heads up for readers: a lot of discussion on this topic happened in GH review comments.
As for me, the approach proposed in PR #1783 looks good in the short term. I'm exploring options for improving CDI in general, and specifically as it relates to the realm context. If that works out, I suppose the config API can be simplified. Cheers, Dmitri. On Tue, Jun 3, 2025 at 1:51 AM yun zou <yunzou.colost...@gmail.com> wrote: > Hi Team, > > In order to get rid of the usage of CallContext.getCurrentContext() call in > the getConfiguration implementation of DefaultConfigurationStore, we did > one attempt leverage CDI to inject request scoped RealmContext into the > application scoped DefaultConfigurationStore. > Commit: > > https://github.com/apache/polaris/commit/d753a95e4ae1724e87db8a0a3a925094f354ee67 > > However, it fails the background task (TaskExecutor) because the > getConfiguration is called outside the active request scope in the > background task, and the request scoped bean is not propagated properly to > the background task thread. While we are still investigating how to > propagate the request scoped bean properly, we pass the RealmContext to the > call of getConfiguration. However, this requires us to pass the > RealmContext as part of the loadTasks interface for > PolarisMetaStoreManager. > > The PR here https://github.com/apache/polaris/pull/1783 adds the > RealmContext as part of the loadTasks call, please help take a look and > provide your valuable feedback. > > Best Regards, > Yun >
