H2 1.4 requires that file paths that are relative to the current
working directory declare this explicitly, i.e. the path
file:/some-dir must now be file:./some-dir.
This commit updates DatabaseTypeIntegrationTests to meet the
above-described requirement, thereby allowing the tests to pass when
run against H2 1.4.x.
While existing functionality provided the ability to override the
components provided in the baseContext.xml (as required for JSR-352
functionality) at a job by job level, this commit allows a user to
specify a base context for all of the JSR-352 based jobs within a JVM.
To set the location of the custom base context, a JVM property
JSR-352-BASE-CONTEXT should be set.
BATCH-2290
When using remote partitioning, each slave worker persists it's current
status in the same job repsository that the master uses. Because of
this, there is no hard need for the master to wait for each worker to
send a formal response once it's work is complete. Instead, the master
(at the cost of polling a db periodically) can determine if the workers
are done by looking up each partition's status in the job repository.
This commit removes the requirement for a reply channel and implements
the polling of the job repository to determine if the workers are done.
BATCH-2332
While the JSR-352 provides only one way to access the JobOperator
(BatchRuntime.getJobOperator()), it is useful for testing and embedding
to be able to wire your own JsrJobOperator instance. This commit
addresses previous issues with using the constructor that provided that
functionality.
Note: This is a breaking change in that there is a new parameter on the
non-default constructor (adding a PlatformTransactionManager reference).
Users using the JsrJobOperator through the BatchRuntime as previously
mentioned should not be impacted. Since this is the *only* method
perscribed by the JSR to consume that class, it is not expected to have
a large impact.
This fix partially addresses BATCH-2290.
As part of the Spring Framework 4.1 updates, the send methods on both
MailSender#send and JavaMailSender#send were converted from taking an
array to vargs. This commit adds a reflection check to determine which
version of Spring it's being built against and uses the apropriate
mocking checks in these tests based on that check.
contexts scenarios
AbstractApplicationContextFactory, when configuring a child context,
copies all of the BeanPostProcessors from the parent to the child.
Normally this is a good thing. However, in the case of teh
ApplicationContextAwareProcessor, the child context ends up with two
instances, one for the parent context and one for the child context.
This brings two issues to light:
1. Unknown which context is being injected - On beans implementing
ApplicationContextAware, it can't be determined which context (the
parent or the child) will be injected since both BPPs will be called in
an indeterimant order.
2. Errors occur with ApplicationObjectSupport - If the child context
contains a bean that extends ApplicationObjectSupport, when the second
ApplicationContextAwareProcessor is called, the bean will throw an
exception since it is only allowed to be initialized with one
application context.
This fix adds additional logic to remove the
ApplicationContextAwareProcessor from the parent context before adding
all the BPPs to the child.
This fix addresses BATCH-2319.
With the upgrade to ActiveMQ 5.9.0 as part of Spring IO, ActiveMQ's
prefetch behavior changed. With each call to
JmsTemplate#receiveAndConvert, the template creates a new consumer. In
these tests, the first call created a consumer which prefetched all the
test messages on the queue, leaving them unavailable for the second
consumer the JmsTemplate created. By setting prefetch to 0, the
messages are now available for the subsequent
JmsTemplate#receiveAndConvert calls. This addresses BATCH-2248.
BATCH-2159 provides a tweak to the insert statement used for
initializing the sequence tables used in the job repository database so
that they don't error if run multiple times as well as don't insert
multiple rows.
If the user explicitly sets the currentItemCount value on a
RepositoryItemReader, prior to this commit, it is ignored. this commit
addresses this by moving the current counter up to the correct value
when that value is set. This fix addresses BATCH-2274.
This commit also addresses a number of compiler warnings around javadoc.
When using java config, the call to SimpleFlow#afterPropertiesSet is not
called in all cases as noted in BATCH-2291. This commit updates the
builder to explicitly make the call, and updates the SimpleFlow to be
idempotent for calls to afterPropertiesSet.
BATCH-2302 documents a scenario where an item throws an exception in a
fault tollerant step in the write, then in the process, throws an
exception as well. This leads to an infinite loop. To address this,
prior to attempting to write items during scanning we need to validate
that the there are items to be written (we were not which was causing a
NoSuchElementException when we did inputs.next().
This commit also removes an eronous System.out left in the
CoreNamespaceUtils.
The SystemCommandTasklet checks a flag to determine if another thread
has requested that it be stopped. Since this flag will be set via
another thread than the executing thread, it needs to be declared
volitile so that it's state is propigated across threads correctly. The
documentation for StoppableTasklet has also been upadted to call out
that scenario.
This commit addresses Jira BATCH-2271
The JdbcPagingItemReader allows the configuration of multiple sort keys.
If a user configures one of those columns with a table alias, it causes
issues. This update strips the aliases off of the sort keys.
Prior to this commit the initialize method (which is annotated with
@PostConstruct) was throwing a checked exception wich goes against the
@PostConstruct contract. We now wrap any checked exceptions in a
BatchConfigurationException and throw that.
This commit fixes (again) BATCH-2276
When using a combination of Java configuration and the XML namespace
configuration, prior to this commit, we would add the batch custom
scopes twice which lead to issues. This commit gives priority to the
custom scopes provided by the java configuration. If you've used
@EnableBatchProcessing and have configured batch components using the
batch namespace, the custom scopes that come with the java config will
take precidence.
This commit addresses Jira BATCH-2266
The ClassifierCompositeItemWriter loops through the items passed to the
write method twice. The first time to classifier each item. The second
tiem to perform the actual writes (this prevents the unnecessicary IO if
an error occurs during classification). A HashMap was previously used
to hold the classified items (key to the writer : item) however this
causes the items to be written out of order. This update changes it to
a LinkedHashMap to honor the order the items come in.
Without this change, when passing an invalid job parameter to a bean
that is created internally by Spring Batch will cause a
BeanCreationException that is essentially ignored. Due to it being
ignored, an infinite loop occurs in processing.
The AsyncItemWriter is used in conjunction with the AsyncItemProcessor
to unwrap the Futures that are returned by that processor.
Traditionally when an ItemProcessor returns null, it's considered having
been filtered out and should not be passed to the ItemWriter. In this
case, the AsyncItemWriter was not checking for nulls so they were being
passed to the delegate ItemWriter. Most of the OOTB ItemWriters do not
perform a null check prior to doing the write so they were throwing NPEs
when using this paradigm.
