ServiceEnhancer Improvements

[Aklakan]

GitHub issue resolved (TDB)

This is a long standing revision of the service enhancer module. It can go into a dot release after jena 6. There are a couple of places that seem odd to me and which I want to give another look.

Pull request Description:

Code maintenance:

• Migrated to Junit 5 / Jupiter
• Migrated to Caffeine Cache (instead of Guava Cache - The Guava dependency is still needed for RangeMap though)
• Refactored the core cache execution class RequestExecutor to support read-transaction-aware concurrent execution.

Bug Fixes:

• The Service Enhancer assembler now supports the standard ja:dataset property instead of the custom ja:baseDataset. Use of the latter still works but now raises a warning.
• Fixed a bug with the plugin that caused colons at the end of URLS to become removed (SERVICE <https://.../foo:> became SERVICE <https://.../foo>)
• Fixed a bug where result sets could become cut off due to concurrent cache eviction. Typically this bug occurred when using SELECT ... LIMIT X where X was slightly larger than the maximum cache size.
• Introduced context constants (that match that of the assembler) to configure cache sizes. As a side effect, Service Enhancer Cache initialization is now lazy in order to have a chance to read the config from the context. Once created, the settings are fixed.
• Iterators that cache data should now honor the abort signal.

New Features:

• Added scoped option for loop (SERVICE <loop+scoped:>) to mimic the behavior of LATERAL - when scoped is set, then only in-scope variables become substituted instead of all variables with the same base name.
• Added experimental concurrent feature. There is an example in the class ServiceConcurrentExample. Essentially, it allows one to do SERVICE <loop+scoped:concurrent+2-10:bulk+10:cache:https://example.org/sparql>: Use 2 additional threads with 10 bindings each, to build bulk requests of 10 bindings, caching the responses to query the specified sparql endpoint. Concurrent requests try to preserve the order of bindings. Of course, one must be cautious not to flood public endpoints. The feature is intended as a tool to improve query federation performance over endpoints that one controls.

---

• Tests are included.
• Documentation change and updates are provided for the Apache Jena website
• Commits have been squashed to remove intermediate development commit messages.
• Key commit messages start with the issue number (GH-xxxx)

By submitting this pull request, I acknowledge that I am making a contribution to the Apache Software Foundation under the terms and conditions of the Contributor's Agreement.

---

See the Apache Jena "Contributing" guide.