Updated April 2023.
At work we use a fork of Marketing Cloud's FuelSDK, generally for keeping its libraries up-to-date as Salesforce doesn't often do so for its main branch. The fork is using the latest CXF 3.5.x branch as of this writing. Other enhancements:
JDK 11 base, above the Java 6 in the main fork.
The Partner API WSDL has been updated. This file is used by CXF to generate Java classes supporting the SOAP requests and responses. The main Salesforce branch is still relying on a copy of the WSDL downloaded in 2017. A comparison of the two WSDLs is showing quite a few new and modified request and response objects since then:
ClientID: CustomerID TaskResult: TblAsyncID SaveOption: TrackChanges AccountUser: IsSendable APIObject: __AdditionalEmailAttribute 1-5. Attribute: no longer inheriting APIObject SubscriberResult: ErrorCodeID TriggeredSendClassEnum TriggeredSendSubClassEnum SenderProfile: FallbackFromAddress deleted: MessagingConfiguration ChatMessagingEventType SalesforceSendActivity ImportDefinition: HasMultipleFiles ImportResultsSummary: NumberRestricted JsonWebKey DirectoryTenant AuditLogUserContext AutomationActivity: SerializedObject AttributeEntityV1 Thumbnail NameIdReference CategorynameIdReference UserBasicsEntity AssetAnyProperty Asset Category ScheduledRequest ScheduledConversation
I found necessary to make two adjustments to the WSDL copied into my fork, as detailed on the project README and commented in the WSDL. For one of them--a mismatch between what the WSDL claims the SOAP response will be for a particular call and what it actually is, causing a validation exception with CXF--I've sent a help ticket to Marketing Cloud requesting they update their WSDL. This particular issue has been around a long time, apparently, as even the 2017 WSDL needed this manual adjustment.
I've found in rare instances that access token requests to Marketing Cloud fail with a MC-side 500 system error with little extra detail provided. In such cases the ETClient in the FuelSDK would throw a generic EtSdkException for the client to handle. Our logging has shown that a second access token request immediately thereafter would usually be successful.
To make it easier for FuelSDK clients to trap and recover from this specific error, in the fork I created a new ETTokenRequestException, subclassed from the current ETSdkException, that the ETClient now instead throws. Subclassing from the current exception maintains backwards compatibility for current clients of the library while allowing them to update as they wish.
The FuelSDK's ETClient now has a close() method to explicitly release various CXF objects prior to making a new connection. While DataDog or another memory profiler can tell you whether that is actually necessary, I've found no harm in calling it.
The fork has been running fine on production for several months now, we're using it to register and update subscribers and send emails, and it is providing us peace of mind that we're using the latest libraries while doing so.