|
ntadmin@xxxxxxxxxxxx wrote:
Ok, I see now how your proof of concept works.
Cool.
I can be a little dense
sometimes..;)
;) Me too.
Do you have any idea what the differences are between using
ifconfig and ip addr?
Yes.
When I add an IP with ifconfig it shows up with ip
addr show...
Correct, but not vice versa. And the real problem is that if you add
multiple IP addresses to a physical interface _without_ a label then
ifconfig won't see it anymore. Also you are kind of restricted on the
labelling with ifconfig. And starting with a certain string length for
the label, ifconfig will simply truncate the names, which can be highly
annoying if you have a VPN cluster with thousands of VPN connections
that are labelled according to a clusterwide semantics. Ohh, and for
example if you use cipe for VPNs (which you seem to be using according
to your kernel) and you try to make a fast failover, ifconfig will be
tripping through races while inserting interfaces which results in some
VPN devices being up and some not (you can see this behavior on fast
machines which try to set up hundreds of VPNs without delays).
But for normal people 'ifconfig ...' == 'ip addr add', basically :).
I'm not sure I know if this means that they do the same thing
or not.
No, they won't. ifconfig uses PF_UNIX socket and an ioctl and ip addr
add simply uses the fast _and_ highly reliable user space to kernel
space transportation mechanism called NETLINK. See below:
# strace -e socket,ioctl ifconfig eth0:4 155.23.5.25 netmask
255.255.255.240 broadcast 155.23.5.31
socket(PF_UNIX, SOCK_DGRAM, 0) = 3
socket(PF_INET, SOCK_DGRAM, IPPROTO_IP) = 4
socket(PF_INET6, SOCK_DGRAM, 0) = 5
ioctl(4, 0x8916, 0xbffff1a0) = 0
ioctl(4, 0x8913, 0xbffff0c0) = 0
ioctl(4, 0x8914, 0xbffff0c0) = 0
ioctl(4, 0x891c, 0xbffff1a0) = 0
ioctl(4, 0x891a, 0xbffff1a0) = 0
ioctl(4, 0x8913, 0xbffff0c0) = 0
ioctl(4, 0x8914, 0xbffff0c0) = 0
# strace -e socket,ioctl ip addr add 155.23.5.25/28 brd + dev eth0 label
eth0:4
socket(PF_NETLINK, SOCK_RAW, 0) = 3
I reckon this speaks for itself ...
In addition, maybe you know, maybe somebody else can help me?
Probably someone else is better.
I am
wondering about the IPaddr script that comes with Heartbeat. When it
executes it runs the gratuitous arp send_arp command 5 times. I am
wondering if it would be safe to reduce this as it is adding additional
minutes to the failover time that we would like to reclaim.
So I went along and checked out the current linux-ha cvs branch. And to
my joy I find that Horms has started IPaddr2 which will be using the
iproute2 framework ;). I'm confident, that with IPaddr2 things will be
working just great.
Checking out ../heartbeat/libnet_util/send_arp.c I would say that the -r
option should do the trick. Now let's see how IPaddr calls send_arp:
Aha:
ARGS="-i $ARP_INTERVAL_MS -r $ARP_REPEAT -p $SENDARPPIDFILE
$TARGET_INTERFACE $BASEIP auto $BASEIP $ARP_NETMASK"
which is in ip_start() which gets called by IPaddr <ip-address> start.
And ARP_REPEAT can be overwritten by presetting that variable. As simple
as that ;).
Best regards,
Roberto Nibali, ratz
--
echo '[q]sa[ln0=aln256%Pln256/snlbx]sb3135071790101768542287578439snlbxq'|dc
|
