[v2,0/2] fetch: speed up mirror-fetches with many refs

Message

Patrick Steinhardt Feb. 2, 2022, 12:51 p.m. UTC

Hi,

this is the second version of my patch series which aims to speed up
mirror-fetches in repos with huge amounts of refs. The only change
compared to v1 is a fixed up commit message: Taylor has pointed out to
me that commit dates retrieved from the commit-graph are not in fact the
corrected commit dates, which are stored separately.

Patrick

Patrick Steinhardt (2):
  fetch-pack: use commit-graph when computing cutoff
  fetch: skip computing output width when not printing anything

 builtin/fetch.c |  8 ++++++--
 fetch-pack.c    | 28 ++++++++++++++++------------
 2 files changed, 22 insertions(+), 14 deletions(-)

Range-diff against v1:
1:  31cf8f87a1 ! 1:  6fac914f0f fetch-pack: use commit-graph when computing cutoff
    @@ Commit message
         the cutoff date. This can be sped up by trying to look up commits via
         the commit-graph first, which is a lot more efficient.
     
    -    One thing to keep in mind though is that the commit-graph corrects
    -    committer dates:
    -
    -        * A commit with at least one parent has corrected committer date
    -          equal to the maximum of its commiter date and one more than the
    -          largest corrected committer date among its parents.
    -
    -    As a result, it may be that the commit date we load via the commit graph
    -    is more recent than it would have been when loaded via the ODB, and as a
    -    result we may also choose a more recent cutoff point. But as the code
    -    documents, this is only a heuristic and it is okay if we determine a
    -    wrong cutoff date. The worst that can happen is that we report more
    -    commits as HAVEs to the server when using corrected dates.
    -
    -    Loading commits via the commit-graph is typically much faster than
    -    loading commits via the object database. Benchmarks in a repository with
    -    about 2,1 million refs and an up-to-date commit-graph show a 20% speedup
    -    when mirror-fetching:
    +    Benchmarks in a repository with about 2,1 million refs and an up-to-date
    +    commit-graph show a 20% speedup when mirror-fetching:
     
             Benchmark 1: git fetch --atomic +refs/*:refs/* (v2.35.0)
               Time (mean ± σ):     75.264 s ±  1.115 s    [User: 68.199 s, System: 10.094 s]
2:  5a3fd3232f = 2:  4b9bbcf795 fetch: skip computing output width when not printing anything