|
Hi,
Two problems:
When a realserver fails, even if it is detected by both keepalived, the
expire_nodest_conn sysctl is not honored.
Is this fixed by now?
Every 4 seconds, the vrrp is going from master to backup and back.
Either configuration issue or misdetection of link state.
The real problem:
Directors are on HP bl10e, small machines (20 nodes on 3U), small hardware
(e100 cards), and a not so configurable enclosure switch. Sometimes, the
Ethernet configuration goes to a degraded state. Even if you fix the ports
(switch and card) to 100baseTx-FD, it seems to loose something.
I thought keepalived does not use the IFF_UP|IFF_RUNNING anymore?
Workaround :
Watch for network entering degraded state.
Use mii-tools with :
mii-tools -A 100baseTx-FD
mii-tools -F 100baseTx-FD
I don't like it, but it works.
I don't feel my explanations very clear :).
This is a can of worms and I would really like to have this cleaned up
once and forever. Every other (commercial) Unix system is capable of DGD
and link state monitoring but within Linux we have to use a handful of
tools (depending on the MII interface chip) to achieve something as
simple as a link state.
I was under the impression that Jeff Garzik was going to clean that all
up and that the ethtool interface architecture would provide use with a
generalised framework for this. I reckon for Alexandre Cassen an event
based netlink message would be the best for vrrpd, correct?
Best regards,
Roberto Nibali, ratz
--
echo
'[q]sa[ln0=aln256%Pln256/snlbx]sb3135071790101768542287578439snlbxq' | dc
| 