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

Performance regression on git fetch in 2.21

From
Orgad Shaneh <orgads@gmail.com>
Date
May 24, 2019, 09:14 UTC
Message-ID
<CAGHpTBKSUJzFSWc=uznSu2zB33qCSmKXM-iAjxRCpqNK5bnhRg@mail.gmail.com>
Hi,

git fetch in my repository *when nothing new is received* takes 2.5x the time when comparing 2.20 against 2.21 (on Windows it's 4x).

I have 5 initialized submodules in this working directory.

I reported this issue on git-for-windows github: https://github.com/git-for-windows/git/issues/2199 but there is also an upstream change related.

I bisected and found this commit to blame:
Building... Fetching... Failed [10.7949124]
Bisecting: 0 revisions left to test after this (roughly 0 steps)
[a62387b3fc9f5aeeb04a2db278121d33a9caafa7] submodule.c: fetch in
submodules git directory instead of in worktree
running git-bisect.rb
Building... Fetching... Success [4.303592]
be76c2128234d94b47f7087152ee55d08bb65d88 is the first bad commit
commit be76c2128234d94b47f7087152ee55d08bb65d88
Author: Stefan Beller <sbeller@google.com>
Date:   Thu Dec 6 13:26:55 2018 -0800
    fetch: ensure submodule objects fetched
    Currently when git-fetch is asked to recurse into submodules, it dispatches
    a plain "git-fetch -C <submodule-dir>" (with some submodule related options
    such as prefix and recusing strategy, but) without any information of the
    remote or the tip that should be fetched.
    But this default fetch is not sufficient, as a newly fetched commit in
    the superproject could point to a commit in the submodule that is not
    in the default refspec. This is common in workflows like Gerrit's.
    When fetching a Gerrit change under review (from refs/changes/??), the
    commits in that change likely point to submodule commits that have not
    been merged to a branch yet.
    Fetch a submodule object by id if the object that the superproject
    points to, cannot be found. For now this object is fetched from the
    'origin' remote as we defer getting the default remote to a later patch.
    A list of new submodule commits are already generated in certain
    conditions (by check_for_new_submodule_commits()); this new feature
    invokes that function in more situations.
    The submodule checks were done only when a ref in the superproject
    changed, these checks were extended to also be performed when fetching
    into FETCH_HEAD for completeness, and add a test for that too.
    Signed-off-by: Stefan Beller <sbeller@google.com>
    Signed-off-by: Junio C Hamano <gitster@pobox.com>

:040000 040000 2bad86c248b79ef1e36ea5edaa423cd73445c9a2 ad989a8f3e6f80d4f5a84ec3db0ff1ab00a7e210 M builtin :100644 100644 d1b6646f42d5d12740a94f50a7d25aad4aa356bf b88343d977d78364b417e2015eaa352dec1501b9 M submodule.c :040000 040000 a3c58919de0796b6467709a0f21fa21e28d4d13b cdf9514c9085efcbfcdba8efc765174f6455ce5d M t

Can you please suggest a fix? Is this the expected behavior?
Thanks,
- Orgad
Next: Orgad Shaneh
Message 1 of 2 in “Performance regression on git fetch in 2.21”
  1. Orgad ShanehMay 24, 2019
  2. Orgad ShanehJun 10, 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.