[j-nsp] RE : Perempt hold-down time on vrrp

[Daniel Roesen]

On Wed, Jul 06, 2011 at 10:09:05PM +0200, david.roy at orange-ftgroup.com wrote:
> Hold-time on Junos is only available when Vrrpd restart (It is usually
> used when VRRP crash or RE reboot or RE switchover without NSR)

I cannot confirm that from own observation. Interface state change is
enough to trigger hold-timer.

We've actually found a problem with hold-time in a corner case
situation where it holds "far too long".

Assume a setup with two routers, A and B. Rtr A has higher VRRP prio
than B, preemption is enabled with hold-timer e.g. 300sec.

Rtr A interface goes down, Rtr B assumes mastership.
Rtr A interface comes up again, and hold-timer kicks in.
Rtr A sees the VRRP PDUs from Rtr B.

Now while in hold-time period, Rtr B VRRP interface goes down and thus
VRRP PDUs sent from Rtr B stop. We would have expected Rtr A to detect
that and immediately assume mastership, but that doesn't happen.
Observing that, we would have expected that it takes until hold-time
expiry for that to happen - but it only took ~1m, which was still ~1.5m
short of expiry.

Didn't have time to dig deeper and confirm the last observation with
more tests, but hold-timer (at least for some time) ignoring sudden
absence of lower-prio master definately caused unexpected outage.

Best regards,
Daniel

-- 
CLUE-RIPE -- Jabber: dr at cluenet.de -- dr at IRCnet -- PGP: 0xA85C8AA0