On Mon, Jul 28, 2008 at 12:46:42PM -0500, Brian Smith wrote: > Glenn Maynard wrote: > > After finishing a transaction and deleting the rollback > > journal, no directory fsync is performed. If the system > > crashes immediately after a commit returns to the caller, the > > unlink may not have been flushed to disk, causing the journal > > to be rolled back and undoing a commit that had already returned. > > > > There's no danger of corruption here, but doesn't this break > > durability? > > Instead of undoing a commit that has already returned, it will *redo* the > commit that has already returned (I think).
I don't get that from http://www.sqlite.org/atomiccommit.html. Is there a detail I'm missing, or that's not documented there? -- Glenn Maynard _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users