Back in e37d0b8730 (builtin/index-pack.c: write reverse indexes,
2021-01-25), Git learned how to read and write a pack's reverse index
from a file instead of in-memory.
A pack's reverse index is a mapping from pack position (that is, the
order that objects appear together in a ".pack") to their position in
lexical order (that is, the order that objects are listed in an ".idx"
file).
Reverse indexes are consulted often during pack-objects, as well as
during auxiliary operations that require mapping between pack offsets,
pack order, and index index.
They are useful in GitHub's infrastructure, where we have seen a
dramatic increase in performance when writing ".rev" files[1]. In
particular:
- an ~80% reduction in the time it takes to serve fetches on a popular
repository, Homebrew/homebrew-core.
- a ~60% reduction in the peak memory usage to serve fetches on that
same repository.
- a collective savings of ~35% in CPU time across all pack-objects
invocations serving fetches across all repositories in a single
datacenter.
Reverse indexes are also beneficial to end-users as well as forges. For
example, the time it takes to generate a pack containing the objects for
the 10 most recent commits in linux.git (representing a typical push) is
significantly faster when on-disk reverse indexes are available:
$ { git rev-parse HEAD && printf '^' && git rev-parse HEAD~10 } >in
$ hyperfine -L v false,true 'git.compile -c pack.readReverseIndex={v} pack-objects --delta-base-offset --revs --stdout <in >/dev/null'
Benchmark 1: git.compile -c pack.readReverseIndex=false pack-objects --delta-base-offset --revs --stdout <in >/dev/null
Time (mean ± σ): 543.0 ms ± 20.3 ms [User: 616.2 ms, System: 58.8 ms]
Range (min … max): 521.0 ms … 577.9 ms 10 runs
Benchmark 2: git.compile -c pack.readReverseIndex=true pack-objects --delta-base-offset --revs --stdout <in >/dev/null
Time (mean ± σ): 245.0 ms ± 11.4 ms [User: 335.6 ms, System: 31.3 ms]
Range (min … max): 226.0 ms … 259.6 ms 13 runs
Summary
'git.compile -c pack.readReverseIndex=true pack-objects --delta-base-offset --revs --stdout <in >/dev/null' ran
2.22 ± 0.13 times faster than 'git.compile -c pack.readReverseIndex=false pack-objects --delta-base-offset --revs --stdout <in >/dev/null'
The same is true of writing a pack containing the objects for the 30
most-recent commits:
$ { git rev-parse HEAD && printf '^' && git rev-parse HEAD~30 } >in
$ hyperfine -L v false,true 'git.compile -c pack.readReverseIndex={v} pack-objects --delta-base-offset --revs --stdout <in >/dev/null'
Benchmark 1: git.compile -c pack.readReverseIndex=false pack-objects --delta-base-offset --revs --stdout <in >/dev/null
Time (mean ± σ): 866.5 ms ± 16.2 ms [User: 1414.5 ms, System: 97.0 ms]
Range (min … max): 839.3 ms … 886.9 ms 10 runs
Benchmark 2: git.compile -c pack.readReverseIndex=true pack-objects --delta-base-offset --revs --stdout <in >/dev/null
Time (mean ± σ): 581.6 ms ± 10.2 ms [User: 1181.7 ms, System: 62.6 ms]
Range (min … max): 567.5 ms … 599.3 ms 10 runs
Summary
'git.compile -c pack.readReverseIndex=true pack-objects --delta-base-offset --revs --stdout <in >/dev/null' ran
1.49 ± 0.04 times faster than 'git.compile -c pack.readReverseIndex=false pack-objects --delta-base-offset --revs --stdout <in >/dev/null'
...and savings on trivial operations like computing the on-disk size of
a single (packed) object are even more dramatic:
$ git rev-parse HEAD >in
$ hyperfine -L v false,true 'git.compile -c pack.readReverseIndex={v} cat-file --batch-check="%(objectsize:disk)" <in'
Benchmark 1: git.compile -c pack.readReverseIndex=false cat-file --batch-check="%(objectsize:disk)" <in
Time (mean ± σ): 305.8 ms ± 11.4 ms [User: 264.2 ms, System: 41.4 ms]
Range (min … max): 290.3 ms … 331.1 ms 10 runs
Benchmark 2: git.compile -c pack.readReverseIndex=true cat-file --batch-check="%(objectsize:disk)" <in
Time (mean ± σ): 4.0 ms ± 0.3 ms [User: 1.7 ms, System: 2.3 ms]
Range (min … max): 1.6 ms … 4.6 ms 1155 runs
Summary
'git.compile -c pack.readReverseIndex=true cat-file --batch-check="%(objectsize:disk)" <in' ran
76.96 ± 6.25 times faster than 'git.compile -c pack.readReverseIndex=false cat-file --batch-check="%(objectsize:disk)" <in'
In the more than two years since e37d0b8730 was merged, Git's
implementation of on-disk reverse indexes has been thoroughly tested,
both from users enabling `pack.writeReverseIndexes`, and from GitHub's
deployment of the feature. The latter has been running without incident
for more than two years.
This patch changes Git's behavior to write on-disk reverse indexes by
default when indexing a pack, which should make the above operations
faster for everybody's Git installation after a repack.
(The previous commit explains some potential drawbacks of using on-disk
reverse indexes in certain limited circumstances, that essentially boil
down to a trade-off between time to generate, and time to access. For
those limited cases, the `pack.readReverseIndex` escape hatch can be
used).
[1]: https://github.blog/2021-04-29-scaling-monorepo-maintenance/#reverse-indexes
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Acked-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since 1615c567b8 (Documentation/config/pack.txt: advertise
'pack.writeReverseIndex', 2021-01-25), we have had the
`pack.writeReverseIndex` configuration option, which tells Git whether
or not it is allowed to write a ".rev" file when indexing a pack.
Introduce a complementary configuration knob, `pack.readReverseIndex` to
control whether or not Git will read any ".rev" file(s) that may be
available on disk.
This option is useful for debugging, as well as disabling the effect of
".rev" files in certain instances.
This is useful because of the trade-off[^1] between the time it takes to
generate a reverse index (slow from scratch, fast when reading an
existing ".rev" file), and the time it takes to access a record (the
opposite).
For example, even though it is faster to use the on-disk reverse index
when computing the on-disk size of a packed object, it is slower to
enumerate the same value for all objects.
Here are a couple of examples from linux.git. When computing the above
for a single object, using the on-disk reverse index is significantly
faster:
$ git rev-parse HEAD >in
$ hyperfine -L v false,true 'git.compile -c pack.readReverseIndex={v} cat-file --batch-check="%(objectsize:disk)" <in'
Benchmark 1: git.compile -c pack.readReverseIndex=false cat-file --batch-check="%(objectsize:disk)" <in
Time (mean ± σ): 302.5 ms ± 12.5 ms [User: 258.7 ms, System: 43.6 ms]
Range (min … max): 291.1 ms … 328.1 ms 10 runs
Benchmark 2: git.compile -c pack.readReverseIndex=true cat-file --batch-check="%(objectsize:disk)" <in
Time (mean ± σ): 3.9 ms ± 0.3 ms [User: 1.6 ms, System: 2.4 ms]
Range (min … max): 2.0 ms … 4.4 ms 801 runs
Summary
'git.compile -c pack.readReverseIndex=true cat-file --batch-check="%(objectsize:disk)" <in' ran
77.29 ± 7.14 times faster than 'git.compile -c pack.readReverseIndex=false cat-file --batch-check="%(objectsize:disk)" <in'
, but when instead trying to compute the on-disk object size for all
objects in the repository, using the ".rev" file is a disadvantage over
creating the reverse index from scratch:
$ hyperfine -L v false,true 'git.compile -c pack.readReverseIndex={v} cat-file --batch-check="%(objectsize:disk)" --batch-all-objects'
Benchmark 1: git.compile -c pack.readReverseIndex=false cat-file --batch-check="%(objectsize:disk)" --batch-all-objects
Time (mean ± σ): 8.258 s ± 0.035 s [User: 7.949 s, System: 0.308 s]
Range (min … max): 8.199 s … 8.293 s 10 runs
Benchmark 2: git.compile -c pack.readReverseIndex=true cat-file --batch-check="%(objectsize:disk)" --batch-all-objects
Time (mean ± σ): 16.976 s ± 0.107 s [User: 16.706 s, System: 0.268 s]
Range (min … max): 16.839 s … 17.105 s 10 runs
Summary
'git.compile -c pack.readReverseIndex=false cat-file --batch-check="%(objectsize:disk)" --batch-all-objects' ran
2.06 ± 0.02 times faster than 'git.compile -c pack.readReverseIndex=true cat-file --batch-check="%(objectsize:disk)" --batch-all-objects'
Luckily, the results when running `git cat-file` with `--unordered` are
closer together:
$ hyperfine -L v false,true 'git.compile -c pack.readReverseIndex={v} cat-file --unordered --batch-check="%(objectsize:disk)" --batch-all-objects'
Benchmark 1: git.compile -c pack.readReverseIndex=false cat-file --unordered --batch-check="%(objectsize:disk)" --batch-all-objects
Time (mean ± σ): 5.066 s ± 0.105 s [User: 4.792 s, System: 0.274 s]
Range (min … max): 4.943 s … 5.220 s 10 runs
Benchmark 2: git.compile -c pack.readReverseIndex=true cat-file --unordered --batch-check="%(objectsize:disk)" --batch-all-objects
Time (mean ± σ): 6.193 s ± 0.069 s [User: 5.937 s, System: 0.255 s]
Range (min … max): 6.145 s … 6.356 s 10 runs
Summary
'git.compile -c pack.readReverseIndex=false cat-file --unordered --batch-check="%(objectsize:disk)" --batch-all-objects' ran
1.22 ± 0.03 times faster than 'git.compile -c pack.readReverseIndex=true cat-file --unordered --batch-check="%(objectsize:disk)" --batch-all-objects'
Because the equilibrium point between these two is highly machine- and
repository-dependent, allow users to configure whether or not they will
read any ".rev" file(s) with this configuration knob.
[^1]: Generating a reverse index in memory takes O(N) time (where N is
the number of objects in the repository), since we use a radix sort.
Reading an entry from an on-disk ".rev" file is slower since each
operation is bound by disk I/O instead of memory I/O.
In order to compute the on-disk size of a packed object, we need to
find the offset of our object, and the adjacent object (the on-disk
size difference of these two). Finding the first offset requires a
binary search. Finding the latter involves a single .rev lookup.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Acked-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The pack bitmap file gained a bitmap-lookup table to speed up
locating the necessary bitmap for a given commit.
* ac/bitmap-lookup-table:
pack-bitmap-write: drop unused pack_idx_entry parameters
bitmap-lookup-table: add performance tests for lookup table
pack-bitmap: prepare to read lookup table extension
pack-bitmap-write: learn pack.writeBitmapLookupTable and add tests
pack-bitmap-write.c: write lookup table extension
bitmap: move `get commit positions` code to `bitmap_writer_finish`
Documentation/technical: describe bitmap lookup table extension
Teach Git to provide a way for users to enable/disable bitmap lookup
table extension by providing a config option named 'writeBitmapLookupTable'.
Default is false.
Also add test to verify writting of lookup table.
Mentored-by: Taylor Blau <me@ttaylorr.com>
Co-Mentored-by: Kaartic Sivaraam <kaartic.sivaraam@gmail.com>
Co-Authored-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Abhradeep Chakraborty <chakrabortyabhradeep79@gmail.com>
Reviewed-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Continue the move of existing Documentation/technical/* protocol and
file-format documentation into our main documentation space by moving
the various documentation pertaining to the *.pack format and related
files, and updating things that refer to it to link to the new
location.
By moving these we can properly link from the newly created
gitformat-commit-graph to a gitformat-chunk-format page.
Integrating "Documentation/technical/bitmap-format.txt" and
"Documentation/technical/cruft-packs.txt" might logically be part of
this change, but as those cover parts of the wider "pack
format" (including associated files) that's documented outside of
"Documentation/technical/pack-format.txt" let's leave those for now,
subsequent commit(s) will address those.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In the previous commit, the bitmap writing code learned to propagate
values from an existing hash-cache extension into the bitmap that it is
writing.
Now that that functionality exists, let's expose it by teaching the 'git
multi-pack-index' builtin to respect the `pack.writeBitmapHashCache`
option so that the hash-cache may be written at all.
Two minor points worth noting here:
- The 'git multi-pack-index write' sub-command didn't previously read
any configuration (instead this is handled in the base command). A
separate handler is added here to respect this write-specific
config option.
- I briefly considered adding a 'bitmap_flags' field to the static
options struct, but decided against it since it would require
plumbing through a new parameter to the write_midx_file() function.
Instead, a new MIDX-specific flag is added, which is translated to
the corresponding bitmap one.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This option is almost never a good idea, as the resulting repository is
larger and slower (see the new explanations in the docs).
I outlined the potential problems. We could go further and make the
option harder to find (or at least, make the command-line option
descriptions a much more terse "you probably don't want this; see
pack.packsizeLimit for details"). But this seems like a minimal change
that may prevent people from thinking it's more useful than it is.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When writing a new pack with a bitmap, it is sometimes convenient to
indicate some reference prefixes which should receive priority when
selecting which commits to receive bitmaps.
A truly motivated caller could accomplish this by setting
'pack.islandCore', (since all commits in the core island are similarly
marked as preferred) but this requires callers to opt into using delta
islands, which they may or may not want to do.
Introduce a new multi-valued configuration, 'pack.preferBitmapTips' to
allow callers to specify a list of reference prefixes. All references
which have a prefix contained in 'pack.preferBitmapTips' will mark their
tips as "preferred" in the same way as commits are marked as preferred
for selection by 'pack.islandCore'.
The choice of the verb "prefer" is intentional: marking the NEEDS_BITMAP
flag on an object does *not* guarantee that that object will receive a
bitmap. It merely guarantees that that commit will receive a bitmap over
any *other* commit in the same window by bitmap_writer_select_commits().
The test this patch adds reflects this quirk, too. It only tests that
a commit (which didn't receive bitmaps by default) is selected for
bitmaps after changing the value of 'pack.preferBitmapTips' to include
it. Other commits may lose their bitmaps as a byproduct of how the
selection process works (bitmap_writer_select_commits() ignores the
remainder of a window after seeing a commit with the NEEDS_BITMAP flag).
This configuration will aide in selecting important references for
multi-pack bitmaps, since they do not respect the same pack.islandCore
configuration. (They could, but doing so may be confusing, since it is
packs--not bitmaps--which are influenced by the delta-islands
configuration).
In a fork network repository (one which lists all forks of a given
repository as remotes), for example, it is useful to set
pack.preferBitmapTips to 'refs/remotes/<root>/heads' and
'refs/remotes/<root>/tags', where '<root>' is an opaque identifier
referring to the repository which is at the base of the fork chain.
Suggested-by: Jeff King <peff@peff.net>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Now that the pack.writeReverseIndex configuration is respected in both
'git index-pack' and 'git pack-objects' (and therefore, all of their
callers), we can safely advertise it for use in the git-config manual.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The pack.useSparse config option was introduced by 3d036eb0
(pack-objects: create pack.useSparse setting, 2019-01-19) and was
first available in v2.21.0. When enabled, the pack-objects process
during 'git push' will use a sparse tree walk when deciding which
trees and blobs to send to the remote. The algorithm was introduced
by d5d2e93 (revision: implement sparse algorithm, 2019-01-16) and
has been in production use by VFS for Git since around that time.
The features.experimental config option also enabled pack.useSparse,
so hopefully that has also increased exposure.
It is worth noting that pack.useSparse has a possibility of
sending more objects across a push, but requires a special
arrangement of exact _copies_ across directories. There is a test
in t5322-pack-objects-sparse.sh that demonstrates this possibility.
This test uses the --sparse option to "git pack-objects" but we
can make it implied by the config value to demonstrate that the
default value has changed.
While updating that test, I noticed that the documentation did not
include an option for --no-sparse, which is now more important than
it was before.
Since the downside is unlikely but the upside is significant, set
the default value of pack.useSparse to true. Remove it from the
set of options implied by features.experimental.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Let's make it possible to configure if we want pack reuse or not.
The main reason it might not be wanted is probably debugging and
performance testing, though pack reuse _might_ cause larger packs,
because we wouldn't consider the reused objects as bases for
finding new deltas.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The 'feature.experimental' setting includes config options that are
not committed to become defaults, but could use additional testing.
Update the following config settings to take new defaults, and to
use the repo_settings struct if not already using them:
* 'pack.useSparse=true'
* 'fetch.negotiationAlgorithm=skipping'
In the case of fetch.negotiationAlgorithm, the existing logic
would load the config option only when about to use the setting,
so had a die() statement on an unknown string value. This is
removed as now the config is parsed under prepare_repo_settings().
In general, this die() is probably misplaced and not valuable.
A test was removed that checked this die() statement executed.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Enabling pack.writebitmaphashcache should always be a performance win.
It costs only 4 bytes per object on disk, and the timings in ae4f07fbcc
(pack-bitmap: implement optional name_hash cache, 2013-12-21) show it
improving fetch and partial-bitmap clone times by 40-50%.
The only reason we didn't enable it by default at the time is that early
versions of JGit's bitmap reader complained about the presence of
optional header bits it didn't understand. But that was changed in
JGit's d2fa3987a (Use bitcheck to check for presence of OPT_FULL option,
2013-10-30), which made it into JGit v3.5.0 in late 2014.
So let's turn this option on by default. It's backwards-compatible with
all versions of Git, and if you are also using JGit on the same
repository, you'd only run into problems using a version that's almost 5
years old.
We'll drop the manual setting from all of our test scripts, including
perf tests. This isn't strictly necessary, but it has two advantages:
1. If the hash-cache ever stops being enabled by default, our perf
regression tests will notice.
2. We can use the modified perf tests to show off the behavior of an
otherwise unconfigured repo, as shown below.
These are the results of a few of a perf tests against linux.git that
showed interesting results. You can see the expected speedup in 5310.4,
which was noted in ae4f07fbcc. Curiously, 5310.8 did not improve (and
actually got slower), despite seeing the opposite in ae4f07fbcc.
I don't have an explanation for that.
The tests from p5311 did not exist back then, but do show improvements
(a smaller pack due to better deltas, which we found in less time).
Test HEAD^ HEAD
-------------------------------------------------------------------------------------
5310.4: simulated fetch 7.39(22.70+0.25) 5.64(11.43+0.22) -23.7%
5310.8: clone (partial bitmap) 18.45(24.83+1.19) 19.94(28.40+1.36) +8.1%
5311.31: server (128 days) 0.41(1.13+0.05) 0.34(0.72+0.02) -17.1%
5311.32: size (128 days) 7.4M 7.0M -4.8%
5311.33: client (128 days) 1.33(1.49+0.06) 1.29(1.37+0.12) -3.0%
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The '--sparse' flag in 'git pack-objects' changes the algorithm
used to enumerate objects to one that is faster for individual
users pushing new objects that change only a small cone of the
working directory. The sparse algorithm is not recommended for a
server, which likely sends new objects that appear across the
entire working directory.
Create a 'pack.useSparse' setting that enables this new algorithm.
This allows 'git push' to use this algorithm without passing a
'--sparse' flag all the way through four levels of run_command()
calls.
If the '--no-sparse' flag is set, then this config setting is
overridden.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
