The IBM FHIR Server implements the HL7 FHIR HTTP API and supports the full set of FHIR-defined resource types. This FHIR server is intended to be a common component for providing FHIR capabilities within health services and solutions.

where <username> is one of the users configured in server.xml (default is fhiruser).
Use single quotes around the URL to prevent $healthcheck from being evaluated as an environment variable on unix-based operating systems.

One should see an empty body in the response with a HTTP Response Code 200.

For more information about the capabilities of the implementation, see Conformance.

The IBM FHIR Server does not include an upgrade installer. To upgrade a server to the next version, install the new version to a separate location and copy the configuration files from your existing installation (reconciling any configuration-related changes from the new release in the process).

To manage database updates over time, the IBM FHIR Server uses custom tools from the fhir-database-utils project. Through the use of a metadata table, the database utilities can detect the currently installed version of the database schema and apply any new changes that are needed to bring the database to the current level.

Complete the following steps to upgrade the server:

The IBM FHIR Server includes a Docker image ibmcom/ibm-fhir-server.

Note, logging for the IBM FHIR Server docker image is to stderr and stdout, and is picked up by Logging agents.

The IBM FHIR Server is configured using Environment variables using:

This chapter contains information about the various ways in which the IBM FHIR Server can be configured by users.

The IBM FHIR Server is built on the open source OpenLiberty project but also works on the commercial IBM WebSphere Liberty. In documentation and elsewhere, we use Liberty to refer to either/or.

As a Liberty application, the IBM FHIR Server is configurable like any other Liberty server. See https://openliberty.io/docs/latest/reference/config/server-configuration-overview.html for more information.

By default, we include:

In the examples within the following sections, you’ll see the default password change-password. In order to secure your server, these values should be changed.

For example, the basic user registry in the default server.xml defines users fhiruser and fhiradmin and variables FHIR_USER_PASSWORD and FHIR_ADMIN_PASSWORD, each with a default value of change-password.

These variables can be changed by setting environment variables by the same name. Alternatively, users can change the server.xml or use Liberty configDropins to change the values.

Optionally, server.xml values can be encoded via the Liberty securityUtility command. For example, to encode a string value with the default {xor} encoding, run the following command:

The output of this command can then be copied and pasted into your server.xml or fhir-server-config.json file as needed. The fhir-server-config.json does not support the securityUtility’s {aes} encoding at this time, but per the limits to protection through password encryption^a, this encoding does not provide significant additional security beyond exclusive or (XOR) encoding.

The default server.xml includes variables for configuring the log output:

By default, INFO level messages will go to both the console and a log file at wlp/usr/servers/fhir-server/logs/messages.log. For the ibmcom/ibm-fhir-server docker image, the default is changed to go just to the console.

To enable tracing, set the TRACE_SPEC variable (e.g. by setting an environment variable by the same name) to a :-delimited list of component = level pair values. For example, to turn on tracing for the SQL generated by fhir-persistence-jdbc, set the TRACE_SPEC to com.ibm.fhir.persistence.jdbc.dao.impl.*=fine:com.ibm.fhir.database.utils.query.QueryUtil=fine.

By default, that trace output appears alongside the messages.log file in a file named trace.log, but the trace output can be sent to the console instead by setting the TRACE_FILE variable to stdout.

For more information on logging and tracing in Liberty, see https://openliberty.io/docs/latest/log-trace-configuration.html.

Configuration properties stored within a fhir-server-config.json file are structured in a hierarchical manner. Here is an example:

Throughout this document, we use a path notation to refer to property names. For example, the name of the defaultPrettyPrint property in the preceding example would be fhirServer/core/defaultPrettyPrint.

The IBM FHIR server supports certain multi-tenant features. One such feature is the ability to set certain configuration properties on a per-tenant basis.

In general, the configuration properties for a particular tenant are stored in the <WLP_HOME>/usr/servers/fhir-server/config/<tenant-id>/fhir-server-config.json file, where <tenant-id> refers to the tenant’s “short name” or tenant id.

The global configuration is considered to be associated with a tenant named default, so those properties are stored in the <WLP_HOME>/usr/servers/fhir-server/config/default/fhir-server-config.json file.

Similarly, tenant-specific search parameters are found at <WLP_HOME>/usr/servers/fhir-server/config/<tenant-id>/extension-search-parameters.json, whereas the global/default extension search parameters are at <WLP_HOME>/usr/servers/fhir-server/config/default/extension-search-parameters.json.

Search parameters are handled like a single configuration properly; providing a tenant-specific file will override the global/default extension search parameters as defined at FHIRSearchConfiguration.

More information about multi-tenant support can be found in Section 4.9 Multi-tenancy.

The IBM FHIR Server supports the ability to compute and store compartment membership values during ingestion. Once stored, these values can help accelerate compartment-related search queries. To use this feature, update the IBM FHIR Server to at least version 4.5.1 and run a reindex operation as described in the fhir-bucket project README. The reindex operation reprocesses the resources stored in the database, computing and storing the new compartment reference values. After the reindex operation has completed, add the useStoredCompartmentParam configuration element to the relevant tenant fhir-server-config.json file to allow the search queries to use the pre-computed values:

Note that this parameter only enables or disables the compartment search query optimization feature. The compartment membership values are always computed and stored during ingestion or reindexing, regardless of the setting of this value. After the reindex operation is complete, it is recommended to set useStoredCompartmentParam to true. No reindex is required if this value is subsequently set to false.

The IBM FHIR Server allows deployers to select a persistence layer implementation that fits their needs. Currently, the server includes a JDBC persistence layer which supports Apache Derby, IBM Db2, and PostgreSQL. However, Apache Derby is not recommended for production usage.

Before you can configure the server to use the JDBC persistence layer implementation, you first need to prepare the database. This step depends on the database product in use and is covered in more detail in Section 3.3.1.1 Supported databases.

The IBM FHIR Server is delivered with a default configuration that is already configured to use the JDBC persistence layer implementation with an Embedded Derby database. This provides the easiest out-of-the-box experience since it requires very little setup.

The IBM FHIR Server persistence configuration is split between fhir-server-config.json and the Liberty server.xml and configDropins.

Within fhir-server-config.json, the value of the fhirServer/persistence/factoryClassname is used to instantiate a FHIRPersistence object. By default, the server is configured to use the FHIRPersistenceJDBCFactory:

When the FHIRPersistenceJDBCFactory is in use, the fhirServer/persistence/datasources property must specify a mapping from datastore-id values to Liberty datasource definitions. For example, here is the configuration for a datastore with id default that is configured for the jdbc/fhir_default_default datasource of type postgresql:

Both the type and the currentSchema properties are required. The type value must be set to one of the supported JDBC types (currently db2, postgresql, or derby) and it must match the actual database type referenced by the datasource definition. If the type does not match, the behavior is undefined. Because the IBM FHIR Server does not quote the schema name in our generated queries, the currentSchema property is effectively case-insensitive.

The jndiName property is optional; it will default to jdbc/fhir_<tenantId>_<dsId> where <tenantId> and <dsId> represent the tenant and datastore ids respectively⁴.

The datasource definitions themselves are configured in accordance with the Liberty documentation on Relational database connections with JDBC. Previous versions of the IBM FHIR Server supported a proxy datasource that allowed for datasource definitions in the fhir-server-config.json, but now Liberty JDBC datasource configuration is used due to benefits related to configuring pool sizes, transaction recovery, and monitoring.

For example, the fhir-server-config snippet from above would have a corresponding Liberty config like this:

Datasource elements should not be defined in the main server.xml file but instead should be defined in the configDropins/overrides directory. See the Liberty Profile Server configuration guide for more details and general guidance on creating modular configurations.

The IBM FHIR Server is packaged with the following sample datasource definitions at configDropins/disabled:

There are 3 libraries defined in the main server.xml that reference the required client drivers for each database type:

When a datasource definition is included in a configDropin under configDropins/overrides, this file is picked up when the server starts as indicated by the following AUDIT message:

The IBM FHIR Server will look up the tenant and datastore id for each request and use the corresponding JNDI name to obtain a connection for the corresponding datasource.

If you are using the ibmcom/ibm-fhir-server docker image, you can ask the entrypoint script to create (bootstrap) the database and the schema during startup by setting the BOOTSTRAP_DB environment variable to true.

This database bootstrap step is only supported for Embedded Derby and will only bootstrap the default datastore of the default tenant (the default for requests with no tenant or datastore headers).

Reminder: the Embedded Derby support is designed to support simple getting started scenarios and is not recommended for production use.

If you configure the FHIR server to use an IBM Db2 database, you must:

An executable fhir-persistence-schema jar can be downloaded from the project’s Releases tab and documentation can be found at https://github.com/IBM/FHIR/tree/main/fhir-persistence-schema.

For a detailed guide on configuring IBM Db2 on Cloud for the IBM FHIR Server, see DB2OnCloudSetup.

Since release 4.3.2 you can use the search.reopt query optimizer hint to improve the performance of certain search queries involving multiple search parameters. This optimization is currently only available for Db2. Valid values are “ALWAYS” and “ONCE”. See Db2 documentation for REOPT for more details.

If you configure the FHIR server to use a PostgreSQL database, you must:

An executable fhir-persistence-schema jar can be downloaded from the project’s Releases tab and documentation can be found at https://github.com/IBM/FHIR/tree/main/fhir-persistence-schema.

Since release 4.5.5 you can set the searchOptimizerOptions/from_collapse_limit and searchOptimizerOptions/join_collapse_limit properties to improve the performance of certain search queries involving multiple search parameters. This optimization is currently only available for PostgreSQL.

To enable the IBM FHIR Server to work with other relational database systems, see https://ibm.github.io/FHIR/guides/BringYourOwnPersistence#adding-support-for-another-relational-database

To understand how the configuration properties are defined for one or more datastores, let’s start off with a couple of examples.

Here is a simple example of a single (default) Derby datastore.

config/default/fhir-server-config.json

configDropins/overrides/datasource-derby.xml

In this example, we define an embedded Derby database named derby/fhirDB (a location relative to the <WLP_HOME>/usr/servers/fhir-server directory). The datastore-id associated with this datastore is default, which is the value that is used if no X-FHIR-DSID request header is found in the incoming request. So, when only a single database is being used, it’s wise to leverage the default datastore-id value to allow REST API consumers to avoid having to set the X-FHIR-DSID request header on each request.

This example shows a slightly more complex scenario. In this scenario, the acme tenant would like to store data in one of two study-specific Db2 databases with datastore-id values study1 and study2.

Furthermore, the REST API consumers associated with Acme applications will be coded to always set the X-FHIR-TENANT-ID request header to be acme and the X-FHIR-DSID request header to the specific datastore-id associated with each request (either study1 or study2). In this case, the following properties would be configured:

config/acme/fhir-server-config.json

configDropins/overrides/datasources-acme.xml

Note that because this example is using the default FHIR server naming scheme for datasource JNDI names, there is no need to include the ‘jndiName’ property in the fhir-server-config.json, although you can specify it should you wish to make the mapping clear.

Additionally, note that each datasource definition in the configDropin gets its own connection pool with properties defined by the ‘connectionManager’ element.

The TransactionManager controls the timeout of database queries.

To modify the default transaction timeout value, set the environment variable FHIR_TRANSACTION_MANAGER_TIMEOUT or enter the value in the server.env file at the root of the WLP fhir-server instance. Example values are 120s (seconds) or 2m (minutes).

Normally, the update operation is invoked with a FHIR resource which represents a new version of an existing resource. The resource specified in the update operation would contain the same id of that existing resource. If a resource containing a non-existent id were specified in the update invocation, an error would result.

The FHIR specification defines optional behavior for the update operation where it can create a new resource if a non-existent resource is specified in the invocation. The FHIR server supports this optional behavior via the fhirServer/persistence/common/updateCreateEnabled configuration parameter. If this configuration parameter is set to true (the default), then the update operation will create a new resource if it is invoked with a resource containing a non-existent id. If the option is set to false, then the optional behavior is disabled and an error would be returned.

The following example shows the JSON for enabling update operations to create resources:

You can modify the default server implementation by taking advantage of the IBM FHIR server’s extensibility. The following extension points are available:

In addition to the standard REST API (create, update, search, and so forth), the IBM FHIR Server supports the FHIR operations framework as described in the FHIR specification.

The FHIR team provides implementations for the standard $validate, $document, $everything, $expand, $lookup, $subsumes, $closure, $export, $import, $convert, $apply and $translate operations, as well as a custom operation named $healthcheck, which queries the configured persistence layer to report its health.

The server also bundles $reindex to reindex instances of Resources so they are searchable, $retrieve-index to retrieve lists of resources available to be reindexed, and $erase to hard delete instances of Resources. To learn more about the $erase operation, read the design document.

To extend the server with additional operations, see Section 4.1.2 Custom operations

The $validate operation checks whether the attached content would be acceptable either generally, or as a create, update, or delete against an existing resource instance or type.

https://www.hl7.org/fhir/r4/resource-operations.html#validate

The $document operation generates a fully bundled document from a composition resource.

https://www.hl7.org/fhir/r4/composition-operations.html#document

The $healthcheck operation returns the health of the FHIR server and its datastore. In the default JDBC persistence layer, this operation creates a connection to the configured database and return its status. The operations returns 200 OK when healthy. Otherwise, it returns an HTTP error code and an OperationOutcome with one or more issues.

In addition to the provided operations, the FHIR server supports user-provided custom operations through a Java Service Provider Interface (SPI).

To contribute an operation:

After you register your operation with the server, it is available via HTTP POST at [base]/api/1/$<yourCode>, where <yourCode> is the value of your OperationDefinition’s code.

The FHIR server provides a notification service that publishes notifications about persistence events, specifically create, update, and delete operations. The notification service can be used by other Healthcare components to trigger specific actions that need to occur as resources are being updated in the FHIR server datastore.

The notification service supports two implementations: WebSocket and Kafka.

The FHIRNotificationEvent class defines the information that is published by the notification service. Each notification event published to the WebSocket or Kafka topic is an instance of FHIRNotificationEvent, serialized as a JSON object. This JSON object will have the following fields:

The following JSON is an example of a serialized notification event:

If the resource is over the limit specified in fhirServer/notifications/common/maxNotificationSizeBytes, the default value is to subset id, meta and resourceType and add the subset to the FHIRNotificationEvent. In alternative configurations, user may set fhirServer/notifications/common/maxNotificationSizeBehavior to omit and subsequently retrieve the resource using the location.

The WebSocket implementation of the notification service will publish notification event messages to a WebSocket. To enable WebSocket notifications, set the fhirServer/notifications/websocket/enabled property to true, as in the following example:

The WebSocket location URI is ws://<host>:<port>/fhir-server/api/v4/notification, where <host> and <port> represent the host and port of the FHIR server’s REST API endpoint. So for example, if the FHIR server endpoint’s base URL is https://localhost:9443/fhir-server/api/v4 then the corresponding location of the WebSocket would be ws://localhost:9443/fhir-server/api/v4/notification.

The Kafka implementation of the notification service will publish notification event messages to a Kafka topic. To configure the Kafka notification publisher, configure properties in the fhir-server-config.json file as indicated in the following example:

The fhirServer/notifications/kafka/enabled property is used to enable or disable the Kafka publisher component.

The fhirServer/notifications/kafka/topicName property is used to configure the appropriate topic name to which the notification events should be published.

The fhirServer/notifications/kafka/connectionProperties property group is used to configure the properties necessary to successfully connect to the Kafka server. You can specify an arbitrary group.id. The bootstrap.servers property is required, but the rest are optional, although if your Kafka server is configured to require an SSL connection and client authentication, then the remaining properties must also be set. For more details about Kafka-related properties, see the Kafka documentation.

In the connectionProperties property group in preceding example, you’ll notice that the password-related properties have encoded values. To store a value requiring security (such as a password), you can use Liberty’s securityUtility command to encode the value. See Section 3.1 Encoded passwords for details.

Before you enable Kafka notifications, it’s important to understand the topology of the environment in which the FHIR server instance will be running. Your topic name selection should be done in consideration of the topology. If you have multiple instances of the FHIR server clustered together to form a single logical endpoint, then each of those instances should be configured to use the same Kafka topic for notifications. This is so that notification consumers (subscribers) can subscribe to a single topic and receive all the notifications published by each of the FHIR server instances within the cluster.

On the other hand, if you have two completely independent FHIR server instances, then you should configure each one with its own topic name.

The FHIRNotificationEvent is asynchronous by default. If you want to specify a synchronous request, you can set fhirServer/notifications/kafka/sync to true, which ensures no message is lost in publishing, however it does add latency in each request.

The NATS implementation of the notification service publishes notification event messages to a NATS streaming cluster. To configure the NATS notification publisher, configure properties in the fhir-server-config.json file as shown in the following example:

Set the fhirServer/notifications/nats/enabled property to true and provide the name of your NATS cluster for the value of fhirServer/notifications/nats/cluster. You may leave fhirServer/notifications/nats/channel and fhirServer/notifications/nats/clientId as defined. Provide the URL for one or more NATS servers as the value for fhirServer/notifications/nats/servers.

To use TLS to connect to your NATS cluster, set fhirServer/notifications/nats/useTLS to true and provide client truststore and keystore locations and passwords as the remaining config values. Ensure that your NATS cluster is configured for TLS client connections.

To store a value requiring security, such as a password, use Liberty’s securityUtility command to encode the value. See Section 3.1 Encoded passwords for details.

By default, notification messages are published for all create and update persistence operations. However, the FHIR server allows you to configure a list of resource types for which notification events will be published. To do this, list the resource types for which you want to generate notifications in an array of strings within the fhirServer/notifications/common/includeResourceTypes property in the fhir-server-config.json file, as in the following example:

With the includeResourceTypesproperty set as in the preceding example, the FHIR server publishes notification events only for Patient and Observation resources. If you omit this property or set its value to [] (an empty array), then the FHIR server publishes notifications for all resource types.

The IBM FHIR Server supports a persistence interceptor feature that enables users to add their own logic to the REST API processing flow around persistence events. This can be used to enforce application-specific business rules associated with resources. Interceptor methods are called immediately before or after each persistence operation.

A persistence interceptor implementation must implement the com.ibm.fhir.persistence.interceptor.FHIRPersistenceInterceptor interface.

Each interceptor method receives a parameter of type FHIRPersistenceEvent, which contains context information related to the request being processed at the time that the interceptor method is invoked. It includes the FHIR resource, security information, request URI information, and the collection of HTTP headers associated with the request.

There are many use cases for persistence interceptors:

It is also possible to modify the incoming resources from the beforeCreate and beforeUpdate methods. For example, an interceptor could be used to add tags to resources on their way into the server. However, it is important to realize that interceptors are called after resource validation. Therefore, interceptor authors must be careful not to alter the resources in a way that breaks conformance with the profiles claimed in Resource.meta.profile or the secondary constraints in the specification. When in doubt, interceptors that modify the incoming resource can use the FHIRValidator to re-validate the resource(s) after they are altered.

To implement a persistence interceptor, complete the following steps:

The IBM FHIR Server includes a dynamic registry of conformance resources. The registry pre-packages all FHIR Definitions from the specification and uses a Java ServiceLoader to discover additional registry resource providers on the classpath.

The server consults this registry for:

One common technique for extending FHIR with a set of conformance resources is to build or reference an Implementation Guide.

The IBM FHIR Server includes a PackageRegistryResourceProvider for registering implementation guide resources.

Additionally, we pre-package a number of popular implementation guides and make those available from both our GitHub Releases and Maven Central.

Finally, the IBM FHIR Server includes a built-in ServerRegistryResourceProvider that can be used to bridge conformance resources from the tenant data store (uploaded through the REST API) to the registry. This provider can be enabled/disabled via the fhirServer/core/serverRegistryResourceProviderEnabled property, but we recommend leaving it disabled for performance-intensive workloads.

As mentioned in the previous section, the IBM FHIR Server registry is consulted during resource validation.

The FHIR specification provides a number of different validation resources including:

The FHIR server validates incoming resources for create and update interactions using the resource definitions and their corresponding FHIRPath constraints. Additionally, the FHIR server provides the following $validate operation that API consumers can use to POST resources and get validation feedback:

FHIR resources can be extended. Each extension has a url and servers can use these urls to validate the contents of the extension.

The IBM FHIR Server performs extension validation by looking up the extension definition in its internal registry. If an instance contains extensions that are unknown to the server, then the server will include a validation warning that indicates this extension could not be validated.

The IBM FHIR Server can be extended with custom profile validation. This allows one to apply validation rules on the basis of the Resource.meta.profile values included on the resource instance.

For example, you can configure a set of FHIRPath Constraints to run for resources that claim conformance to the http://ibm.com/fhir/profile/partner profile when you see an input resource like the following:

It is also possible to configure a set of profiles, one or more of which a resource must claim conformance to and be successfully validated against in order to be persisted to the FHIR server. The FHIR server supports this optional behavior via the fhirServer/resources/<resourceType>/profiles/atLeastOne configuration parameter. If this configuration parameter is set to a non-empty list of profiles, then the FHIR server will perform the following validation, returning FAILURE if not successful:

If a profile in the list specified by the configuration parameter contains a version, for example http://ibm.com/fhir/profile/partner|1.0, then a profile of the same name specified in the resource’s meta.profile element will only be considered a match if it contains exactly the same version. However, if a profile in the list specified by the configuration parameter does not contain a version, for example http://ibm.com/fhir/profile/partner, then a profile of the same name specified in the resource’s meta.profile element will be considered a match whether it contains a version or not.

If this configuration parameter is not set or is set to an empty list, then the FHIR server will perform its standard validation.

The IBM FHIR Server pre-packages all conformance resources from the core specification.

See Validation Guide - Optional profile support for a list of pre-built Implementation Guide resources and how to load them into the IBM FHIR server.

See Validation Guide - Making profiles available to the fhir registry for information about how to extend the server with additional Implementation Guide artifacts.

In addition to supporting tenant-specific search parameter extensions as described in Section 3.2.2 Tenant-specific configuration properties, the IBM FHIR Server also supports loading search parameters from the registry.

For performance reasons, the registry search parameters are retrieved once and only once during startup.

The set of search parameters can filtered / refined via fhirServer/resources/[resourceType]/searchParameters as described in the Search configuration guide.

In addition to the server, we also offer a Java API for invoking the FHIR REST APIs. The IBM FHIR Client API is based on the JAX-RS 2.0 standard and provides a simple properties-driven client that can be easily configured for a given endpoint, mutual authentication, request/response logging, and more.

To use the FHIR Client from your application, specify the fhir-client artifact as a dependency within your pom.xml file, as in the following example:

For examples on how to use the IBM FHIR Client, look for tests like com.ibm.fhir.client.test.mains.FHIRClientSample from the fhir-client project in git. Additionally, the FHIR Client is heavilly used from our integration tests in fhir-server-test.

Inter-dependencies between resources are typically defined by one resource containing a field of type Reference which contains an external reference⁵ to another resource. For example, an Observation resource could reference a Patient resource via the Observation’s subject field. The value that is stored in the Reference-type field (for example, subject in the case of the Observation resource) could be an absolute URL, such as https://fhirserver1:9443/fhir-server/api/v4/Patient/12345, or a relative URL (for example, Patient/12345).

In order to establish a reference to a resource, you must first know its resource identifier. However, if you are using a request bundle to create both the referenced resource (Patient in this example) and the resource which references it (Observation), then it is impossible to know the Patientresource identifier before the request bundle has been processed (that is, before the new Patient resource is created).

Thankfully, the HL7 FHIR specification defines a way to express a dependency between two resources within a request bundle by using a local reference⁶. In the following example, a request bundle contains a POST request to create a new Patient resource, along with a POST request to create a new Observation resource that references that Patient:

In order to reference a resource via a local reference, you must first define a local identifier for that resource by specifying the fullUrl field within the request entry. To define a local identifier, use a value that starts with urn: as in the preceding example. The HL7 FHIR specification highly recommends that the value of the identifier should be a UUID-type value, although the FHIR server does not require the use of a UUID value. You can use any value you like after the urn: prefix as long as it is unique within the bundle. For example, you can use urn:Patient_1, urn:ABC, or urn:Foo. Just make sure that no two request entries share the same fullUrl value within the request bundle.

After you define a local identifier for the referenced resource, you can then define one or more references to that resource by using the local identifier instead of an external identifier. In the preceding example, you can see that the Observation’s subject.reference field specifies the Patient’s local identifier as specified in the fullUrl field of the Patient’s request entry.

The following processing rules apply for the use of local references within a request bundle:

In the example in Section 4.8.0.1, you can see that there are two POST requests and the Patient request entry appears in the bundle before the Observation request entry. However, based on rule 4, it would still be a valid request bundle if the Observation request entry appeared before the Patient request entry, unless the Patient request entry specified a conditional create (rule 5).

If those entries were reversed, and the Patient request entry specified a conditional create, then the FHIR server would return an error when processing the Observation request entry, because the Patient local identifier would not be defined yet.

The following examples also satisfy the local reference processing rules:

In Example 2, if the Patient request entry was a conditional create request, this would still be a valid request bundle, because POST requests are processed before PUT requests (rule 3). This means the Patient request entry would be processed before the Observation request entry, and thus the Patient local identifier would be defined when the Observation request entry was processed.

While processing a request bundle, but before processing individual request entries, the IBM FHIR server detects the use of a local identifier within any POST or PUT request entry’s fullUrl field, and establishes a mapping between that local identifier and the corresponding external identifier that results from performing the POST or PUT operation.

Using Example 3, the FHIR server detects the use of local identifiers in the Encounter request entry (urn:Encounter_1) and in the Procedure request entry (urn:Procedure_1), and establishes a mapping between the local identifiers and the external references to be associated with the new Encounter and Procedure resources (for example, Encounter/1cc5d299-d2be-4f93-8745-a121232ffe5b and Procedure/22b21fcf-8d00-492d-9de0-e25ddd409eaf).

Then when the FHIR server processes the POST requests for the Encounter and Procedure resources, it detects the use of the local references and substitutes the corresponding external references for them before creating the new resources. Here is an example of a response bundle for the request bundle depicted in Example 3 in which we can see that the Encounter’s reasonReference.reference field now contains a proper external reference to the newly-created Procedure resource, and the Procedure’s encounter.reference field now contains a proper external reference to the newly-created Encounter resource:

The FHIR server includes features that allow a single instance of the server to simultaneously support multiple tenants. A tenant is defined as a group of one or more FHIR REST API consumers that share a FHIR server configuration along with one or more data stores associated with that configuration. A tenant could be a single application using the FHIR REST API, or it could be a group of applications belonging to a single customer. The main idea behind multi-tenancy is that each tenant can experience its own customized FHIR server runtime behavior and its data can be physically isolated from other tenants’ data for increased security and privacy.

To support multi-tenancy, the FHIR server must know which tenant an incoming REST API request is intended for. To provide the tenant id to the FHIR server, a REST API consumer must set a request header in each REST API request. The name of this request header is itself configurable by setting the fhirServer/core/tenantIdHeaderName configuration property in the FHIR server’s global configuration file (located at $⁠{server.config.dir}/config/default/fhir-server-config.json). The following example shows the default setting for this configuration parameter:

With this configuration in place, each REST API consumer would need to set the X-FHIR-TENANT-ID request header to the appropriate tenant id. Each tenant’s tenant id value is assigned by the deployer. It is simply a short name associated with the tenant and must be unique among all tenants supported by a single FHIR server instance. For example, let’s suppose Acme Healthcare, Inc. is using the FHIR server and their tenant id has been assigned by the deployer as “acme”. In this case, Acme-related applications would set the following request header in each REST API request: X-FHIR-TENANT-ID: acme

As mentioned earlier, the name of the request header is configurable in the FHIR server’s global configuration file. For example, the FHIR server deployer could configure the request header name to be X-WHCLSF-tenant-id by setting tenantIdHeaderName as shown in the following example:

This would be useful in an environment where the applications might already be using a similar type of request header. With this configuration in place, the “Acme Healthcare, Inc.” tenant would set the following request header in each REST API request: X-WHCLSF-tenant-id: acme

The FHIR server allows a deployer to configure a subset of the supported configuration properties on a tenant-specific basis. For a complete list of configuration properties supported on a per-tenant basis, see Section 5.1.3 Property attributes.

When the FHIR server needs to retrieve any of the tenant-specific configuration properties, it does so dynamically each time the property value is needed. This means that a deployer can change the value of a tenant-specific property within a tenant’s configuration file on disk, and the FHIR server will immediately “see” the new value the next time it tries to retrieve it. For example, suppose the deployer initially defines the acme tenant’s fhir-server-config.json file such that the fhirServer/core/defaultPrettyPrint property is set to true.

Requests from the acme tenant would result in pretty-printed responses (with newlines and indentation), making it easier for humans to read. Now suppose the deployer changes the value of that property to true within the acme tenant’s fhir-server-config.json file. A subsequent REST API request would then see the output condensed into a single line with minimal whitespace.

This section contains examples of both a global (default) configuration and a tenant-specific configuration.

The global configuration contains non-tenant specific configuration parameters (configuration parameters that are not resolved or used on a tenant-specific basis), as well as default values for tenant-specific configuration parameters.

${server.config.dir}/config/default/fhir-server-config.json

The Acme Healthcare, Inc tenant configuration overrides the fhirServer/core/defaultPrettyPrint property so that the development team can more easilly read the messages.

${server.config.dir}/config/acme/fhir-server-config.json

It is also possible to configure the persistence properties for a specific tenant, for example to set an alternate database hostname or database schema name.

The IBM FHIR Server implements bulk data export according to the HL7 FHIR BulkDataAccess IG: STU1, and bulk data import is implemented according to the Proposal for $import Operation.

There are 2 modules involved:

The fhir-operation-bulkdata module implements the REST APIs for bulk data export, import and status as FHIR Operations. There are five operations:

Each operation queues a job with the Open Liberty JavaBatch framework. Each job instance unit-of-work is executed as a job with the fhir-bulkdata-webapp module. There are 5 JavaBatch jobs defined in the fhir-bulkdata-webapp module for the above 3 export operations and 1 import operation:

The fhir-bulkdata-webapp module is a wrapper for the whole BulkData web application, which is the build artifact - fhir-bulkdata-webapp.war. This web archive is copied to the apps/ directory of the liberty server. The feature is configured using the configDropins/default/bulkdata.xml, such as:

The Bulk Data web application writes the exported FHIR resources to an IBM Cloud Object Storage (COS), any Amazon S3-Compatible bucket (e.g, Amazon S3, minIO etc), or directory as configured in the per-tenant server configuration under fhirServer/bulkdata. The following is an example configuration for bulkdata; please refer to section 5 for the detailed description of these properties:

Each tenant’s configuration may define multiple storageProviders. The default is assumed, unless specified with the X-FHIR-BULKDATA-PROVIDER and X-FHIR-BULKDATA-PROVIDER-OUTCOME. Each tenant’s configuration may mix the different providers, however each provider is only of a single type. For instance, minio is aws-s3 and default is file. Note, type http is only applicable to $import operations. Export is only supported with s3 and file.

To use Amazon S3 bucket for exporting, please set accessKeyId to S3 access key, and set secretAccessKey to the S3 secret key, and the auth type to hmac.

Basic system exports to S3 without typeFilters use a streamlined implementation which bypasses the IBM FHIR Server Search API for direct access to the data enabling better throughput. The fhirServer/bulkdata/core/systemExportImpl property can be used to disable the streamlined system export implementation. To use the legacy implementation based on IBM FHIR Server search, set the value to “legacy”. The new system export implementation is used by default for any export not using typeFilters. Exports using typeFilters use FHIR Search, and cannot use the streamlined export.

To import using the $import operation with https, one must additionally configure the fhirServer/bulkdata/validBaseUrls. For example, if one stores bulk data at https://test-url1.cos.ibm.com/bucket1/test.ndjson and https://test-url2.cos.ibm.com/bucket2/test2.ndjson you must specify both baseUrls in the configuration:

These base urls are not checked when using cloud object store and bulk-import. If you need to disable the validBaseUrls feature you may add fhirServer/bulkdata/storageProviders/(source)/disableBaseUrlValidation as true.

For Bulk Data import, the fhirServer/bulkdata/core/maxInputs is used to configure a maximum number of inputs supported by the instance. The default number is 5. There is a hard character limit on the total input type and url must be under 4096 characters, as such the configuration may be tuned for each url scheme.

Note: When $import is executed, if a resource to import includes a Resource.id then this id is honored (via create-on-update). If Resource.id is not valued, the server will perform a create and assign a new Resource.id for this resource.

Following is the beautified response of sample polling location request after the export is finished:

For the Import Operation, the polled status includes an indication of $import and the location of the OperationOutcome NDJSON files and the corresponding failure and success counts.

For S3 exports, the bulk data feature may use presigned urls. To enable presigned urls, the authentication type must be hmac and fhirServer/bulkdata/storageProviders/(source)/presigned must be true. The urls become:

The presigned URL is valid for 86400 seconds (1 day).

Note, the deletion of an a job is split into two phases, ACCEPTED (202) response and DELETED (204). 202 is returned until the operation is stopped or removed, and then 204.

Prior to version 4.8.1, the exported ndjson file is configured with public access automatically and with 2 hours expiration time using fhirServer/bulkdata/storageProviders/(source)/exportPublic. The exported content is best made available with presigned urls with the hmac authentication type.

In 4.8.1, the randomly generated path is used to uniquely identify the exported files in a single folder.

JavaBatch feature must be enabled in server.xml as following on the Liberty server:

The JavaBatch user is configured in bulkdata.xml and the fhir-server-config.json:

Note: The batch-user referenced in the fhir-server-config.json must have a role of at least batchSubmitter.

By default, in-memory Derby database is used for persistence of the JavaBatch Jobs as configured in fhir-server/configDropins/defaults/bulkdata.xml. This database is destroyed on the restart of the IBM FHIR Server, and does not support load balancing.

To support IBM Db2 on IBM Cloud, copy fhir-server/configDropins/disabled/db2-cloud/bulkdata.xml to fhir-server/configDropins/defaults/bulkdata.xml replacing the existing bulkdata.xml. One can configure the Datasource by setting the following environment variables:

Instruction is also provided in ‘Configuring a Liberty Datasource with API Key’ section of the DB2OnCloudSetup guide to configure DB2 service in IBM Clouds as JavaBatch persistence store. The JavaBatch schema is created using the fhir-persistence-schema command line interface jar.

To support IBM Db2 with a user-name and password , copy fhir-server/configDropins/disabled/db2/bulkdata.xml to fhir-server/configDropins/defaults/bulkdata.xml replacing the existing bulkdata.xml. One can configure the Datasource by setting the following environment variables:

If one wants to support Postgres with a user-name and password , one should copy fhir-server/configDropins/disabled/postgres/bulkdata.xml to fhir-server/configDropins/defaults/bulkdata.xml replacing the existing bulkdata.xml. One can configure the Datasource by setting the following environment variables:

Note: If you use PostgreSQL database as IBM FHIR Server data store or the JavaBatch job repository, please enable max_prepared_transactions in postgresql.conf, otherwise the import/export JavaBatch jobs fail.

For more information about Liberty JavaBatch configuration, please refer to IBM WebSphere Liberty Java Batch White paper.

For BulkData storage types of ibm-cos and aws-s3, the IBM FHIR Server supports two styles of accessing the s3 bucket - virtual host and path. In the IBM FHIR Server, path is the default access. Link

With path style access, the objects in a bucket are accessed using the pattern - https://s3.region.host-name.com/bucket-name/object-key. To configure path style access, one needs to configure the fhir-server-config.json.

There are three critical elements in the configuration to configure path style:

Example of path based access:

With virtual host style access, the objects in a bucket are accessed using the pattern - https://bucket-name.s3.region.host-name.com/object-key. To configure virtual host style access, one needs to configure the API.

There are three critical elements in the configuration to configure virtual host style:

Note, while the endpointInternal is specified with the S3 region endpoint, the calls to the API will use the virtual host directly.

Example of host based access:

When Importing from an S3 Bucket, the IBM FHIR Server identifies the matching file segments. The parameter.input.url is used to query the S3 API to find the matching files. For instance, the following import of Patient.ndjson matches Patient.ndjson_seg0 and Patient.ndjson_seg1 which are imported to the IBM FHIR Server.

This feature is useful for imports which follow a prefix pattern:

To integration test, there are tests in ExportOperationTest.java in fhir-server-test module with server integration test cases for system, patient and group export. Further, there are tests in ImportOperationTest.java in fhir-server-test module. These tests rely on the fhir-server-config-db2.json which specifies two storageProviders.

Version 4.4 of the IBM FHIR Server introduced experimental support for exporting to Parquet format (as an alternative to the default NDJSON export). However, due to the size of the dependencies needed to make this work, this feature is disabled by default.

To enable export to parquet, an administrator must:

An alternative way to accomplish the first part of this is to change the scope of these dependencies from the fhir-bulkdata-webapp pom.xml and rebuild the webapp to include them.

Because the bulk import and export operations are built on Liberty’s java batch implementation, users may need to check the Liberty batch job logs for detailed step information / troubleshooting.

In a standard installation, these logs will be at wlp/usr/servers/fhir-server/logs/joblogs. In the ibmcom/ibm-fhir-server docker image, these logs will be at /logs/joblogs.

Note, if you are using the default derby, the logs are overwritten upon restart of the server. You should use Db2 or Postgres for production purposes.

The Audit logging service pushes FHIR server audit events for FHIR operations in Cloud Auditing Data Federation (CADF) standard format to a Kafka backend, such as IBM Cloud Event Streams service.

There is early support for the FHIR Standard: AuditEvent format.

Each FHIR interaction triggers a CADF audit log entry to be logged. The mapping of FHIR interaction to CADF action is as follows:

The following table describes the JSON fields of the CADF audit log entries logged by the FHIR server:

Note for Batch/Transactions, attachments/content includes counts of the number of C-R-U-D-E actions.

Please refer to the property names that start with fhirServer/audit/ in 5.1 Configuration properties reference for how to enable and configure the CADF audit logging service.

There are two types of configuration for the Audit Logging Service. The first type is using the environment variable, and the second is the configuration driven from the fhir-server-config.json.

For example, the following uses the ‘config’ in order to config the Kafka publisher as part of the audit logging service.

Or, you could load from an environment, note, this is the default behavior.

The audit logging service gets the event streams service credential from environment variable EVENT_STREAMS_AUDIT_BINDING with values like this:

If you are using the IBM EventStreams on IBM Cloud, the service credential can be generated automatically when you run:

to bind your event streams service instance to your Kubernetes cluster.

And then in the YAML file for your Kubernetes deployment, specify the environment variable EVENT_STREAMS_AUDIT_BINDING that references the binding key of the generated secret(binding-<event_streams_service_instance_name>) as following:

Please refer to https://cloud.ibm.com/docs/containers?topic=containers-service-binding for detailed instructions if needed.

The service can map to the CADF format or the FHIR AuditEvent resource format by declaring a mapper type - ‘cadf’ or ‘auditevent’.

Watson Studio stream flow can be created to push those FHIR Audit CADF events from the Event Streams service to a COS bucket (e.g. fhir-audit-dev) in CSV format. Another option is to configure Event Streams (Kafka) S3 connect to push those CADF events to a COS bucket (e.g. fhir-audit-dev) but in raw CADF json format.

A service instance of the IBM Cloud SQL Query service can be created to allow you to query those CADF audit events in COS with SQL queries. Before running an SQL query, it’s recommended to first create a COS bucket to store your query results, otherwise the query results will be stored in a bucket which is automatically created by the SQL query service.

Sample queries for CSV records expanded from the JSON CADF events:

Sample queries for the raw JSON CADF events:

By default, the IBM FHIR Server allows the following RESTful interactions associated with the FHIR REST API: create, read, vread, history, search, update, patch, delete. However, it is possible to configure which of these interactions are allowed on a per resource basis through a set of interaction rules specified via the fhirServer/resources/<resourceType>/interactions property in fhir-server-config.json. The following snippet shows the general form for specifying interaction rules:

The fhirServer/resources/<resourceType>/interactions property is a JSON array of strings that represent the RESTful interactions allowed for the given resource type. If an interaction is not in the list of strings specified for a resource type, that interaction will not be allowed for that resource type. In the example above, the following interactions are allowed for the Observation resource type: [create, read, vread, history, delete]. This means a user will not be able to search for Observation resources because the search interaction is not specified in the list of allowed interactions.

Omitting the fhirServer/resources/<resourceType>/interactions property is equivalent to allowing all interactions for a given resource type. An empty array, [], can be used to indicate that no interactions are allowed. Additionally, to define the set of interactions allowed for resource types which are not specifically configured in the fhirServer/resources property group, or for resource types which are configured, but which do not specify the fhirServer/resources/<resourceType>/interactions property, the base type of Resource may be specified:

In the example above, for any resource type which is not specifically configured, such as Encounter, or for any resource type which is configured but does not specify the fhirServer/resources/<resourceType>/interactions property, such as Procedure, all of the interactions listed for the Resource resource type will be allowed.

One final consideration when configuring interactions is the fhirServer/resources/open property. If this property is specified and its value is set to false, then no interactions will be allowed for resource types which are not specifically listed in the fhirServer/resources property group. Assume the following configuration:

In this case, since the fhirServer/resources/open property is set to false, only the resource types listed (Condition, Observation, Patient) are allowed to be interacted with via the FHIR REST API. For example, a create request of a Procedure resource will fail since that resource type is not specified.

Whole-system search and whole-system history are special cases. Since no resource type is specified on a whole-system request, validation will be done against the Resource resource type. In the above configuration example, a whole-system search request such as GET [base]?_lastUpdated=gt2020-01-01 will fail because the Resource resource type is not specified. If the configuration were to have the fhirServer/resources/open property set to true, or if the Resource resource type were specified in the fhirServer/resources property group, then the whole-system search request would be allowed, assuming the search interaction was valid for the Resource resource type.

In addition to interaction configuration, the fhirServer/resources property group also provides the ability to configure search parameter filtering and profile validation. See Search configuration and Resource validation respectively for details.

It is possible to run the IBM FHIR Server behind a reverse proxy such as Kubernetes Ingress or an API Gateway.

Because the FHIR API relies on links and references between resources (both absolute and relative), operators must ensure that the IBM FHIR Server is configured appropriately for the front-end URL being used by the FHIR clients.

This can be accomplished by configuring the fhirServer/core/originalRequestUriHeaderName property in the default fhir-server-config.json. When this parameter is configured, the IBM FHIR Server will use the value of the corresponding header to set the “originalRequestUri” for the scope of the request.

For example, consider a FHIR Server that is listening at https://fhir:9443/fhir-server/api/v4 and is configured with an originalRequestUriHeaderName of X-FHIR-FORWARDED-URL. If this server is proxied by a server at https://example.com/fhir, then the proxy must set the X-FHIR-FORWARDED-URL header to the value of the front-end request URL (e.g. https://example.com/fhir/Patient/abc-123). In alternative deployments, the fhirServer/core/externalBaseUrl may be used in-lieu of the X-FHIR-FORWARDED-URL.

The originalRequestUriHeader is expected to contain the full path of the original request. Values with no scheme (e.g. https://) will be handled like relative URLs, but full URL values (including scheme, hostname, optional port, and path) are recommended. Query string values can be included in the header value but will be ignored by the server; the server will use the query string of the actual request to process the request.

This section contains reference information about each of the configuration properties supported by the FHIR server.

Depending on the context of their use, config properties can be:

The following table tracks which properties can be set on a tenant-specific basis and which properties are loaded dynamically. If you change a properties that has an N in the Dynamic? column, it means you must restart the server for that change to take effect.

As stated earlier, the FHIR server is installed with a default configuration in server.xml which includes the definition of a keystore (fhirKeyStore.p12) and a truststore (fhirTrustStore.p12)⁷. These files are provided only as examples and while they may suffice in a test environment, the FHIR server deployer should generate a new keystore and truststore for any installations where security is a concern. Review the information in the following topics to learn how to configure a secure keystore and truststore.

Additionally, the server has a trustDefault.xml config dropin that references the SEC_TLS_TRUSTDEFAULTCERTS variable (defaultValue = true) to indicate whether or not the JVM truststore should be used in combination with the configured trust store.

By default, the FHIR server REST API is only available via HTTPS on port 9443 and is protected by HTTP basic authentication. Alternatively, the server can use OpenID Connect and OAuth 2.0 via a Bearer Token as described in Section 5.3 OpenID Connect and OAuth 2.0. In addition, the FHIR server web application can be secured via client certificate-based authentication.

Here are some notes related to these authentication schemes:

To properly configure the FHIR server’s keystore and truststore files, perform the following steps.

At this point, you should have a client keystore that contains a client certificate whose Distinguished Name’s Common Name component is set to the username. You should also have a client truststore which contains the server’s public key certificate. Essentially, the server and client both have a keystore that contains their own private and public key certificate and they both have a truststore which contains the public key certificate of their counterpart.

Copy the server keystore (serverKeystore.jks) and truststore (serverTruststore.jks) files to the appropriate directory (<WLP_HOME>/usr/servers/fhir-server/resources/security). Then configure the server.xml file correctly to reference your new keystore and truststore files.

The precise steps required to configure certificate-based authentication for a client application depend on the specific REST API client framework, but these are the general rules:

The FHIR specification recommends the use of OAuth 2.0. The IBM FHIR Server supports OAuth 2.0 through the use of WebSphere Liberty / OpenLiberty features.

While it is possible to configure Liberty as an OpenID Connect Client, more typically the IBM FHIR Server will be configured as a generic OAuth 2.0 “Protected Resource Server” that works with JWT access tokens that have been issued by a trusted Authorization Server like Keycloak.

Liberty can be configured to act as an OAuth 2.0 Protected Resource Server via either the openidConnectClient feature or the mpJwt feature.

One advantage of the mpJwt approach is that users can be mapped into pre-defined JEE security roles. To enable this feature without modifying the default server.xml, move the jwtRS.xml config snippet from configDropins/disabled/ to configDropins/defaults/ and modify as desired.

A copy of this snippet is provided here for illustrative purposes:

In the snippet above, the mpJwt element is configured to obtain JWK information from http://keycloak:8080/auth/realms/test/protocol/openid-connect/certs and use this to validate JWT tokens that are passed to the server.

Additionally, the server will validate the iss (issuer) and aud (audiences) claims of the JWT and use the sub claim as the user principal (for audit logging).

Finally, the value(s) of the group claim are used to map this user into a corresponding JEE security-role as defined in the application-bnd section of the webApplication element.

To make the FHIR Server advertise the OAuth endpoints of the configured provider, supply values for at least the following properties in the default fhir-server-config.json file:

These values will be used to populate the corresponding entries in both the server capability statement (GET [base]/metadata) and the smart-configuration (GET [base]/.well-known/smart-configuration).

For example, the following excerpt from a CapabilityStatement shows sample OAuth-related URLs (register, authorize, and token) values in the valueUri elements.

SMART on FHIR applications should use the .well-known/smart-configuration endpoint to determine the OAuth URLs to use for authorization, but the entries in the Capability Statement are needed for backwards compatibility.

To support SMART App Launch, the IBM FHIR Server can be used with a SMART-enabled authorization server. For an example of a SMART-enabled Authorization Server, see the Alvearie Keycloak extensions for FHIR.

The OAuth configuration described in the previous sections will restrict API access to clients with a valid access token. However, SMART defines additional access controls via OAuth 2.0 scopes and context parameters.

To enforce authorization policy on the server, drop the fhir-smart module into the server’s userlib directory.

This component uses the IBM FHIR Server’s PersistenceInterceptor feature to automatically scope searches to the compartments for which the user has access (as indicated by a special patient_id claim in the access token).

Additionally, before returning resources to the client, the fhir-smart component performs authorization policy enforcement based on the list of SMART scopes included in the token’s scope claim and the list of patient compartments in the patient_id claim.

For an example of using the IBM FHIR Server together with a SMART-enabled Keycloak authorization server, please see the data-access pattern at https://github.com/Alvearie/health-patterns/tree/main/data-access.

IBM FHIR Server Supports the following custom HTTP Headers:

For more information about topics related to configuring a FHIR server, see the following documentation:

FHIR® is the registered trademark of HL7 and is used with the permission of HL7.