In that case, might I suggest an alternative configuration. Do DHCP with
reservations and setup both sets of IPs based on where the VMs will live. The
DHCP configuration will control the gateway based on what subnet they’re
actually attached to and avoid the issue. Even for machines with static IPs I
normally create DHCP reservations in my configuration just to make it easier to
track what IPs are in use.
--
There are 10 kinds of people in the world...
those who understand binary and those who don't.
From: [email protected] [mailto:[email protected]] On
Behalf Of Michael Leone
Sent: Friday, June 24, 2016 12:31 PM
To: [email protected]
Subject: Re: [NTSysADM] Fwd: Confused about multiple gateways
On Fri, Jun 24, 2016 at 12:04 PM, Rubens Almeida
<[email protected]<mailto:[email protected]>> wrote:
Here's my 2 cents on this matter: I'm still waiting to see when a Windows
server host will handle 2 gateways without trouble. I'm used to see on every
customer I'm assigned to work as SME on my day job. Every one of them have this
kind of issue on one degree or another. What I do is: on the production NIC I
set the customer's gateway. On all other NICs no gateway at all. If needed, I
then set a persistent routes pointing to the respective gateway handling that
specific network. Hope that helps!
As I said, there are no other NICs. Also, in case of disaster, I don't want to
have to edit 175 VMs, to set addressing on a previously unused NIC
(script-based or not). I need an automatic dead-gateway detection and failover,
apparently.
Rubens
On Fri, Jun 24, 2016 at 12:23 PM, Michael Leone
<[email protected]<mailto:[email protected]>> wrote:
Here's my setup: I have a lot of VMware VMs. We also use their SRM (Site
Recovery Manager) for Disaster Recovery. Basically, SRM lets the VMs fail over
to another site, in case of disaster. They will keep their current IP
addressing.
So what we did was set 2 gateways on each VM - first entry is x.x.x.1, which is
the gateway at the production site. Second entry is x.x.x.2, which is the
gateway at the recovery site. This way, if the VMs did fail over, they would
still be able to find a gateway and continue to work (since theoretically
x.x.x.1 would not be available, being a smoldering pile of ash or whatever).
Note that these are all 1 NIC machines, no multi-homing. And all static
addressing, no DHCP.
I seem to recall testing this a couple years ago, and it worked fine. However,
I'm old, so who knows how faulty my memory is ...
Here's the problem - yesterday the recovery site went down. Mind you, the main
production site stayed up, and in fact, has never gone down. But then I started
getting weird calls - I couldn't ping some VMs, yet other on the same subnet as
I am had no difficulties.
Eventually, what I had to do was delete the x.x.x.2 gateway entry from the
problematical machines, flush their DNS cache, and then everyone could access
these VMs again.
But why?. Since the main production site switch never went down, none of the
VMs should have been using the recovery site as a gateway; they should all have
been using x.x.x.1, and the fact that x.x.x.2 was unavailable should not have
matter to them in the slightest.
And even if they were using the recovery site x.x.x.2 as gateway, once it
dropped, the VM should have still been able to use the other entry, the
production site switch x.x.x.1, as a gateway and continued to be available.
So, 3 questions then:
1. Am I wrong in believing that a Windows machine (Win 2008 R2 and Win 2012 R2)
will use the gateways in the order listed? (i.e., use x.x.x.1 first, and not
try to use x.x.x.2 unless x.x.x.1 is unavailable). Seems most of my VMs worked
this way, but not all, yet all are configured the same way.
2. And, if the gateway in use (for example, x.x.x.2) becomes unavailable, I
thought Windows would automatically try the other entry, without any user
intervention. Is this not so?
3. What I want is that for the VMs to use the first gateway listed. If it can't
reach or use that, then I want it to automatically use the next entry in the
gateway list. Is this possible? If so, then how?
Thanks for any help.
