File Support

Spring Integration’s file support extends the Spring Integration core with a dedicated vocabulary to deal with reading, writing, and transforming files.

You need to include this dependency into your project:

Maven
<dependency>
    <groupId>org.springframework.integration</groupId>
    <artifactId>spring-integration-file</artifactId>
    <version>6.2.0-M1</version>
</dependency>
Gradle
compile "org.springframework.integration:spring-integration-file:6.2.0-M1"

It provides a namespace that enables elements defining channel adapters dedicated to files and support for transformers that can read file contents into strings or byte arrays.

This section explains the workings of FileReadingMessageSource and FileWritingMessageHandler and how to configure them as beans. It also discusses the support for dealing with files through file-specific implementations of Transformer. Finally, it explains the file-specific namespace.

Reading Files

A FileReadingMessageSource can be used to consume files from the filesystem. This is an implementation of MessageSource that creates messages from a file system directory. The following example shows how to configure a FileReadingMessageSource:

<bean id="pollableFileSource"
    class="org.springframework.integration.file.FileReadingMessageSource"
    p:directory="${input.directory}"/>

To prevent creating messages for certain files, you can supply a FileListFilter. By default, we use the following filters:

  • IgnoreHiddenFileListFilter

  • AcceptOnceFileListFilter

The IgnoreHiddenFileListFilter ensures that hidden files are not processed. Note that the exact definition of hidden is system-dependent. For example, on UNIX-based systems, a file beginning with a period character is considered to be hidden. Microsoft Windows, on the other hand, has a dedicated file attribute to indicate hidden files.

Version 4.2 introduced the IgnoreHiddenFileListFilter. In prior versions, hidden files were included. With the default configuration, the IgnoreHiddenFileListFilter is triggered first, followed by the AcceptOnceFileListFilter.

The AcceptOnceFileListFilter ensures files are picked up only once from the directory.

The AcceptOnceFileListFilter stores its state in memory. If you wish the state to survive a system restart, you can use the FileSystemPersistentAcceptOnceFileListFilter. This filter stores the accepted file names in a MetadataStore implementation (see Metadata Store). This filter matches on the filename and modified time.

Since version 4.0, this filter requires a ConcurrentMetadataStore. When used with a shared data store (such as Redis with the RedisMetadataStore), it lets filter keys be shared across multiple application instances or across a network file share being used by multiple servers.

Since version 4.1.5, this filter has a new property (flushOnUpdate), which causes it to flush the metadata store on every update (if the store implements Flushable).

The persistent file list filters now have a boolean property forRecursion. Setting this property to true, also sets alwaysAcceptDirectories, which means that the recursive operation on the outbound gateways (ls and mget) will now always traverse the full directory tree each time. This is to solve a problem where changes deep in the directory tree were not detected. In addition, forRecursion=true causes the full path to files to be used as the metadata store keys; this solves a problem where the filter did not work properly if a file with the same name appears multiple times in different directories. IMPORTANT: This means that existing keys in a persistent metadata store will not be found for files beneath the top level directory. For this reason, the property is false by default; this may change in a future release.

The following example configures a FileReadingMessageSource with a filter:

<bean id="pollableFileSource"
    class="org.springframework.integration.file.FileReadingMessageSource"
    p:inputDirectory="${input.directory}"
    p:filter-ref="customFilterBean"/>

A common problem with reading files is that a file may be detected before it is ready (that is, some other process may still be writing the file). The default AcceptOnceFileListFilter does not prevent this. In most cases, this can be prevented if the file-writing process renames each file as soon as it is ready for reading. A filename-pattern or filename-regex filter that accepts only files that are ready (perhaps based on a known suffix), composed with the default AcceptOnceFileListFilter, allows for this situation. The CompositeFileListFilter enables the composition, as the following example shows:

<bean id="pollableFileSource"
    class="org.springframework.integration.file.FileReadingMessageSource"
    p:inputDirectory="${input.directory}"
    p:filter-ref="compositeFilter"/>

<bean id="compositeFilter"
    class="org.springframework.integration.file.filters.CompositeFileListFilter">
    <constructor-arg>
        <list>
            <bean class="o.s.i.file.filters.AcceptOnceFileListFilter"/>
            <bean class="o.s.i.file.filters.RegexPatternFileListFilter">
                <constructor-arg value="^test.*$"/>
            </bean>
        </list>
    </constructor-arg>
</bean>

If it is not possible to create the file with a temporary name and rename to the final name, Spring Integration provides another alternative. Version 4.2 added the LastModifiedFileListFilter. This filter can be configured with an age property so that only files older than this value are passed by the filter. The age defaults to 60 seconds, but you should choose an age that is large enough to avoid picking up a file early (due to, say, network glitches). The following example shows how to configure a LastModifiedFileListFilter:

<bean id="filter" class="org.springframework.integration.file.filters.LastModifiedFileListFilter">
    <property name="age" value="120" />
</bean>

Starting with version 4.3.7, a ChainFileListFilter (an extension of CompositeFileListFilter) has been introduced to allow scenarios when subsequent filters should only see the result of the previous filter. (With the CompositeFileListFilter, all filters see all the files, but it passes only files that have passed all filters). An example of where the new behavior is required is a combination of LastModifiedFileListFilter and AcceptOnceFileListFilter, when we do not wish to accept the file until some amount of time has elapsed. With the CompositeFileListFilter, since the AcceptOnceFileListFilter sees all the files on the first pass, it does not pass it later when the other filter does. The CompositeFileListFilter approach is useful when a pattern filter is combined with a custom filter that looks for a secondary file to indicate that file transfer is complete. The pattern filter might only pass the primary file (such as something.txt) but the “done” filter needs to see whether (for example) something.done is present.

Say we have files a.txt, a.done, and b.txt.

The pattern filter passes only a.txt and b.txt, while the “done” filter sees all three files and passes only a.txt. The final result of the composite filter is that only a.txt is released.

With the ChainFileListFilter, if any filter in the chain returns an empty list, the remaining filters are not invoked.

Version 5.0 introduced an ExpressionFileListFilter to execute SpEL expression against a file as a context evaluation root object. For this purpose, all the XML components for file handling (local and remote), along with an existing filter attribute, have been supplied with the filter-expression option, as the following example shows:

<int-file:inbound-channel-adapter
        directory="${inputdir}"
        filter-expression="name matches '.text'"
        auto-startup="false"/>

Version 5.0.5 introduced the DiscardAwareFileListFilter implementations that have an interest in rejected files. For this purpose, such a filter implementation should be supplied with a callback through addDiscardCallback(Consumer<File>). In the framework, this functionality is used from the FileReadingMessageSource.WatchServiceDirectoryScanner, in combination with LastModifiedFileListFilter. Unlike the regular DirectoryScanner, the WatchService provides files for processing according to the events on the target file system. At the moment of polling an internal queue with those files, the LastModifiedFileListFilter may discard them because they are too young relative to its configured age. Therefore, we lose the file for future possible considerations. The discard callback hook lets us retain the file in the internal queue so that it is available to be checked against the age in subsequent polls. The CompositeFileListFilter also implements a DiscardAwareFileListFilter and populates a discard callback to all its DiscardAwareFileListFilter delegates.

Since CompositeFileListFilter matches the files against all delegates, the discardCallback may be called several times for the same file.

Starting with version 5.1, the FileReadingMessageSource doesn’t check a directory for existence and doesn’t create it until its start() is called (typically via wrapping SourcePollingChannelAdapter). Previously, there was no simple way to prevent an operation system permissions error when referencing the directory, for example from tests, or when permissions are applied later.

Message Headers

Starting with version 5.0, the FileReadingMessageSource (in addition to the payload as a polled File) populates the following headers to the outbound Message:

  • FileHeaders.FILENAME: The File.getName() of the file to send. Can be used for subsequent rename or copy logic.

  • FileHeaders.ORIGINAL_FILE: The File object itself. Typically, this header is populated automatically by framework components (such as splitters or transformers) when we lose the original File object. However, for consistency and convenience with any other custom use cases, this header can be useful to get access to the original file.

  • FileHeaders.RELATIVE_PATH: A new header introduced to represent the part of file path relative to the root directory for the scan. This header can be useful when the requirement is to restore a source directory hierarchy in the other places. For this purpose, the DefaultFileNameGenerator (see "`Generating File Names) can be configured to use this header.

Directory Scanning and Polling

The FileReadingMessageSource does not produce messages for files from the directory immediately. It uses an internal queue for 'eligible files' returned by the scanner. The scanEachPoll option is used to ensure that the internal queue is refreshed with the latest input directory content on each poll. By default (scanEachPoll = false), the FileReadingMessageSource empties its queue before scanning the directory again. This default behavior is particularly useful to reduce scans of large numbers of files in a directory. However, in cases where custom ordering is required, it is important to consider the effects of setting this flag to true. The order in which files are processed may not be as expected. By default, files in the queue are processed in their natural (path) order. New files added by a scan, even when the queue already has files, are inserted in the appropriate position to maintain that natural order. To customize the order, the FileReadingMessageSource can accept a Comparator<File> as a constructor argument. It is used by the internal (PriorityBlockingQueue) to reorder its content according to the business requirements. Therefore, to process files in a specific order, you should provide a comparator to the FileReadingMessageSource rather than ordering the list produced by a custom DirectoryScanner.

Version 5.0 introduced RecursiveDirectoryScanner to perform file tree visiting. The implementation is based on the Files.walk(Path start, int maxDepth, FileVisitOption…​ options) functionality. The root directory (DirectoryScanner.listFiles(File)) argument is excluded from the result. All other sub-directories inclusions and exclusions are based on the target FileListFilter implementation. For example, the SimplePatternFileListFilter filters out directories by default. See AbstractDirectoryAwareFileListFilter and its implementations for more information.

Starting with version 5.5, the FileInboundChannelAdapterSpec of the Java DSL has a convenient recursive(boolean) option to use a RecursiveDirectoryScanner in the target FileReadingMessageSource instead of the default one.

Namespace Support

The configuration for file reading can be simplified by using the file-specific namespace. To do so, use the following template:

<?xml version="1.0" encoding="UTF-8"?>
<beans xmlns="http://www.springframework.org/schema/beans"
  xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
  xmlns:int="http://www.springframework.org/schema/integration"
  xmlns:int-file="http://www.springframework.org/schema/integration/file"
  xsi:schemaLocation="http://www.springframework.org/schema/beans
    https://www.springframework.org/schema/beans/spring-beans.xsd
    http://www.springframework.org/schema/integration
    https://www.springframework.org/schema/integration/spring-integration.xsd
    http://www.springframework.org/schema/integration/file
    https://www.springframework.org/schema/integration/file/spring-integration-file.xsd">
</beans>

Within this namespace, you can reduce the FileReadingMessageSource and wrap it in an inbound Channel Adapter, as follows:

<int-file:inbound-channel-adapter id="filesIn1"
    directory="file:${input.directory}" prevent-duplicates="true" ignore-hidden="true"/>

<int-file:inbound-channel-adapter id="filesIn2"
    directory="file:${input.directory}"
    filter="customFilterBean" />

<int-file:inbound-channel-adapter id="filesIn3"
    directory="file:${input.directory}"
    filename-pattern="test*" />

<int-file:inbound-channel-adapter id="filesIn4"
    directory="file:${input.directory}"
    filename-regex="test[0-9]+\.txt" />

The first channel adapter example relies on the default FileListFilter implementations:

  • IgnoreHiddenFileListFilter (do not process hidden files)

  • AcceptOnceFileListFilter (prevent duplication)

Therefore, you can also leave off the prevent-duplicates and ignore-hidden attributes, as they are true by default.

Spring Integration 4.2 introduced the ignore-hidden attribute. In prior versions, hidden files were included.

The second channel adapter example uses a custom filter, the third uses the filename-pattern attribute to add an AntPathMatcher based filter, and the fourth uses the filename-regex attribute to add a regular expression pattern-based filter to the FileReadingMessageSource. The filename-pattern and filename-regex attributes are each mutually exclusive with the regular filter reference attribute. However, you can use the filter attribute to reference an instance of CompositeFileListFilter that combines any number of filters, including one or more pattern-based filters to fit your particular needs.

When multiple processes read from the same directory, you may want to lock files to prevent them from being picked up concurrently. To do so, you can use a FileLocker. There is a java.nio-based implementation available, but it is also possible to implement your own locking scheme. The nio locker can be injected as follows:

<int-file:inbound-channel-adapter id="filesIn"
    directory="file:${input.directory}" prevent-duplicates="true">
    <int-file:nio-locker/>
</int-file:inbound-channel-adapter>

You can configure a custom locker as follows:

<int-file:inbound-channel-adapter id="filesIn"
    directory="file:${input.directory}" prevent-duplicates="true">
    <int-file:locker ref="customLocker"/>
</int-file:inbound-channel-adapter>
When a file inbound adapter is configured with a locker, it takes responsibility for acquiring a lock before the file is allowed to be received. It does not assume the responsibility to unlock the file. If you have processed the file and keep the locks hanging around, you have a memory leak. If this is a problem, you should call FileLocker.unlock(File file) yourself at the appropriate time.

When filtering and locking files is not enough, you might need to control the way files are listed entirely. To implement this type of requirement, you can use an implementation of DirectoryScanner. This scanner lets you determine exactly what files are listed in each poll. This is also the interface that Spring Integration uses internally to wire FileListFilter instances and FileLocker to the FileReadingMessageSource. You can inject a custom DirectoryScanner into the <int-file:inbound-channel-adapter/> on the scanner attribute, as the following example shows:

<int-file:inbound-channel-adapter id="filesIn" directory="file:${input.directory}"
     scanner="customDirectoryScanner"/>

Doing so gives you full freedom to choose the ordering, listing, and locking strategies.

It is also important to understand that filters (including patterns, regex, prevent-duplicates, and others) and locker instances are actually used by the scanner. Any of these attributes set on the adapter are subsequently injected into the internal scanner. For the case of an external scanner, all filter and locker attributes are prohibited on the FileReadingMessageSource. They must be specified (if required) on that custom DirectoryScanner. In other words, if you inject a scanner into the FileReadingMessageSource, you should supply filter and locker on that scanner, not on the FileReadingMessageSource.

By default, the DefaultDirectoryScanner uses an IgnoreHiddenFileListFilter and an AcceptOnceFileListFilter. To prevent their use, you can configure your own filter (such as AcceptAllFileListFilter) or even set it to null.

WatchServiceDirectoryScanner

The FileReadingMessageSource.WatchServiceDirectoryScanner relies on file-system events when new files are added to the directory. During initialization, the directory is registered to generate events. The initial file list is also built during initialization. While walking the directory tree, any subdirectories encountered are also registered to generate events. On the first poll, the initial file list from walking the directory is returned. On subsequent polls, files from new creation events are returned. If a new subdirectory is added, its creation event is used to walk the new subtree to find existing files and register any new subdirectories found.

There is an issue with WatchKey when its internal events queue is not drained by the program as quickly as the directory modification events occur. If the queue size is exceeded, a StandardWatchEventKinds.OVERFLOW is emitted to indicate that some file system events may be lost. In this case, the root directory is re-scanned completely. To avoid duplicates, consider using an appropriate FileListFilter (such as the AcceptOnceFileListFilter) or removing files when processing is complete.

The WatchServiceDirectoryScanner can be enabled through the FileReadingMessageSource.use-watch-service option, which is mutually exclusive with the scanner option. An internal FileReadingMessageSource.WatchServiceDirectoryScanner instance is populated for the provided directory.

In addition, now the WatchService polling logic can track the StandardWatchEventKinds.ENTRY_MODIFY and StandardWatchEventKinds.ENTRY_DELETE.

If you need to track the modification of existing files as well as new files, you should implement the ENTRY_MODIFY events logic in the FileListFilter. Otherwise, the files from those events are treated the same way.

The ResettableFileListFilter implementations pick up the ENTRY_DELETE events. Consequently, their files are provided for the remove() operation. When this event is enabled, filters such as the AcceptOnceFileListFilter have the file removed. As a result, if a file with the same name appears, it passes the filter and is sent as a message.

For this purpose, the watch-events property (FileReadingMessageSource.setWatchEvents(WatchEventType…​ watchEvents)) has been introduced. (WatchEventType is a public inner enumeration in FileReadingMessageSource.) With such an option, we can use one downstream flow logic for new files and use some other logic for modified files. The following example shows how to configure different logic for create and modify events in the same directory:

It is worth mentioning that the ENTRY_DELETE event is involved in the rename operation of sub-directory of the watched directory. More specifically, ENTRY_DELETE event, which is related to the previous directory name, precedes ENTRY_CREATE event which notifies about the new (renamed) directory. On some operating systems (like Windows), the ENTRY_DELETE event has to be registered to deal with that situation. Otherwise, renaming watched sub-directory in the File Explorer could result in the new files not being detected in that sub-directory.

<int-file:inbound-channel-adapter id="newFiles"
     directory="${input.directory}"
     use-watch-service="true"/>

<int-file:inbound-channel-adapter id="modifiedFiles"
     directory="${input.directory}"
     use-watch-service="true"
     filter="acceptAllFilter"
     watch-events="MODIFY"/> <!-- The default is CREATE. -->

Starting with version 6.1, the FileReadingMessageSource exposes two new WatchService-related options:

  • watchMaxDepth - an argument for the Files.walkFileTree(Path root, Set attributes, int maxDepth, FileVisitor visitor) API;

  • watchDirPredicate - a Predicate<Path> to test if a directory in the scanned tree should be walked and registered with the WatchService and the configured watch event kinds.

Limiting Memory Consumption

You can use a HeadDirectoryScanner to limit the number of files retained in memory. This can be useful when scanning large directories. With XML configuration, this is enabled by setting the queue-size property on the inbound channel adapter.

Prior to version 4.2, this setting was incompatible with the use of any other filters. Any other filters (including prevent-duplicates="true") overwrote the filter used to limit the size.

The use of a HeadDirectoryScanner is incompatible with an AcceptOnceFileListFilter. Since all filters are consulted during the poll decision, the AcceptOnceFileListFilter does not know that other filters might be temporarily filtering files. Even if files that were previously filtered by the HeadDirectoryScanner.HeadFilter are now available, the AcceptOnceFileListFilter filters them.

Generally, instead of using an AcceptOnceFileListFilter in this case, you should remove the processed files so that the previously filtered files are available on a future poll.

Configuring with Java Configuration

The following Spring Boot application shows an example of how to configure the outbound adapter with Java configuration:

@SpringBootApplication
public class FileReadingJavaApplication {

    public static void main(String[] args) {
        new SpringApplicationBuilder(FileReadingJavaApplication.class)
            .web(false)
            .run(args);
    }

    @Bean
    public MessageChannel fileInputChannel() {
        return new DirectChannel();
    }

    @Bean
    @InboundChannelAdapter(value = "fileInputChannel", poller = @Poller(fixedDelay = "1000"))
    public MessageSource<File> fileReadingMessageSource() {
         FileReadingMessageSource source = new FileReadingMessageSource();
         source.setDirectory(new File(INBOUND_PATH));
         source.setFilter(new SimplePatternFileListFilter("*.txt"));
         return source;
    }

    @Bean
    @Transformer(inputChannel = "fileInputChannel", outputChannel = "processFileChannel")
    public FileToStringTransformer fileToStringTransformer() {
        return new FileToStringTransformer();
    }

}

Configuring with the Java DSL

The following Spring Boot application shows an example of how to configure the outbound adapter with the Java DSL:

@SpringBootApplication
public class FileReadingJavaApplication {

    public static void main(String[] args) {
        new SpringApplicationBuilder(FileReadingJavaApplication.class)
            .web(false)
            .run(args);
    }

    @Bean
    public IntegrationFlow fileReadingFlow() {
         return IntegrationFlow
                  .from(Files.inboundAdapter(new File(INBOUND_PATH))
                              .patternFilter("*.txt"),
                          e -> e.poller(Pollers.fixedDelay(1000)))
                  .transform(Files.toStringTransformer())
                  .channel("processFileChannel")
                  .get();
    }

}

'tail’ing Files

Another popular use case is to get 'lines' from the end (or tail) of a file, capturing new lines when they are added. Two implementations are provided. The first, OSDelegatingFileTailingMessageProducer, uses the native tail command (on operating systems that have one). This is generally the most efficient implementation on those platforms. For operating systems that do not have a tail command, the second implementation, ApacheCommonsFileTailingMessageProducer, uses the Apache commons-io Tailer class.

In both cases, file system events, such as files being unavailable and other events, are published as ApplicationEvent instances by using the normal Spring event publishing mechanism. Examples of such events include the following:

[message=tail: cannot open '/tmp/somefile' for reading:
               No such file or directory, file=/tmp/somefile]

[message=tail: '/tmp/somefile' has become accessible, file=/tmp/somefile]

[message=tail: '/tmp/somefile' has become inaccessible:
               No such file or directory, file=/tmp/somefile]

[message=tail: '/tmp/somefile' has appeared;
               following end of new file, file=/tmp/somefile]

The sequence of events shown in the preceding example might occur, for example, when a file is rotated.

Starting with version 5.0, a FileTailingIdleEvent is emitted when there is no data in the file during idleEventInterval. The following example shows what such an event looks like:

[message=Idle timeout, file=/tmp/somefile] [idle time=5438]
Not all platforms that support a tail command provide these status messages.

Messages emitted from these endpoints have the following headers:

  • FileHeaders.ORIGINAL_FILE: The File object

  • FileHeaders.FILENAME: The file name (File.getName())

In versions prior to version 5.0, the FileHeaders.FILENAME header contained a string representation of the file’s absolute path. You can now obtain that string representation by calling getAbsolutePath() on the original file header.

The following example creates a native adapter with the default options ('-F -n 0', meaning to follow the file name from the current end).

<int-file:tail-inbound-channel-adapter id="native"
	channel="input"
	task-executor="exec"
	file="/tmp/foo"/>

The following example creates a native adapter with '-F -n +0' options (meaning follow the file name, emitting all existing lines).

<int-file:tail-inbound-channel-adapter id="native"
	channel="input"
	native-options="-F -n +0"
	task-executor="exec"
	file-delay=10000
	file="/tmp/foo"/>

If the tail command fails (on some platforms, a missing file causes the tail to fail, even with -F specified), the command is retried every 10 seconds.

By default, native adapters capture from standard output and send the content as messages. They also capture from standard error to raise events. Starting with version 4.3.6, you can discard the standard error events by setting the enable-status-reader to false, as the following example shows:

<int-file:tail-inbound-channel-adapter id="native"
	channel="input"
	enable-status-reader="false"
	task-executor="exec"
	file="/tmp/foo"/>

In the following example, IdleEventInterval is set to 5000, meaning that, if no lines are written for five seconds, FileTailingIdleEvent is triggered every five seconds:

<int-file:tail-inbound-channel-adapter id="native"
	channel="input"
	idle-event-interval="5000"
	task-executor="exec"
	file="/tmp/somefile"/>

This can be useful when you need to stop the adapter.

The following example creates an Apache commons-io Tailer adapter that examines the file for new lines every two seconds and checks for existence of a missing file every ten seconds:

<int-file:tail-inbound-channel-adapter id="apache"
	channel="input"
	task-executor="exec"
	file="/tmp/bar"
	delay="2000"
	end="false"             (1)
	reopen="true"           (2)
	file-delay="10000"/>
1 The file is tailed from the beginning (end="false") instead of the end (which is the default).
2 The file is reopened for each chunk (the default is to keep the file open).
Specifying the delay, end or reopen attributes forces the use of the Apache commons-io adapter and makes the native-options attribute unavailable.

Dealing With Incomplete Data

A common problem in file-transfer scenarios is how to determine that the transfer is complete so that you do not start reading an incomplete file. A common technique to solve this problem is to write the file with a temporary name and then atomically rename it to the final name. This technique, together with a filter that masks the temporary file from being picked up by the consumer, provides a robust solution. This technique is used by Spring Integration components that write files (locally or remotely). By default, they append .writing to the file name and remove it when the transfer is complete.

Another common technique is to write a second “marker” file to indicate that the file transfer is complete. In this scenario, you should not consider somefile.txt (for example) to be available for use until somefile.txt.complete is also present. Spring Integration version 5.0 introduced new filters to support this mechanism. Implementations are provided for the file system (FileSystemMarkerFilePresentFileListFilter), FTP and SFTP. They are configurable such that the marker file can have any name, although it is usually related to the file being transferred. See the Javadoc for more information.

Writing files

To write messages to the file system, you can use a FileWritingMessageHandler. This class can deal with the following payload types:

  • File

  • String

  • byte array

  • InputStream (since version 4.2)

For a String payload, you can configure the encoding and the charset.

To make things easier, you can configure the FileWritingMessageHandler as part of an outbound channel adapter or outbound gateway by using the XML namespace.

Starting with version 4.3, you can specify the buffer size to use when writing files.

Starting with version 5.1, you can provide a BiConsumer<File, Message<?>> newFileCallback which is triggered if you use FileExistsMode.APPEND or FileExistsMode.APPEND_NO_FLUSH and a new file has to be created. This callback receives a newly created file and the message which triggered it. This callback could be used to write a CSV header defined in the message header, for an example.

Generating File Names

In its simplest form, the FileWritingMessageHandler requires only a destination directory for writing the files. The name of the file to be written is determined by the handler’s FileNameGenerator. The default implementation looks for a message header whose key matches the constant defined as FileHeaders.FILENAME.

Alternatively, you can specify an expression to be evaluated against the message to generate a file name — for example, headers['myCustomHeader'] + '.something'. The expression must evaluate to a String. For convenience, the DefaultFileNameGenerator also provides the setHeaderName method, letting you explicitly specify the message header whose value is to be used as the filename.

Once set up, the DefaultFileNameGenerator employs the following resolution steps to determine the filename for a given message payload:

  1. Evaluate the expression against the message and, if the result is a non-empty String, use it as the filename.

  2. Otherwise, if the payload is a java.io.File, use the File object’s filename.

  3. Otherwise, use the message ID appended with .msg as the filename.

When you use the XML namespace support, both the file outbound channel adapter and the file outbound gateway support the following mutually exclusive configuration attributes:

  • filename-generator (a reference to a FileNameGenerator implementation)

  • filename-generator-expression (an expression that evaluates to a String)

While writing files, a temporary file suffix is used (its default is .writing). It is appended to the filename while the file is being written. To customize the suffix, you can set the temporary-file-suffix attribute on both the file outbound channel adapter and the file outbound gateway.

When using the APPEND file mode, the temporary-file-suffix attribute is ignored, since the data is appended to the file directly.

Starting with ,version 4.2.5, the generated file name (as a result of filename-generator or filename-generator-expression evaluation) can represent a child path together with the target file name. It is used as a second constructor argument for File(File parent, String child) as before. However, in the past we did not create (mkdirs()) directories for the child path, assuming only the file name. This approach is useful for cases when we need to restore the file system tree to match the source directory — for example, when unzipping the archive and saving all the files in the target directory in the original order.

Specifying the Output Directory

Both, the file outbound channel adapter and the file outbound gateway provide two mutually exclusive configuration attributes for specifying the output directory:

  • directory

  • directory-expression

Spring Integration 2.2 introduced the directory-expression attribute.
Using the directory Attribute

When you use the directory attribute, the output directory is set to a fixed value, which is set when the FileWritingMessageHandler is initialized. If you do not specify this attribute, you must use the directory-expression attribute.

Using the directory-expression Attribute

If you want to have full SpEL support, you can use the directory-expression attribute. This attribute accepts a SpEL expression that is evaluated for each message being processed. Thus, you have full access to a message’s payload and its headers when you dynamically specify the output file directory.

The SpEL expression must resolve to either a String, java.io.File or org.springframework.core.io.Resource. (The latter is evaluated into a File anyway.) Furthermore, the resulting String or File must point to a directory. If you do not specify the directory-expression attribute, then you must set the directory attribute.

Using the auto-create-directory Attribute

By default, if the destination directory does not exist, the respective destination directory and any non-existing parent directories are automatically created. To prevent that behavior, you can set the auto-create-directory attribute to false. This attribute applies to both the directory and the directory-expression attributes.

When using the directory attribute and auto-create-directory is false, the following change was made starting with Spring Integration 2.2:

Instead of checking for the existence of the destination directory when the adapter is initialized, this check is now performed for each message being processed.

Furthermore, if auto-create-directory is true and the directory was deleted between the processing of messages, the directory is re-created for each message being processed.

Dealing with Existing Destination Files

When you write files and the destination file already exists, the default behavior is to overwrite that target file. You can change this behavior by setting the mode attribute on the relevant file outbound components. The following options exist:

  • REPLACE (Default)

  • REPLACE_IF_MODIFIED

  • APPEND

  • APPEND_NO_FLUSH

  • FAIL

  • IGNORE

Spring Integration 2.2 introduced the mode attribute and the APPEND, FAIL, and IGNORE options.
REPLACE

If the target file already exists, it is overwritten. If the mode attribute is not specified, this is the default behavior when writing files.

REPLACE_IF_MODIFIED

If the target file already exists, it is overwritten only if the last modified timestamp differs from that of the source file. For File payloads, the payload lastModified time is compared to the existing file. For other payloads, the FileHeaders.SET_MODIFIED (file_setModified) header is compared to the existing file. If the header is missing or has a value that is not a Number, the file is always replaced.

APPEND

This mode lets you append message content to the existing file instead of creating a new file each time. Note that this attribute is mutually exclusive with the temporary-file-suffix attribute because, when it appends content to the existing file, the adapter no longer uses a temporary file. The file is closed after each message.

APPEND_NO_FLUSH

This option has the same semantics as APPEND, but the data is not flushed and the file is not closed after each message. This can provide a significant performance at the risk of data loss in the event of a failure. See Flushing Files When Using APPEND_NO_FLUSH for more information.

FAIL

If the target file exists, a MessageHandlingException is thrown.

IGNORE

If the target file exists, the message payload is silently ignored.

When using a temporary file suffix (the default is .writing), the IGNORE option applies if either the final file name or the temporary file name exists.

Flushing Files When Using APPEND_NO_FLUSH

The APPEND_NO_FLUSH mode was added in version 4.3. Using it can improve performance because the file is not closed after each message. However, this can cause data loss in the event of a failure.

Spring Integration provides several flushing strategies to mitigate this data loss:

  • Use flushInterval. If a file is not written to for this period of time, it is automatically flushed. This is approximate and may be up to 1.33x this time (with an average of 1.167x).

  • Send a message containing a regular expression to the message handler’s trigger method. Files with absolute path names matching the pattern are flushed.

  • Provide the handler with a custom MessageFlushPredicate implementation to modify the action taken when a message is sent to the trigger method.

  • Invoke one of the handler’s flushIfNeeded methods by passing in a custom FileWritingMessageHandler.FlushPredicate or FileWritingMessageHandler.MessageFlushPredicate implementation.

The predicates are called for each open file. See the Javadoc for these interfaces for more information. Note that, since version 5.0, the predicate methods provide another parameter: the time that the current file was first written to if new or previously closed.

When using flushInterval, the interval starts at the last write. The file is flushed only if it is idle for the interval. Starting with version 4.3.7, an additional property (flushWhenIdle) can be set to false, meaning that the interval starts with the first write to a previously flushed (or new) file.

File Timestamps

By default, the destination file’s lastModified timestamp is the time when the file was created (except that an in-place rename retains the current timestamp). Starting with version 4.3, you can now configure preserve-timestamp (or setPreserveTimestamp(true) when using Java configuration). For File payloads, this transfers the timestamp from the inbound file to the outbound (regardless of whether a copy was required). For other payloads, if the FileHeaders.SET_MODIFIED header (file_setModified) is present, it is used to set the destination file’s lastModified timestamp, as long as the header is a Number.

File Permissions

Starting with version 5.0, when writing files to a file system that supports Posix permissions, you can specify those permissions on the outbound channel adapter or gateway. The property is an integer and is usually supplied in the familiar octal format — for example, 0640, meaning that the owner has read/write permissions, the group has read-only permission, and others have no access.

File Outbound Channel Adapter

The following example configures a file outbound channel adapter:

<int-file:outbound-channel-adapter id="filesOut" directory="${input.directory.property}"/>

The namespace-based configuration also supports a delete-source-files attribute. If set to true, it triggers the deletion of the original source files after writing to a destination. The default value for that flag is false. The following example shows how to set it to true:

<int-file:outbound-channel-adapter id="filesOut"
    directory="${output.directory}"
    delete-source-files="true"/>
The delete-source-files attribute has an effect only if the inbound message has a File payload or if the FileHeaders.ORIGINAL_FILE header value contains either the source File instance or a String representing the original file path.

Starting with version 4.2, the FileWritingMessageHandler supports an append-new-line option. If set to true, a new line is appended to the file after a message is written. The default attribute value is false. The following example shows how to use the append-new-line option:

<int-file:outbound-channel-adapter id="newlineAdapter"
	append-new-line="true"
    directory="${output.directory}"/>

Outbound Gateway

In cases where you want to continue processing messages based on the written file, you can use the outbound-gateway instead. It plays a role similar to that of the outbound-channel-adapter. However, after writing the file, it also sends it to the reply channel as the payload of a message.

The following example configures an outbound gateway:

<int-file:outbound-gateway id="mover" request-channel="moveInput"
    reply-channel="output"
    directory="${output.directory}"
    mode="REPLACE" delete-source-files="true"/>

As mentioned earlier, you can also specify the mode attribute, which defines the behavior of how to deal with situations where the destination file already exists. See Dealing with Existing Destination Files for further details. Generally, when using the file outbound gateway, the result file is returned as the message payload on the reply channel.

This also applies when specifying the IGNORE mode. In that case the pre-existing destination file is returned. If the payload of the request message was a file, you still have access to that original file through the message header. See FileHeaders.ORIGINAL_FILE.

The 'outbound-gateway' works well in cases where you want to first move a file and then send it through a processing pipeline. In such cases, you may connect the file namespace’s inbound-channel-adapter element to the outbound-gateway and then connect that gateway’s reply-channel to the beginning of the pipeline.

If you have more elaborate requirements or need to support additional payload types as input to be converted to file content, you can extend the FileWritingMessageHandler, but a much better option is to rely on a Transformer.

Configuring with Java Configuration

The following Spring Boot application shows an example of how to configure the inbound adapter with Java configuration:

@SpringBootApplication
@IntegrationComponentScan
public class FileWritingJavaApplication {

    public static void main(String[] args) {
        ConfigurableApplicationContext context =
                      new SpringApplicationBuilder(FileWritingJavaApplication.class)
                              .web(false)
                              .run(args);
             MyGateway gateway = context.getBean(MyGateway.class);
             gateway.writeToFile("foo.txt", new File(tmpDir.getRoot(), "fileWritingFlow"), "foo");
    }

    @Bean
    @ServiceActivator(inputChannel = "writeToFileChannel")
    public MessageHandler fileWritingMessageHandler() {
         Expression directoryExpression = new SpelExpressionParser().parseExpression("headers.directory");
         FileWritingMessageHandler handler = new FileWritingMessageHandler(directoryExpression);
         handler.setFileExistsMode(FileExistsMode.APPEND);
         return handler;
    }

    @MessagingGateway(defaultRequestChannel = "writeToFileChannel")
    public interface MyGateway {

        void writeToFile(@Header(FileHeaders.FILENAME) String fileName,
                       @Header(FileHeaders.FILENAME) File directory, String data);

    }
}

Configuring with the Java DSL

The following Spring Boot application shows an example of how to configure the inbound adapter with the Java DSL:

@SpringBootApplication
public class FileWritingJavaApplication {

    public static void main(String[] args) {
        ConfigurableApplicationContext context =
                 new SpringApplicationBuilder(FileWritingJavaApplication.class)
                         .web(false)
                         .run(args);
        MessageChannel fileWritingInput = context.getBean("fileWritingInput", MessageChannel.class);
        fileWritingInput.send(new GenericMessage<>("foo"));
    }

    @Bean
   	public IntegrationFlow fileWritingFlow() {
   	    return IntegrationFlow.from("fileWritingInput")
   		        .enrichHeaders(h -> h.header(FileHeaders.FILENAME, "foo.txt")
   		                  .header("directory", new File(tmpDir.getRoot(), "fileWritingFlow")))
   	            .handle(Files.outboundGateway(m -> m.getHeaders().get("directory")))
   	            .channel(MessageChannels.queue("fileWritingResultChannel"))
   	            .get();
    }

}

File Transformers

To transform data read from the file system to objects and the other way around, you need to do some work. Unlike FileReadingMessageSource and to a lesser extent FileWritingMessageHandler, you probably need your own mechanism to get the job done. For this, you can implement the Transformer interface. Alternatively, you can extend the AbstractFilePayloadTransformer for inbound messages. Spring Integration provides some obvious implementations.

See the Javadoc for the Transformer interface to see which Spring Integration classes implement it. Similarly, you can check the Javadoc for the AbstractFilePayloadTransformer class to see which Spring Integration classes extend it.

FileToByteArrayTransformer extends AbstractFilePayloadTransformer and transforms a File object into a byte[] by using Spring’s FileCopyUtils. It is often better to use a sequence of transformers than to put all transformations in a single class. In that case the File to byte[] conversion might be a logical first step.

FileToStringTransformer extends AbstractFilePayloadTransformer convert a File object to a String. If nothing else, this can be useful for debugging (consider using it with a wire tap).

To configure file-specific transformers, you can use the appropriate elements from the file namespace, as the following example shows:

<int-file:file-to-bytes-transformer  input-channel="input" output-channel="output"
    delete-files="true"/>

<int-file:file-to-string-transformer input-channel="input" output-channel="output"
    delete-files="true" charset="UTF-8"/>

The delete-files option signals to the transformer that it should delete the inbound file after the transformation is complete. This is in no way a replacement for using an AcceptOnceFileListFilter when the FileReadingMessageSource is being used in a multi-threaded environment (such as when you use Spring Integration in general).

File Splitter

The FileSplitter was added in version 4.1.2, and its namespace support was added in version 4.2. The FileSplitter splits text files into individual lines, based on BufferedReader.readLine(). By default, the splitter uses an Iterator to emit lines one at a time as they are read from the file. Setting the iterator property to false causes it to read all the lines into memory before emitting them as messages. One use case for this might be if you want to detect I/O errors on the file before sending any messages containing lines. However, it is only practical for relatively short files.

Inbound payloads can be File, String (a File path), InputStream, or Reader. Other payload types are emitted unchanged.

The following listing shows possible ways to configure a FileSplitter:

Java DSL
@SpringBootApplication
public class FileSplitterApplication {

    public static void main(String[] args) {
        new SpringApplicationBuilder(FileSplitterApplication.class)
            .web(false)
            .run(args);
    }

    @Bean
    public IntegrationFlow fileSplitterFlow() {
        return IntegrationFlow
            .from(Files.inboundAdapter(tmpDir.getRoot())
                 .filter(new ChainFileListFilter<File>()
                        .addFilter(new AcceptOnceFileListFilter<>())
                        .addFilter(new ExpressionFileListFilter<>(
                             new FunctionExpression<File>(f -> "foo.tmp".equals(f.getName()))))))
            .split(Files.splitter()
                     .markers()
                     .charset(StandardCharsets.US_ASCII)
                     .firstLineAsHeader("fileHeader")
                     .applySequence(true))
            .channel(c -> c.queue("fileSplittingResultChannel"))
            .get();
    }

}
Kotlin DSL
@Bean
fun fileSplitterFlow() =
    integrationFlow(
        Files.inboundAdapter(tmpDir.getRoot())
            .filter(
                ChainFileListFilter<File?>()
                    .addFilter(AcceptOnceFileListFilter())
                    .addFilter(ExpressionFileListFilter(FunctionExpression { f: File? -> "foo.tmp" == f!!.name }))
            )
    ) {
        split(
            Files.splitter()
                .markers()
                .charset(StandardCharsets.US_ASCII)
                .firstLineAsHeader("fileHeader")
                .applySequence(true)
        )
        channel { queue("fileSplittingResultChannel") }
    }
Java
@Splitter(inputChannel="toSplitter")
@Bean
public MessageHandler fileSplitter() {
    FileSplitter splitter = new FileSplitter(true, true);
    splitter.setApplySequence(true);
    splitter.setOutputChannel(outputChannel);
    return splitter;
}
XML
<int-file:splitter id="splitter" (1)
    iterator=""                  (2)
    markers=""                   (3)
    markers-json=""              (4)
    apply-sequence=""            (5)
    requires-reply=""            (6)
    charset=""                   (7)
    first-line-as-header=""      (8)
    input-channel=""             (9)
    output-channel=""            (10)
    send-timeout=""              (11)
    auto-startup=""              (12)
    order=""                     (13)
    phase="" />                  (14)
1 The bean name of the splitter.
2 Set to true (the default) to use an iterator or false to load the file into memory before sending lines.
3 Set to true to emit start-of-file and end-of-file marker messages before and after the file data. Markers are messages with FileSplitter.FileMarker payloads (with START and END values in the mark property). You might use markers when sequentially processing files in a downstream flow where some lines are filtered. They enable the downstream processing to know when a file has been completely processed. In addition, a file_marker header that contains START or END is added to these messages. The END marker includes a line count. If the file is empty, only START and END markers are emitted with 0 as the lineCount. The default is false. When true, apply-sequence is false by default. See also markers-json (the next attribute).
4 When markers is true, set this to true to have the FileMarker objects be converted to a JSON string. (Uses a SimpleJsonSerializer underneath).
5 Set to false to disable the inclusion of sequenceSize and sequenceNumber headers in messages. The default is true, unless markers is true. When true and markers is true, the markers are included in the sequencing. When true and iterator is true, the sequenceSize header is set to 0, because the size is unknown.
6 Set to true to cause a RequiresReplyException to be thrown if there are no lines in the file. The default is false.
7 Set the charset name to be used when reading text data into String payloads. The default is the platform charset.
8 The header name for the first line to be carried as a header in the messages emitted for the remaining lines. Since version 5.0.
9 Set the input channel used to send messages to the splitter.
10 Set the output channel to which messages are sent.
11 Set the send timeout. Only applies if the output-channel can block — such as a full QueueChannel.
12 Set to false to disable automatically starting the splitter when the context is refreshed. The default is true.
13 Set the order of this endpoint if the input-channel is a <publish-subscribe-channel/>.
14 Set the startup phase for the splitter (used when auto-startup is true).

The FileSplitter also splits any text-based InputStream into lines. Starting with version 4.3, when used in conjunction with an FTP or SFTP streaming inbound channel adapter or an FTP or SFTP outbound gateway that uses the stream option to retrieve a file, the splitter automatically closes the session that supports the stream when the file is completely consumed See FTP Streaming Inbound Channel Adapter and SFTP Streaming Inbound Channel Adapter as well as FTP Outbound Gateway and SFTP Outbound Gateway for more information about these facilities.

When using Java configuration, an additional constructor is available, as the following example shows:

public FileSplitter(boolean iterator, boolean markers, boolean markersJson)

When markersJson is true, the markers are represented as a JSON string (using a SimpleJsonSerializer).

Version 5.0 introduced the firstLineAsHeader option to specify that the first line of content is a header (such as column names in a CSV file). The argument passed to this property is the header name under which the first line is carried as a header in the messages emitted for the remaining lines. This line is not included in the sequence header (if applySequence is true) nor in the lineCount associated with FileMarker.END. NOTE: Starting with version 5.5, the lineCount` is also included as a FileHeaders.LINE_COUNT into headers of the FileMarker.END message, since the FileMarker could be serialized into JSON. If a file contains only the header line, the file is treated as empty and, therefore, only FileMarker instances are emitted during splitting (if markers are enabled — otherwise, no messages are emitted). By default (if no header name is set), the first line is considered to be data and becomes the payload of the first emitted message.

If you need more complex logic about header extraction from the file content (not first line, not the whole content of the line, not one particular header, and so on), consider using header enricher ahead of the FileSplitter. Note that the lines that have been moved to the headers might be filtered downstream from the normal content process.

Idempotent Downstream Processing a Split File

When apply-sequence is true, the splitter adds the line number in the SEQUENCE_NUMBER header (when markers is true, the markers are counted as lines). The line number can be used with an Idempotent Receiver to avoid reprocessing lines after a restart.

For example:

@Bean
public ConcurrentMetadataStore store() {
    return new ZookeeperMetadataStore();
}

@Bean
public MetadataStoreSelector selector() {
    return new MetadataStoreSelector(
            message -> message.getHeaders().get(FileHeaders.ORIGINAL_FILE, File.class)
                    .getAbsolutePath(),
            message -> message.getHeaders().get(IntegrationMessageHeaderAccessor.SEQUENCE_NUMBER)
                    .toString(),
            store())
                    .compareValues(
                            (oldVal, newVal) -> Integer.parseInt(oldVal) < Integer.parseInt(newVal));
}

@Bean
public IdempotentReceiverInterceptor idempotentReceiverInterceptor() {
    return new IdempotentReceiverInterceptor(selector());
}

@Bean
public IntegrationFlow flow() {
    ...
    .split(new FileSplitter())
    ...
    .handle("lineHandler", e -> e.advice(idempotentReceiverInterceptor()))
    ...
}

File Aggregator

Starting with version 5.5, a FileAggregator is introduced to cover other side of FileSplitter use-case when START/END markers are enabled. For convenience the FileAggregator implements all three sequence details strategies:

  • The HeaderAttributeCorrelationStrategy with the FileHeaders.FILENAME attribute is used for correlation key calculation. When markers are enabled on the FileSplitter, it does not populate sequence details headers, since START/END marker messages are also included into the sequence size. The FileHeaders.FILENAME is still populated for each line emitted, including START/END marker messages.

  • The FileMarkerReleaseStrategy - checks for FileSplitter.FileMarker.Mark.END message in the group and then compare a FileHeaders.LINE_COUNT header value with the group size minus 2 - FileSplitter.FileMarker instances. It also implements a convenient GroupConditionProvider contact for conditionSupplier function to be used in the AbstractCorrelatingMessageHandler. See Message Group Condition for more information.

  • The FileAggregatingMessageGroupProcessor just removes FileSplitter.FileMarker messages from the group and collect the rest of messages into a list payload to produce.

The following listing shows possible ways to configure a FileAggregator:

Java DSL
@Bean
public IntegrationFlow fileSplitterAggregatorFlow(TaskExecutor taskExecutor) {
    return f -> f
            .split(Files.splitter()
                    .markers()
                    .firstLineAsHeader("firstLine"))
            .channel(c -> c.executor(taskExecutor))
            .filter(payload -> !(payload instanceof FileSplitter.FileMarker),
                    e -> e.discardChannel("aggregatorChannel"))
            .<String, String>transform(String::toUpperCase)
            .channel("aggregatorChannel")
            .aggregate(new FileAggregator())
            .channel(c -> c.queue("resultChannel"));
}
Kotlin DSL
@Bean
fun fileSplitterAggregatorFlow(taskExecutor: TaskExecutor?) =
    integrationFlow {
        split(Files.splitter().markers().firstLineAsHeader("firstLine"))
        channel { executor(taskExecutor) }
        filter<Any>({ it !is FileMarker }) { discardChannel("aggregatorChannel") }
        transform(String::toUpperCase)
        channel("aggregatorChannel")
        aggregate(FileAggregator())
        channel { queue("resultChannel") }
    }
Java
@serviceActivator(inputChannel="toAggregateFile")
@Bean
public AggregatorFactoryBean fileAggregator() {
    AggregatorFactoryBean aggregator = new AggregatorFactoryBean();
    aggregator.setProcessorBean(new FileAggregator());
    aggregator.setOutputChannel(outputChannel);
    return aggregator;
}
XML
<int:chain input-channel="input" output-channel="output">
    <int-file:splitter markers="true"/>
    <int:aggregator>
        <bean class="org.springframework.integration.file.aggregator.FileAggregator"/>
    </int:aggregator>
</int:chain>

If default behavior of the FileAggregator does not satisfy the target logic, it is recommended to configure an aggregator endpoint with individual strategies. See FileAggregator JavaDocs for more information.

Remote Persistent File List Filters

Inbound and streaming inbound remote file channel adapters (FTP, SFTP, and other technologies) are configured with corresponding implementations of AbstractPersistentFileListFilter by default, configured with an in-memory MetadataStore. To run in a cluster, these can be replaced with filters using a shared MetadataStore (see Metadata Store for more information). These filters are used to prevent fetching the same file multiple times (unless it’s modified time changes). Starting with version 5.2, a file is added to the filter immediately before the file is fetched (and reversed if the fetch fails).

In the event of a catastrophic failure (such as power loss), it is possible that the file currently being fetched will remain in the filter and won’t be re-fetched when restarting the application. In this case you would need to manually remove this file from the MetadataStore.

In previous versions, the files were filtered before any were fetched, meaning that several files could be in this state after a catastrophic failure.

In order to facilitate this new behavior, two new methods have been added to FileListFilter.

boolean accept(F file);

boolean supportsSingleFileFiltering();

If a filter returns true in supportsSingleFileFiltering, it must implement accept().

If a remote filter does not support single file filtering (such as the AbstractMarkerFilePresentFileListFilter), the adapters revert to the previous behavior.

If multiple filters are in used (using a CompositeFileListFilter or ChainFileListFilter), then all of the delegate filters must support single file filtering in order for the composite filter to support it.

The persistent file list filters now have a boolean property forRecursion. Setting this property to true, also sets alwaysAcceptDirectories, which means that the recursive operation on the outbound gateways (ls and mget) will now always traverse the full directory tree each time. This is to solve a problem where changes deep in the directory tree were not detected. In addition, forRecursion=true causes the full path to files to be used as the metadata store keys; this solves a problem where the filter did not work properly if a file with the same name appears multiple times in different directories. IMPORTANT: This means that existing keys in a persistent metadata store will not be found for files beneath the top level directory. For this reason, the property is false by default; this may change in a future release.