My problem is a bit easier then you mentioned with the mail server
example because I dont have many processes to check. The application is
handling well the active/passive state and I dont have to care about the
data incosistency.
The problem comes from this situation. Both of the servers are alive and
heartbeat (that doesnt support app level HA) is changing the IP address
when the active server failed.
Andras
On 2014.11.15. 17:21, Miles Fidelman wrote:
Andras POTOCZKY wrote:
Hi List,
I would like to configure an application based HA solution. I mean if
my important application stops working (but the active server is
still alive), the HA configuration is passing the shared IP to the
standby node.
I've found pacemaker is the tool for it but I couldnt find any
document or example related to the application level HA.
Can somebody point me how can I figure this out?
The short answer: Look for documentation on resource agents. If your
application has a pre-existing resource agent defined, then you set
that up and it handles failover.
The longer answer is that application-level failover is difficult,
because it involves things like:
- IP addresses
- process state
- work queues
- data files
- etc.
All of which are unique to a specific application.
Hence, it becomes a lot easier to:
- address failover at the virtual machine level (IP addresses,
applications, processes)
- use either a storage area network, or DRBD to make sure that your
data remains available and consistent
Some applications provide their own application-layer failover - such
as databases with replication logic built in:
- put them behind a load leveler, and you're done (assuming your
load-leveler isn't a single point of failure), or,
- if your public-facing DNS name points to multiple IP addresses, then
you have a level of failover there, or,
- use CARP to support IP address failover
For other applications, it's quite a bit harder: I've been trying to
come up with a way to make our email and mailing-list infrastructure
HA, at the application layer, but there are so many distinct processes
running, and ques, and such, that just VM-level failover, with DRBD
disk replication has always been far simpler. (The reason I keep
looking is that I'd like to migrate to either a BSD or illumos based
platform, but there's no Xen and no DRBD.)
Miles Fidelman
HAST
I've been wrestling with this one for a while - more in terms of
setting up HA applications
_______________________________________________
Linux-HA mailing list
Linux-HA@lists.linux-ha.org
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems