Details
-
Bug
-
Resolution: Fixed
-
Major
-
2.0-developer-preview-4
-
Security Level: Public
-
None
Description
If we have really many documents (and relatively large), and if
compaction ends up writing many of them at file offsets that cause
the documents to be split into 2 or more blocks, the MVCC snapshot
byte size post compaction can exceed the file size. This happens
because it's reusing the old doc info sizes, which don't account for
the new block prefix bytes.
This behaviour was introduced with the new database compaction approach
(btree copying + disk sorting).