Chris Cox opened DATAMONGO-719 and commented

Lazily ensuring indexes based off dbName.collectionName (instead of class type) would be incredibly useful in cases such as collection based multi-tenant systems.

Currently, the MongoPersistenEntityIndexCreator tracked index ensured collections classesSeen. This works very well when the entityClass:collection is 1:1. In cases such as the one mentioned above, a new tenant means a new collection and new indexes. In the current code, Spring Data will at best create a defaulted or "tenantless" collection to index which will likely never be used.

If the index creator instead tracked by a resolved dbName.collectionName pattern, this would cover all mulit-tenant instances. However, it can no longer listen for an MappingContextEvent from the mapping context as this will no longer cover all cases where a new collection is required. It would most likely need to connect to an events on both a (yet to exist) BeforeFindEvent and BeforeSaveEvent and check to see if the target collection has been accessed yet. This cache of accessed collections can be exposed to a distributed caching, but ensuring indexes that already exist should be a minor operation for mongo.

See the referenced URL. The project's workaround branch has a test showing desired functionality and workarounds to the current code to make it possible

---

Affects: 1.2.1

Reference URL: https://bitbucket.org/ccox/collectionmultitenantpoc

Issue Links:

• DATAMONGO-841 Indexing do not work properly after collection is dropped

• DATAMONGO-660 Index not automatically created

• DATAMONGO-2188 Add configuration option to enable/disable automatic index creation

Referenced from: pull request #754

1 votes, 5 watchers