SSH Authentication is quite expensive (~120ms on localhost against
Gerrit with LFS plugin). The SSH authentication typically also sends a
validity time of the returned token, which allows to re-use it for a
certain time, avoiding the expensive authentication on every download
request. This improves checkout times by large factors depending on the
LFS object amount/sizes.
Also make sure that all instances of Gson used by LFS are configured in
the same way.
Change-Id: I422c94c37021b4322789b3829fa0185e25d683f2
Signed-off-by: Markus Duft <markus.duft@ssi-schaefer.com>
Commit 9530c10192 (2018-02-11)
"Add a minimum negotiation feature for fetch" made fetch
negotiation cheaper for repos with huge numbers of
references (we are seeing a 15x reduction in maximum fetch
times for chromium/chromium/src on trans-Pacific links).
But it inadvertently broke the handling of stateless RPC
connections, so fix that here.
Change-Id: I0090aa76ffecc55801ebb833ac2e0c933a4a7c54
Signed-off-by: Terry Parker <tparker@google.com>
If JGit built in LFS support is enabled for the current repository (or
user/system), any existing pre-push hook will cause an exception for the
time beeing, as only a single pre-push hook is supported.
Thus either native pre-push hooks OR JGit built-in LFS support may be
enabled currently, but not both.
Change-Id: Ie7d2b90e26e948d9cca3d05a7a19489488c75895
Signed-off-by: Markus Duft <markus.duft@ssi-schaefer.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
This is based on the ObjectIdSerialization class written by Shawn Pearce
for the Gerrit Code Review project in 2009 [1]. As mentioned in the
commit message there, it should be part of core JGit.
This implementation is slightly different to Shawn's version. Rather
than having separate methods for null/non-null ids, single methods are
implemented with @Nullable annotations.
[1] https://gerrit-review.googlesource.com/c/gerrit/+/9792
Change-Id: I7599cf8bd1ecd546e2252783d6d672eb76804060
Signed-off-by: David Pursehouse <david.pursehouse@gmail.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
Processing of negated rules, like !bin/ was not working correctly: they
were interpreted too broad, resulting in unexpected untracked files
which should actually be ignored
Bug: 409664
Change-Id: I0a422fd6607941461bf2175c9105a0311612efa0
Signed-off-by: Marc Strapetz <marc.strapetz@syntevo.com>
The upgrade was already done 3 weeks ago, this version number was
missing, however.
Change-Id: Ibc794ba8035c46ea2160925b344ded5714b7c987
Signed-off-by: Michael Keppler <Michael.Keppler@gmx.de>
The previous wording was ambiguous as to whether these were blocks
requested from the cache (hits + misses) or read from underlying storage
(misses only).
They are in fact recording only misses:
Accumulator#{readBlock,readBlockBytes,readBlockMicros} are only
incremented from BlockBasedFile#readOneBlock, which is only called from
the cache miss path in DfsBlockCache#getOrLoad (line 391).
Change-Id: I0135cd1e76d09c1e28e0f1833b34c312511c66ce
**/ should match only directories, but not files
Change-Id: I885c83e5912cac5bff338ba657faf6bb9ec94064
Signed-off-by: Marc Strapetz <marc.strapetz@syntevo.com>
* changes:
RepoCommand: generate relative submodule URLs from absolute URLs.
RepoCommand: don't record new commit if tree did not change
RepoCommand: persist unreadable submodules in .gitmodules
If an object can be found in a non-garbage pack, favor that pack over
paging in the garbage pack's idx and pack content.
Only fall back to garbage packs if an object cannot be found and there
are garbage packs present in the repository. This fallback is
required to correct race conditions during GC.
Change-Id: Ia7c123975bc069b8e6e713eda2d357303b71e329
If a manifest file specifies an absolute URL on the same host on which
the superproject resides, rewrite the URLs to be relative.
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Change-Id: Id616611e5195998fb665c8e7806539a3a02e219a
In cases where a manifest file mixes different remotes, a Gerrit
server process may not have access to all remotes, and won't be able
to produce a full submodule tree.
Preserving this information in .gitmodules will let downstream clients
reconstruct the full tree.
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Change-Id: I52f5d3f288e771dca0af2b4dd3f3fa0f940dcf15
Transfer data in chunks of 8k Transferring data byte per byte is slow,
running checkout with CleanFilter on a 2.9MB file takes 20 seconds.
Using a buffer of 8k shrinks this time to 70ms.
Also register the filter commands in a way that the native GIT LFS can
be used alongside with JGit.
Implements auto-discovery of LFS server URL when cloning from a Gerrit
LFS server.
Change-Id: I452a5aa177dcb346d92af08b27c2e35200f246fd
Also-by: Christian Halstrick <christian.halstrick@sap.com>
Signed-off-by: Markus Duft <markus.duft@ssi-schaefer.com>
When the command is run on a non-bare repository, an instance of
Git is created to execute the commit, and is left open when the
command has finished.
Refactor to not use a class scope Git instance, and make sure it
gets closed before returning.
Change-Id: Ic623ae0fd8b9e264b5dfd434da0de6bb4f910984
Signed-off-by: David Pursehouse <david.pursehouse@gmail.com>
* stable-4.10:
Fix ssh host name handling for Jsch
Jsch overrides the port in the URI with the one in ~/.ssh/config
Change-Id: I860fc61ceb12ae792b1ee7421046ecd32373b9f8
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
These problem usually occur when you subclass ThreadLocal (usually to
implement initialValue). Those classes reference the webapp's
classloader. The ThreadLocal subclass in turn is referenced by each
Thread instance (that's how ThreadLocals are implemented, they have a
"helper-Map" in each Thread instance, so the leak is actually not a tiny
Random instance but the whole webapp's classloader with a bunch of class
definitions and statically referenced parts of the webapp.
Bug: 449321
Change-Id: Ie7a8b0b90e40229e2471202f2a12637b9e0b1d11
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
* stable-4.9:
Fix ssh host name handling for Jsch
Jsch overrides the port in the URI with the one in ~/.ssh/config
Change-Id: Iff9076f65e767bbe8df016337b631bdaeb40ad98
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
If we give Jsch access to the ssh config file, we must _not_ resolve
the host name from the alias. Instead we must give the alias (i.e.,
the host name as is in the URI) to Jsch, so that it finds the same
ssh config entry.
Otherwise if the hostname in the URI, which is taken as an alias in
ssh config ("Host" line), is unequal to the "Hostname" line, and
there happens to be another ssh config entry with that translated
host name as alias, Jsch will pick up that second entry, and we end
up with a strange mixture of both.
Add tests for this case.
Bug: 531118
Change-Id: I249d8c073b0190ed110a69dca5b9be2a749822c3
Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
Jsch unconditionally overwrites the port from the ssh config
file (if a port is specified there), even if the URI explicitly does
give a different port.
Fix this, and add tests.
Change-Id: I7b014543c7ece26270e366db39d7647f82d64f0d
Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
The InitCommand returns a Git that is instantiated with the newly
created Repository, but the Repository is not closed with the Git
resulting in resource leaks.
Create the Git with `closeRepo` set to true, such that the Repository
is also closed when the Git is closed.
Adjust the tests to use try-with-resource on the Git instance.
Change-Id: Ib26e7428c7d8840956d1edb09e53b93e23e6fe5a
Signed-off-by: David Pursehouse <david.pursehouse@gmail.com>
Android an Chrome have several repos with >300k refs. We sometimes see
negotiations of >100k rounds. This change provides a "minimal negotiation"
feature on the client side that limits how many "have" lines the client
sends. The client extracts the current SHA-1 values for the refs in its
wants set, and terminates negotiation early when all of those values have
been sent as haves. If a new branch is being fetched then that set will
be empty and the client will terminate after current default minimum
of two rounds.
This feature is gated behind a "fetch.useminimalnegotiation" configuration
flag, which defaults to false.
Change-Id: Ib12b095cac76a59da6e8f72773c4129e3b32ff2b
Signed-off-by: Terry Parker <tparker@google.com>
Add fetch statistics for the counts of advertised refs, wants and haves.
Also add the duration in milliseconds for the negotiation phase. For
non-bidirectional transports like HTTP, this is the time for the final
round that sends the pack back to the user.
Change-Id: I1af7ffd3cb7b62182340682e2a243691ea24ec2e
Signed-off-by: Terry Parker <tparker@google.com>
This would allow compact and GC process to clean up duplicate ref names in the reftables.
Change-Id: I2b9df0bf72dba63cc3525e374982e60559a776c2
Signed-off-by: Minh Thai <mthai@google.com>
When CleanCommand is collecting the files and folders to be deleted
it may happen that the list of directories contains obsolete entries.
E.g. a folder and its parent folder may be in the list. Only the
parent folder would be sufficient.
This was a reason for hitting FileNotFoundExceptions when finally
trying to delete the files and folders. Improve CleanCommand
to ignore files to be deleted which are already gone.
Bug: 514434
Change-Id: I10caa01bfb9cec5967dfdaea50c6e4a713eeeabd
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
After packaging references, the folders containing these references are
not deleted. In a busy repository, this causes operations to slow down
as traversing the references tree becomes longer.
Delete empty reference folders after the loose references have been
packed.
To avoid deleting a folder that was just created by another concurrent
operation, only delete folders that were not modified in the last 30
seconds.
Signed-off-by: Hector Oswaldo Caballero <hector.caballero@ericsson.com>
Change-Id: Ie79447d6121271cf5e25171be377ea396c7028e0
Bazel@HEAD supports Java 9.
The current code has one single issue with Java 9 compliance: the usage
of javax.xml.bind.DatatypeConverter class for printHexBinary() method.
This class is not available on Java 9. One alternative is to use guava
library. Something similar was done here: [1]. But unlike the case with
checkstyle library, JGit currently doesn't use guava. Instead, we add
java.xml.bind module with --add-modules compiler option.
To build (or test) with Java 9, build custom bazel version and issue:
$ bazel --host_javabase=/usr/lib64/jvm/java-9-openjdk build \
--javacopt='--release 9' \
--java_toolchain=@bazel_tools//tools/jdk:toolchain_jdk9 :all
The Java 9 support is backwards compatible.
* [1] https://github.com/checkstyle/checkstyle/issues/5027
Change-Id: I2c5203fc4e65885ce7b210f824fda85ba6d6c51d
Signed-off-by: David Ostrovsky <david@ostrovsky.org>
Markus Duft
Terry Parker
David Pursehouse
Matthias Sohn
Marc Strapetz
Michael Keppler
Dave Borowitz
Shawn Pearce
Han-Wen Nienhuys
Thomas Wolf
Minh Thai
Christian Halstrick
Hector Caballero
David Ostrovsky