From: Shmulik Hen <shmulik.hen@in...> - 2003-06-19 13:42:21
There seem to be two related problematic issues of IP settings on
slaves. The first has to do with slaves being left with the bonding
settings after being released, which might cause a problem if they are
brought up as stand alone interfaces. The second issue is that ifenslave
doesn't bother to check is an interface is actually a valid slave when
using the -c option. If the ethX is a stand alone interface and not part
of the bond, it will still be set by ifenslave to all the settings of the
bond causing existing connections to break.
I know both issues result in errors only if the user make a
mistake, but shouldn't bonding take more care to prevent such problems?
| Shmulik Hen |
| Israel Design Center (Jerusalem) |
| LAN Access Division |
| Intel Communications Group, Intel corp. |
From: Peter C. Norton <spacey-bonding@le...> - 2003-06-19 13:59:42
On Thu, Jun 19, 2003 at 04:42:16PM +0300, Shmulik Hen wrote:
> There seem to be two related problematic issues of IP settings on
> slaves. The first has to do with slaves being left with the bonding
> settings after being released, which might cause a problem if they are
> brought up as stand alone interfaces.
Yes, this is an old issue. We need to store the original MAC address
for the bond and restore it when the card is un-bonded. IMO it would
also be better to make up a bogus MAC address in a private space,
since that way the first card in the bond could be faild out of the
bond and restored without incurring the wrath of the switch its
sitting on when it comes back up.
The 5 year plan:
In five years we'll make up another plan.
Or just re-use this one.