I'm developing a db-driven web site for a client.
So far the solution happens to use a lot of open sources software (best tool
for the job).
But when looking at areas of high-availability and performance in relation
to our database back-end, I'm trying to find a solution that will fit the
clients need (say, 4 "nines" of reliability or so). The application the db
server is running is mostly SELECTs, but a fair share of inserts
(interchange e-commerce is the application). The open source
performance/reliability solution I came up with:
- master database server (high end box) is read/write.- primary slave database server (high end box) is read-only, and
getsit's
data by means of replication from master database server. This box is
specially marked to take over for the master in the event that the master
fails (hot failover).- many slave database servers (low end boxes) are read-only. These get
their data from the primary slave database server, instead of the master
database server, so that the master only has to replicate once (and then,
only to one machine: the primary slave db server).
What do you guys think of my solution? It's more complicated than Oracle's
parallel clustering, etc. But Oracle costs $30,000 (for our install,
anyway). So I would like to implement the above on open source software.
But, I've read that postgresql replication code is not yet in "usable"
status. MySQL on the other hand claims their replication has "alpha" code
quality, but that many customers use it successfully on a day-to-day basis
(that was the feeling I got, anyway). And neither pgsql or mysql have
claimed any hot failover abilities. So my questions are twofold:
1) What is the status of the features I described? (replication, seamless
failover).
2) My client is able to "donate" several thousand dollars to the development
of said features (heck, I might kick in a few bucks). What are our options
for this? Anyone willing to step up to the plate and say, "yes, I'll do it
on a contract for 10k!". Or is there already an established way of getting
X feature for Y dollars?
3) Or, should I just bite the bullet and use Mysql? (minus foreign keys,
minus transactions, minus ....)
Thanks,
Dan Browning
Network & Database Administrator
Cyclone Computer Systems
