Philip Gwyn wrote:
On 18-Sep-2007 Dave Blaschke wrote:
True, but if it is not a hardware issue (i.e. network hiccup) then a
reset brings the node back. If a node cannot be contacted and fencing
is setup properly, it will be stonith'd to prevent it from corrupting
any shared resources - if it is a software issue (service crash,
network, etc.) then your node could come back and resume its tasks if it
is only reset, while poweroff certainly requires operator intervention.
Of course, that's your call, but that is why the stonith actions is
configurable. :-)
OK, I've been thinking over stonith and fencing. And I can see that if all
services are being run by heartbeat, they won't get reactivated after a stonith
reset.
So, the question becomes does a stonith RESET *have* to be an SNMP RESET? I'll
have to admit that ideally, yes. So I'll go modify the code.
A stonith reset should be whatever it takes for the device to reboot the
node. For some devices, this is simply invoking it with a built-in
reset/cycle command (see ibmhmc), for others that don't have this (for
example, see the bladehpi plugin when softreset is false - the plugin
does have the capability to do a reset, but is configurable to also do
an off/wait/on if the use chooses) it should be a power off/on combo
with any necessary delay.
-Philip
_______________________________________________________
Linux-HA-Dev: [email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
_______________________________________________________
Linux-HA-Dev: [email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
