Hello,

I could find no mention of version control when using the Elasticsearch backend. Does Hibernate Search use any version checks when indexing documents? If there is version control, how are conflicts resolved? Elasticsearch supports the 'version' and 'version_type' parameter when indexing to either use optimistic concurrency control or rely on an external version number.

I am asking because I am a little worried that concurrent threads updating the same document in separate transactions can easily overwrite each other, leading to missing updates to the index.
Also, if in a single transaction there are multiple updates performed on a document, are those updates merged into one update request when building the batch request to Elasticsearch?

Keep in mind that I aim to use asynchronous worker execution to minimise performance impact.

Thank you in advance!

Hibernate Search doesn't use versions.

Concurrent threads updating the same entities will benefit from the Transaction and Hibernate Session: updates to Elasticsearch will be applied in the same order as operations get committed on your database.

Essentially the problems you highlight could theoretically happen but only if you allow weak locking semantics on your database; you should be fine when relying on correct isolation levels from the database.

The asynchronous worker should be safe against reorederings as while the operations are "async" relatively to the transaction commit, the event queue which handles them is handled by a single thread which will batch them all periodically, yet still maintaining the same order of operations.

All right, if I understand correctly since I use the default READ_COMMITTED isolation level, I should be fine. Thank you!