Обсуждение: suggestions/hints for multimaster solution
Hello, we're looking for a multi-master solution (whole database not only some tables) for our production database (Postgres 8.3.5 64bit on OpenSuse), and I wanted to know what are your experiences/suggestions with/for such a scenario (multimaster for failover purpose). I know there are some tools, but : * bucardo -> not for failover * pg_cluster -> frozen, currently unavailable, old ..?!? Perhaps we should have a look at * pg_pool(II) * clustering solution from Cybertec, called "cybercluster" We want to implement it on 2 servers with identical hardware -> shared nothing architecture. One of them contains the production database already. any help appreciated...GERD...
On Fri, Jan 9, 2009 at 11:32 AM, Gerd König <koenig@transporeon.com> wrote: > Perhaps we should have a look at > * pg_pool(II) I can only speak of pgpool-II. It's working fine for me in a setup which is exactly the same one you have (change OpenSuse for Debian, but the rest is the same). The thing is that pgpool-II itself can become a single point of failure unless you provide HA for it (I am doing so with Heartbeat), but IMHO that would then be a failover solution, not a multimaster (from pgpool-II's point of view, not the databases behind pgpool-II). Next thing I want to do is try Slony-I with pgpool-II, but I think that is not a multimaster solution either. Of course we may have different points of view on what multimaster mean :) If you plan on using pgpool-II, take into consideration that replication is done at a SQL query level, which is cool, but has its limitations. If you use Slony-I inside pgpool-II I think you can take over those limitations, but the solution is not so... neat, so to speak :) /discuss -- Jaume Sabater http://linuxsilo.net/ "Ubi sapientas ibi libertas"
2009/1/9 Gerd König <koenig@transporeon.com>: > we want to have a scenario where (at least) 2 servers are r/w accessible > and if one of the server crashes this should cause (almost) no downtime > for the database. > Yes, you're right, it's more like "keeping two databases in sync". I would call this replication with failover (active-passive). To me, multimaster sounds like active-active, where clients can send queries to any of the nodes (that is why all nodes are masters). Of course, it's just my point of view. :) > We also use heartbeat for pinging the server themselves. Is it also > possible to use heartbeat for checking the availability of a service > (I'm not that familiar with heartbeat, so long) ? > Does this mean you have two identical installations of pgpool-II on each > of the db-servers ? Yes, I have two almost identical installations of pgpool-II on each of the PostgreSQL servers. pgpool-II handles the pgsql servers. Heartbeat provides HA to pgpool-II. Hope it helps, at least clarifying. :) -- Jaume Sabater http://linuxsilo.net/ "Ubi sapientas ibi libertas"
-----BEGIN PGP SIGNED MESSAGE----- Hash: RIPEMD160 > we're looking for a multi-master solution (whole database not only some > tables) for our production database (Postgres 8.3.5 64bit on OpenSuse), > and I wanted to know what are your experiences/suggestions with/for such > a scenario (multimaster for failover purpose). > I know there are some tools, but : > * bucardo -> not for failover Not sure what you mean by failover in this context, as a true multi-master solution does not need failover as there are always two or more nodes that are read/write at the same time. (Bucardo probably fails your other requirement though, of non-table items, as it does not do DDL). - -- Greg Sabino Mullane greg@turnstep.com End Point Corporation PGP Key: 0x14964AC8 200901091856 http://biglumber.com/x/web?pk=2529DF6AB8F79407E94445B4BC9B906714964AC8 -----BEGIN PGP SIGNATURE----- iEYEAREDAAYFAkln5C8ACgkQvJuQZxSWSsg2fgCgxCwxWMh/kNKpP5A8q8Pg3ajD K6cAoMn8YjfuMHP0t72DS6TIr7v/3jl5 =uZwi -----END PGP SIGNATURE-----