Uploaded image for project: 'Core Server'
  1. Core Server
  2. SERVER-8972

extent layout differences can cause incorrect capped collection replication



    • Icon: Bug Bug
    • Resolution: Won't Fix
    • Icon: Major - P3 Major - P3
    • None
    • None
    • MMAPv1, Replication, Storage
    • None
    • Storage Execution
    • ALL


      There is a potential problem in cases where the extent layout of capped collections on primary and secondary is not consistent - for example if the total sizes of the collections are allowed to be slightly different. Or if the total extent size on the secondary is the same as on the primary, but the extent sizes and ordering differs. In either case I think there are situations where the capped collection allocation policy for allocating a new record by removing the oldest record(s) can cause a different number of documents to be deleted on a secondary than on a primary in cases where the physical extent layout differs. This could result in data inconsistencies between primary and secondary.

      Also, I'm not sure it might be possible for such an inconsistency to trigger subsequent inconsistencies. (For example an upsert applied to the secondary for a doc that was deleted by allocating a new doc might incorrectly insert the doc.)

      Also it might make sense to do a complete analysis of replicating capped collections at some point.




            backlog-server-execution Backlog - Storage Execution Team
            aaron Aaron Staple
            1 Vote for this issue
            8 Start watching this issue