Most of these are straight forward. GETTEXT_POISON does catch the last
string in cmd_pack_objects(), but since this is --progress output, it's
not supposed to be machine-readable.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Many messages will be marked for translation in the following
commits. This commit updates some of them to be more consistent and
reduce diff noise in those commits. Changes are
- keep the first letter of die(), error() and warning() in lowercase
- no full stop in die(), error() or warning() if it's single sentence
messages
- indentation
- some messages are turned to BUG(), or prefixed with "BUG:" and will
not be marked for i18n
- some messages are improved to give more information
- some messages are broken down by sentence to be i18n friendly
(on the same token, combine multiple warning() into one big string)
- the trailing \n is converted to printf_ln if possible, or deleted
if not redundant
- errno_errno() is used instead of explicit strerror()
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Let's start with some background about oe_delta_size() and
oe_set_delta_size(). If you already know, skip the next paragraph.
These two are added in 0aca34e826 (pack-objects: shrink delta_size
field in struct object_entry - 2018-04-14) to help reduce 'struct
object_entry' size. The delta size field in this struct is reduced to
only contain max 1MB. So if any new delta is produced and larger than
1MB, it's dropped because we can't really save such a large size
anywhere. Fallback is provided in case existing packfiles already have
large deltas, then we can retrieve it from the pack.
While this should help small machines repacking large repos without
large deltas (i.e. less memory pressure), dropping large deltas during
the delta selection process could end up with worse pack files. And if
existing packfiles already have >1MB delta and pack-objects is
instructed to not reuse deltas, all of them will be dropped on the
floor, and the resulting pack would be definitely bigger.
There is also a regression in terms of CPU/IO if we have large on-disk
deltas because fallback code needs to parse the pack every time the
delta size is needed and just access to the mmap'd pack data is enough
for extra page faults when memory is under pressure.
Both of these issues were reported on the mailing list. Here's some
numbers for comparison.
Version Pack (MB) MaxRSS(kB) Time (s)
------- --------- ---------- --------
2.17.0 5498 43513628 2494.85
2.18.0 10531 40449596 4168.94
This patch provides a better fallback that is
- cheaper in terms of cpu and io because we won't have to read
existing pack files as much
- better in terms of pack size because the pack heuristics is back to
2.17.0 time, we do not drop large deltas at all
If we encounter any delta (on-disk or created during try_delta phase)
that is larger than the 1MB limit, we stop using delta_size_ field for
this because it can't contain such size anyway. A new array of delta
size is dynamically allocated and can hold all the deltas that 2.17.0
can. This array only contains delta sizes that delta_size_ can't
contain.
With this, we do not have to drop deltas in try_delta() anymore. Of
course the downside is we use slightly more memory, even compared to
2.17.0. But since this is considered an uncommon case, a bit more
memory consumption should not be a problem.
Delta size limit is also raised from 1MB to 16MB to better cover
common case and avoid that extra memory consumption (99.999% deltas in
this reported repo are under 12MB; Jeff noted binary artifacts topped
out at about 3MB in some other private repos). Other fields are
shuffled around to keep this struct packed tight. We don't use more
memory in common case even with this limit update.
A note about thread synchronization. Since this code can be run in
parallel during delta searching phase, we need a mutex. The realloc
part in packlist_alloc() is not protected because it only happens
during the object counting phase, which is always single-threaded.
Access to e->delta_size_ (and by extension
pack->delta_size[e - pack->objects]) is unprotected as before, the
thread scheduler in pack-objects must make sure "e" is never updated
by two different threads.
The area under the new lock is as small as possible, avoiding locking
at all in common case, since lock contention with high thread count
could be expensive (most blobs are small enough that delta compute
time is short and we end up taking the lock very often). The previous
attempt to always hold a lock in oe_delta_size() and
oe_set_delta_size() increases execution time by 33% when repacking
linux.git with with 40 threads.
Reported-by: Elijah Newren <newren@gmail.com>
Helped-by: Elijah Newren <newren@gmail.com>
Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When running the same reproduction script as the previous patch,
it turns out the stack is too small, which can be easily avoided.
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The is_descendant_of method previously used in_merge_bases() to check if
the commit can reach any of the commits in the provided list. This had
two performance problems:
1. The performance is quadratic in worst-case.
2. A single in_merge_bases() call requires walking beyond the target
commit in order to find the full set of boundary commits that may be
merge-bases.
The can_all_from_reach method avoids this quadratic behavior and can
limit the search beyond the target commits using generation numbers. It
requires a small prototype adjustment to stop using commit-date as a
cutoff, as that optimization is no longer appropriate here.
Since in_merge_bases() uses paint_down_to_common(), is_descendant_of()
naturally found cutoffs to avoid walking the entire commit graph. Since
we want to always return the correct result, we cannot use the
min_commit_date cutoff in can_all_from_reach. We then rely on generation
numbers to provide the cutoff.
Since not all repos will have a commit-graph file, nor will we always
have generation numbers computed for a commit-graph file, create a new
method, generation_numbers_enabled(), that checks for a commit-graph
file and sees if the first commit in the file has a non-zero generation
number. In the case that we do not have generation numbers, use the old
logic for is_descendant_of().
Performance was meausured on a copy of the Linux repository using the
'test-tool reach is_descendant_of' command using this input:
A:v4.9
X:v4.10
X:v4.11
X:v4.12
X:v4.13
X:v4.14
X:v4.15
X:v4.16
X:v4.17
X.v3.0
Note that this input is tailored to demonstrate the quadratic nature of
the previous method, as it will compute merge-bases for v4.9 versus all
of the later versions before checking against v4.1.
Before: 0.26 s
After: 0.21 s
Since we previously used the is_descendant_of method in the ref_newer
method, we also measured performance there using
'test-tool reach ref_newer' with this input:
A:v4.9
B:v3.19
Before: 0.10 s
After: 0.08 s
By adding a new commit with parent v3.19, we test the non-reachable case
of ref_newer:
Before: 0.09 s
After: 0.08 s
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The can_all_from_reach_with_flags() algorithm is currently quadratic in
the worst case, because it calls the reachable() method for every 'from'
without tracking which commits have already been walked or which can
already reach a commit in 'to'.
Rewrite the algorithm to walk each commit a constant number of times.
We also add some optimizations that should work for the main consumer of
this method: fetch negotitation (haves/wants).
The first step includes using a depth-first-search (DFS) from each
'from' commit, sorted by ascending generation number. We do not walk
beyond the minimum generation number or the minimum commit date. This
DFS is likely to be faster than the existing reachable() method because
we expect previous ref values to be along the first-parent history.
If we find a target commit, then we mark everything in the DFS stack as
a RESULT. This expands the set of targets for the other 'from' commits.
We also mark the visited commits using 'assign_flag' to prevent re-
walking the same commits.
We still need to clear our flags at the end, which is why we will have a
total of three visits to each commit.
Performance was measured on the Linux repository using
'test-tool reach can_all_from_reach'. The input included rows seeded by
tag values. The "small" case included X-rows as v4.[0-9]* and Y-rows as
v3.[0-9]*. This mimics a (very large) fetch that says "I have all major
v3 releases and want all major v4 releases." The "large" case included
X-rows as "v4.*" and Y-rows as "v3.*". This adds all release-candidate
tags to the set, which does not greatly increase the number of objects
that are considered, but does increase the number of 'from' commits,
demonstrating the quadratic nature of the previous code.
Small Case:
Before: 1.52 s
After: 0.26 s
Large Case:
Before: 3.50 s
After: 0.27 s
Note how the time increases between the two cases in the two versions.
The new code increases relative to the number of commits that need to be
walked, but not directly relative to the number of 'from' commits.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The ref_newer method is used by 'git push' to check if a force-push is
required. This method does not use any kind of cutoff when walking, so
in the case of a force-push will walk all reachable commits.
The is_descendant_of method already uses paint_down_to_common along with
cutoffs. By translating the ref_newer arguments into the commit and
commit_list required by is_descendant_of, we can have one fewer commit
walk and also improve our performance!
For a copy of the Linux repository, 'test-tool reach ref_newer' presents
the following improvements with the specified input. In the case that
ref_newer returns 1, there is no improvement. The improvement is in the
second case where ref_newer returns 0.
Input:
A:v4.9
B:v3.19
Before: 0.09 s
After: 0.09 s
To test the negative case, add a new commit with parent v3.19,
regenerate the commit-graph, and then run with B pointing at that
commit.
Before: 0.43 s
After: 0.09 s
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The commit_contains method has two modes which depend on the given
ref_filter struct. We have the "normal" algorithm (which is also the
typically-slow operation) and the "tag" algorithm. This difference is
essentially what changes performance for 'git branch --contains' versus
'git tag --contains'. There are thoughts that the data shapes used by
these two applications justify the different implementations.
Create tests using 'test-tool reach commit_contains [--tag]' to cover
both methods.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The can_all_from_reach_with_flags method is used by ok_to_give_up in
upload-pack.c to see if we have done enough negotiation during a fetch.
This method is intentionally created to preserve state between calls to
assist with stateful negotiation, such as over SSH.
To make this method testable, add a new can_all_from_reach method that
does the initial setup and final tear-down. We will later use this
method in production code. Call the method from 'test-tool reach' for
now.
Since this is a many-to-many reachability query, add a new type of input
to the 'test-tool reach' input format. Lines "Y:<committish>" create a
list of commits to be the reachability targets from the commits in the
'X' list. In the context of fetch negotiation, the 'X' commits are the
'want' commits and the 'Y' commits are the 'have' commits.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The get_merge_bases_many method returns a list of merge bases for a
single commit (A) against a list of commits (X). Some care is needed in
constructing the expected behavior because the result is not the
expected merge-base for an octopus merge with those parents but instead
the set of maximal commits that are reachable from A and at least one of
the commits in X.
Add get_merge_bases_many to 'test-tool reach' and create a test that
demonstrates that this output returns multiple results. Specifically, we
select a list of three commits such that we output two commits that are
reachable from one of the first two, respectively, and none are
reachable from the third.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The is_descendant_of method takes a single commit as its first parameter
and a list of commits as its second parameter. Extend the input of the
'test-tool reach' command to take multiple lines of the form
"X:<committish>" to construct a list of commits. Pass these to
is_descendant_of and create tests that check each result.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
As we prepare to change the behavior of the algorithms in
commit-reach.c, create a new test-tool subcommand 'reach' to test these
methods on interesting commit-graph shapes.
To use the new test-tool, use 'test-tool reach <method>' and provide
input to stdin that describes the inputs to the method. Currently, we
only implement the ref_newer method, which requires two commits. Use
lines "A:<committish>" and "B:<committish>" for the two inputs. We will
expand this input later to accommodate methods that take lists of
commits.
The test t6600-test-reach.sh creates a repo whose commits form a
two-dimensional grid. This grid makes it easy for us to determine
reachability because commit-A-B can reach commit-X-Y if and only if A is
at least X and B is at least Y. This helps create interesting test cases
for each result of the methods in commit-reach.c.
We test all methods in three different states of the commit-graph file:
Non-existent (no generation numbers), fully computed, and mixed (some
commits have generation numbers and others do not).
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are several commit walks in the codebase. Group them together into
a new commit-reach.c file and corresponding header. After we group these
walks into one place, we can reduce duplicate logic by calling
equivalent methods.
The can_all_from_reach_with_flags method is used in a stateful way by
upload-pack.c. The parameters are very flexible, so we will be able to
use its commit walking logic for many other callers.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The ok_to_give_up() method uses the commit date as a cutoff to avoid
walking the entire reachble set of commits. Before moving the
reachable() method to commit-reach.c, pull out the dependence on the
global constant 'oldest_have' with a 'min_commit_date' parameter.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In anticipation of consolidating all commit reachability algorithms,
refactor ok_to_give_up() in order to allow splitting its logic into
an external method.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In anticipation of moving the reachable() method to commit-reach.c,
modify the prototype to be more generic to flags known outside of
upload-pack.c. Also rename 'want' to 'from' to make the statement
more clear outside of the context of haves/wants negotiation.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are several commit walks in the codebase. Group them together into
a new commit-reach.c file and corresponding header. After we group these
walks into one place, we can reduce duplicate logic by calling
equivalent methods.
All methods are direct moves, except we also make the commit_contains()
method public so its consumers in ref-filter.c can still call it. We can
also test this method in a test-tool in a later commit.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are several commit walks in the codebase. Group them together into
a new commit-reach.c file and corresponding header. After we group these
walks into one place, we can reduce duplicate logic by calling
equivalent methods.
The ref_newer() method is used by 'git push -f' to check if a force-push
is necessary. By making the method public, we make it possible to test
the method directly without setting up an envieronment where a 'git
push' call makes sense.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
These methods are now declared in commit-reach.h. Remove them from
commit.h and add new include statements in all files that require these
declarations.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are several commit walks in the codebase. Group them together into
a new commit-reach.c file and corresponding header. After we group these
walks into one place, we can reduce duplicate logic by calling
equivalent methods.
The method declarations in commit.h are not touched by this commit and
will be moved in a following commit. Many consumers need to point to
commit-reach.h and that would bloat this commit.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Teach clone to send a list of ref-prefixes, when using protocol v2, to
allow the server to filter out irrelevant references from the
ref-advertisement. This reduces wasted time and bandwidth when cloning
repositories with a larger number of references.
Signed-off-by: Brandon Williams <bmwill@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If a 'git repack' command replaces existing packfiles, then we must
clear the existing multi-pack-index before moving the packfiles it
references.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The multi-pack-index, when present, tracks the existence of objects and
their offsets within a list of packfiles. This allows us to use the
multi-pack-index for object lookups, abbreviations, and object counts.
When the multi-pack-index tracks a packfile, then we do not need to add
that packfile to the packed_git linked list or the MRU list.
We still need to load the packfiles that are not tracked by the
multi-pack-index.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Due to how Windows handles replacing a lockfile when there is an open
handle, create the close_midx() method to close the existing midx before
writing the new one.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The core.multiPackIndex config setting controls the multi-pack-
index (MIDX) feature. If false, the setting will disable all reads
from the multi-pack-index file.
Read this config setting in the new prepare_multi_pack_index_one()
which is called during prepare_packed_git(). This check is run once
per repository.
Add comparison commands in t5319-multi-pack-index.sh to check
typical Git behavior remains the same as the config setting is turned
on and off. This currently includes 'git rev-list' and 'git log'
commands to trigger several object database reads. Currently, these
would only catch an error in the prepare_multi_pack_index_one(), but
with later commits will catch errors in object lookups, abbreviations,
and approximate object counts.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The final pair of chunks for the multi-pack-index file stores the object
offsets. We default to using 32-bit offsets as in the pack-index version
1 format, but if there exists an offset larger than 32-bits, we use a
trick similar to the pack-index version 2 format by storing all offsets
at least 2^31 in a 64-bit table; we use the 32-bit table to point into
that 64-bit table as necessary.
We only store these 64-bit offsets if necessary, so create a test that
manipulates a version 2 pack-index to fake a large offset. This allows
us to test that the large offset table is created, but the data does not
match the actual packfile offsets. The multi-pack-index offset does match
the (corrupted) pack-index offset, so a future feature will compare these
offsets during a 'verify' step.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
