Hi Minh
On 11/6/19 10:33 am, Minh Hon Chau wrote:
Hi Gary,
Those variables e.g node_sync_window_closed have been used before
headless sync complete. If there is a failover during the headless
sync, the new active will start the headless sync again, so those
variables have not been needed to checkpoint. But here the scenario
happens in split brain, in which the new active is in separated
network instead of coming from headless, so I guess we do need
checkpoint it, but the checkpoint should be done after the headless
sync ?
I will checkpoint node_sync_window_closed in a new version. As you
pointed out, using the timer alone isn't sufficient as sync messages
could come before the active controller's amfnd has sent node_up (and
therefore starting the timer).
And the change in timer.h seems not much relates to this ticket?
The values in the timer structure aren't initialized at startup. So
things like is_active has random values. It would be good just to set
them to known values.
Thanks
Gary
_______________________________________________
Opensaf-devel mailing list
Opensaf-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/opensaf-devel
