To aid in ingest or to provide services for external content, Fedora 4 has the ability to expose that content as if it were included in the repository. Federation may be useful for migrating content into Fedora 4 or serving large files already on disk.

Modeshape's federation overview provides more background on how federation works and the underlying concepts.

Note: The term "projection" is sometimes used as a synonym for the "federation" feature.

Limits of Federation

Federated content is accessible through the REST APi – however, some features will not work:

Filesystem federation is currently read-only (though we plan to enable writable filesystem federation soon once a few outstanding issues are resolved).
Access control can currently only be applied to an entire federated filesystem, not different access controls for different directories/files.
Transactions are not available.
Versioning is not available.
Querying using the internal search functionality (including calculating object and byte counts in the repository root) results in every file and directory being accessed each time the caching timeout expires (this is set with the cacheTtlSeconds parameter). For federations with large numbers of files, this can take a long time and result in "Too many open files" errors as files are opened faster than Java garbage-collects them.

Filesystem Federation

Filesystem federation maps a node in the repository to a directory on disk. This allows files on disk to be served and updated by Fedora 4 as though they were in the repository. Filesystem federation avoids having to transfer files using HTTP – and with larger file sizes (or with larger numbers of files being processed), this can improve performance significantly. If you are ingesting a large number of multi-gigabyte files, we recommend you consider filesystem federation.

Another use for filesystem federation is interoperability with another system. If you have files on disk managed by another application or workflow, you can use filesystem federation to serve them with Fedora 4 without having to ingest them using the REST API or create another copy of the files.

Configuration

Once you have run Fedora with a federated filesystem configuration, do not change the configuration name ("federated-directory" in the example below), or federation will not be loaded correctly.

An example filesystem federation configuration to include in your Modeshape repository.json :

"externalSources" : {
    "federated-directory" : {
        "classname" : "org.fcrepo.connector.file.FedoraFileSystemConnector",
        "directoryPath" : "/path/to/files",
        "propertiesDirectoryPath" : "/path/to/external/properties",
        "projections" : [ "default:/federated => /" ],
        "contentBasedSha1" : "false",
        "readonly" : true,
        "extraPropertiesStorage" : "none",
        "cacheTtlSeconds" : 5
    }
},

directoryPath - base directory for all files shared with the repository
propertiesDirectoryPath - (optional) a path, that if specified causes computed properties to be stored in an external file structure
projections - lists one or more mappings from the repository to the filesystem. The format is "{workspace}:{repository path} => {path relative to directoryPath}". See Multiple Directories below for how to handle multiple mappings.
contentBasedSha1 - controls how internal identifiers are computed for files. By default (contentBaseSha1 = true), Modeshape computes the SHA-1 checksum of a file's content every time the file is accessed. For small files this creates a modest overhead. For large files, however, this dramatically reduces performance, since generating the checksum can take several seconds per gigabyte of data. For this reason, we recommend setting contentBasedSha1 to false when serving files larger than 100MB
readonly - controls whether the contents of the filesbase directory for all files shared with the repository are read-only ( currently read-only is the only supported mode)
extraPropertiesStorage - sets the format for storing "extra" properties (properties that can't be set using filesystem attributes). Recommended values are "json" for the current JSON properties format, or "none" for disabling extra properties. This property is ignored if propertiesDirectoryPath is set, since an external properties store will be used. (A warning or notice will appear in the logs indicating that the asserted preference here is overridden.)
cacheTtlSeconds - the maximum time that cached entr ies are held before being refreshed. Setting to a low value will make changes to the filesystem (like adding new files) show up more quickly in the REST API. Setting to a higher value will improve performance for files that don't change often.

Modeshape's FileSystemConnector documentation and configuration provide additional information about configuring the filesystem connector.

Multiple Directories

If you want to map multiple directories, the first entry in the projections array should map the parent directory (i.e, the directory in directoryPath). Subsequent entries can map subdirectories to other repository paths. For example, if you have a directory /pub/ that contains two directories (/pub/project1/ and /pub/project2/) which you want to map the project1 and project2 directories to the top level of the repository:

"externalSources" : {
    "federated-1" : {
        "classname" : "org.fcrepo.connector.file.FedoraFileSystemConnector",
        "directoryPath" : "/pub",
        "projections" : [ "default:/pub => /", "default:/project1 => /project1", "default:/project2 => /project2" ],
        "contentBasedSha1" : "false",
        "readonly" : true,
        "extraPropertiesStorage" : "none",
        "cacheTtlSeconds" : 5
    }
},

This configuration would provide the following mappings:

Repository URL	Filesystem Path
http://localhost:8080/rest/pub/	/pub/
http://localhost:8080/rest/project1/	/pub/project1/
http://localhost:8080/rest/project2/	/pub/project2/

External Datastreams

One use case for filesystem federation is for external datastreams – storing objects in the repository, but linking to large files on disk instead of ingesting them as datastream content. To do this, copy the file to the federated filesystem directory, create the object in Fedora using the REST API, and then link from the object to the federated filesystem content using the fedorarelsext:hasExternalContent property (see PATCH ):

insert data { <> <http://fedora.info/definitions/v4/rels-ext#hasExternalContent> <http://localhost:8080/rest/federated/file/fcr:content> . }

Separate Properties Store

For a connector, even a read-only connector, some properties are computed and stored. These include a modification date and a checksum (if enabled) which is only recomputed when the file is altered. If it is not desirable or possible to store these properties on the same file system path that is being projected over you may specify an alternate location (propertiesDirectoryPath) in which those properties will be cached. This is especially important when using checksums in directories of large files.

Transactions

Basic testing has been completed with transactions over a read/write filesystem federation. Content will not be written to the filesystem federation until a transaction is complete. For transactions that are never completed (due to rollback or timeout) content is never written to the filesystem.

How-Tos

Other Connectors

In addition to the filesystem connector, Modeshape includes several other connectors, for federating content from Git, CMIS repositories, and relational databases.

Custom connectors can also be developed to support other system. The filesystem connector is a good reference implementation, particularly for file-based resources.

Page tree