[Linux-HA] Process "resume" after failover
beekhof at gmail.com
Sun Sep 30 11:38:15 MDT 2007
On 9/29/07, Radu Handorean <xraduhx at yahoo.com> wrote:
> So we have NodeA (active) and NodeB (standby). When NodeA fails (say one of the monitored processes dies and the others are killed and the node rebooted) will the processes on NodeB start from scratch? I wonder about the transfer of state information
heartbeat is service-agnostic
which means we dont want/need to understand what it is we're managing
so the only state that is transfered is state which you or your
application has transfered itself
to some extent we support the ability to migrate resources, but that
only really makes sense when A is healthy... (its hard to migrate data
from a dead machine)
so for the scenario you mentioned below to work, your application
would have had to have already replicated enough information before A
but just to re-iterate, your application needs to be cluster-aware for
this to work.
> from A to B which I don't believe is possible and thus processes on NodeB would look like newly started processes. If one has a webserver that starts on NodeB it may not be such a big deal to start it fresh, but I am thinking about the following scenario: NodeA plays back some mp3 file and about 50% into the song the process crashes and NodeB would have to continue playing from that point forward. In this case the "resume" or "continue" behavior is clearly preferred (required :) ). However, it seems unlikely that this will happen without the application being aware of such scenarios (e.g. use checkpoints). The entire HA mechanism is definitely oblivious of what happens "above", at the app level.
> Can anybody comment on this? Maybe pointers to literature I missed?
> Got a little couch potato?
> Check out fun summer activities for kids.
> Linux-HA mailing list
> Linux-HA at lists.linux-ha.org
> See also: http://linux-ha.org/ReportingProblems
More information about the Linux-HA