On Wed, Jun 21, 2006 at 11:30:09AM +0800, Syan Tan wrote: > mvcc means no read will block, but a write may block or abort, or if > configured, block and timeout. you need to start the transaction at the > time of reading for mvcc to detect a concurrent write.
> In the past, once a transaction aborted, all writes in the transaction also > abort, > without heirarchical transaction control. The problem is that getting an emr > is > a group fetch of many items that may get updated. What's needed is a a > subcommit, > that says it is trying to commit a couple for rows of a mass read within in a > transaction, > ( a subtransaction). Karsten really has implemented the only solution if you > can't > have compound transactions. Yep. The question, however, is whether we need explicit "select for update", too. Probably not, it seems. Karsten -- GPG key ID E4071346 @ wwwkeys.pgp.net E167 67FD A291 2BEA 73BD 4537 78B9 A9F9 E407 1346 _______________________________________________ Gnumed-devel mailing list [email protected] http://lists.gnu.org/mailman/listinfo/gnumed-devel
