Now it’s turned out that the developers have found a critical bug in the application, potentially causing many users to be using a large volume of data.
The bug was discovered in the delayed commit feature of the database. The feature allows a speed increase in the database by only commiting writes to disk in batches, while sacrificing a small amount of durability (not taking account of the bug, of course); the feature is turned on by default.
When the code path is triggered, subsequent database writes are never committed to disk, risking data loss. The developers recommend that the urgent priority here is for the users not to restart their CouchDB instance; they have provided a hotfix for a running application to allow it to flush the data to disk and be shut down safely. There is a detailed guide on the Couch DB website on how to rescue your data if you are affected by this problem, and temporarily stop the issue until the next version.
Earlier releases of Couch DB are not affected by the bug, and it will be resolved in the imminent 1.0.1 release.