On Wed, Feb 5, 2014 at 9:03 AM, Brett Mcdonald <aheavensbantert...@gmail.com > wrote:
> >The only consequence of losing -mjXXXX file is that in a multi-database > >transaction (which can only be in rollback mode, not in WAL mode) if you > >lose power and come back up without the -mjXXXX file, then the transaction > >might commit to some of the database files, but not to others. > > Are you only stating that an -mjXXXX file cannot support transaction > rollback across WAL dbs in a recovery situation (e.g power failure), or are > you also stating that cross database atomic behavior cannot be supported > for WAL databases, period? > See disadvantage #3 at http://www.sqlite.org/wal.html If the process doing the COMMIT is killed off (for example using "kill -9") in the middle of the COMMIT, then after recovery the WAL database might see the commit and the RAM-disk database might not. Or it might be the other way around. Usually they would both either commit or both rollback. But there is a very narrow window of vulnerability, during which the death of the committing process might cause only one or the other of the databases to commit. -- D. Richard Hipp d...@sqlite.org _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
