|
On Mon, 11 Jun 2001, Roberto Nibali wrote:
>
> I think there is a solution which is not too complex. It's
> publicly referred to as end-to-end health check with event
> generation. What you basically have are an event table an
> action table and a severity table. The severity table
> describes how important the service is and what kind of
> event should be generated. For exmple: w3 is a http server
> which is checked with a html-check. Because it is not so
> mission critical, if it fails the severity is low and the
> event generated is 'take out template VIP:RIP'. The action
> table knows that 'take out template VIP:RIP' is associated
> with 'ipvsadm -d ...'. Now for w1 we define an end-2-end
> check for the db connectivity.
I will put something like that into ldiectord, instead of starting
afresh with a control daemon.
> This server is very mission
> critical and therefore in the severity table it would be
> 'take out all RIPs of this service, namely the DBMS'. The
> action would be 'ipvsadm -C' in Alois' case.
Not quite so drastic, I still want to maintain ssh access,
for example, which I also have set up as a virtual service
(to w1 only).
I think I will define a flag for each virtual service in the ldirectord
configuration, which controls whether this service depends on the
availability of the 'dbms master service', and take down each service
if it has this flag and the master service is down.
Alois
|
