VIP Manager suggestions

[CS_DBA]

On 03/20/2014 11:27 AM, Mark Haney wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
>
>
> On 03/20/14 12:51, CS_DBA wrote:
>
>> So we're looking at database failover, application connections to
>> the db are via I.P.'s (or aliases) we can of course use ifconfig
>> and create/destroy aliases and thus "move" the IP aliases at
>> failover time, however I would like to leverage a tool so the IP
>> alias move can be both seamless and executable via command(s) on a
>> single server.
>>
>> With ifconfig aliases we need to do the following in the failover
>> script: - execute the ifdown  on the old / failed master server
>> (probably via commands over ssh) - validate that the ip alias came
>> down - execute the ifup of the alias on the slave (going to be the
>> new master shortly) - verify that it came up - then finish the db
>> failover
>
> I've been in a similar situation before, and here's what I did.
> Instead of failing over from one to another, use round-robin DNS and
> replicate the DB between servers.  My original setup was with 2 DB
> servers with a backend GigE fiber connection with as close to
> real-time replication as I could make it.  This kept the data
> consistent and let me take one down if I needed to, or if one failed
> round robin would send connections to the live DB.
>
> I ended up scaling it up from 2 Db servers to 4, 2 live servers, 2
> 'backup' servers that I could make live if my primaries were hit
> really hard, or would be backups if one of my primaries died.
>
> RR DNS and DB replication is your friend here.  Mucking with IP
> aliases might sound good initially, and it may be fine for a while,
> but I would consider long-term improvements.  My situation was kind of
> special due to the demands of MilSpec with the SGI Altix HPCCs and DB
> servers I managed, but this setup will work in almost every situation.
>
> I'll be glad to offer any assistance off-list if you like.

we're running PostgreSQL and Streaming replication which means that we 
can only write to the master.
Were both of your db's read/write nodes?  What db platform?


>
> - -- 
> Mark Haney
> Network/Systems Administrator
> Practichem
> W: (919) 714-8428
> Fedora release 20 (Heisenbug) 3.13.6-200.fc20.x86_64
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1
> Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
>
> iQEcBAEBAgAGBQJTKyT6AAoJEM/YzwEAv6e7KQQH/RHZ0OE3Ni+Mu+ChDkkZ0z2C
> NwZt4Br11vYfHq/80rPGnFm/W7m4i1Sfg1cunwN6LqKGGkp1B0KRIDDuuM26oxd6
> LZMrUPhGOGWPFxjruV9r658USVsKqbLbFNNFmTapoDK6XyuB+URYbYdz1JCordEy
> 4vHBnTIyPi4H5qMjG6cfcIHVXieda6+/v7pm4JZIfIE3Hv6kztWXUOvhKJAG8gCl
> DLloSCxEABiaOq/M/v/zaK1NEc5Ge4OCHQT+WqY76FveM7QbNcF9k/4Oda3sZfcZ
> 38Wn/qQdWNaVnqfjPJeQlWmYGImnok7sKIgfQpVfLOlfCnD2I+4T0dngm8jsyug=
> =oayJ
> -----END PGP SIGNATURE-----