Thread: RE: [ADMIN] Question about multiple postmasters
> > Has anyone tried running multiple postmasters w/ the same data
> > directory.
> > Any issues to watch out for? Problems with locking?
> > All of the systems would be running on Red Hat 5.1 x86.
>
> Hard to imagine why you would want to do that. pg_log would be shared
> by all backends, even if they connect to different databases, so this
> is
> very bad.
Where is the pg_log located? The only thing they would share would be
data directories. I plan on running each backing on different machines
sharing a data directory. It's more for load distribution than anything
else.
-DEJ
> > > Has anyone tried running multiple postmasters w/ the same data > > > directory. > > > Any issues to watch out for? Problems with locking? > > > All of the systems would be running on Red Hat 5.1 x86. > > > > Hard to imagine why you would want to do that. pg_log would be shared > > by all backends, even if they connect to different databases, so this > > is > > very bad. > Where is the pg_log located? The only thing they would share would be > data directories. I plan on running each backing on different machines > sharing a data directory. It's more for load distribution than anything > else. It will not work. pg_log is in the toplevel data directory, and some of the stuff is stored in shared memory, which is not shared among postmasters. -- Bruce Momjian | http://www.op.net/~candle maillist@candle.pha.pa.us | (610) 853-3000 + If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania 19026
On Thu, 1 Oct 1998, Bruce Momjian wrote: > > > > Has anyone tried running multiple postmasters w/ the same data > > > > directory. > > > > Any issues to watch out for? Problems with locking? > > > > All of the systems would be running on Red Hat 5.1 x86. > > > > > > Hard to imagine why you would want to do that. pg_log would be shared > > > by all backends, even if they connect to different databases, so this > > > is > > > very bad. > > Where is the pg_log located? The only thing they would share would be > > data directories. I plan on running each backing on different machines > > sharing a data directory. It's more for load distribution than anything > > else. > > It will not work. pg_log is in the toplevel data directory, and some of > the stuff is stored in shared memory, which is not shared among > postmasters. I believe what he is asking is if he can use an NFS mount or some other network based file system to house the data directory used by a few machines thereby spreading the queries across them. Now I don't know much about the internals but I can suggest that he be very careful... NFS is basically stateless and it would be *REALLY* easy to screw up a data file. Regardless his original question still stands - can it be done? BTW replication would probably be the real solution he wants. I could see trying to NFS mount a common disk area as a poor man's replication, but then the reliability of the entire server array becomes focused on a single server (the NFS master) and speed becomes based on the slowest link- transfering huge files over NFS. Sort of defeats the purpose of an sharing the load across a set of servers doesn't it? Chris -- Useless Fact #3: Donald Duck's middle name is Fauntleroy.
> On Thu, 1 Oct 1998, Bruce Momjian wrote: > > > > > > Has anyone tried running multiple postmasters w/ the same data > > > > > directory. > > > > > Any issues to watch out for? Problems with locking? > > > > > All of the systems would be running on Red Hat 5.1 x86. > > > > > > > > Hard to imagine why you would want to do that. pg_log would be shared > > > > by all backends, even if they connect to different databases, so this > > > > is > > > > very bad. > > > Where is the pg_log located? The only thing they would share would be > > > data directories. I plan on running each backing on different machines > > > sharing a data directory. It's more for load distribution than anything > > > else. > > > > It will not work. pg_log is in the toplevel data directory, and some of > > the stuff is stored in shared memory, which is not shared among > > postmasters. > > > I believe what he is asking is if he can use an NFS mount or some other > network based file system to house the data directory used by a few > machines thereby spreading the queries across them. Now I don't know much > about the internals but I can suggest that he be very careful... NFS is > basically stateless and it would be *REALLY* easy to screw up a data file. > > Regardless his original question still stands - can it be done? > > BTW replication would probably be the real solution he wants. I could see > trying to NFS mount a common disk area as a poor man's replication, but > then the reliability of the entire server array becomes focused on a > single server (the NFS master) and speed becomes based on the slowest > link- transfering huge files over NFS. Sort of defeats the purpose of an > sharing the load across a set of servers doesn't it? Can't be done. Period. There is a shared memory are that holds the most recent table contents. No way of one backend under one postmaster knowing about changes another backend under another postmaster is making. -- Bruce Momjian | http://www.op.net/~candle maillist@candle.pha.pa.us | (610) 853-3000 + If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania 19026