This introduces an additional guard for platforms where `unsigned long`
and `size_t` are not of the same size. If the size of an object in the
database would overflow `unsigned long`, instead we now exit with an
error.
A complete fix will have to update _many_ other functions throughout the
codebase to use `size_t` instead of `unsigned long`. It will have to be
implemented at some stage.
This commit puts in a stop-gap for the time being.
Helped-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Matt Cooper <vtbassmatt@gmail.com>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
When adding many objects to a repo with core.fsyncObjectFiles set to
true, the cost of fsync'ing each object file can become prohibitive.
One major source of the cost of fsync is the implied flush of the
hardware writeback cache within the disk drive. Fortunately, Windows,
and macOS offer mechanisms to write data from the filesystem page cache
without initiating a hardware flush. Linux has the sync_file_range API,
which issues a pagecache writeback request reliably after version 5.2.
This patch introduces a new 'core.fsyncObjectFiles = batch' option that
batches up hardware flushes. It hooks into the bulk-checkin plugging and
unplugging functionality and takes advantage of tmp-objdir.
When the new mode is enabled do the following for each new object:
1. Create the object in a tmp-objdir.
2. Issue a pagecache writeback request and wait for it to complete.
At the end of the entire transaction when unplugging bulk checkin:
1. Issue an fsync against a dummy file to flush the hardware writeback
cache, which should by now have processed the tmp-objdir writes.
2. Rename all of the tmp-objdir files to their final names.
3. When updating the index and/or refs, we assume that Git will issue
another fsync internal to that operation. This is not the case today,
but may be a good extension to those components.
On a filesystem with a singular journal that is updated during name
operations (e.g. create, link, rename, etc), such as NTFS, HFS+, or XFS
we would expect the fsync to trigger a journal writeout so that this
sequence is enough to ensure that the user's data is durable by the time
the git command returns.
This change also updates the macOS code to trigger a real hardware flush
via fnctl(fd, F_FULLFSYNC) when fsync_or_die is called. Previously, on
macOS there was no guarantee of durability since a simple fsync(2) call
does not flush any hardware caches.
_Performance numbers_:
Linux - Hyper-V VM running Kernel 5.11 (Ubuntu 20.04) on a fast SSD.
Mac - macOS 11.5.1 running on a Mac mini on a 1TB Apple SSD.
Windows - Same host as Linux, a preview version of Windows 11.
This number is from a patch later in the series.
Adding 500 files to the repo with 'git add' Times reported in seconds.
core.fsyncObjectFiles | Linux | Mac | Windows
----------------------|-------|-------|--------
false | 0.06 | 0.35 | 0.61
true | 1.88 | 11.18 | 2.47
batch | 0.15 | 0.41 | 1.53
Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When creating a subprocess with a temporary ODB, we set the
GIT_QUARANTINE_ENVIRONMENT env var to tell child Git processes not
to update refs, since the tmp-objdir may go away.
Introduce a similar mechanism for in-process temporary ODBs when
we call tmp_objdir_replace_primary_odb. Now both mechanisms set
the disable_ref_updates flag on the odb, which is queried by
the ref_transaction_prepare function.
Note: This change adds an assumption that the state of
the_repository is relevant for any ref transaction that might
be initiated. Unwinding this assumption should be straightforward
by saving the relevant repository to query in the transaction or
the ref_store.
Peff's test case was invoking ref updates via the cachetextconv
setting. That particular code silently does nothing when a ref
update is forbidden. See the call to notes_cache_put in
fill_textconv where errors are ignored.
Reported-by: Jeff King <peff@peff.net>
Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The tmp_objdir API provides the ability to create temporary object
directories, but was designed with the goal of having subprocesses
access these object stores, followed by the main process migrating
objects from it to the main object store or just deleting it. The
subprocesses would view it as their primary datastore and write to it.
Here we add the tmp_objdir_replace_primary_odb function that replaces
the current process's writable "main" object directory with the
specified one. The previous main object directory is restored in either
tmp_objdir_migrate or tmp_objdir_destroy.
For the --remerge-diff usecase, add a new `will_destroy` flag in `struct
object_database` to mark ephemeral object databases that do not require
fsync durability.
Add 'git prune' support for removing temporary object databases, and
make sure that they have a name starting with tmp_ and containing an
operation-specific name.
Based-on-patch-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Neeraj Singh <neerajsi@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Build fix.
* cb/many-alternate-optim-fixup:
object-file: use unsigned arithmetic with bit mask
object-store: avoid extra ';' from KHASH_INIT
oidtree: avoid nested struct oidtree_node
33f379eee6 (make object_directory.loose_objects_subdir_seen a bitmap,
2021-07-07) replaced a wasteful 256-byte array with a 32-byte array
and bit operations. The mask calculation shifts a literal 1 of type
int left by anything between 0 and 31. UndefinedBehaviorSanitizer
doesn't like that and reports:
object-file.c:2477:18: runtime error: left shift of 1 by 31 places cannot be represented in type 'int'
Make sure to use an unsigned 1 instead to avoid the issue.
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Optimization for repositories with many alternate object store.
* ew/many-alternate-optim:
oidtree: a crit-bit tree for odb_loose_cache
oidcpy_with_padding: constify `src' arg
make object_directory.loose_objects_subdir_seen a bitmap
avoid strlen via strbuf_addstr in link_alt_odb_entry
speed up alt_odb_usable() with many alternates
Prepare the internals for lazily fetching objects in submodules
from their promisor remotes.
* jt/partial-clone-submodule-1:
promisor-remote: teach lazy-fetch in any repo
run-command: refactor subprocess env preparation
submodule: refrain from filtering GIT_CONFIG_COUNT
promisor-remote: support per-repository config
repository: move global r_f_p_c to repo struct
This saves 8K per `struct object_directory', meaning it saves
around 800MB in my case involving 100K alternates (half or more
of those alternates are unlikely to hold loose objects).
This is implemented in two parts: a generic, allocation-free
`cbtree' and the `oidtree' wrapper on top of it. The latter
provides allocation using alloc_state as a memory pool to
improve locality and reduce free(3) overhead.
Unlike oid-array, the crit-bit tree does not require sorting.
Performance is bound by the key length, for oidtree that is
fixed at sizeof(struct object_id). There's no need to have
256 oidtrees to mitigate the O(n log n) overhead like we did
with oid-array.
Being a prefix trie, it is natively suited for expanding short
object IDs via prefix-limited iteration in
`find_short_object_filename'.
On my busy workstation, p4205 performance seems to be roughly
unchanged (+/-8%). Startup with 100K total alternates with no
loose objects seems around 10-20% faster on a hot cache.
(800MB in memory savings means more memory for the kernel FS
cache).
The generic cbtree implementation does impose some extra
overhead for oidtree in that it uses memcmp(3) on
"struct object_id" so it wastes cycles comparing 12 extra bytes
on SHA-1 repositories. I've not yet explored reducing this
overhead, but I expect there are many places in our code base
where we'd want to investigate this.
More information on crit-bit trees: https://cr.yp.to/critbit.html
Signed-off-by: Eric Wong <e@80x24.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There's no point in using 8 bits per-directory when 1 bit
will do. This saves us 224 bytes per object directory, which
ends up being 22MB when dealing with 100K alternates.
Signed-off-by: Eric Wong <e@80x24.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We can save a few milliseconds (across 100K odbs) by using
strbuf_addbuf() instead of strbuf_addstr() by passing `entry' as
a strbuf pointer rather than a "const char *".
Signed-off-by: Eric Wong <e@80x24.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
With many alternates, the duplicate check in alt_odb_usable()
wastes many cycles doing repeated fspathcmp() on every existing
alternate. Use a khash to speed up lookups by odb->path.
Since the kh_put_* API uses the supplied key without
duplicating it, we also take advantage of it to replace both
xstrdup() and strbuf_release() in link_alt_odb_entry() with
strbuf_detach() to avoid the allocation and copy.
In a test repository with 50K alternates and each of those 50K
alternates having one alternate each (for a total of 100K total
alternates); this speeds up lookup of a non-existent blob from
over 16 minutes to roughly 2.7 seconds on my busy workstation.
Note: all underlying git object directories were small and
unpacked with only loose objects and no packs. Having to load
packs increases times significantly.
Signed-off-by: Eric Wong <e@80x24.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Linux users may benefit from additional information on how to
avoid ENOMEM from mmap despite the system having enough RAM to
accomodate them. We can't reliably unmap pack windows to work
around the issue since malloc and other library routines may
mmap without our knowledge.
Signed-off-by: Eric Wong <e@80x24.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is one step towards supporting partial clone submodules.
Even after this patch, we will still lack partial clone submodules
support, primarily because a lot of Git code that accesses submodule
objects does so by adding their object stores as alternates, meaning
that any lazy fetches that would occur in the submodule would be done
based on the config of the superproject, not of the submodule. This also
prevents testing of the functionality in this patch by user-facing
commands. So for now, test this mechanism using a test helper.
Besides that, there is some code that uses the wrapper functions
like has_promisor_remote(). Those will need to be checked to see if they
could support the non-wrapper functions instead (and thus support any
repository, not just the_repository).
Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Reviewed-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git clean" and "git ls-files -i" had confusion around working on
or showing ignored paths inside an ignored directory, which has
been corrected.
* en/dir-traversal:
dir: introduce readdir_skip_dot_and_dotdot() helper
dir: update stale description of treat_directory()
dir: traverse into untracked directories if they may have ignored subfiles
dir: avoid unnecessary traversal into ignored directory
t3001, t7300: add testcase showcasing missed directory traversal
t7300: add testcase showing unnecessary traversal into ignored directory
ls-files: error out on -i unless -o or -c are specified
dir: report number of visited directories and paths with trace2
dir: convert trace calls to trace2 equivalents
Many places in the code were doing
while ((d = readdir(dir)) != NULL) {
if (is_dot_or_dotdot(d->d_name))
continue;
...process d...
}
Introduce a readdir_skip_dot_and_dotdot() helper to make that a one-liner:
while ((d = readdir_skip_dot_and_dotdot(dir)) != NULL) {
...process d...
}
This helper particularly simplifies checks for empty directories.
Also use this helper in read_cached_dir() so that our statistics are
consistent across platforms. (In other words, read_cached_dir() should
have been using is_dot_or_dotdot() and skipping such entries, but did
not and left it to treat_path() to detect and mark such entries as
path_none.)
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Up until recently, object IDs did not have an algorithm member, only a
hash. Consequently, it was possible to share one null (all-zeros)
object ID among all hash algorithms. Now that we're going to be
handling objects from multiple hash algorithms, it's important to make
sure that all object IDs have a correct algorithm field.
Introduce a per-algorithm null OID, and add it to struct hash_algo.
Introduce a wrapper function as well, and use it everywhere we used to
use the null_oid constant.
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Now that struct object_id has an algorithm field, we should populate it.
This will allow us to handle object IDs in any supported algorithm and
distinguish between them. Ensure that the field is written whenever we
write an object ID by storing it explicitly every time we write an
object. Set values for the empty blob and tree values as well.
In addition, use the algorithm field to compare object IDs. Note that
because we zero-initialize struct object_id in many places throughout
the codebase, we default to the default algorithm in cases where the
algorithm field is zero rather than explicitly initialize all of those
locations.
This leads to a branch on every comparison, but the alternative is to
compare the entire buffer each time and padding the buffer for SHA-1.
That alternative ranges up to 3.9% worse than this approach on the perf
t0001, t1450, and t1451.
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we're hashing a value which is going to be an object ID, we want to
zero-pad that value if necessary. To do so, use the final_oid_fn
instead of the final_fn anytime we're going to create an object ID to
ensure we perform this operation.
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
To avoid the penalty of having to branch in hash comparison functions,
we'll want to always compare the full hash member in a struct object_id,
which will require that SHA-1 object IDs be zero-padded. To do so, add
a function which finalizes a hash context and writes it into an object
ID that performs this padding.
Move the definition of struct object_id and the constant definitions
higher up so we they are available for us to use.
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add and apply a semantic patch for converting code that open-codes
CALLOC_ARRAY to use it instead. It shortens the code and infers the
element size automatically.
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change all remnants of "sha1" in hash-lookup.c and .h and rename them to
reflect that we're not just able to handle SHA-1 these days.
Signed-off-by: Martin Ågren <martin.agren@gmail.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Drop the last remnant of "sha1" in this file and rename it to reflect
that we're not just able to handle SHA-1 these days.
Signed-off-by: Martin Ågren <martin.agren@gmail.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
