Originally, when designing pluggable object databases the goal was that
the object database can have multiple sources, and every source attached
to it could use a different backend. This would have allowed for quite a
lot of flexibility, as you could trivially mix and match different kinds
of object storages in whatever way you like.
But while well-intentioned, this design led to a bunch of conceptual
problems:
- We're now trying to read objects in source order, whereas we
previously tried to read objects via packfiles before trying to read
them via loose objects. This led to a performance regression when
using alternates or when using a quarantine directory.
- Some data structures are supposed to only ever exist once, like for
example bitmaps and commit graphs. At the same time, those data
structures also span across the union of all objects, so they may
cross sources.
- It is unclear how we can extend GIT_OBJECT_DIRECTORY or
GIT_ALTERNATE_OBJECT_DIRECTORIES to become backend-agnostic in a
backwards-compatible way. In general, introducing an object storage
extension into the current status quo where alternates may have to
be extended to become generic was proving to be painful.
- Some mechanisms of alternates assume way too much about how exactly
their backends work. Alternate refs for example assume that the
alternate is backed by a filesystem path, and that this filesystem
path may also allow us to read references. This is not a given
though, as backends may not even have local data at all.
In short, there are a bunch of conceptual mismatches when we have
alternates and pluggable object databases coexist. So while the original
idea was nice, it does not result in a system that is easy to reason
about.
Correct course by moving alternates into the "files" source itself so
that it becomes an implementation detail thereof so that we can avoid
all of these shortcomings. While it's unfortunate that we cannot easily
mix and match sources now, that ability doesn't go away. It's still very
much feasible to introduce a new backend that allows for exactly that
use case, and such a backend may also be a lot more flexible as we can
now add new logic to determine which objects should be stored where. So
the original motivation for having per-source backends can still be
realized with the new architecture.
Note that as part of this move, we also handle the GIT_OBJECT_DIRECTORY
and GIT_ALTERNATE_OBJECT_DIRECTORIES environment variables in the
"files" backend. This may be surprising at first, but object directories
are very much a concept of that backend, too. So these variables would
have bad interactions with other backends, and they create a bit of a
mismatch with the eventual object storage extension that we plan to
introduce.
Note that this commit is way larger than I'd like it to be. I'm sorry, I
couldn't find a way to split it up further. That being said, most of the
changes are straight-forward conversions that go from iterating over
sources to iterating over object directories. The more involved changes
are moving the infrastructure to track individual sources from "odb.c"
and moving them into "odb/source-files.c".
This also serves to show a bit of a who's-who of commands that don't
work properly with pluggable object databases. Almost all of these
commands are related to housekeeping though -- some of them will be
converted eventually, like for example commit graphs or MIDXs. But many
of them will stay incompatible going forward as they are simply too
specific to the "files" backend.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The "files" object database source consists of two separate backends for
storing loose and packed objects. These are managed as somewhat separate
entities even though they derive from the same object directory, whether
it's the primary object directory or that one from an alternate.
In a subsequent commit we'll move the handling of alternates into the
"files" backend completely so that it becomes another implementation
detail thereof. As part of this, we'll want to keep track of both of
these sub-sources as a single entity derived from their respective
object directory.
Prepare for this change by introducing a new `struct odb_files_dir` that
encapsulates them. For now, every "files" source has exactly one such
directory. In a subsequent commit though, we'll make it a linked list of
directories so that we can manage multiple such directories in a single
"files" source.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Our interfaces used to interact with MIDXs all work on top of the
generic `struct odb_source`. This doesn't make much sense though: a MIDX
is strictly tied to the "packed" source, so passing in a generic source
gives the false sense that it may also work with a different type of
source.
Fix this conceptual weirdness and instead require the caller to pass in
a "packed" source explicitly. This also makes the next commit easier to
implement, where we drop the pointer to the "files" source in the
"packed" source.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Expose the incremental MIDX repacking mode (implemented in an earlier
commit) via a new --write-midx=incremental option for `git repack`.
Add "incremental" as a recognized argument to the --write-midx
OPT_CALLBACK, mapping it to REPACK_WRITE_MIDX_INCREMENTAL. When this
mode is active and --geometric is in use, set the midx_layer_threshold
on the pack geometry so that only packs in sufficiently large tip layers
are considered for repacking.
Two new configuration options control the compaction behavior:
- repack.midxSplitFactor (default: 2): the factor used in the
geometric merging condition for MIDX layers.
- repack.midxNewLayerThreshold (default: 8): the minimum number of
packs in the tip MIDX layer before its packs are considered as
candidates for geometric repacking.
Add tests exercising the new mode across a variety of scenarios
including basic geometric violations, multi-round chain integrity,
branching and merging histories, cross-layer object uniqueness, and
threshold-based compaction.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Store the ODB source in the `existing_packs` struct and use that in
place of the raw `repo->objects->sources` access within `cmd_repack()`.
The source used is still assigned from the first source in the list, so
there are no functional changes in this commit. The changes instead
serve two purposes (one immediate, one not):
- The incremental MIDX-based repacking machinery will need to know what
source is being used to read the existing MIDX/chain (should one
exist).
- In the future, if "git repack" is taught how to operate on other
object sources, this field will serve as the authoritative value for
that source.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We have a bunch of different sites that want to iterate through all
packs of a given `struct packfile_store`. This pattern is somewhat
verbose and repetitive, which makes it somewhat cumbersome.
Introduce a new macro `repo_for_each_pack()` that removes some of the
boilerplate.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The "pack_kept_objects" variable is defined as static to the repack
builtin, but is inherently related to the pack-objects arguments that
the builtin uses when generating new packs.
Move that field into the "struct pack_objects_args", and shuffle around
where we append the corresponding command-line option when preparing a
pack-objects process. Specifically:
- `write_cruft_pack()` always wants to pass "--honor-pack-keep", so
explicitly set the `pack_kept_objects` field to "0" when initializing
the `write_pack_opts` struct before calling `write_cruft_pack()`.
- `write_filtered_pack()` no longer needs to handle writing the
command-line option "--honor-pack-keep" when preparing a pack-objects
process, since its call to `prepare_pack_objects()` will have already
taken care of that.
`write_filtered_pack()` also reads the `pack_kept_objects` field to
determine whether to write the existing kept packs with a leading "^"
character, so update that to read through the `po_args` pointer
instead.
- `cmd_repack()` also no longer has to write the "--honor-pack-keep"
flag explicitly, since this is also handled via its call to
`prepare_pack_objects()`.
Since there is a default value for "pack_kept_objects" that relies on
whether or not we are writing a bitmap (and not writing a MIDX), extract
a default initializer for `struct pack_objects_args` that keeps this
conditional default behavior.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In a similar spirit as the previous commit(s), now that the function
`finish_pack_objects_cmd()` has no explicit dependencies within the
repack builtin, let's extract it.
This prepares us to extract the remaining two functions within the
repack builtin that explicitly write packfiles, which are
`write_cruft_pack()` and `write_filtered_pack()`, which will be done in
the future commits.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Similar to the previous commit, the functions `write_cruft_pack()` and
`write_filtered_pack()` both compute a "local" variable via the exact
same mechanism:
const char *scratch;
int local = skip_prefix(opts->destination, opts->packdir, &scratch);
Not only does this cause us to repeat the same pair of lines, it also
introduces an unnecessary "scratch" variable that is common between both
functions.
Instead of repeating ourselves, let's extract that functionality into a
new function in the repack.h API called "write_pack_opts_is_local()".
That function takes a pointer to a "struct write_pack_opts" (which has
as fields both "destination" and "packdir"), and can encapsulate the
dangling "scratch" field.
Extract that function and make it visible within the repack.h API, and
use it within both `write_cruft_pack()` and `write_filtered_pack()`.
While we're at it, match our modern conventions by returning a "bool"
instead of "int", and use `starts_with()` instead of `skip_prefix()` to
avoid storing the dummy "scratch" variable.
The remaining duplication (that is, that both `write_cruft_pack()` and
`write_filtered_pack()` still both call `write_pack_opts_is_local()`)
will be addressed in the following commit.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Both callers within the repack builtin which call functions that take a
'write_pack_opts' structure have the following pattern:
struct write_pack_opts opts = {
.packdir = packdir,
.packtmp = packtmp,
.pack_prefix = find_pack_prefix(packdir, packtmp),
/* ... */
};
int ret = write_some_kind_of_pack(&opts, /* ... */);
, but both "packdir" and "packtmp" are fields within the write_pack_opts
struct itself!
Instead of also computing the pack_prefix ahead of time, let's have the
callees compute it themselves by moving `find_pack_prefix()` out of the
repack builtin, and have it take a write_pack_opts pointer instead of
the "packdir" and "packtmp" fields directly.
This avoids the callers having to do some prep work that is common
between the two of them, but also avoids the potential pitfall of
accidentally writing:
.pack_prefix = find_pack_prefix(packtmp, packdir),
(which is well-typed) when the caller meant to instead write:
.pack_prefix = find_pack_prefix(packdir, packtmp),
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Instead of storing the list of MIDX pack names separately, let's inline
it into the existing_packs struct, further reducing the number of
parameters we have to pass around.
This amounts to adding a new string_list to the existing_packs struct,
and populating it via `existing_packs_collect()`. This is fairly
straightforward to do, since we are already looping over all packs, all
we need to do is:
if (p->multi_pack_index)
string_list_append(&existing->midx_packs, pack_basename(p));
Note, however, that this check *must* come before other conditions where
we discard and do not keep track of a pack, including the condition "if
(!p->pack_local)" immediately below. This is because the existing
routine which collects MIDX pack names does so blindly, and does not
discard, for example, non-local packs.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Now that we have factored the "generated_pack" API, we can move it to
repack.ch, further slimming down builtin/repack.c.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The repack builtin defines an API for keeping track of which packs
were found in the repository at the beginning of the repack operation.
This is used to classify what state a pack was in (kept, non-kept, or
cruft), and is also used to mark which packs to delete (or keep) at the
end of a repack operation.
Now that the prerequisite refactoring is complete, this API is isolated
enough that it can be moved out to repack.[ch] and removed from the
builtin entirely.
As a result, some of its functions become static within repack.c,
cleaning up the visible API.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Extract "remove_redundant_pack()" as generic repack-related
functionality by moving its implementation to the repack.[ch]
compilation unit.
This is a prerequisite to moving the "existing_packs" API, which is one
of the callers of this function. (The remaining caller in the pack
geometry code will eventually move to its own compilation unit as well,
and will likewise rely on this function.)
While moving it over, prefix the function name with "repack_" to
indicate that it belongs to the repack-subsystem.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Now that the 'prepare_pack_objects' function no longer refers to
external, static variables, move it out to repack.h as generic
functionality.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Over the years, builtin/repack.c has turned into a grab-bag of
functionality powering the 'git repack' builtin. Among its many
capabilities, it:
- can build and spawn 'git pack-objects' commands, which in turn
generate new packs
- has infrastructure to manage the set of existing packs in a
repository
- has infrastructure to split a sequence of packs into a geometric
progression based on object size
- can manage both generating and combining cruft packs together
- can write new MIDXs
to name a few.
As a result, this builtin has accumulated a lot of code, making adding
new functionality difficult. In the future, 'repack' will learn how to
manage a chain of incremental MIDXs, adding yet more functionality into
the builtin.
As a prerequisite step, let's first move some of the functionality in
the builtin into its own repack.[ch].
This will be done over the course of many steps, since there are many
individual components, some of which will end up in other, yet-to-exist
compilation units of their own. Some of the code movement here is also
non-trivial, so performing it in individual steps will make it easier to
verify.
Let's start by migrating 'struct pack_objects_args' (and the related
corresponding pack_objects_args_release() function) into repack.h, and
teach both the Makefile and Meson how to build the new compilation unit.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Patrick Steinhardt