| Age | Commit message (Collapse) | Author |
|---|
|
Test #29 ('ref transaction: corrupted tables cause failure') started to
fail intermittently for me (from v2.52.0-rc0) when running the testsuite
with '-j8'. (Also, having moved to a new laptop and windows 11, rather
than windows 10). If the test is run by hand, or without any parallelism,
then it passes without issue.
When the test fails (e.g. 1 out of 32 parallel runs) the cause is due to
a permission error while corrupting a table file:
./test-lib.sh: line 1010: .git/reftable/0x000000000001-0x000000000002-d89bb8ee.ref: Permission denied
This corruption is done in a shell loop, directly after a 'test_commit',
which uses an ': >"$f"' expression to truncate the file. Adding a sleep
of one second after the 'test_commit' and before the shell loop fixes
the test (it is not clear why). Replacing the redirection shell expression
with a 'test-tool truncate "$f" 0' invocation also provides a fix, which
could simply be another way to change the timing sufficiently to win the
race.
During a debug session, I tried looking at the strace output for the
shell redirection:
$ rm /tmp/hello; echo hello >/tmp/hello; ls -l /tmp/hello
-rw-r--r-- 1 ramsay None 6 Nov 10 17:25 /tmp/hello
$
$ strace -o zzz bash -c ': >/tmp/hello'
$
Similarly, for the test-tool solution:
$ strace -o xxx ./t/helper/test-tool truncate /tmp/hello 0
$
When comparing the output, the differences seemed to be what you would
expect and, if anything, the shell redirect probably would have taken
longer than the test-tool solution (many fcntl() calls to dup the stdout
to the <fd>). The call to the win32 api NtCreateFile() was identical,
apart from the first (FileHandle) parameter, of course.
In order to fix this flaky test on cygwin, despite not knowing why it
works, replace the shell redirection with the above 'test-tool truncate'
invocation.
Helped-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Ramsay Jones <ramsay@ramsayjones.plus.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Both the "files" and "reftable" backend store peeled object IDs for
references that point to tags:
- The "files" backend stores the value when packing refs, where each
peeled object ID is prefixed with "^".
- The "reftable" backend stores the value whenever writing a new
reference that points to a tag via a special ref record type.
Both of these backends use `peel_object()` to find the peeled object ID.
But as explained in the preceding commit, that function does not detect
the case where the tag's tagged object and its claimed type mismatch.
The consequence of storing these bogus peeled object IDs is that we're
less likely to detect such corruption in other parts of Git.
git-for-each-ref(1) for example does not notice anymore that the tag is
broken when using "--format=%(*objectname)" to dereference tags.
One could claim that this is good, because it still allows us to mostly
use the tag as intended. But the biggest problem here is that we now
have different behaviour for such a broken tag depending on whether or
not we have its peeled value in the refdb.
Fix the issue by verifying the object type when peeling the object. If
that verification fails we simply skip storing the peeled value in
either of the reference formats.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The "seq" tool has a "-f" option to produce printf-style formatted
lines. Let's teach our test_seq helper the same trick. This lets us get
rid of some shell loops in test snippets (which are particularly verbose
in our test suite because we have to "|| return 1" to keep the &&-chain
going).
This converts a few call-sites I found by grepping around the test
suite. A few notes on these:
- In "seq", the format specifier is a "%g" float. Since test_seq only
supports integers, I've kept the more natural "%d" (which is what
these call sites were using already).
- Like "seq", test_seq automatically adds a newline to the specified
format. This is what all callers are doing already except for t0021,
but there we do not care about the exact format. We are just trying
to printf a large number of bytes to a file. It's not worth
complicating other callers or adding an option to avoid the newline
in that caller.
- Most conversions are just replacing a shell loop (which does get rid
of an extra fork, since $() requires a subshell). In t0612 we can
replace an awk invocation, which I think makes the end result more
readable, as there's less quoting.
- In t7422 we can replace one loop, but sadly we have to leave the
loop directly above it. This is because that earlier loop wants to
include the seq value twice in the output, which test_seq does not
support (nor does regular seq). If you run:
test_seq -f "foo-%d %d" 10
the second "%d" will always be the empty string. You might naively
think that test_seq could add some extra arguments, like:
# 3 ought to be enough for anyone...
printf "$fmt\n" "$i "$i" $i"
but that just triggers printf to format multiple lines, one per
extra set of arguments.
So we'd have to actually parse the format string, figure out how
many "%" placeholders are there, and then feed it that many
instances of the sequence number. The complexity isn't worth it.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
"make test" used to have a hard dependency on (basic) Perl; tests
have been rewritten help environment with NO_PERL test the build as
much as possible.
* ps/test-wo-perl-prereq:
t5703: refactor test to not depend on Perl
t5316: refactor `max_chain()` to not depend on Perl
t0210: refactor trace2 scrubbing to not use Perl
t0021: refactor `generate_random_characters()` to not depend on Perl
t/lib-httpd: refactor "one-time-perl" CGI script to not depend on Perl
t/lib-t6000: refactor `name_from_description()` to not depend on Perl
t/lib-gpg: refactor `sanitize_pgp()` to not depend on Perl
t: refactor tests depending on Perl for textconv scripts
t: refactor tests depending on Perl to print data
t: refactor tests depending on Perl substitution operator
t: refactor tests depending on Perl transliteration operator
Makefile: stop requiring Perl when running tests
meson: stop requiring Perl when tests are enabled
t: adapt existing PERL prerequisites
t: introduce PERL_TEST_HELPERS prerequisite
t: adapt `test_readlink()` to not use Perl
t: adapt `test_copy_bytes()` to not use Perl
t: adapt character translation helpers to not use Perl
t: refactor environment sanitization to not use Perl
t: skip chain lint when PERL_PATH is unset
|
|
A bunch of tests rely on Perl to print data in various different ways.
These usages fall into the following categories:
- Print data conditionally by matching patterns. These usecases can be
converted to use awk(1) rather easily.
- Print data repeatedly. These usecases can typically be converted to
use a combination of `test-tool genzeros` and sed(1).
- Print data in reverse. These usecases can be converted to use
awk(1) or `sort -r`.
Refactor the tests accordingly so that we can drop a couple of
PERL_TEST_HELPERS prerequisites.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
In the early days of Git, Perl was used quite prominently throughout the
project. This has changed significantly as almost all of the executables
we ship nowadays have eventually been rewritten in C. Only a handful of
subsystems remain that require Perl:
- gitweb, a read-only web interface.
- A couple of scripts that allow importing repositories from GNU Arch,
CVS and Subversion.
- git-send-email(1), which can be used to send mails.
- git-request-pull(1), which is used to request somebody to pull from
a URL by sending an email.
- git-filter-branch(1), which uses Perl with the `--state-branch`
option. This command is typically recommended against nowadays in
favor of git-filter-repo(1).
- Our Perl bindings for Git.
- The netrc Git credential helper.
None of these subsystems can really be considered to be part of the
"core" of Git, and an installation without them is fully functional.
It is more likely than not that an end user wouldn't even notice that
any features are missing if those tools weren't installed. But while
Perl nowadays very much is an optional dependency of Git, there is a
significant limitation when Perl isn't available: developers cannot run
our test suite.
Preceding commits have started to lift this restriction by removing the
strict dependency on Perl in many central parts of the test library. But
there are still many tests that rely on small Perl helpers to do various
different things.
Introduce a new PERL_TEST_HELPERS prerequisite that guards all tests
that require Perl. This prerequisite is explicitly different than the
preexisting PERL prerequisite:
- PERL records whether or not features depending on the Perl
interpreter are built.
- PERL_TEST_HELPERS records whether or not a Perl interpreter is
available for our tests.
By having these two separate prerequisites we can thus distinguish
between tests that inherently depend on Perl because the underlying
feature does, and those tests that depend on Perl because the test
itself is using Perl.
Adapt all tests to set the PERL_TEST_HELPERS prerequisite as needed.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Remove the_repository global variable in favor of the repository
argument that gets passed in "builtin/pack-refs.c".
When `-h` is passed to the command outside a Git repository, the
`run_builtin()` will call the `cmd_pack_refs()` function with `repo` set
to NULL and then early in the function, `parse_options()` call will give
the options help and exit.
Mentored-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Usman Akinyemi <usmanakinyemi202@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Leakfixes.
* ps/leakfixes-part-10: (27 commits)
t: remove TEST_PASSES_SANITIZE_LEAK annotations
test-lib: unconditionally enable leak checking
t: remove unneeded !SANITIZE_LEAK prerequisites
t: mark some tests as leak free
t5601: work around leak sanitizer issue
git-compat-util: drop now-unused `UNLEAK()` macro
global: drop `UNLEAK()` annotation
t/helper: fix leaking commit graph in "read-graph" subcommand
builtin/branch: fix leaking sorting options
builtin/init-db: fix leaking directory paths
builtin/help: fix leaks in `check_git_cmd()`
help: fix leaking return value from `help_unknown_cmd()`
help: fix leaking `struct cmdnames`
help: refactor to not use globals for reading config
builtin/sparse-checkout: fix leaking sanitized patterns
split-index: fix memory leak in `move_cache_to_base_index()`
git: refactor builtin handling to use a `struct strvec`
git: refactor alias handling to use a `struct strvec`
strvec: introduce new `strvec_splice()` function
line-log: fix leak when rewriting commit parents
...
|
|
Now that the default value for TEST_PASSES_SANITIZE_LEAK is `true` there
is no longer a need to have that variable declared in all of our tests.
Drop it.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
By default, Windows restricts access to files when those files have been
opened by another process. As explained in the preceding commits, these
restrictions can be loosened such that reads, writes and/or deletes of
files with open handles _are_ allowed.
While we set up those sharing flags in most relevant code paths now, we
still don't properly handle POSIX-style atomic renames in case the
target path is open. This is failure demonstrated by t0610, where one of
our tests spawns concurrent writes in a reftable-enabled repository and
expects all of them to succeed. This test fails most of the time because
the process that has acquired the "tables.list" lock is unable to rename
it into place while other processes are busy reading that file.
Windows 10 has introduced the `FILE_RENAME_FLAG_POSIX_SEMANTICS` flag
that allows us to fix this usecase [1]. When set, it is possible to
rename a file over a preexisting file even when the target file still
has handles open. Those handles must have been opened with the
`FILE_SHARE_DELETE` flag, which we have ensured in the preceding
commits.
Careful readers might have noticed that [1] does not mention the above
flag, but instead mentions `FILE_RENAME_POSIX_SEMANTICS`. This flag is
not for use with `SetFileInformationByHandle()` though, which is what we
use. And while the `FILE_RENAME_FLAG_POSIX_SEMANTICS` flag exists, it is
not documented on [2] or anywhere else as far as I can tell.
Unfortunately, we still support Windows systems older than Windows 10
that do not yet have this new flag. Our `_WIN32_WINNT` SDK version still
targets 0x0600, which is Windows Vista and later. And even though that
Windows version is out-of-support, bumping the SDK version all the way
to 0x0A00, which is Windows 10 and later, is not an option as it would
make it impossible to compile on Windows 8.1, which is still supported.
Instead, we have to manually declare the relevant infrastructure to make
this feature available and have fallback logic in place in case we run
on a Windows version that does not yet have this flag.
On another note: `mingw_rename()` has a retry loop that is used in case
deleting a file failed because it's still open in another process. One
might be pressed to not use this loop anymore when we can use POSIX
semantics. But unfortunately, we have to keep it around due to our
dependence on the `FILE_SHARE_DELETE` flag. While we know to set that
sharing flag now, other applications may not do so and may thus still
cause sharing violations when we try to rename a file.
This fixes concurrent writes in the reftable backend as demonstrated in
t0610, but may also end up fixing other usecases where Git wants to
perform renames.
[1]: https://learn.microsoft.com/en-us/windows-hardware/drivers/ddi/ntifs/ns-ntifs-_file_rename_information
[2]: https://learn.microsoft.com/en-us/windows/win32/api/winbase/ns-winbase-file_rename_info
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
In 6241ce2170 (refs/reftable: reload locked stack when preparing
transaction, 2024-09-24) we have introduced a new test that exercises
how the reftable backend behaves with many concurrent writers all racing
with each other. This test was introduced after a couple of fixes in
this context that should make concurrent writes behave gracefully. As it
turns out though, Windows systems do not yet handle concurrent writes
properly, as we've got two reports for Cygwin and MinGW failing in this
newly added test.
The root cause of this is how we update the "tables.list" file: when
writing a new stack of tables we first write the data into a lockfile
and then rename that file into place. But Windows forbids us from doing
that rename when the target path is open for reading by another process.
And as the test races both readers and writers with each other we are
quite likely to hit this edge case.
This is not a regression: the logic didn't work before the mentioned
commit, and after the commit it performs well on Linux and macOS, and
the situation on Windows should have at least improved a bit. But the
test shows that we need to put more thought into how to make this work
properly there.
Work around the issue by disabling the test on Windows for now. While at
it, increase the locking timeout to address reported timeouts when using
either the address or memory sanitizer, which also tend to significantly
extend the runtime of this test.
This should be revisited after Git v2.47 is out.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When starting a reftable transaction we lock all stacks we are about to
modify. While it may happen that the stack is out-of-date at this point
in time we don't really care: transactional updates encode the expected
state of a certain reference, so all that we really want to verify is
that the _current_ value matches that expected state.
Pass `REFTABLE_STACK_NEW_ADDITION_RELOAD` when locking the stack such
that an out-of-date stack will be reloaded after having been locked.
This change is safe because all verifications of the expected state
happen after this step anyway.
Add a testcase that verifies that many writers are now able to write to
the stack concurrently without failures and with a deterministic end
result.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When multiple concurrent processes try to update references in a
repository they may try to lock the same lockfiles. This can happen even
when the updates are non-conflicting and can both be applied, so it
doesn't always make sense to abort the transaction immediately. Both the
"loose" and "packed" backends thus have a grace period that they wait
for the lock to be released that can be controlled via the config values
"core.filesRefLockTimeout" and "core.packedRefsTimeout", respectively.
The reftable backend doesn't have such a setting yet and instead fails
immediately when it sees such a lock. But the exact same concepts apply
here as they do apply to the other backends.
Introduce a new "reftable.lockTimeout" config that controls how long we
may wait for a "tables.list" lock to be released. The default value of
this config is 100ms, which is the same default as we have it for the
"loose" backend.
Note that even though we also lock individual tables, this config really
only applies to the "tables.list" file. This is because individual
tables are only ever locked when we already hold the "tables.list" lock
during compaction. When we observe such a lock we in fact do not want to
compact the table at all because it is already in the process of being
compacted by a concurrent process. So applying the same timeout here
would not make any sense and only delay progress.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When compacting tables, it may happen that we want to compact a set of
tables which are already locked by a concurrent process that compacts
them. In the case where we wanted to perform a full compaction of all
tables it is sensible to bail out in this case, as we cannot fulfill the
requested action.
But when performing auto-compaction it isn't necessarily in our best
interest of us to abort the whole operation. For example, due to the
geometric compacting schema that we use, it may be that process A takes
a lot of time to compact the bulk of all tables whereas process B
appends a bunch of new tables to the stack. B would in this case also
notice that it has to compact the tables that process A is compacting
already and thus also try to compact the same range, probably including
the new tables it has appended. But because those tables are locked
already, it will fail and thus abort the complete auto-compaction. The
consequence is that the stack will grow longer and longer while A isn't
yet done with compaction, which will lead to a growing performance
impact.
Instead of aborting auto-compaction altogether, let's gracefully handle
this situation by instead compacting tables which aren't locked. To do
so, instead of locking from the beginning of the slice-to-be-compacted,
we start locking tables from the end of the slice. Once we hit the first
table that is locked already, we abort. If we succeeded to lock two or
more tables, then we simply reduce the slice of tables that we're about
to compact to those which we managed to lock.
This ensures that we can at least make some progress for compaction in
said scenario. It also helps in other scenarios, like for example when a
process died and left a stale lockfile behind. In such a case we can at
least ensure some compaction on a best-effort basis.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
There are a bunch of tests which do not have any leaks:
- t0411: Introduced via 5c5a4a1c05 (t0411: add tests for cloning from
partial repo, 2024-01-28), passes since its inception.
- t0610: Introduced via 57db2a094d (refs: introduce reftable backend,
2024-02-07), passes since its inception.
- t2405: Passes since 6741e917de (repository: avoid leaking
`fsmonitor` data, 2024-04-12).
- t7423: Introduced via b20c10fd9b (t7423: add tests for symlinked
submodule directories, 2024-01-28), passes since e8d0608944
(submodule: require the submodule path to contain directories only,
2024-03-26). The fix is not obviously related, but probably works
because we now die early in many code paths.
- t9xxx: All of these are exercising CVS-related tooling and pass
since at least Git v2.40. It's likely that these pass for a long
time already, but nobody ever noticed because Git developers do not
tend to have CVS on their machines.
Mark all of these tests as passing.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Tests to ensure interoperability between reftable written by jgit
and our code have been added and enabled in CI.
* ps/ci-test-with-jgit:
t0612: add tests to exercise Git/JGit reftable compatibility
t0610: fix non-portable variable assignment
t06xx: always execute backend-specific tests
ci: install JGit dependency
ci: make Perforce binaries executable for all users
ci: merge scripts which install dependencies
ci: fix setup of custom path for GitLab CI
ci: merge custom PATH directories
ci: convert "install-dependencies.sh" to use "/bin/sh"
ci: drop duplicate package installation for "linux-gcc-default"
ci: skip sudo when we are already root
ci: expose distro name in dockerized GitHub jobs
ci: rename "runs_on_pool" to "distro"
|
|
Code to write out reftable has seen some optimization and
simplification.
* ps/reftable-write-optim:
reftable/block: reuse compressed array
reftable/block: reuse zstream when writing log blocks
reftable/writer: reset `last_key` instead of releasing it
reftable/writer: unify releasing memory
reftable/writer: refactorings for `writer_flush_nonempty_block()`
reftable/writer: refactorings for `writer_add_record()`
refs/reftable: don't recompute committer ident
reftable: remove name checks
refs/reftable: skip duplicate name checks
refs/reftable: perform explicit D/F check when writing symrefs
refs/reftable: fix D/F conflict error message on ref copy
|
|
The strategy to compact multiple tables of reftables after many
operations accumulate many entries has been improved to avoid
accumulating too many tables uncollected.
* jt/reftable-geometric-compaction:
reftable/stack: use geometric table compaction
reftable/stack: add env to disable autocompaction
reftable/stack: expose option to disable auto-compaction
|
|
Document and apply workaround for a buggy version of dash that
mishandles "local var=val" construct.
* jc/local-extern-shell-rules:
t1016: local VAR="VAL" fix
t0610: local VAR="VAL" fix
t: teach lint that RHS of 'local VAR=VAL' needs to be quoted
t: local VAR="VAL" (quote ${magic-reference})
t: local VAR="VAL" (quote command substitution)
t: local VAR="VAL" (quote positional parameters)
CodingGuidelines: quote assigned value in 'local var=$val'
CodingGuidelines: describe "export VAR=VAL" rule
|
|
The "shared repository" test in the t0610 reftable test failed
under restrictive umask setting (e.g. 007), which has been
corrected.
* ps/t0610-umask-fix:
t0610: execute git-pack-refs(1) with specified umask
t0610: make `--shared=` tests reusable
|
|
Older versions of the Dash shell fail to parse `local var=val`
assignments in some cases when `val` is unquoted. Such failures can be
observed e.g. with Ubuntu 20.04 and older, which has a Dash version that
still has this bug.
Such an assignment has been introduced in t0610. The issue wasn't
detected for a while because this test used to only run when the
GIT_TEST_DEFAULT_REF_FORMAT environment variable was set to "reftable".
We have dropped that requirement now though, meaning that it runs
unconditionally, including on jobs which use such older versions of
Ubuntu.
We have worked around such issues in the past, e.g. in ebee5580ca
(parallel-checkout: avoid dash local bug in tests, 2021-06-06), by
quoting the `val` side. Apply the same fix to t0610.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The tests in t06xx exercise specific ref formats. Next to probing some
basic functionality, these tests also exercise other low-level details
specific to the format. Those tests are only executed though in case
`GIT_TEST_DEFAULT_REF_FORMAT` is set to the ref format of the respective
backend-under-test.
Ideally, we would run the full test matrix for ref formats such that our
complete test suite is executed with every supported format on every
supported platform. This is quite an expensive undertaking though, and
thus we only execute e.g. the "reftable" tests on macOS and Linux. As a
result, we basically have no test coverage for the "reftable" format at
all on other platforms like Windows.
Adapt these tests so that they override `GIT_TEST_DEFAULT_REF_FORMAT`,
which means that they'll always execute. This increases test coverage on
platforms that don't run the full test matrix, which at least gives us
some basic test coverage on those platforms for the "reftable" format.
This of course comes at the cost of running those tests multiple times
on platforms where we do run the full test matrix. But arguably, this is
a good thing because it will also cause us to e.g. run those tests with
the address sanitizer and other non-standard parameters.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The tests for git-pack-refs(1) with the `core.sharedRepository` config
execute git-pack-refs(1) outside of the shell that has the expected
umask set. This is wrong because we want to test the behaviour of that
command with different umasks. The issue went unnoticed because most
distributions have a default umask of 0022, and we only ever test with
`--shared=true`, which re-adds the group write bit.
Fix the issue by moving git-pack-refs(1) into the umask'd shell and add
a bunch of test cases that exercise behaviour more thoroughly.
Note that we drop the check for whether `core.sharedRepository` was set
to the correct value to make the test setup a bit easier. We should be
able to rely on git-init(1) doing its thing correctly. Furthermore, to
help readability, we convert tests that pass `--shared=true` to instead
pass the equivalent `--shared=group`.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
We have two kinds of `--shared=` tests, one for git-init(1) and one for
git-pack-refs(1). Merge them into a reusable function such that we can
easily add additional testcases with different umasks and flags for the
`--shared=` switch.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
We already perform explicit D/F checks in all reftable callbacks which
write refs, except when writing symrefs. For one this leads to an error
message which isn't perfectly actionable because we only tell the user
that there was a D/F conflict, but not which refs conflicted with each
other. But second, once all ref updating callbacks explicitly check for
D/F conflicts, we can disable the D/F checks in the reftable library
itself and thus avoid some duplicated efforts.
Refactor the code that writes symref tables to explicitly call into
`refs_verify_refname_available()` when writing symrefs.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The `write_copy_table()` function is shared between the reftable
implementations for renaming and copying refs. The only difference
between those two cases is that the rename will also delete the old
reference, whereas copying won't.
This has resulted in a bug though where we don't properly verify refname
availability. When calling `refs_verify_refname_available()`, we always
add the old ref name to the list of refs to be skipped when computing
availability, which indicates that the name would be available even if
it already exists at the current point in time. This is only the right
thing to do for renames though, not for copies.
The consequence of this bug is quite harmless because the reftable
backend has its own checks for D/F conflicts further down in the call
stack, and thus we refuse the update regardless of the bug. But all the
user gets in this case is an uninformative message that copying the ref
has failed, without any further details.
Fix the bug and only add the old name to the skip-list in case we rename
the ref. Consequently, this error case will now be handled by
`refs_verify_refname_available()`, which knows to provide a proper error
message.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
To reduce the number of on-disk reftables, compaction is performed.
Contiguous tables with the same binary log value of size are grouped
into segments. The segment that has both the lowest binary log value and
contains more than one table is set as the starting point when
identifying the compaction segment.
Since segments containing a single table are not initially considered
for compaction, if the table appended to the list does not match the
previous table log value, no compaction occurs for the new table. It is
therefore possible for unbounded growth of the table list. This can be
demonstrated by repeating the following sequence:
git branch -f foo
git branch -d foo
Each operation results in a new table being written with no compaction
occurring until a separate operation produces a table matching the
previous table log value.
Instead, to avoid unbounded growth of the table list, the compaction
strategy is updated to ensure tables follow a geometric sequence after
each operation by individually evaluating each table in reverse index
order. This strategy results in a much simpler and more robust algorithm
compared to the previous one while also maintaining a minimal ordered
set of tables on-disk.
When creating 10 thousand references, the new strategy has no
performance impact:
Benchmark 1: update-ref: create refs sequentially (revision = HEAD~)
Time (mean ± σ): 26.516 s ± 0.047 s [User: 17.864 s, System: 8.491 s]
Range (min … max): 26.447 s … 26.569 s 10 runs
Benchmark 2: update-ref: create refs sequentially (revision = HEAD)
Time (mean ± σ): 26.417 s ± 0.028 s [User: 17.738 s, System: 8.500 s]
Range (min … max): 26.366 s … 26.444 s 10 runs
Summary
update-ref: create refs sequentially (revision = HEAD) ran
1.00 ± 0.00 times faster than update-ref: create refs sequentially (revision = HEAD~)
Some tests in `t0610-reftable-basics.sh` assert the on-disk state of
tables and are therefore updated to specify the correct new table count.
Since compaction is more aggressive in ensuring tables maintain a
geometric sequence, the expected table count is reduced in these tests.
In `reftable/stack_test.c` tests related to `sizes_to_segments()` are
removed because the function is no longer needed. Also, the
`test_suggest_compaction_segment()` test is updated to better showcase
and reflect the new geometric compaction behavior.
Signed-off-by: Justin Tobler <jltobler@gmail.com>
Acked-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
In future tests it will be neccesary to create repositories with a set
number of tables. To make this easier, introduce the
`GIT_TEST_REFTABLE_AUTOCOMPACTION` environment variable that, when set
to false, disables autocompaction of reftables.
Signed-off-by: Justin Tobler <jltobler@gmail.com>
Acked-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The series was based on maint and fixes all the tests that exist
there, but we have acquired a few more.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When running `git maintenance run --auto`, then the various subtasks
will only run as needed. Thus, we for example end up only packing loose
objects if we hit a certain threshold.
Interestingly enough, the "pack-refs" task is actually _never_ executed
when the auto-flag is set because it does not have a condition at all.
As 41abfe15d9 (maintenance: add pack-refs task, 2021-02-09) mentions:
The 'auto_condition' function pointer is left NULL for now. We could
extend this in the future to have a condition check if pack-refs
should be run during 'git maintenance run --auto'.
It is not quite clear from that quote whether it is actually intended
that the task doesn't run at all in this mode. Also, no test was added
to verify this behaviour. Ultimately though, it feels quite surprising
that `git maintenance run --auto --task=pack-refs` would quietly never
do anything at all.
In any case, now that we do have the logic in place to let ref backends
decide whether or not to repack refs, it does make sense to wire it up
accordingly. With the "reftable" backend we will thus now perform
auto-compaction, which optimizes the refdb as needed.
But for the "files" backend we now unconditionally pack refs as it does
not yet know to handle the "auto" flag. Arguably, this can be seen as a
bug fix given that previously the task never did anything at all.
Eventually though we should amend the "files" backend to use some
heuristics for auto compaction, as well.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Forward the `--auto` flag to git-pack-refs(1) when it has been invoked
with this flag itself. This does not change anything for the "files"
backend, which will continue to eagerly pack refs. But it does ensure
that the "reftable" backend only compacts refs as required.
This change does not impact git-maintenance(1) because this command will
in fact never run the pack-refs task when run with `--auto`. This issue
will be addressed in a subsequent commit.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Calling git-pack-refs(1) will unconditionally cause it to pack all
requested refs regardless of the current state of the ref database. For
example:
- With the "files" backend we will end up rewriting the complete
"packed-refs" file even if only a single ref would require
compaction.
- With the "reftable" backend we will end up always compacting all
tables into a single table.
This behaviour can be completely unnecessary depending on the backend
and is thus wasteful.
With the introduction of the `PACK_REFS_AUTO` flag in the preceding
commit we can improve this and let the backends decide for themselves
whether to pack refs in the first place. Expose this functionality via a
new "--auto" flag in git-pack-refs(1), which mirrors the same flag in
both git-gc(1) and git-maintenance(1).
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When git-pack-refs(1) fails in the reftable backend we end up printing
no error message at all, leaving the caller puzzled as to why compaction
has failed. Fix this.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Whenever we commit a new table to the reftable stack we will end up
invoking auto-compaction of the stack to keep the total number of tables
at bay. This auto-compaction may fail though in case at least one of the
tables which we are about to compact is locked. This is indicated by the
compaction function returning `REFTABLE_LOCK_ERROR`. We do not handle
this case though, and thus bubble that return value up the calling
chain, which will ultimately cause a failure.
Fix this bug by ignoring `REFTABLE_LOCK_ERROR`.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
In b0f6b6b523 (refs/reftable: don't fail empty transactions in repo
without HEAD, 2024-02-27), we have added a new test to t0610. This test
contains a useless assignment to a variable that is never actually used.
Remove it.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Under normal circumstances, it shouldn't ever happen that a repository
has no HEAD reference. In fact, git-update-ref(1) would fail any request
to delete the HEAD reference, and a newly initialized repository always
pre-creates it, too.
We have however changed git-clone(1) to partially initialize the
refdb just up to the point where remote helpers can find the
repository. With that change, we are going to run into a situation
where repositories have no refs at all.
Now there is a very particular edge case in this situation: when
preparing an empty ref transacton, we end up returning whatever value
`read_ref_without_reload()` returned to the caller. Under normal
conditions this would be fine: "HEAD" should usually exist, and thus the
function would return `0`. But if "HEAD" doesn't exist, the function
returns a positive value which we end up returning to the caller.
Fix this bug by resetting the return code to `0` and add a test.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Due to scalability issues, Shawn Pearce has originally proposed a new
"reftable" format more than six years ago [1]. Initially, this new
format was implemented in JGit with promising results. Around two years
ago, we have then added the "reftable" library to the Git codebase via
a4bbd13be3 (Merge branch 'hn/reftable', 2021-12-15). With this we have
landed all the low-level code to read and write reftables. Notably
missing though was the integration of this low-level code into the Git
code base in the form of a new ref backend that ties all of this
together.
This gap is now finally closed by introducing a new "reftable" backend
into the Git codebase. This new backend promises to bring some notable
improvements to Git repositories:
- It becomes possible to do truly atomic writes where either all refs
are committed to disk or none are. This was not possible with the
"files" backend because ref updates were split across multiple loose
files.
- The disk space required to store many refs is reduced, both compared
to loose refs and packed-refs. This is enabled both by the reftable
format being a binary format, which is more compact, and by prefix
compression.
- We can ignore filesystem-specific behaviour as ref names are not
encoded via paths anymore. This means there is no need to handle
case sensitivity on Windows systems or Unicode precomposition on
macOS.
- There is no need to rewrite the complete refdb anymore every time a
ref is being deleted like it was the case for packed-refs. This
means that ref deletions are now constant time instead of scaling
linearly with the number of refs.
- We can ignore file/directory conflicts so that it becomes possible
to store both "refs/heads/foo" and "refs/heads/foo/bar".
- Due to this property we can retain reflogs for deleted refs. We have
previously been deleting reflogs together with their refs to avoid
file/directory conflicts, which is not necessary anymore.
- We can properly enumerate all refs. With the "files" backend it is
not easily possible to distinguish between refs and non-refs because
they may live side by side in the gitdir.
Not all of these improvements are realized with the current "reftable"
backend implementation. At this point, the new backend is supposed to be
a drop-in replacement for the "files" backend that is used by basically
all Git repositories nowadays. It strives for 1:1 compatibility, which
means that a user can expect the same behaviour regardless of whether
they use the "reftable" backend or the "files" backend for most of the
part.
Most notably, this means we artificially limit the capabilities of the
"reftable" backend to match the limits of the "files" backend. It is not
possible to create refs that would end up with file/directory conflicts,
we do not retain reflogs, we perform stricter-than-necessary checks.
This is done intentionally due to two main reasons:
- It makes it significantly easier to land the "reftable" backend as
tests behave the same. It would be tough to argue for each and every
single test that doesn't pass with the "reftable" backend.
- It ensures compatibility between repositories that use the "files"
backend and repositories that use the "reftable" backend. Like this,
hosters can migrate their repositories to use the "reftable" backend
without causing issues for clients that use the "files" backend in
their clones.
It is expected that these artificial limitations may eventually go away
in the long term.
Performance-wise things very much depend on the actual workload. The
following benchmarks compare the "files" and "reftable" backends in the
current version:
- Creating N refs in separate transactions shows that the "files"
backend is ~50% faster. This is not surprising given that creating a
ref only requires us to create a single loose ref. The "reftable"
backend will also perform auto compaction on updates. In real-world
workloads we would likely also want to perform pack loose refs,
which would likely change the picture.
Benchmark 1: update-ref: create refs sequentially (refformat = files, refcount = 1)
Time (mean ± σ): 2.1 ms ± 0.3 ms [User: 0.6 ms, System: 1.7 ms]
Range (min … max): 1.8 ms … 4.3 ms 133 runs
Benchmark 2: update-ref: create refs sequentially (refformat = reftable, refcount = 1)
Time (mean ± σ): 2.7 ms ± 0.1 ms [User: 0.6 ms, System: 2.2 ms]
Range (min … max): 2.4 ms … 2.9 ms 132 runs
Benchmark 3: update-ref: create refs sequentially (refformat = files, refcount = 1000)
Time (mean ± σ): 1.975 s ± 0.006 s [User: 0.437 s, System: 1.535 s]
Range (min … max): 1.969 s … 1.980 s 3 runs
Benchmark 4: update-ref: create refs sequentially (refformat = reftable, refcount = 1000)
Time (mean ± σ): 2.611 s ± 0.013 s [User: 0.782 s, System: 1.825 s]
Range (min … max): 2.597 s … 2.622 s 3 runs
Benchmark 5: update-ref: create refs sequentially (refformat = files, refcount = 100000)
Time (mean ± σ): 198.442 s ± 0.241 s [User: 43.051 s, System: 155.250 s]
Range (min … max): 198.189 s … 198.670 s 3 runs
Benchmark 6: update-ref: create refs sequentially (refformat = reftable, refcount = 100000)
Time (mean ± σ): 294.509 s ± 4.269 s [User: 104.046 s, System: 190.326 s]
Range (min … max): 290.223 s … 298.761 s 3 runs
- Creating N refs in a single transaction shows that the "files"
backend is significantly slower once we start to write many refs.
The "reftable" backend only needs to update two files, whereas the
"files" backend needs to write one file per ref.
Benchmark 1: update-ref: create many refs (refformat = files, refcount = 1)
Time (mean ± σ): 1.9 ms ± 0.1 ms [User: 0.4 ms, System: 1.4 ms]
Range (min … max): 1.8 ms … 2.6 ms 151 runs
Benchmark 2: update-ref: create many refs (refformat = reftable, refcount = 1)
Time (mean ± σ): 2.5 ms ± 0.1 ms [User: 0.7 ms, System: 1.7 ms]
Range (min … max): 2.4 ms … 3.4 ms 148 runs
Benchmark 3: update-ref: create many refs (refformat = files, refcount = 1000)
Time (mean ± σ): 152.5 ms ± 5.2 ms [User: 19.1 ms, System: 133.1 ms]
Range (min … max): 148.5 ms … 167.8 ms 15 runs
Benchmark 4: update-ref: create many refs (refformat = reftable, refcount = 1000)
Time (mean ± σ): 58.0 ms ± 2.5 ms [User: 28.4 ms, System: 29.4 ms]
Range (min … max): 56.3 ms … 72.9 ms 40 runs
Benchmark 5: update-ref: create many refs (refformat = files, refcount = 1000000)
Time (mean ± σ): 152.752 s ± 0.710 s [User: 20.315 s, System: 131.310 s]
Range (min … max): 152.165 s … 153.542 s 3 runs
Benchmark 6: update-ref: create many refs (refformat = reftable, refcount = 1000000)
Time (mean ± σ): 51.912 s ± 0.127 s [User: 26.483 s, System: 25.424 s]
Range (min … max): 51.769 s … 52.012 s 3 runs
- Deleting a ref in a fully-packed repository shows that the "files"
backend scales with the number of refs. The "reftable" backend has
constant-time deletions.
Benchmark 1: update-ref: delete ref (refformat = files, refcount = 1)
Time (mean ± σ): 1.7 ms ± 0.1 ms [User: 0.4 ms, System: 1.2 ms]
Range (min … max): 1.6 ms … 2.1 ms 316 runs
Benchmark 2: update-ref: delete ref (refformat = reftable, refcount = 1)
Time (mean ± σ): 1.8 ms ± 0.1 ms [User: 0.4 ms, System: 1.3 ms]
Range (min … max): 1.7 ms … 2.1 ms 294 runs
Benchmark 3: update-ref: delete ref (refformat = files, refcount = 1000)
Time (mean ± σ): 2.0 ms ± 0.1 ms [User: 0.5 ms, System: 1.4 ms]
Range (min … max): 1.9 ms … 2.5 ms 287 runs
Benchmark 4: update-ref: delete ref (refformat = reftable, refcount = 1000)
Time (mean ± σ): 1.9 ms ± 0.1 ms [User: 0.5 ms, System: 1.3 ms]
Range (min … max): 1.8 ms … 2.1 ms 217 runs
Benchmark 5: update-ref: delete ref (refformat = files, refcount = 1000000)
Time (mean ± σ): 229.8 ms ± 7.9 ms [User: 182.6 ms, System: 46.8 ms]
Range (min … max): 224.6 ms … 245.2 ms 6 runs
Benchmark 6: update-ref: delete ref (refformat = reftable, refcount = 1000000)
Time (mean ± σ): 2.0 ms ± 0.0 ms [User: 0.6 ms, System: 1.3 ms]
Range (min … max): 2.0 ms … 2.1 ms 3 runs
- Listing all refs shows no significant advantage for either of the
backends. The "files" backend is a bit faster, but not by a
significant margin. When repositories are not packed the "reftable"
backend outperforms the "files" backend because the "reftable"
backend performs auto-compaction.
Benchmark 1: show-ref: print all refs (refformat = files, refcount = 1, packed = true)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.5 ms … 2.0 ms 1729 runs
Benchmark 2: show-ref: print all refs (refformat = reftable, refcount = 1, packed = true)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.5 ms … 1.8 ms 1816 runs
Benchmark 3: show-ref: print all refs (refformat = files, refcount = 1000, packed = true)
Time (mean ± σ): 4.3 ms ± 0.1 ms [User: 0.9 ms, System: 3.3 ms]
Range (min … max): 4.1 ms … 4.6 ms 645 runs
Benchmark 4: show-ref: print all refs (refformat = reftable, refcount = 1000, packed = true)
Time (mean ± σ): 4.5 ms ± 0.2 ms [User: 1.0 ms, System: 3.3 ms]
Range (min … max): 4.2 ms … 5.9 ms 643 runs
Benchmark 5: show-ref: print all refs (refformat = files, refcount = 1000000, packed = true)
Time (mean ± σ): 2.537 s ± 0.034 s [User: 0.488 s, System: 2.048 s]
Range (min … max): 2.511 s … 2.627 s 10 runs
Benchmark 6: show-ref: print all refs (refformat = reftable, refcount = 1000000, packed = true)
Time (mean ± σ): 2.712 s ± 0.017 s [User: 0.653 s, System: 2.059 s]
Range (min … max): 2.692 s … 2.752 s 10 runs
Benchmark 7: show-ref: print all refs (refformat = files, refcount = 1, packed = false)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.5 ms … 1.9 ms 1834 runs
Benchmark 8: show-ref: print all refs (refformat = reftable, refcount = 1, packed = false)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.4 ms … 2.0 ms 1840 runs
Benchmark 9: show-ref: print all refs (refformat = files, refcount = 1000, packed = false)
Time (mean ± σ): 13.8 ms ± 0.2 ms [User: 2.8 ms, System: 10.8 ms]
Range (min … max): 13.3 ms … 14.5 ms 208 runs
Benchmark 10: show-ref: print all refs (refformat = reftable, refcount = 1000, packed = false)
Time (mean ± σ): 4.5 ms ± 0.2 ms [User: 1.2 ms, System: 3.3 ms]
Range (min … max): 4.3 ms … 6.2 ms 624 runs
Benchmark 11: show-ref: print all refs (refformat = files, refcount = 1000000, packed = false)
Time (mean ± σ): 12.127 s ± 0.129 s [User: 2.675 s, System: 9.451 s]
Range (min … max): 11.965 s … 12.370 s 10 runs
Benchmark 12: show-ref: print all refs (refformat = reftable, refcount = 1000000, packed = false)
Time (mean ± σ): 2.799 s ± 0.022 s [User: 0.735 s, System: 2.063 s]
Range (min … max): 2.769 s … 2.836 s 10 runs
- Printing a single ref shows no real difference between the "files"
and "reftable" backends.
Benchmark 1: show-ref: print single ref (refformat = files, refcount = 1)
Time (mean ± σ): 1.5 ms ± 0.1 ms [User: 0.4 ms, System: 1.0 ms]
Range (min … max): 1.4 ms … 1.8 ms 1779 runs
Benchmark 2: show-ref: print single ref (refformat = reftable, refcount = 1)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.4 ms … 2.5 ms 1753 runs
Benchmark 3: show-ref: print single ref (refformat = files, refcount = 1000)
Time (mean ± σ): 1.5 ms ± 0.1 ms [User: 0.3 ms, System: 1.1 ms]
Range (min … max): 1.4 ms … 1.9 ms 1840 runs
Benchmark 4: show-ref: print single ref (refformat = reftable, refcount = 1000)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.5 ms … 2.0 ms 1831 runs
Benchmark 5: show-ref: print single ref (refformat = files, refcount = 1000000)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.5 ms … 2.1 ms 1848 runs
Benchmark 6: show-ref: print single ref (refformat = reftable, refcount = 1000000)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.5 ms … 2.1 ms 1762 runs
So overall, performance depends on the usecases. Except for many
sequential writes the "reftable" backend is roughly on par or
significantly faster than the "files" backend though. Given that the
"files" backend has received 18 years of optimizations by now this can
be seen as a win. Furthermore, we can expect that the "reftable" backend
will grow faster over time when attention turns more towards
optimizations.
The complete test suite passes, except for those tests explicitly marked
to require the REFFILES prerequisite. Some tests in t0610 are marked as
failing because they depend on still-in-flight bug fixes. Tests can be
run with the new backend by setting the GIT_TEST_DEFAULT_REF_FORMAT
environment variable to "reftable".
There is a single known conceptual incompatibility with the dumb HTTP
transport. As "info/refs" SHOULD NOT contain the HEAD reference, and
because the "HEAD" file is not valid anymore, it is impossible for the
remote client to figure out the default branch without changing the
protocol. This shortcoming needs to be handled in a subsequent patch
series.
As the reftable library has already been introduced a while ago, this
commit message will not go into the details of how exactly the on-disk
format works. Please refer to our preexisting technical documentation at
Documentation/technical/reftable for this.
[1]: https://public-inbox.org/git/CAJo=hJtyof=HRy=2sLP0ng0uZ4=S-DpZ5dR1aF+VHVETKG20OQ@mail.gmail.com/
Original-idea-by: Shawn Pearce <spearce@spearce.org>
Based-on-patch-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
