On Tue, Feb 2, 2010 at 11:50 AM, Steven Dake <[email protected]> wrote:

>
> Here is what we plan to do to satisfy community requirements:
> If the network interface is down, totempg.c will operate in single node
> mode passing incoming messages directly back out the delivery mechanism.
> totemudp/totemiba will continue to monitor the network interfaces for an
> up event from the hardware.  Once hardware becomes available,
> totemudp/iba will bind to the network device and then totempg will enter
> multinode mode.  If the network interface later becomes unavailable,
> totempg will enter single user mode.
>
> In my humble opinion, it will be easy to implement this in totemsrp, not
totempg. What's basic criteria which layer a code should be implemented?

This is what I am thinking to fix this issue in a short time:
- Monitor ring interfaces in a regular interval. The interval can be
configured by downcheck.
- If ring 0 interface is down, then bind it to 127.0.0.1. Fixing ring 0 here
is to simplfy the implementation. If ring 0 interface is up, then bind back
to the oriiginal interface.
- If other ring interface is down, make this interface faulty. This faulty
is not recoverable, meaning reseting faullty can't fix it.
- The reseting faluty should trigger the checking of ring interfaces
immediately, so if interface is up, then it will recover quickly.

Thanks
_______________________________________________
Openais mailing list
[email protected]
https://lists.linux-foundation.org/mailman/listinfo/openais

Reply via email to