On Thu, 24 Nov 2016 22:59:32 +0100 Florian Weimer <fwei...@redhat.com> wrote:
> Does LMDB perform lock-free optimistic reads and > retroactively verifies that the entire read operation was consistent? In LMDB there are readers and reader-writers. A reader never writes; a reader-writer may read, but that read counts as a "write" for purposes of isolation. Each actor -- reader or reader-writer -- is isolated from all others. There are no write-after-read errors or other SQL isolation anomalies because there is only ever at most 1 transaction in existence capable of writing. Any call to begin a second writable transaction blocks until the first one completes. LMDB uses MVCC to provide each reader with a database snapshot. If a writer modifies a row that a reader sees, that modification happens "elsewhere" to a version that the writer sees and the reader does not. After the writer commits and no transactions remain that refer to the superseded version, it ceases to exist. New readers thenceforward see the modified version. --jkl _______________________________________________ sqlite-users mailing list sqlite-users@mailinglists.sqlite.org http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users
