What data is actually stored in a B-tree database in CouchDB?
- by Andrey Vlasovskikh
I'm wondering what is actually stored in a CouchDB database B-tree? The CouchDB: The Definitive Guide tells that a database B-tree is used for append-only operations and that a database is stored in a single B-tree (besides per-view B-trees).
So I guess the data items that are appended to the database file are revisions of documents, not the whole documents:
+---------|### ...
| |
+------|###|------+ ... ---+
| | | |
+------+ +------+ +------+ +------+
| doc1 | | doc2 | | doc1 | ... | doc1 |
| rev1 | | rev1 | | rev2 | | rev7 |
+------+ +------+ +------+ +------+
Is it true?
If it is true, then how the current revision of a document is determined based on such a B-tree?
Doesn't it mean, that CouchDB needs a separate "view" database for indexing current revisions of documents to preserve O(log n) access? Wouldn't it lead to race conditions while building such an index? (as far as I know, CouchDB uses no write locks).