Conditional updates and ETag concurrency are separate FHIR mechanisms that solve different problems, and vendor documentation does not treat them as interchangeable. A conditional update selects a resource by search criteria rather than by ID, while an ETag check validates that the client is modifying the version it previously read. Google Cloud Healthcare API documentation describes an ETag as a unique identifier for the current state of a FHIR resource on the server, similar to a version number, generated each time the resource is created or modified [1]. Azure Health Data Services documentation lists conditional update as a supported interaction and describes the If-Match header as a way to validate the version being updated [3]. The practical distinction matters because a conditional update can still overwrite a newer version if no ETag precondition is supplied, and an ETag check alone does not resolve which resource a search should target. Teams evaluating these mechanisms should treat the documented behavior as a starting point for their own validation rather than as a complete concurrency contract.

What Conditional Update Actually Selects

Oracle Healthcare Data Repository documentation describes conditional update as a mechanism where the client supplies a search URL instead of an ID. Before performing the update, the server first performs the search. If no resources match, a new resource is created; if a resource already matches, it is updated using the contents in Bundle.entry.resource [6]. The documented example uses a transaction bundle with a PUT method and a search URL such as Patient?identifier=acme.org/mrns|013872 [6]. This behavior makes conditional update useful when a client knows a business identifier but not the server-assigned logical ID.

Azure Health Data Services documentation lists conditional update among supported create and update interactions as defined by the FHIR specification [3]. The documentation does not describe the create-if-absent behavior in the same detail as Oracle, but it does state that conditional patch versions patch resources by search criteria instead of resource IDs [3]. Boomi documentation describes an Operation Mode that indicates whether the connector should perform a standard or a conditional update based on determined criteria [5]. The connector documentation notes that availability of object types and operation properties may differ between FHIR versions and servers [5].

ETag Validation as a Separate Concurrency Control

Google Cloud Healthcare API documentation explains that an ETag is a unique identifier for the current state of a FHIR resource on the server, similar to a version number. Each time a FHIR resource is created or modified, a new ETag value is generated [1]. The documentation states that the ETag matches the Meta.versionId in the FHIR resource [1]. When a client includes an If-Match header, the request only succeeds if the provided ETag matches the current ETag on the server [1].

If the ETag does not match, the update fails with a 412 Precondition Failed error, indicating another client has modified the resource since the original ETag was fetched [1]. Google Cloud documentation describes this as preventing data loss from accidental overwrites [2]. Azure Health Data Services documentation similarly states that the If-Match header validates the version being updated before making the update, and that a mismatch returns a 412 Precondition Failed error [3]. Azure API for FHIR documentation repeats this behavior for the If-Match header [4].

A Suggested Validation Scenario for Conditional Update with ETag

A useful hypothetical validation scenario would test whether a conditional update can be combined with an If-Match precondition in a single request. A team could create a Patient resource with identifier acme.org/mrns|013872, read it to obtain the ETag, then submit a conditional update using the search URL Patient?identifier=acme.org/mrns|013872 with the If-Match header set to the fetched ETag. The expected documented behavior would be a successful update if the ETag matches and a 412 Precondition Failed if another client modified the resource in the interim [1][3]. This scenario is a suggestion, not a tested result.

A second scenario could verify the create-if-absent path. A team could submit a conditional update with a search URL that matches no existing resource and observe whether the server creates a new resource, as Oracle documentation describes [6]. The team should also record whether the created resource receives a new ETag and whether the response includes the Meta.versionId. These observations would establish the deployment-specific behavior that vendor documentation leaves open.

Weak Validation and Cross-Instance Boundaries

Google Cloud Healthcare API documentation states that FHIR ETags use weak validation, meaning they might not be identical across different server instances, but they still effectively track resource changes [1]. This is a consequential boundary for distributed deployments. A client that caches an ETag from one server instance cannot assume that the same value will be accepted by another instance, even if the underlying resource has not changed. The documentation does not specify how many instances a deployment may use or how ETag values are generated across them.

The weak validation note also affects client-side caching. Google Cloud documentation describes using If-None-Match to retrieve a previously cached resource, where a matching ETag produces a 304 Not Modified response and the client uses its cached copy [1]. This saves bandwidth and reduces unnecessary network traffic [1]. However, the documentation does not state whether the 304 behavior is guaranteed across instances or whether a weak ETag comparison may produce a 304 when the cached copy is semantically equivalent but byte-different. Teams should test this behavior in their own deployment topology.

Conditional Operations and Performance Trade-offs

Azure API for FHIR documentation includes a performance consideration for conditional interactions. It states that conditional interactions can be complex and performance-intensive, and that the request header x-conditionalquery-processing-logic can be set to parallel to allow concurrent execution of queries with conditional interactions [4]. A separate header, x-ms-query-latency-over-efficiency, when set to true, executes queries using maximum supported parallelism, forcing the scan of physical partitions to be executed concurrently [4]. The documentation notes that such queries can take longer due to the number of physical segments that need to be scanned [4].

This is a documented trade-off specific to Azure API for FHIR, not a general FHIR requirement. The documentation does not state whether the same headers apply to Azure Health Data Services or whether the parallel processing logic changes conditional update semantics. Google Cloud and Oracle documentation do not describe equivalent performance headers for conditional updates. Teams comparing platforms should not assume that a performance tuning option documented for one Azure product transfers to another, even within the same vendor family.

Bundles, Patch, and the Limits of Documentation

Azure Health Data Services documentation states that batch and transaction bundles can submit actions on multiple resources of the same or different types, such as create, update, or delete [3]. A batch or transaction bundle interaction is performed with HTTP POST at the base URL [3]. Oracle documentation shows a transaction bundle example where a Patient and an Observation are conditionally updated using search URLs, with a placeholder ID reference between them [6]. The Oracle example demonstrates that conditional update can participate in a transaction, but the documentation does not state whether ETag preconditions can be combined with conditional update search URLs in the same request.

Azure Health Data Services supports JSON Patch and FHIRPath Patch, including conditional versions that patch resources by search criteria instead of resource IDs [3]. Azure API for FHIR documentation adds that when using PATCH against STU3 and requesting a History bundle, the patched resource's Bundle.entry.request.method maps to PUT because STU3 does not contain a definition for the PATCH verb in the HTTPVerb value set [4]. This version-specific mapping is a documented detail that teams working with STU3 should verify in their own history responses.