git/list[1] front-page[2] threads[3] people[4] search[5] about
 

[PATCH v6 00/15] Many promisor remotes

From
Christian Couder <christian.couder@gmail.com>
Date
Jun 25, 2019, 13:40 UTC
Message-ID
<20190625134039.21707-1-chriscool@tuxfamily.org>
This patch series is based on master as of 8dca754b1e (The third batch, 2019-06-21).

It was previously based on jt/batch-fetch-blobs-in-diff, but this has been merged into master.

Introduction ~~~~~~~~~~~~

This path series is a follow up from the "remote odb" patch series that I sent last year, which were a follow up from previous series. See the links section for more information.

The goal of this patch series is to make it possible to have and to fetch missing objects from multiple remotes instead of only one.

For now the fetch order is the order of the remotes in the config, except for the remote specified by extensions.partialclone config option which comes last in the fetch order.

I selected the name "Promisor remote" over "Partial clone remote" because it is shorter and because it is not only about cloning.

The existing extensions.partialclone config option is respected, but it is not written in the config when a partial clone or fetch is made. Instead remote.<name>.promisor is set to "true". This may create a compatibility issue, but it makes it possible to start using many promisor remotes by just cloning and fetching from different remotes with partial clone filters. The compatibility issue could be resolved in a future iteration by just setting extensions.partialclone instead of remote.<name>.promisor the first time a promisor remote is used.

The code might not work with many promisor remotes that don't all have all the promised objects, as that would require the fetch protocol to send packs with best effort, as described by Junio in:

https://public-inbox.org/git/xmqqpnqve71d.fsf@gitster-ct.c.googlers.com/
I plan to take a look at improving that soon.

In general I have tried to change as few things as possible in the first patches of the series, though the last patches try to hide the old features that only made sense for the general code to use when there was only one promisor remote.

High level view of changes since the V5 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

The only patchs that changed since V5 are:  
  - Patch 4/15 (promisor-remote: implement promisor_remote_get_direct())

The new OBJECT_INFO_SKIP_FETCH_OBJECT flag is used instead of using the ugly fetch_if_missing global variable as suggested by Stolee.

Also, "missing" in variable names has been replaced with "remaining" as suggested by Stolee.

  - Patch 7/15 (Use promisor_remote_get_direct() and
    has_promisor_remote())

The changes to diff.c that were in patch 8/16 in V5 have been squashed into this patch as the squashed patch applied to diff.c the same changes that are made in this patch. They were in a separate patch because the changes in diff.c needed to be made only if the jt/batch-fetch-blobs-in-diff series was merged, which has happened.

Some similar changes in builtin/index-pack.c were also added to this patch. They were necessary since 8a30a1efd1 (index-pack: prefetch missing REF_DELTA bases, 2019-05-14) has been merged.

High level overview of old patches in this patch series ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

   - Patch 1/15 (t0410: remove pipes after git commands)

This is a preparatory cleanup patch to improve t0410 as suggested by Szeder Gábor.

  - Patch 2/15 (fetch-object: make functions return an error code)

This patch, makes functions in fetch-object.c return an error code, which is necessary to later tell that they failed and try another promisor remote when there is more than one. This could also just be seen as a fix to these functions.

  - Patch 3/15 (Add initial support for many promisor remotes)
This introduces the minimum infrastructure for promisor remotes.
  - Patch 4/15 (promisor-remote: implement promisor_remote_get_direct()) 
  - Patch 5/15 (promisor-remote: add promisor_remote_reinit())
  - Patch 6/15 (promisor-remote: use repository_format_partial_clone)

These patches add a few missing bits in the promisor remote infrastructure that will be needed in the following patches.

  - Patch 7/15 (Use promisor_remote_get_direct() and has_promisor_remote())

This replaces the previous interface that used only one promisor remote defined in extensions.partialclone with the new interface created by the previous patches.

  - Patch 8/15 (promisor-remote: parse remote.*.partialclonefilter)

This replaces the way a partial clone filter was handled by a new way based on the previous patches that support more than one partial clone filter.

  - Patch 9/15 (builtin/fetch: remove unique promisor remote limitation)

This patch removes the limitation in builtin/fetch.c to have only one promisor remote.

  - Patch 10/15 (t0410: test fetching from many promisor remotes)

This adds test cases that shows that now more than one promisor remote can be used and that remote.<name>.promisor is set to "true" when fetching from a new promisor remote.

  - Patch 11/15 (partial-clone: add multiple remotes in the doc)
  - Patch 12/15 (remote: add promisor and partial clone config to the doc)

These documentation patches explain how things can work with more than one promisor remote.

  - Patch 13/15 (Remove fetch-object.{c,h} in favor of promisor-remote.{c,h})
  - Patch 14/15 (Move repository_format_partial_clone to promisor-remote.c)
  - Patch 15/15 (Move core_partial_clone_filter_default to promisor-remote.c)

These patches try to hide the old features (fetch_objects(), repository_format_partial_clone and core_partial_clone_filter_default) that only made sense for the general code to use when there was only one promisor remote. This ensures that there will be compilation errors rather than bugs or test failures if the old features are used in the old fashion way.

Links ~~~~~

This patch series on GitHub:
V6: https://github.com/chriscool/git/commits/many-promisor-remotes
V5: https://github.com/chriscool/git/commits/many-promisor-remotes68
V4: https://github.com/chriscool/git/commits/many-promisor-remotes58
V3: https://github.com/chriscool/git/commits/many-promisor-remotes40
V2: https://github.com/chriscool/git/commits/many-promisor-remotes35
V1: https://github.com/chriscool/git/commits/many-promisor-remotes17
On the mailing list:
V5: https://public-inbox.org/git/20190409161116.30256-1-chriscool@tuxfamily.org/
V4: https://public-inbox.org/git/20190401164045.17328-1-chriscool@tuxfamily.org/
V3: https://public-inbox.org/git/20190312132959.11764-1-chriscool@tuxfamily.org/
V2: https://public-inbox.org/git/20190122144212.15119-1-chriscool@tuxfamily.org/
V1: https://public-inbox.org/git/20181211052746.16218-1-chriscool@tuxfamily.org/

This patch series is a follow up from the discussions related to the remote odb V4 patch series:

https://public-inbox.org/git/20180802061505.2983-1-chriscool@tuxfamily.org/
Especially in:
https://public-inbox.org/git/CAP8UFD3nrhjANwNDqTwx5ZtnZNcnbAFqUN=u=LrvzuH4+3wQQA@mail.gmail.com/
I said that I would like to work on things in the following order:
  1) Teaching partial clone to attempt to fetch missing objects from
multiple remotes instead of only one using the order in the config.
  2) Simplifying the protocol for fetching missing objects so that it
can be satisfied by a lighter weight object storage system than a full
Git server.
  3) Making it possible to explicitly define an order in which the
remotes are accessed.
  4) Making the criteria for what objects can be missing more
aggressive, so that I can "git add" a large file and work with it
using Git without even having a second copy of that object in my local
object store.
And this patch series is about the 1).
The previous remote odb patch series on GitHub:
V5: https://github.com/chriscool/git/commits/remote-odb
V4: https://github.com/chriscool/git/commits/remote-odb5
V3: https://github.com/chriscool/git/commits/remote-odb3
V2: https://github.com/chriscool/git/commits/remote-odb2
V1: https://github.com/chriscool/git/commits/remote-odb1
Discussions related to previous versions of the odb patch series:
V4: https://public-inbox.org/git/20180802061505.2983-1-chriscool@tuxfamily.org/
V3: https://public-inbox.org/git/20180713174959.16748-1-chriscool@tuxfamily.org/
V2: https://public-inbox.org/git/20180630083542.20347-1-chriscool@tuxfamily.org/
V1: https://public-inbox.org/git/20180623121846.19750-1-chriscool@tuxfamily.org/
Christian Couder (15):
  t0410: remove pipes after git commands
  fetch-object: make functions return an error code
  Add initial support for many promisor remotes
  promisor-remote: implement promisor_remote_get_direct()
  promisor-remote: add promisor_remote_reinit()
  promisor-remote: use repository_format_partial_clone
  Use promisor_remote_get_direct() and has_promisor_remote()
  promisor-remote: parse remote.*.partialclonefilter
  builtin/fetch: remove unique promisor remote limitation
  t0410: test fetching from many promisor remotes
  partial-clone: add multiple remotes in the doc
  remote: add promisor and partial clone config to the doc
  Remove fetch-object.{c,h} in favor of promisor-remote.{c,h}
  Move repository_format_partial_clone to promisor-remote.c
  Move core_partial_clone_filter_default to promisor-remote.c
 Documentation/config/remote.txt           |   8 +
 Documentation/technical/partial-clone.txt | 117 +++++++---
 Makefile                                  |   2 +-
 builtin/cat-file.c                        |   5 +-
 builtin/fetch.c                           |  29 +--
 builtin/gc.c                              |   3 +-
 builtin/index-pack.c                      |   8 +-
 builtin/repack.c                          |   3 +-
 cache-tree.c                              |   3 +-
 cache.h                                   |   2 -
 config.c                                  |   5 -
 connected.c                               |   3 +-
 diff.c                                    |   9 +-
 environment.c                             |   2 -
 fetch-object.c                            |  40 ----
 fetch-object.h                            |   9 -
 list-objects-filter-options.c             |  51 +++--
 list-objects-filter-options.h             |   3 +-
 packfile.c                                |   3 +-
 promisor-remote.c                         | 265 ++++++++++++++++++++++
 promisor-remote.h                         |  31 +++
 setup.c                                   |   3 +-
 sha1-file.c                               |  15 +-
 t/t0410-partial-clone.sh                  |  61 ++++-
 t/t5601-clone.sh                          |   3 +-
 t/t5616-partial-clone.sh                  |   4 +-
 unpack-trees.c                            |   8 +-
 27 files changed, 523 insertions(+), 172 deletions(-)
 delete mode 100644 fetch-object.c
 delete mode 100644 fetch-object.h
 create mode 100644 promisor-remote.c
 create mode 100644 promisor-remote.h
-- 
2.22.0.229.ga13d9ffdf7.dirty
Next: Christian Couder
Message 1 of 16 in “Many promisor remotes”
  1. 00/15 Many promisor remotesChristian Couder, Jun 25, 2019
  2. 01/15 t0410: remove pipes after git commandsChristian Couder, Jun 25, 2019
  3. 02/15 fetch-object: make functions return an error codeChristian Couder, Jun 25, 2019
  4. 03/15 Add initial support for many promisor remotesChristian Couder, Jun 25, 2019
  5. 05/15 promisor-remote: add promisor_remote_reinit()Christian Couder, Jun 25, 2019
  6. 06/15 promisor-remote: use repository_format_partial_cloneChristian Couder, Jun 25, 2019
  7. 04/15 promisor-remote: implement promisor_remote_get_direct()Christian Couder, Jun 25, 2019
  8. 09/15 builtin/fetch: remove unique promisor remote limitationChristian Couder, Jun 25, 2019
  9. 08/15 promisor-remote: parse remote.*.partialclonefilterChristian Couder, Jun 25, 2019
  10. 10/15 t0410: test fetching from many promisor remotesChristian Couder, Jun 25, 2019
  11. 11/15 partial-clone: add multiple remotes in the docChristian Couder, Jun 25, 2019
  12. 14/15 Move repository_format_partial_clone to promisor-remote.cChristian Couder, Jun 25, 2019
  13. 15/15 Move core_partial_clone_filter_default to promisor-remote.cChristian Couder, Jun 25, 2019
  14. 12/15 remote: add promisor and partial clone config to the docChristian Couder, Jun 25, 2019
  15. 13/15 Remove fetch-object.{c,h} in favor of promisor-remote.{c,h}Christian Couder, Jun 25, 2019
  16. 07/15 Use promisor_remote_get_direct() and has_promisor_remote()Christian Couder, Jun 25, 2019

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.