Carl wrote:
>> 1) multiple app servers sharing the same
>> MySQL master-master backend
This is exactly the situation I'm using mmm in. I have a master-master
pair (no slaves) and 8 app servers.
On 13 May 2009, at 21:25, Rick James wrote:
> * "Automatic" is not possible.
Well, it's working with mmm for me.
> * Need to monitor for failure, then be able to double check that it
> is not a false alarm.
mmm monitors by connecting to back-ends and running queries. This
works, but we've had some problems with false alarms when queries are
held up by long-running (> 1 hour) transactions, or if a server is
iowait saturated.
> * DNS for redirecting traffic --> can take many minutes (TTL +
> caching)
I agree, this is not much use for DBs.
> * Port (3306) forwarding -- a significant possibility
This is effectively what mmm does - it accepts connections on a
floating IP and forwards to the appropriate back-end.
> * Software or hardware load balancing, such that the passive master
> appears to be "not available" -- a significant possibility. Change
> the health check after establishing it as the read-write master.
The monitoring needs to be pretty intimate with the back-ends if it's
not going to give false alarms - simple ping or connect tests are not
enough; you need to select databases and run queries.
> * Different port or different logical host name -- helpful in
> segregating readonly connects versus read-write connects.
Again, mmm does this - you can use a separate IP for each role, and
have any number of back-ends supporting each.
I can't comment on the redundant networking aspect as I've not tried
that, but I would expect that this to be dealt with separately at a
lower level.
Marcus
--
Marcus Bointon
Synchromedia Limited: Creators of http://www.smartmessages.net/
UK resellers of info@hand CRM solutions
marcus@stripped | http://www.synchromedia.co.uk/
