On 3/4/2015 9:10 PM, AI Rumman wrote:
> I am working on setting up a new database server with shared disk and
> cluster failover.
> In this environment, only one node will be active at a certain time.
> Underneath, we are planning to have shared storage with CIFS protocol.
>
> As I am newbie with this storag, can anyone please help me with some
> info what are the database issues I can face with this kind of file
> system protocol with Postgresql 9.1
>
why are you building a NEW system with the 2nd oldest release of
postgres? within a year or so, 9.1 will be obsolete and unsupported.
CIFS will be pretty slow at the sorts of random writes that a database
server does a lot of, and there's all sorts of room for hard-to-diagnose
issues with unsafe write cache buffering in the file server, depending
on the specifics of the CIFS server implementation. Not sure how you
implement a high availability CIFS server without single points of
failure, either... thats hard enough with shared block storage
implementations (requiring redundant storage networks, switches, and
dual storage controllers with shared cache, dual homing the actual
physical block storage, which is dual ported and all raid 10 typically).
ISCSI or a proper SAN (fiberchannel) would be a much better choice for a
shared storage active/passive cluster, just implement some sort of
storage fencing to ensure only one node can have the file system mounted
at a time.
with postgres, its usually better to implement a HA cluster via
streaming replication, the master and slave each with their own
dedicated storage, and promoting the slave to master if/when the master
dies.
--
john r pierce 37N 122W
somewhere on the middle of the left coast
