Yesterday the Windows Server team released an important hotfix for anyone using static routes on a Windows Server 2008 or Windows Server 2008 R2 cluster. The issue is that manually added routes, like static routes that many of us use, are deleted from the route table when you delete an IP address from the network adapter. This is exactly what happens when a cluster IP address resource moves from one node to another.
If you use Windows Server 2008 or Windows Server 2008 R2 and static routes, you should read this article and determine if you need to apply this hotfix to your servers.
Note: the hotfix also applies to Windows Vista and Windows 7
Hotfix: Manually added route table entries are deleted unexpectedly when you delete an additional IP address in Windows Vista, in Windows 7, in Windows Server 2008 or in Windows Server 2008 R2
Wow. I don’t even know where to start this post. Let me just say that I am having nightmares of whiteboards and that my employees have joked that they are going to draw clusters in permanent marker on all of our boards so that I don’t have to redraw them for every meeting.
SQL Server “Denali” has radically changed the way that we view high availability and disaster recovery, and how we architect the environment and the procedures around it. Without posting anything that I cannot disclose due to my NDA, let me just say that the key to your success with Denali’s HA/DR features is terminology. As soon as the new featureset is public, I will post about terminology and try to describe everything in detail. I can’t tell you how many times I have been in meetings or on calls where we first had to spend a significant amount of time understanding what we were even talking about. In one case, I met for 30 minutes with two very intelligent people, both of them familiar with the technologies in the current builds. It took us 27 minutes to understand what we were trying to communicate, and 3 minutes to arrive at the answer to the question!
The bottom line is this – try to forget everything that you know about current HA/DR technologies built into SQL Server, and try to start with an open mind and an open notebook. Listen a lot, read a lot, and try to understand the terminology before you start asking questions or figuring out implementations. A huge part of the learning curve is simply related to the new terminology and understanding what it means, all without comparing it to existing technologies.
Oh, and a whiteboard really helps…
Last month Microsoft released an important update that probably escaped your attention – it certainly did mine. The KB article is titled “Validate SCSI Device Vital Product Data (VPD) test fails aver you install Windows Server 2008 R2 SP1”, which didn’t seem to fit any problem that I had experienced. Until I read the details.
The hotfix will fix several different cluster validation issues, but the important part to me is that it fixes an issue where even though you choose to keep the services/applications online when running the cluster validation, the validation incorrectly runs against disks that are being used by those services/applications. Validation then fails because it cannot get a persistent reservation on the disk. Note: it will pass on the node that owns the the disk because it can get a persistent reservation on that node only.
This means that either you can’t add nodes to the cluster, or that you have to take all services/applications offline – which is not very helpful on a production cluster. To resolve this issue, install the hotfix mentioned in this KB article: http://support.microsoft.com/kb/2531907.
Several people have asked me if I have heard when SQL Server “Denali” CTP2 will be released. I have not heard anything official (and couldn’t comment if I had!), but unofficially, it sounds like the next public CTP may be CTP3.
In the builds that we are currently testing there are some major changes in HA technologies. I think most people will be happy to see some of the long-standing feature requests being addressed in this version. A very significant effort has been devoted to mission critical systems in the Denali release, and the current builds are really starting to highlight the work that has been devoted to this area.
Hopefully there will be a public CTP2, but whether there is or not, be ready to download the next CTP as soon as it is released. It will be full of new features and changes, and you will want to get familiar with them quickly.
Windows Server cluster installations have typically required shared storage, meaning that all nodes in the cluster must be able to see the storage. SQL Server Denali uses Windows Server Clustering for some of its new high availability technologies, but requires that there be a single Windows Server Failover Cluster (WSFC) between sites. The problem with that is that if you have two sites that are separate geographically, there is no shared storage. When you attempt to add the remote nodes, th clustering validation tool will fail with errors.
The Windows team has released a fix for this so that Windows clusters can now support Asymmetric Storage – some nodes can see some of the storage, and some nodes can see other storage. Clustering validation will still show warnings, but the validation will pass and allow you to add the remote nodes to the cluster.
Here is an example of the warnings that are shown. There will be a warning for each disk:
Windows Server 2008 R2 requires SP1, and Windows Server 2008 requires a hotfix. More information about this can be found in the following KB article: http://support.microsoft.com/kb/976097.
Last week at the SQL PASS Community Summit, Microsoft released SQL Server code name “Denali” CTP 1 (Community Technology Preview). Unfortunately, many of the cool new features are not available in this release, but you can begin to work with improvements in SSMS, SSIS, and Full-Text Search and with new features, like Column-based Query Accelerator, Contained Databases (CDBs) and some of the emerging HADRON features (High Availability, Disaster Recovery, Always On).
What should be clear to you if you being looking at Denali is that there is definitely a learning curve, and you should start now to get familiar with the new features. I expect the next CTP to have a lot more “meat” to it, and it is going to take you some real-time to get up-to-speed with the terminology and functionality of the new features, especially HADRON.
If you haven’t yet, I highly suggest you get a copy of CTP1 and start playing with it. This should shorten your learning curve when CTP2 is released in the next few months.