In practice the DHT storage layer has not been performing as well as
large scale server environments want to see from a Git server.
The performance of the DHT schema degrades rapidly as small changes
are pushed into the repository due to the chunk size being less than
1/3 of the pushed pack size. Small chunks cause poor prefetch
performance during reading, and require significantly longer prefetch
lists inside of the chunk meta field to work around the small size.
The DHT code is very complex (>17,000 lines of code) and is very
sensitive to the underlying database round-trip time, as well as the
way objects were written into the pack stream that was chunked and
stored on the database. A poor pack layout (from any version of C Git
prior to Junio reworking it) can cause the DHT code to be unable to
enumerate the objects of the linux-2.6 repository in a completable
time scale.
Performing a clone from a DHT stored repository of 2 million objects
takes 2 million row lookups in the DHT to locate the OBJECT_INDEX row
for each object being cloned. This is very difficult for some DHTs to
scale, even at 5000 rows/second the lookup stage alone takes 6 minutes
(on local filesystem, this is almost too fast to bother measuring).
Some servers like Apache Cassandra just fall over and cannot complete
the 2 million lookups in rapid fire.
On a ~400 MiB repository, the DHT schema has an extra 25 MiB of
redundant data that gets downloaded to the JGit process, and that is
before you consider the cost of the OBJECT_INDEX table also being
fully loaded, which is at least 223 MiB of data for the linux kernel
repository. In the DHT schema answering a `git clone` of the ~400 MiB
linux kernel needs to load 248 MiB of "index" data from the DHT, in
addition to the ~400 MiB of pack data that gets sent to the client.
This is 193 MiB more data to be accessed than the native filesystem
format, but it needs to come over a much smaller pipe (local Ethernet
typically) than the local SATA disk drive.
I also never got around to writing the "repack" support for the DHT
schema, as it turns out to be fairly complex to safely repack data in
the repository while also trying to minimize the amount of changes
made to the database, due to very common limitations on database
mutation rates..
This new DFS storage layer fixes a lot of those issues by taking the
simple approach for storing relatively standard Git pack and index
files on an abstract filesystem. Packs are accessed by an in-process
buffer cache, similar to the WindowCache used by the local filesystem
storage layer. Unlike the local file IO, there are some assumptions
that the storage system has relatively high latency and no concept of
"file handles". Instead it looks at the file more like HTTP byte range
requests, where a read channel is a simply a thunk to trigger a read
request over the network.
The DFS code in this change is still abstract, it does not store on
any particular filesystem, but is fairly well suited to the Amazon S3
or Apache Hadoop HDFS. Storing packs directly on HDFS rather than
HBase removes a layer of abstraction, as most HBase row reads turn
into an HDFS read.
Most of the DFS code in this change was blatently copied from the
local filesystem code. Most parts should be refactored to be shared
between the two storage systems, but right now I am hesistent to do
this due to how well tuned the local filesystem code currently is.
Change-Id: Iec524abdf172e9ec5485d6c88ca6512cd8a6eafb
BlameGenerator digs through history and discovers the origin of each
line of some result file. BlameResult consumes the stream of regions
created by the generator and lays them out in a table for applications
to display alongside of source lines.
Applications may optionally push in the working tree copy of a file
using the push(String, byte[]) method, allowing the application to
receive accurate line annotations for the working tree version. Lines
that are uncommitted (difference between HEAD and working tree) will
show up with the description given by the application as the author,
or "Not Committed Yet" as a default string.
Applications may also run the BlameGenerator in reverse mode using the
reverse(AnyObjectId, AnyObjectId) method instead of push(). When
running in the reverse mode the generator annotates lines by the
commit they are removed in, rather than the commit they were added in.
This allows a user to discover where a line disappeared from when they
are looking at an older revision in the repository. For example:
blame --reverse 16e810b2..master -L 1080, org.eclipse.jgit.test/tst/org/eclipse/jgit/storage/file/RefDirectoryTest.java
( 1080) }
2302a6d3 (Christian Halstrick 2011-05-20 11:18:20 +0200 1081)
2302a6d3 (Christian Halstrick 2011-05-20 11:18:20 +0200 1082) /**
2302a6d3 (Christian Halstrick 2011-05-20 11:18:20 +0200 1083) * Kick the timestamp of a local file.
Above we learn that line 1080 (a closing curly brace of the prior
method) still exists in branch master, but the Javadoc comment below
it has been removed by Christian Halstrick on May 20th as part of
commit 2302a6d3. This result differs considerably from that of C
Git's blame --reverse feature. JGit tells the reader which commit
performed the delete, while C Git tells the reader the last commit
that still contained the line, leaving it an exercise to the reader
to discover the descendant that performed the removal.
This is still only a basic implementation. Quite notably it is
missing support for the smart block copy/move detection that the C
implementation of `git blame` is well known for. Despite being
incremental, the BlameGenerator can only be run once. After the
generator runs it cannot be reused. A better implementation would
support applications browsing through history efficiently.
In regards to CQ 5110, only a little of the original code survives.
CQ: 5110
Bug: 306161
Change-Id: I84b8ea4838bb7d25f4fcdd540547884704661b8f
Signed-off-by: Kevin Sawicki <kevin@github.com>
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Chris Aniszczyk <caniszczyk@gmail.com>
Using a resolver and factory pattern for the anonymous git:// Daemon
class makes transport.Daemon more useful on non-file storage systems,
or in embedded applications where the caller wants more precise
control over the work tasks constructed within the daemon.
Rather than defining new interfaces, move the existing HTTP ones
into transport.resolver and make them generic on the connection
handle type. For HTTP, continue to use HttpServletRequest, and
for transport.Daemon use DaemonClient.
To remain compatible with transport.Daemon, FileResolver needs to
learn how to use multiple base directories, and how to export any
Repository instance at a fixed name.
Change-Id: I1efa6b2bd7c6567e983fbbf346947238ea2e847e
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
The NoteMap makes it easy to read a small notes tree as created by
the `git notes` command in C Git. To make the initial implementation
simple a notes tree is read recursively into a map in memory.
This is reasonable if the application will need to access all notes,
or if there are less than 256 notes in the tree, but doesn't behave
well when the number of notes exceeds 256 and the application
doesn't need to access all of them.
We can later add support for lazily loading different subpaths,
thus fixing the large note tree problem described above.
Currently the implementation only supports reading. Writing notes
is more complex because trees need to be expanded or collapsed at
the exact 256 entry cut-off in order to retain the same tree SHA-1
that C Git would use for the same content. It also needs to retain
non-note tree entries such as ".gitignore" or ".gitattribute" files
that might randomly appear within a notes tree. We can also add
writing support later.
Change-Id: I93704bd84ebf650d51de34da3f1577ef0f7a9144
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Chris Aniszczyk <caniszczyk@gmail.com>
The maven 2 build for jgit source bundle didn't create a correct
OSGi version string, instead of
org.eclipse.jgit.source_0.10.0.<timestamp>
the generated OSGi version was
org.eclipse.jgit.source_0.10.0.SNAPSHOT
This caused trouble when trying to install it from p2 repository.
Bug: 327616
Change-Id: Ic27c763ae9a4bcbb5bd6ed9562cd98bb4da3386b
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
Create a new 'org.eclipse.jgit.api.errors' package to contain
exceptions related to using the Git porcelain API.
Change-Id: Iac1781bd74fbd520dffac9d347616c3334994470
Signed-off-by: Chris Aniszczyk <caniszczyk@gmail.com>
This patch adds ignore compatibility to jgit. It encompasses
exclude files as well as .gitignore. Uses TreeWalk and
FileTreeIterator to find nodes and parses .gitignore
files when required. The patch includes a simple cache that
can be used to save results and avoid excessive gitignore
parsing.
CQ: 4302
Bug: 303925
Change-Id: Iebd7e5bb534accca4bf00d25bbc1f561d7cad11b
Signed-off-by: Chris Aniszczyk <caniszczyk@gmail.com>
Signed-off-by: Stefan Lay <stefan.lay@sap.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
Similar to what we did with the file code, move the pack writer
into its own package so the related classes and their package
private methods are hidden from the rest of the library.
Change-Id: Ic1b5c7c8c8d266e90c910d8d68dfc8e93586854f
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
This move isolates all of the local file specific implementation code
into a single package, where their package-private methods and support
classes are properly hidden away from the rest of the core library.
Because of the sheer number of files impacted, I have limited this
change to only the renames and the updated imports.
Change-Id: Icca4884e1a418f83f8b617d0c4c78b73d8a4bd17
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Replace the old crude event listener system with a much more generic
implementation, patterned after the event dispatch techniques used
in Google Web Toolkit 1.5 and later.
Each event delivers to an interface that defines a single method,
and the event itself is what performs the delivery in a type-safe
way through its own dispatch method.
Listeners are registered in a generic listener list, indexed by
the interface they implement and wish to receive an event for.
Delivery of events is performed by looping through all listeners
implementing the event's corresponding listener interface, and using
the event's own dispatch method to deliver the event. This is the
classical "double dispatch" pattern for event delivery.
Listeners can be unregistered by invoking remove() on their
registration handle. This change therefore requires application
code to track the handle if it wishes to remove the listener at a
later point in time.
Event delivery is now exposed as a generic public method on the
Repository class, making it easier for any type of message to
be sent out to any type of listener that has registered, without
needing to pre-arrange for type-safe fireFoo() methods.
New event types can be added in the future simply by defining a
new RepositoryEvent subclass and a corresponding RepositoryListener
interface that it dispatches to. By always adding new events through
a new interface, we never need to worry about defining an Adapter
to provide default no-op implementations of new event methods.
Change-Id: I651417b3098b9afc93d91085e9f0b2265df8fc81
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Created wrong tags for 0.8.3 hence creating another version.
Change-Id: I4e00bbcffe1cf872e2d7e3f3d88d068701fb5330
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
Matthias Sohn
Shawn O. Pearce
Chris Aniszczyk
Charley Wang