From: Pavan D. <pav...@gm...> - 2013-10-09 06:42:35
|
On Wed, Oct 9, 2013 at 12:09 PM, Andrei Martsinchyk < and...@gm...> wrote: > I believe even plain vacuum may move tuples within a page and purge empty > pages, therefore CTIDs could be changed. > Plain vacuum does not move tuples around.. Note that all indexes must be updated if a tuple is moved, even within a page. But besides that all your points below still hold true. > Autovacuum on Datamodes are not in sync and concurrent writes may occur in > different order on Datanodes. > CTID can not be used to identify tuple replicas. Primary key could be > used, or special system column. Otherwise not shippable updates against > replicated tables should not be allowed. > > > 2013/10/9 Pavan Deolasee <pav...@gm...> > >> >> >> >> On Wed, Oct 9, 2013 at 11:49 AM, Koichi Suzuki <koi...@gm...>wrote: >> >>> Hmm.. Yes, it could be serious. I suppose that CTID changes only when >>> vacuum full occurs. >>> >> >> CTID will also change every time a row is updated. So the bug should be >> very easy to reproduce. One easiest way to compile different datanodes with >> diferent block size. Another way could be to update/delete from a >> replicated table and vacuum the table on one datanode but not other and >> then again update several rows in the table. Most likely the new versions >> on two different nodes will not have the same CTID anymore. >> >> >>> Vacuum freeze will updates xmim to the smallest value and usual vacuum >>> recycles "dead" row area. We may have a chance to assign different CTIDs >>> to the same rows in replicated table if vacuum full runs locally in >>> datanodes. (My assumption may be wrong. Does vacuum freeze move tuples >>> and change their CTID?) >>> >>> >> No, it does not. >> >> Thanks, >> Pavan >> >> -- >> Pavan Deolasee >> http://www.linkedin.com/in/pavandeolasee >> >> >> ------------------------------------------------------------------------------ >> October Webinars: Code for Performance >> Free Intel webinars can help you accelerate application performance. >> Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most >> from >> the latest Intel processors and coprocessors. See abstracts and register > >> >> http://pubads.g.doubleclick.net/gampad/clk?id=60134071&iu=/4140/ostg.clktrk >> >> _______________________________________________ >> Postgres-xc-developers mailing list >> Pos...@li... >> https://lists.sourceforge.net/lists/listinfo/postgres-xc-developers >> >> > > > -- > Andrei Martsinchyk > > > StormDB - http://www.stormdb.com > The Database Cloud > > -- Pavan Deolasee http://www.linkedin.com/in/pavandeolasee |