Re: [Draft PATCH net-next] Bonding: add missed_max option

[Hangbin Liu <liuhangbin@gmail.com>]

Hi Jay, Denis,

Thanks for the comments. Please see my replies below.

On Fri, Oct 29, 2021 at 03:11:18PM -0700, Jay Vosburgh wrote:
> >+missed_max
> >+
> >+        Maximum number of arp_interval for missed ARP replies.
> >+        If this number is exceeded, link is reported as down.
> >+
> >+        Note a small value means a strict time. e.g. missed_max is 1 means
> >+        the correct arp reply must be recived during the interval.
> >+
> >+        default 3
> 
> 	I'd suggest "arp" in the option name to make the scope more
> obvious.

I didn't add arp prefix in purpose because I'd like to re-use this parameter
after adding bonding IPv6 NS/NA support. I will add this reason in the commit
description.

Or if you like to make a difference for ARP and IPv6 NS, I can add the arp_
prefix.

> >@@ -3224,7 +3224,7 @@ static int bond_ab_arp_inspect(struct bonding *bond)
> > 
> > 		/* Backup slave is down if:
> > 		 * - No current_arp_slave AND
> >-		 * - more than 3*delta since last receive AND
> >+		 * - more than missed_max*delta since last receive AND
> > 		 * - the bond has an IP address
> > 		 *
> > 		 * Note: a non-null current_arp_slave indicates
> >@@ -3236,20 +3236,20 @@ static int bond_ab_arp_inspect(struct bonding *bond)
> > 		 */
> > 		if (!bond_is_active_slave(slave) &&
> > 		    !rcu_access_pointer(bond->current_arp_slave) &&
> >-		    !bond_time_in_interval(bond, last_rx, 3)) {
> >+		    !bond_time_in_interval(bond, last_rx, bond->params.missed_max)) {
> > 			bond_propose_link_state(slave, BOND_LINK_DOWN);
> > 			commit++;
> > 		}
> > 
> > 		/* Active slave is down if:
> >-		 * - more than 2*delta since transmitting OR
> >-		 * - (more than 2*delta since receive AND
> >+		 * - more than missed_max*delta since transmitting OR
> >+		 * - (more than missed_max*delta since receive AND
> > 		 *    the bond has an IP address)
> > 		 */
> > 		trans_start = dev_trans_start(slave->dev);
> > 		if (bond_is_active_slave(slave) &&
> >-		    (!bond_time_in_interval(bond, trans_start, 2) ||
> >-		     !bond_time_in_interval(bond, last_rx, 2))) {
> >+		    (!bond_time_in_interval(bond, trans_start, bond->params.missed_max) ||
> >+		     !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) {
> > 			bond_propose_link_state(slave, BOND_LINK_DOWN);
> > 			commit++;
> > 		}
> 
> 	The above two changes make the backup and active logic both
> switch to using the missed_max value (i.e., both set to the same value),
> when previously these two cases used differing values (2 for active, 3
> for backup).
> 
> 	Historically, these intervals were staggered deliberately; an
> old comment removed by b2220cad583c9b states:
> 
> 			if ((slave != bond->curr_active_slave) &&
> 			    (!bond->current_arp_slave) &&
> 			    (time_after_eq(jiffies, slave_last_rx(bond, slave) + 3*delta_in_ticks))) {
> 				/* a backup slave has gone down; three times
> 				 * the delta allows the current slave to be
> 				 * taken out before the backup slave.
> 
> 	I think it would be prudent to insure that having the active and
> backup timeouts set in lockstep does not result in an undesirable change
> of behavior.

Yes, I'm also a little concern about this. What about make the backup slave
timeout 1 plus delta then active slave. i.e. for backup slave

bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)

> >@@ -270,6 +272,15 @@ static const struct bond_option bond_opts[BOND_OPT_LAST] = {
> > 		.values = bond_intmax_tbl,
> > 		.set = bond_option_arp_interval_set
> > 	},
> >+	[BOND_OPT_MISSED_MAX] = {
> >+		.id = BOND_OPT_MISSED_MAX,
> >+		.name = "missed_max",
> >+		.desc = "Maximum number of missed ARP interval",
> >+		.unsuppmodes = BIT(BOND_MODE_8023AD) | BIT(BOND_MODE_TLB) |
> >+			       BIT(BOND_MODE_ALB),
> >+		.values = bond_intmax_tbl,
> 
> 	This allows missed_max to be set to 0; is that intended to be a
> valid setting?

In bond_time_in_interval() there is an arp_interval/2 extra time. Do you think
if this is enough for fast network when we set missed_max to 0?

Of course in the very fast network the value should be at lease 1 in case
the ARP send/recv time frame is same.

Thanks
Hangbin