Re: Long failover time...

Well, I've been setting the IPs in the host file for 7 years now and have not run into one issue. Again, you only set the IPs of the private interfaces and, AFAIK, all best practices show the private interfaces as not being routed anyway, so there should be no issue whatsoever.

I think you're arguing on documentation and best practices which don't always consider the optimal solution. They consider the *best* solution for the majority of users based on simplicity and stability.

As for the hosts file. A NIC doesn't use the host file. The routing stack in the host uses it in the following order: lmhosts, hosts, dns and wins. There is no issue with bindings or nic configurations. This is a very simple task that effects nothing except for the IP for the hosts you put in the file. So, the normal lookup for an IP addr is like this:

Node1 -> where is Node2 -> is it cached? nope-> how do I find it? -> go to hosts/lmhosts -> nothing there (they are empty by default) -> use dns/wins -> found it, use IP returned by gethostbyname() ... The only difference now, is it finds the host IP in the hosts file and no network requests are required.

Again, this does not effect anything except for how an IP address is returned to from the function gethostbyname(), doesn't effect routing, nic config, etc etc. And it only effects those hosts you put in the file, hence you only add the cluster Nodes in the file.

~kenny

"Rodney R. Fournier [MVP]" <rod@xxxxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message news:uOpezee3HHA.5796@xxxxxxxxxxxxxxxxxxxxxxx
Yeah I missed the 90% one. But I will still contain that it (hosts file) does nothing with a properly configured systems and DNS. I think you need to check into bindings, cause I believe your are not setup correctly.

As I stated both nic will use the hosts files and each will need the address of the other one. In your example you hard code the heartbeat? I would love to see your nic configurations, bindings, etc. because you have to be missing a best practice or 3 to have to use this kind of host file! I hope nobody tries this that reads your post!

MSCS is only Highly Available. Parts of any system, Exchange, SQL can be FT - Raid hardware, redundant power supplies, etc. The complete system will only be HA.

Cheers,

Rodney R. Fournier

MVP - Windows Server - Clustering
http://www.nw-america.com - Clustering Website
http://msmvps.com/clustering - Blog
http://www.clusterhelp.com - Cluster Training
ClusterHelp.com is a Microsoft Certified Gold Partner


"Kenny Speer" <kenny.speer@xxxxxxxxx> wrote in message news:OolSpPe3HHA.1204@xxxxxxxxxxxxxxxxxxxxxxx
Ah, first it's not 90% of clusters, it's 90% of clusters with long takeover times. Your context is wrong.

Second, in your hosts file, you put a hostname and IP address. Each NIC only has one IP address (in this config, yes it's possible to add more), therefore, the hosts file is used for that network only. For example:

Public Network: 10.x.x.x
Private Cluster Network: 192.168.x.x

In your hosts file, you add: Node1 192.168.x.1 and Node2 192.168.x.2

Pretty simple. If Node1 wants to communicate with Node2 for any reason, it goes over the private cluster network. No DNS required. This will ensure that your cluster is running regardless of your DNS status. For any other network comm, normal DNS lookups are used.

You distinguish between Highly Available and Fault Tolerant (yadda yadda) here is a Microsoft page which uses Fault Tolerant quite often:

http://technet.microsoft.com/en-us/library/aa997507.aspx

http://technet.microsoft.com/en-us/library/aa997234.aspx

As for where I read it, /etc/hosts and drivers/etc/hosts have been used forever to bypass DNS in mission critical scenarios.


"Rodney R. Fournier [MVP]" <rod@xxxxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message news:e9oQs%23d3HHA.4400@xxxxxxxxxxxxxxxxxxxxxxx
Kenny, I am not about to argue that DNS can be an issue. I am firmly stating that I have never needed a host file on a clustered node. Does not matter who runs DNS, they just need to ensure it works. DNS is really easy stuff to get right or wrong. If you really have had a need on 90% of clusters for DNS, then they had DNS issues, plain and simple. It is not a best practice to create a hosts file for a clustered node. Does not matter if DNS is Microsoft or not, it needs to be configured correctly.

Next, you mentioned a hosts file for the private only? Where did you read about this? How do you tell a hosts file to only work for one NIC and not the other(s)?

MSCS is not and never will be fault tolerant, it is considered highly available. And Windows 2003 with or without clustering is dependant on DNS, Active Directory has for a long time now.

Cheers,

Rodney R. Fournier

MVP - Windows Server - Clustering
http://www.nw-america.com - Clustering Website
http://msmvps.com/clustering - Blog
http://www.clusterhelp.com - Cluster Training
ClusterHelp.com is a Microsoft Certified Gold Partner


"Kenny Speer" <kenny.speer@xxxxxxxxx> wrote in message news:eJwgbid3HHA.5852@xxxxxxxxxxxxxxxxxxxxxxx
Not true. If the DNS server is running properly you won't have an issue. A perfectly configured Cluster can have any number of issues due to a misconfigured/non functioning DNS server. To say that 90% of clusters were misconfigured is complete crap. Often, an admin who owns the clusters does not own the DNS servers and often those DNS servers are not Microsoft servers. It is absolutely an option to put the cluster host names ONLY in the hosts file. Scalability is not an issue, how many nodes are in your cluster? The majority of clusters are 2-4 nodes, 4 lines isn't very many to add to a host file which should not change.

Also, remember, the hosts file has the private interface addresses ONLY, not the public. This insures that all cluster comm from one node to the other will use the cluster comm interface (i'm not just talking about heartbeat here) without requiring or *depending* on an outside service (DNS).

Anyway, I think your argument doesn't hold any water, since we're talking about making MSCS as fault tolerant as possible but then you make it *depend* on DNS for proper operation (not access from clients, but even then, we've used IP addresses for 30 years when DNS goes down).

~kenny

"Rodney R. Fournier [MVP]" <rod@xxxxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message news:eDsCWQd3HHA.4436@xxxxxxxxxxxxxxxxxxxxxxx
Than 90% of clusters were configured wrong to begin with :) Honestly, if DNS is running properly you won't ever need a hosts file. Host files don't scale very well or allow for easy changes!

Cheers,

Rodney R. Fournier

MVP - Windows Server - Clustering
http://www.nw-america.com - Clustering Website
http://msmvps.com/clustering - Blog
http://www.clusterhelp.com - Cluster Training
ClusterHelp.com is a Microsoft Certified Gold Partner


"Kenny Speer" <kenny.speer@xxxxxxxxx> wrote in message news:OvjOjNd3HHA.1208@xxxxxxxxxxxxxxxxxxxxxxx
Really? Hmmm. I've seen 90% of long failover times resolved by doing this.

~kenny




"Rodney R. Fournier [MVP]" <rod@xxxxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message news:OdC0XMc3HHA.5984@xxxxxxxxxxxxxxxxxxxxxxx
Hosts file! No, DNS works nicely :)

Cheers,

Rodney R. Fournier

MVP - Windows Server - Clustering
http://www.nw-america.com - Clustering Website
http://msmvps.com/clustering - Blog
http://www.clusterhelp.com - Cluster Training
ClusterHelp.com is a Microsoft Certified Gold Partner


"Kenny Speer" <kenny.speer@xxxxxxxxx> wrote in message news:uaFNDsH2HHA.5884@xxxxxxxxxxxxxxxxxxxxxxx
I thought it was recommended to put all cluster names/ips in the hosts file located at:
%WINDIR%\system32\drivers\etc\hosts

By default, you don't want your cluster communicating over the public (client access) interface and you don't want your cluster to fail just because your DNS server goes down or is not accessible.

By adding each node to this file on both nodes, you won't even do a name lookup via DNS since Windows uses hosts/lmhosts then DNS.

~kenny

"LOVEBEINGDBA" <LOVEBEINGDBA@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message news:EC8A3B7D-E8D7-4B1E-BF4A-DD63E31750E9@xxxxxxxxxxxxxxxx
Thanks and I appretiate your reply...

Is it required to create a PTR record for the Cluster Name in the DNS
Server??? We already have a cluster whose name is registered in the DNS only
for forward lookup...

Thanks.
Arun M

"John Fullbright" wrote:

"The server for 143.3.16.172.in-addr.arpa. could not be contacted over
adapter
'Public' to determine whether it accepts DNS registration updates. Retrying
at a later time"

Looks like problems contacting an authoratative DNS server for the reverse
lookup zone that holds the PTR record.

..

"LOVEBEINGDBA" <LOVEBEINGDBA@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:A70ABBFB-6DDA-495D-B715-542869DC021C@xxxxxxxxxxxxxxxx
> Environment:
> Windows Server 2003 R2 x64 SP2
> MSCS 2 node failover cluster
>
> Failover takes about 2 minutes. When doing the failover the > clustername
> take
> a long time to come back up.
>
> Clustername has been registered in the DNS for forward look up > .
>
> We have another production cluster and that works fine in the > same setup.
>
> These are the lines I found that are related to this issue from > cluster
> log:
>
> 00000840.00000bb4::2007/08/06-16:57:27.835 WARN Network Name > <Cluster
> Name>:
> The server for 143.3.16.172.in-addr.arpa. could not be > contacted over
> adapter
> 'Public' to determine whether it accepts DNS registration > updates.
> Retrying
> at a later time.
> 00000840.00000bb4::2007/08/06-16:57:27.835 INFO Network Name > <Cluster
> Name>:
> Replaced DNS name <clustername.domain> with IP Address > 172.16.3.143 over
> adapter 'Public'.
> 00000840.00000b08::2007/08/06-16:57:28.679 INFO Network Name: > time until
> next DNS reg: 2007/08/06-22:17:35 (128309122556769715)
> 00000840.00000bb4::2007/08/06-16:58:26.795 WARN Network Name > <Cluster
> Name>:
> Failed to register DNS PTR record 143.3.16.172.in-addr.arpa. > for host
> <clustername.domain> over adapter 'Public', status 1460
> 00000840.00000bb4::2007/08/06-16:58:26.795 INFO Network Name > <Cluster
> Name>:
> Modified DNS name <clustername.domain> with IP Address > 172.16.3.143 over
> adapter 'Public'.
> 00000840.00000b08::2007/08/06-16:58:28.638 INFO Network Name: > time until
> next DNS reg: 2007/08/06-22:17:35 (128309122556769715)
>
> Any help will be greatly appretiated.
















.

Relevant Pages

  • Re: Long failover time...
    ... MVP - Windows Server - Clustering ... As for the hosts file. ... hence you only add the cluster Nodes in the file. ... does nothing with a properly configured systems and DNS. ...
    (microsoft.public.windows.server.clustering)
  • Re: Long failover time...
    ... MVP - Windows Server - Clustering ... http://www.clusterhelp.com - Cluster Training ... As for the hosts file. ... file) does nothing with a properly configured systems and DNS. ...
    (microsoft.public.windows.server.clustering)
  • Re: Hosts file ignored
    ... > any of the entries I have made to the server's hosts file. ... > Successfully flushed the DNS Resolver Cache. ... > I am not running a DNS server on my system. ... > I can ping IP addresses without any problem, both on the LAN and on ...
    (microsoft.public.windows.server.general)
  • Re: Long failover time...
    ... I was explicitly stating that the cluster comm was NOT used just for the hearbeat. ... MVP - Windows Server - Clustering ... The routing stack in the host uses it in the following order: lmhosts, hosts, dns and wins. ... is it finds the host IP in the hosts file and no network requests are required. ...
    (microsoft.public.windows.server.clustering)
  • Re: Long failover time...
    ... MVP - Windows Server - Clustering ... Second, in your hosts file, you put a hostname and IP address. ... Private Cluster Network: 192.168.x.x ... No DNS required. ...
    (microsoft.public.windows.server.clustering)