selected subset of
  44 the source tree, they may not encounter any missing objects.
  45 Alternatively, the user could try to pre-fetch various objects if they
  46 know that they are going offline.
  47
  48
  49 Non-Goals
  50 ---------
  51
  52 Partial clone is a mechanism to limit the number of blobs and trees downloaded
  53 *within* a given range of commits -- and is therefore independent of and not
  54 intended to conflict with existing DAG-level mechanisms to limit the set of
  55 requested commits (i.e. shallow clone, single branch, or fetch '<refspec>').
  56
  57
  58 Design Overview
  59 ---------------
  60
  61 Partial clone logically consists of the following parts:
  62
  63 - A mechanism for the client to describe unneeded or unwanted objects to
  64   the server.
  65
  66 - A mechanism for the server to omit such unwanted objects from packfiles
  67   sent to the client.
  68
  69 - A mechanism for the client to gracefully handle missing objects (that
  70   were previously omitted by the server).
  71
  72 - A mechanism for the client to backfill missing objects as needed.
  73
  74
  75 Design Details
  76 --------------
  77
  78 - A new pack-protocol capability "filter" is added to the fetch-pack and
  79   upload-pack negotiation.
  80 +
  81 This uses the existing capability discovery mechanism.
  82 See "filter" in Documentation/technical/pack-protocol.txt.
  83
  84 - Clients pass a "filter-spec" to clone and fetch which is passed to the
  85   server to request filtering during packfile construction.
  86 +
  87 There are various filters available to accommodate different situations.
  88 See "--filter=<filter-spec>" in Documentation/rev-list-options.txt.
  89
  90 - On the server pack-objects applies the requested filter-spec as it
  91   creates "filtered" packfiles for the client.
  92 +
  93 These filtered packfiles are *incomplete* in the traditional sense because
  94 they may contain objects that reference objects not contained in the
  95 packfile and that the client doesn't already have.  For example, the
  96 filtered packfile may contain trees or tags that reference missing blobs
  97 or commits that reference missing trees.
  98
  99 - On the client these incomplete packfiles are marked as "promisor packfiles"
 100   and treated differently by various commands.
 101
 102 - On the client a repository extension is added to the local config to
 103   prevent older versions of git from failing mid-operation because of
 104   missing objects that they cannot handle.
 105   See "extensions.partialClone" in Documentation/technical/repository-version.txt"
 106
 107
 108 Handling Missing Objects
 109 ------------------------
 110
 111 - An object may be missing due to a partial clone or fetch, or missing
 112   due to repository corruption.  To differentiate these cases, the
 113   local repository specially indicates such filtered packfiles
 114   obtained from promisor remotes as "promisor packfiles".
 115 +
 116 These promisor packfiles consist of a "<name>.promisor" file with
 117 arbitrary contents (like the "<name>.keep" files), in addition to
 118 their "<name>.pack" and "<name>.idx" files.
 119
 120 - The local repository considers a "promisor object" to be an object that
 121   it knows (to the best of its ability) that promisor remotes have promised
 122   that they have, either because the local repository has that object in one of
 123   its promisor packfiles, or because another promisor object refers to it.
 124 +
 125 When Git encounters a missing object, Git can see if it is a promisor object
 126 and handle it appropriately.  If not, Git can report a corruption.
 127 +
 128 This means that there is no need for the client to explicitly maintain an
 129 expensive-to-modify list of missing objects.[a]
 130
 131 - Since almost all Git code currently expects any referenced object to be
 132   present locally and because we do not want to force every command to do
 133   a dry-run first, a fallback mechanism is added to allow Git to attempt
 134   to dynamically fetch missing objects from promisor remotes.
 135 +
 136 When the normal object lookup fails to find an object, Git invokes
 137 promisor_remote_get_direct() to try to get the object from a promisor
 138 remote and then retry the object lookup.  This allows objects to be
 139 "faulted in" without complicated prediction algorithms.
 140 +
 141 For efficiency reasons, no check as to whether the missing object is
 142 actually a promisor object is performed.
 143 +
 144 Dynamic object fetching tends to be slow as objects are fetched one at
 145 a time.
 146
 147 - `checkout` (and any other command using `unpack-trees`) has been taught
 148   to bulk pre-fetch all required missing blobs in a single batch.
 149
 150 - `rev-list` has been taught to print missing objects.
 151 +
 152 This can be used by other commands to bulk prefetch objects.
 153 For example, a "git log -p A..B" may internally want to first do
 154 something like "git rev-list --objects --quiet --missing=print A..B"
 155 and prefetch those objects in bulk.
 156
 157 - `fsck` has been updated to be fully aware of promisor objects.
 158
 159 - `repack` in GC has been updated to not touch promisor packfiles at all,
 160   and to only repack other objects.
 161
 162 - The global variable "fetch_if_missing" is used to control whether an
 163   object lookup will attempt to dynamically fetch a missing object or
 164   report an error.
 165 +
 166 We are not happy with this global variable and would like to remove it,
 167 but that requires significant refactoring of the object code to pass an
 168 additional flag.
 169
 170
 171 Fetching Missing Objects
 172 ------------------------
 173
 174 - Fetching of objects is done by invoking a "git fetch" subprocess.
 175
 176 - The local repository sends a request with the hashes of all requested
 177   objects, and does not perform any packfile negotiation.
 178   It then receives a packfile.
 179
 180 - Because we are reusing the existing fetch mechanism, fetching
 181   currently fetches all objects referred to by the requested objects, even
 182   though they are not necessary.
 183
 184
 185 Using many promisor remotes
 186 ---------------------------
 187
 188 Many promisor remotes can be configured and used.
 189
 190 This allows for example a user to have multiple geographically-close
 191 cache servers for fetching missing blobs while continuing to do
 192 filtered `git-fetch` commands from the central server.
 193
 194 When fetching objects, promisor remotes are tried one after the other
 195 until all the objects have been fetched.
 196
 197 Remotes that are considered "promisor" remotes are those specified by
 198 the following configuration variables:
 199
 200 - `extensions.partialClone = <name>`
 201
 202 - `remote.<name>.promisor = true`
 203
 204 - `remote.<name>.partialCloneFilter = ...`
 205
 206 Only one promisor remote can be configured using the
 207 `extensions.partialClone` config variable. This promisor remote will
 208 be the last one tried when fetching objects.
 209
 210 We decided to make it the last one we try, because it is likely that
 211 someone using many promisor remotes is doing so because the other
 212 promisor remotes are better for some reason (maybe they are closer or
 213 faster for some kind of objects) than the origin, and the origin is
 214 likely to be the remote specified by extensions.partialClone.
 215
 216 This justification is not very strong, but one choice had to be made,
 217 and anyway the long term plan should be to make the order somehow
 218 fully configurable.
 219
 220 For now though the other promisor remotes will be tried in the order
 221 they appear in the config file.
 222
 223 Current Limitations
 224 -------------------
 225
 226 - It is not possible to specify the order in which the promisor
 227   remotes are tried in other ways than the order in which they appear
 228   in the config file.
 229 +
 230 It is also not possible to specify an order to be used when fetching
 231 from one remote and a different order when fetching from another
 232 remote.
 233
 234 - It is not possible to push only specific objects to a promisor
 235   remote.
 236 +
 237 It is not possible to push at the same time to multiple promisor
 238 remote in a specific order.
 239
 240 - Dynamic object fetching will only ask promisor remotes for missing
 241   objects.  We assume that promisor remotes have a complete view of the
 242   repository and can satisfy all such requests.
 243
 244 - Repack essentially treats promisor and non-promisor packfiles as 2
 245   distinct partitions and does not mix them.  Repack currently only works
 246   on non-promisor packfiles and loose objects.
 247
 248 - Dynamic object fetching invokes fetch-pack once *for each item*
 249   because most algorithms stumble upon a missing object and need to have
 250   it resolved before continuing their work.  This may incur significant
 251   overhead -- and multiple authentication requests -- if many objects are
 252   needed.
 253
 254 - Dynamic object fetching currently uses the existing pack protocol V0
 255   which means that each object is requested via fetch-pack.  The server
 256   will send a full set of info/refs when the connection is established.
 257   If there are large number of refs, this may incur significant overhead.
 258
 259
 260 Future Work
 261 -----------
 262
 263 - Improve the way to specify the order in which promisor remotes are
 264   tried.
 265 +
 266 For example this could allow to specify explicitly something like:
 267 "When fetching from this remote, I want to use these promisor remotes
 268 in this order, though, when pushing or fetching to that remote, I want
 269 to use those promisor remotes in that order."
 270
 271 - Allow pushing to promisor remotes.
 272 +
 273 The user might want to work in a triangular work flow with multiple
 274 promisor remotes that each have an incomplete view of the repository.
 275
 276 - Allow repack to work on promisor packfiles (while keeping them distinct
 277   from non-promisor packfiles).
 278
 279 - Allow non-pathname-based filters to make use of packfile bitmaps (when
 280   present).  This was just an omission during the initial implementation.
 281
 282 - Investigate use of a long-running process to dynamically fetch a series
 283   of objects, such as proposed in [5,6] to reduce process startup and
 284   overhead costs.
 285 +
 286 It would be nice if pack protocol V2 could allow that long-running
 287 process to make a series of requests over a single long-running
 288 connection.
 289
 290 - Investigate pack protocol V2 to avoid the info/refs broadcast on
 291   each connection with the server to dynamically fetch missing objects.
 292
 293 - Investigate the need to handle loose promisor objects.
 294 +
 295 Objects in promisor packfiles are allowed to reference missing objects
 296 that can be dynamically fetched from the server.  An assumption was
 297 made that loose objects are only created locally and therefore should
 298 not reference a missing object.  We may need to revisit that assumption
 299 if, for example, we dynamically fetch a missing tree and store it as a
 300 loose object rather than a single object packfile.
 301 +
 302 This does not necessarily mean we need to mark loose objects as promisor;
 303 it may be sufficient to relax the object lookup or is-promisor functions.
 304
 305
 306 Non-Tasks
 307 ---------
 308
 309 - Every time the subject of "demand loading blobs" comes up it seems
 310   that someone suggests that the server be allowed to "guess" and send
 311   additional objects that may be related to the requested objects.
 312 +
 313 No work has gone into actually doing that; we're just documenting that
 314 it is a common suggestion.  We're not sure how it would work and have
 315 no plans to work on it.
 316 +
 317 It is valid for the server to send more objects than requested (even
 318 for a dynamic object fetch), but we are not building on that.
 319
 320
 321 Footnotes
 322 ---------
 323
 324 [a] expensive-to-modify list of missing objects:  Earlier in the design of
 325     partial clone we discussed the need for a single list of missing objects.
 326     This would essentially be a sorted linear list of OIDs that the were
 327     omitted by the server during a clone or subsequent fetches.
 328
 329 This file would need to be loaded into memory on every object lookup.
 330 It would need to be read, updated, and re-written (like the .git/index)
 331 on every explicit "git fetch" command *and* on any dynamic object fetch.
 332
 333 The cost to read, update, and write this file could add significant
 334 overhead to every command if there are many missing objects.  For example,
 335 if there are 100M missing blobs, this file would be at least 2GiB on disk.
 336
 337 With the "promisor" concept, we *infer* a missing object based upon the
 338 type of packfile that references it.
 339
 340
 341 Related Links
 342 -------------
 343 [0] https://crbug.com/git/2
 344     Bug#2: Partial Clone
 345
 346 [1] https://lore.kernel.org/git/20170113155253.1644-1-benpeart@microsoft.com/ +
 347     Subject: [RFC] Add support for downloading blobs on demand +
 348     Date: Fri, 13 Jan 2017 10:52:53 -0500
 349
 350 [2] https://lore.kernel.org/git/cover.1506714999.git.jonathantanmy@google.com/ +
 351     Subject: [PATCH 00/18] Partial clone (from clone to lazy fetch in 18 patches) +
 352     Date: Fri, 29 Sep 2017 13:11:36 -0700
 353
 354 [3] https://lore.kernel.org/git/20170426221346.25337-1-jonathantanmy@google.com/ +
 355     Subject: Proposal for missing blob support in Git repos +
 356     Date: Wed, 26 Apr 2017 15:13:46 -0700
 357
 358 [4] https://lore.kernel.org/git/1488999039-37631-1-git-send-email-git@jeffhostetler.com/ +
 359     Subject: [PATCH 00/10] RFC Partial Clone and Fetch +
 360     Date: Wed,  8 Mar 2017 18:50:29 +0000
 361
 362 [5] https://lore.kernel.org/git/20170505152802.6724-1-benpeart@microsoft.com/ +
 363     Subject: [PATCH v7 00/10] refactor the filter process code into a reusable module +
 364     Date: Fri,  5 May 2017 11:27:52 -0400
 365
 366 [6] https://lore.kernel.org/git/20170714132651.170708-1-benpeart@microsoft.com/ +
 367     Subject: [RFC/PATCH v2 0/1] Add support for downloading blobs on demand +
 368     Date: Fri, 14 Jul 2017 09:26:50 -0400