|
To answer the first question, I'm pretty sure you can have heartbeat fake as
many addresses as needed and specify which interface to bring up the fake
addresses on.
To answer the second question, I know that the program I use as a substitute
to heartbeat (I explain later) exhibits the same characteristics because
when the server portion is turned off the connection is refused and the
client does not have to wait for a timeout. When the linuxdirector which was
running the server portion is actually disconnected from the network, the
client linuxdirector has to wait to timeout every time it tries to connect.
Just as a note -- I achieve high-availability between dual linuxdirectors
without heartbeat (if you noticed that the rest of this message wasn't very
helpful it was because my real point of this message was to mention this).
Basically I have two linuxdirectors (I use LVS-DR, BTW):
216.69.192.194 (linuxdirector-a.netwalk.net)
216.69.192.195 (linuxdirector-b.netwalk.net)
And another IP:
216.69.192.196 (linuxdirector.netwalk.net)
In the router directly connected to 216.69.192.192/26 I route all of my
virtual servers to 216.69.192.196.
Now all I have to do is bring up 216.69.192.196 as an IP alias on either
linuxdirector. Whenever the switchover occurs, I also have an automatic
script which contacts the router and clears its arp-cache.
Rather than using heartbeat to handle the switchover, I wrote a tiny
client-server program which runs on both linuxdirectors and determines if it
is a master or a slave (rather than using a serial connection to determine
if the linuxdirector is alive, it connects over a TCP socket and has a
little question-answer session and then disconnects) (I called this program
"etherbeat" :) ).
While this might sound more complicated, I like it better because that way I
don't have to have any subordinate linuxdirectors and I don't have to fake
anything during downtime. My two linuxdirectors are equal -- the image I use
to build linuxdirectors is exactly the same. I just feel better about this
configuration than the typical heartbeat one.
I actually took MOST of this idea from the preconfigured actions of my
telephone switch. It uses three IPs and brings up an alias on the
appropriate switch when necessary, just like my linuxdirectors.
----- Original Message -----
From: "???" <conan@xxxxxxxxxxxxxxxxxxx>
To: "Lars Marowsky-Bree" <lmb@xxxxxxx>
Cc: <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Sent: Tuesday, July 11, 2000 3:42 AM
Subject: Re: question about heartbeat
> I see.
> Then how about the case when servers have multiple NIC with different
subnet ?
> Does it send ARP reply for each segment ?
> The test result from tcpdump make me confused, cause it didn't show any
ARP packets.
>
> Another thing is that IP takeover took place immediatly when I stoped
heartbeat of primary
> while it took much time when I disconnected all the network connections of
primary.
> What's the difference ?
>
>
> ----- Original Message -----
> From: "Lars Marowsky-Bree" <lmb@xxxxxxx>
> To: "???" <conan@xxxxxxxxxxxxxxxxxxx>
> Cc: <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
> Sent: Tuesday, July 11, 2000 4:04 PM
> Subject: Re: question about heartbeat
>
>
> > On 2000-07-11T15:49:07,
> > ??? <conan@xxxxxxxxxxxxxxxxxxx> said:
> >
> > > I wonder the mechanism of IP takeover.
> > > How the packets find secondary when primary is dead ?
> > > I know that secondary takes the IP (which may called VIP) of the
primary.
> > > And what next ?
> > > Is ARP broadcasted by secondary ? Is it done automatically or done by
heartbeat ?
> >
> > Yes. The system sends out an unsolicited ARP reply and the other hosts
on the
> > same ethernet segment update their arp caches.
> >
> > Sincerely,
> > Lars Marowsky-Brée <lmb@xxxxxxx>
> > Development HA
> >
> > --
> > Perfection is our goal, excellence will be tolerated. -- J. Yahl
> >
> >
> >
> >
>
>
|
