The AbstractChannelAdapterParser creates an implicit DirectChannel
if the adapter has no 'channel' attribute.
The Event, TCP, and UDP channel adapter parsers did not bind
this channel to the adapter and AC initialization failed with
'outputChannel is required'.
Further, the event schema marked the channel as being 'required',
precluding this feature.
INT-2407 Remove Channel use="required"
Parsers automatically generate the channel when none is provided.
- JMX
- JDBC
- SFTP
- Redis
- Feed
- XMPP
- Mail
- FTP
- HTTP
* MessageChannel was moved from core to the base package.
* ChannelResolver moved from core to support.
* TcpConnectionInterceptorFactoryChain was incorrect.
This is a significant update to the build system, including the changes
listed below. README.md has been updated with instructions on the most
important day-to-day commands.
- Eliminate buildSrc submodule
In favor of using the new bundlor and docbook-reference plugins. The
net effect is a large reduction in number of lines of build code.
Common docbook resources, stylesheets, etc are stored directly in the
docbook plugin.
This means that --recursive is no longer required when cloning and
there will never be a need to use `git submodule` commands. README
files have been updated to reflect.
Use of the new bundlor plugin also means the removal of template.mf
files from the source tree in favor of an inline approach. See
build.gradle for details. Bundlor 'import templates' are built up
programmatically and kept physically close to gradle dependency
declarations, leading to more convenience when changing these values
and hopefully fewer errors / version inconsistencies over time.
Certain tests depended on the presence of template.mf files, all of
which have recently been removed from the source tree in favor of the
new bundlor plugin which allows for inlining bundlor configuration
within the Gradle build script. These tests now create temp files
using the java.io.File API instead.
- Upgrade to Gradle 1.0-milestone-6
The m6 release is significantly faster when resolving dependencies
and has a number of valuable new features over the earlier m3
version. Review the release notes for Gradle 1.0-milestone-6 online
for full details.
- Switch to repo.springsource.org repository
Previously the project build declared as many repositories as
necessary to resolve all project dependencies.
Now depending on a single 'virtual repository' defined within the
SpringSource Artifactory instance at http://repo.springsource.org.
Currently, the virtual repository in use is 'libs-milestone', which
allows for the resolution of all "milestone-or-better" versions of
all S2 and third-party dependencies.
Should snapshot dependencies become required, this value may be
changed from 'libs-milestone' to 'libs-snapshot'. To build only
against GA releases, change the value to 'libs-release'.
- New build plan(s)
Spring Integration build plans have been updated to use the
Artifactory Bamboo plugin and publish to repo.springsource.org.
Build plans have names like 2.1.x to reflect the version under
development, not necessarily the name of the branch, as this may
change over time and across major releases.
- Improve release process
As mentioned above, Spring Integration will now use the Artifactory
Bamboo plugin to publish releases and also use Artifactory's support
for pushing builds directly into Maven Central via oss.sonatype.org.
Generate poms that contain all necessary fields for onboarding at
Maven central (scm, developers, organization, licenses, etc).
Generate -source and -javadoc poms to comply with Maven Central
onboarding rules (and for general good practice anyway).
Generation of PGP signatures, sha1 and md5 checksums are all handled
automatically by Artifactory. These are also requirements for
automated entry into Maven Central.
- Remove source-level pom generation
Automatic generation of Maven poms suitable for use in building
Spring Integration is no longer supported. Generation and
publication of poms for the purpose of dependency management remains
supported.
Sonar support has to date depended on these poms, but will be
switched over to use the Gradle Sonar plugin shortly.
- Eliminate docs subproject
Move docs/src to the root of the project and eliminate docs as a
formal subproject. This simplifies the build in a number of ways,
including removing the need for distinguishing between 'subprojects'
and 'javaprojects' as well as allowing users to build both 'api' and
'reference' docs without qualifying with a ':docs' prefix.
Also rename the src/info directory to src/dist to better reflect that
these files are packaged with the distribution. For example, the
readme.txt there is really the distribution readme, distinct from the
README.md at the root of the project which is for building from source,
etc.
Mac OSX problem was encountered only on the client; similar
code exists on the server side, so the workaround should
be implemented there too; just in case.
Selector.select() throws CancelledKeyException when selecting
on a closed socket.
http://java.net/jira/browse/MACOSX_PORT-526
Work around is to catch the exception and continue.
The IP module produced a significant number of benign exceptions
during a maven build. Due to auto-startup adapters with a
small retry-interval for client-mode adapters.
If a tcp client connection factory was configured for single-use
connections (one socket per message), and it was used by an outbound
channel adapter (send and forget), the socket is closed immediately
after sending the message.
This could cause issues with the connection handling because certain
failures could occur. For example, the close could occur before we
ever registered the channel for read selection, resulting in a
ClosedChannelException. Secondly, there was a small timing hole
where a selection key could be invalidated between the isValid()
call and isReadable().
These problems have been corrected.
Occasional test failures caused because test HelloWorldInterceptor
could prematurely close the connection, depending on timing.
Also added some minor cleanup to the tests themselves.
Autostartup defaulted to false and isRunning() returned
the status of the associated connection factory instead
of its own status. Precluded it being started because
LifeCycle processor thought it was already started.
Previously, for single-use server sockets, and for client sockets
that have a collaborating inbound adapter, the socket timeout
was set to 10 seconds, with no way to set to infinity. The work
around was to set it to Integer.MAX_VALUE.
The intent was to avoid DoS attacks and mis-behaved clients, but
some users have expressed a desire to be able to set an infinite
timeout in both scenarios.
A future release may remove the 10 second default from the client
socket that has a collaborating inbound adapter, and default to
infinity. For this reason, that case now emits a warning.
The default 10 second timeout on single-use server sockets will
likely remain.
INT-1872 Remove WARN Log
We decided to eliminate the log message regarding a possible
future change to the default timeout for client sockets used
for request/reply.
Add client mode for inbound endpoints, and for the
outbound adapter.
INT-1770 Allow Inbound Adapter to Open Connection
Normally, inbound adapters use server sockets and wait for incoming
connection. There are use cases where the adapter should establish
the connection and wait for inbound messages.
INT-1770 Open Connection on Start
Allow configuration of outbound adapter to permit
connection establishment when the adapter is started
rather than when the first message arrives.
INT-1770 Allow Inbound Gateway to Open Connection
Normally, inbound gateways listen for connections. There are
use cases where an inbound gateway might open the connection
and then wait for incoming requests.
INT-1770 Parsers
Update parsers and tests to support attributes for setting
endpoints in client-mode.
INT-1770 Docs
Update reference with client-mode information.
INT-1770 Add Control Bus for Client Mode
Enable control bus commands to check status and to attempt
connection establishment.
ip_connection_id header is used to determine to which socket
a message should be written. If this header is missing, we
can't send the message and we log an error. Previously, an
NPE was thrown.
This applies to inbound gateway replies and asynchronous
outbound server messages.
INT-1876 Add AutoStart and Phase to TCP Endpoints
There was no way for an application to control the startup
of TCP endpoints. This is particularly an issue for controlling
availability of inbound endpoints.
The TCP Endpoints are mostly static, with dynamic behavior
being contained in connection factories. The factories
implement SmartLifecycle but there is no way to
configure auto-startup.
Adds auto-startup and phase attributes to all TCP endpoints
Starting/stopping an endpoint will start/stop the underlying
connection factory. When a connection factory is stopped,
all open connections are closed, and no new connections will
be acceptedi (inbound) or allowex (outbound).
While the factories implement SmartLifecycle, they return
false to auto-startup. We rely on the fact that
the endpoints are dependent on factories and
the factories will be started because of this dependency.
Finally, with collaborating endpoints (where an inbound and
outbound adapter are bound to the same factory), starting/
stopping one endpoint will affect the collaborating
endpoint too.
INT-1876 Polish Schema Docs
Doc was incorrectly talking about connection factories instead
of endpoints.
INT-2162 Add apply-sequence to Connection Factory
When messages are generated as a result of incoming tcp messages, and this
attribute is true (default=false), the sequenceNumber and
correlationId headers will be set. This will facilitate
downstream resequencing if necessary.
The 'ip_connection_id' header is still required for internal correlation
of reply messages to the right socket (regardless of whether downstream
components have altered the correlationId), and will contain the same value
as the correlationId (if set). This will not use up more memory for the
header content because they will both point to the same heap object.
The 'ip_connection_seq' header is deprecated and may be removed
in a future release.
TcpConnection.getConnectionSeq() has been renamed to
incrementAndGetConnectionSequence() because it is not a simple
getter method, it increments the connection's sequence number.
INT-2162 Update Docs for apply-sequence
Add documentation for apply-sequence attribute on
TCP Connection Factories.
Factor out socket creation to a protected method, to facilitate
easier subclassing, enabling something other than the default
Socket factories to be used.
2 modules still had Spring 3.1 transitive dependencies:
* Spring Integration Redis
* Spring Integration Gemfire
1 module was lacking a Jackson dependency:
* Spring Integration AMQP
