Re: [Sqlgrey-users] multiple read/write database support patch

[Paul B. H. <he...@ac...>]

On Fri, 2 Feb 2007, Dan Faerch wrote:

> 1: Would allowing you to set the READ_HOSTS_NEVER_WRITE = 0 from config
> do the same for you? So ALL hosts defined are read & write hosts and
> switched between in a round robin manner.

I'm not sure, I didn't really look into DBIx. I don't really want to round
robin though; for any given instance of sqlgrey, I want it to use the same
database for all queries until a failure, and only switch to a different
one in the case of such a failure. I'm mostly interested in fault tolerance
at this point, not load balancing.


> 2: How scalable is your setup? We currently run 1 write-host and 7
> read-hosts. Can multi-master be used in such a setup? (a little off
> topic i guess;)

I believe you can scale to higher numbers of multimaster servers, but if I
remember correctly it is not a full mesh but they end up replicating
circularly so the failure of any one breaks the replication chain.


> If we need to specify more than 1 "write" hosts, would there be any
> logic in simply allowing multiple hosts to be defined on "db_host"
> instead (comma separated)? Just so we dont need an extra config
> directive to do the job that an existing directive could do.

That works for me, I just didn't want to change existing functionality.
Given that the degenerative case of setting db_hostlist to just one
instance is exactly the same as using db_host, changing the definition of
that configuration variable to accept a list does seem the simpler thing to
do.


-- 
Paul B. Henson  |  (909) 979-6361  |  http://www.csupomona.edu/~henson/
Operating Systems and Network Analyst  |  he...@cs...
California State Polytechnic University  |  Pomona CA 91768