Hi Steve, this does sound amazing, I think it's certainly worth trying to have the most efficient strategy chosen on a per-entity base.
One question: could we benefit from "cascade delete" rules defined in the table structure? If Hibernate ORM was able parse the cascading rules from existing schemas, and possibly generate them for new schemas, you could have an entire crop of additional such optimisations. I'm not sure if I'm talking about material for ORM 7+ but if you could make sure that such future optimisations aren't ruled out that would be awesome :) Thanks, Sanne On Tue, 26 Mar 2019 at 13:25, Steve Ebersole <st...@hibernate.org> wrote: > > While working on 6 I discovered that we maybe do not do the best job in > regards to "bulk update/delete" queries against multi-table entities > (secondary tables, joined inheritance, etc). This short-coming exists in > previous versions. Essentially, when a strategy is not explicitly > specified we fallback to using an "id table" and performing the SQL updates > or deletes using the id table as a sub-query. The problem is that for some > databases this breaks when the ids are composite due to the database not > having complete support for tuples. For example, H2 does not allow the > sub-query for an in-predicate to return more than one column: so a query > like `... where (id1, id2) in (select val1, val2 from ...)` will not work. > This tuple concern is something I had not considered when I first wrote > this feature. > > Obviously with 6 we have a chance to improve this. So I have been thinking > about some ways this might be improved. The guiding principle here was to > make this as implicit as possible... > > Because certain strategies will not work when the ids are composite, I > think the first consideration is whether we want to allow the strategy to > be definable per-entity. The idea would be to allow for the most efficient > strategy to be used generally (whatever that might be for the given > app/database) but to pick an alternative strategy in cases where that > "fallback" one will not work. The Dialect should certainly play a part in > this strategy selection. > > And lastly, we should consider whether it is beneficial to leverage these > strategies when performing normal mutations (merge, persist, etc). I think > really this comes down to whether we think there is a benefit to handling > these via CTE if the database supports that as opposed to sending > individual updates or deletes against each table. > > Thoughts? > _______________________________________________ > hibernate-dev mailing list > hibernate-dev@lists.jboss.org > https://lists.jboss.org/mailman/listinfo/hibernate-dev _______________________________________________ hibernate-dev mailing list hibernate-dev@lists.jboss.org https://lists.jboss.org/mailman/listinfo/hibernate-dev
