VxFS multiple hosts HP-UX 11.31 (Mar-2010 VSE)

[aeronet]

Hi,

our customer have a EVA4400 and two BL860c I2 blades in a c7000 enclosure with Ethernet passthru modules and 2 Brocade 8/12 FC-Switches. The two blades are installed with HP-UX 11.31 VSE edition. No vpars/npars configuration on it, no ServiceGuard installed (customer didn't bought a license for it) . One machine is the production server and the other server is the test machine. Every server has different VDISKs presented from the EVA for the for the application using LVM 2.1 and VxFS. HP-UX is installed on the local disk . The request of our customer is, if the production server is gone, the test server must assume the role of the production server without a great effort. It's not a problem, that the customer run some scripts to take the failover, but the customer don't want to change the presentation of the EVA and import the VGs to the test server, etc….

Thats happened if I present the VDISKs from the production server also to the test server and mount the LVs of the production server on the test server and do some writes on it by mistake.
Has HP-UX in version 11.31 a solution to handle multiple writes from different nodes on a VxFS volume?

Has anybody an other idea to find a solution for the described scenario?

Thanks

[Steven E. Protter]

Shalom,

You ask:
Has HP-UX in version 11.31 a solution to handle multiple writes from different nodes on a VxFS volume?

SEP: The solution is to use a Cluster File System. That along with Service Guard would have to be licensed.

Also the applications you plan to access the data must be certified for an active-active cluster environment.

What you might try is providing access to the test servers, but only writing from one node at a time, to avoid some heavy license and support costs.

SEP

[aeronet]

Hi SEP,

> What you might try is providing access to
> the test servers, but only writing from one
> node at a time, to avoid some heavy license > and support costs.

Yes, correct. The customer don't have bought a license for HP ServiceGuard etc.... to reduce the costs.

The test server must be handle the role of the production server on a hardware failure e.g.. It don't must be a transparent failover. The customer can launch some scripts to take the failover of the application by hand.

The question exactly is: is any new feature bundled in the HP-UX 11.31 (VSE Edition) to protect against data loss if both servers write to a VxFS volume by a mistake.

I'am not really sure thats happend, when I present the VDISK from the EVA to both servers.

Cheers,
Andre

[Bill Hassell]

Sharing filesystems between two systems is not possible without special (priced) software. There is nothing to prevent either system from destroying the data by presenting the same disks to two different servers. What your customer wants is a zero cost version of Service Guard. If this were a simple matter of scripting, then Service Guard would not be a priced product. So vgimport and EVA presentation changes are indeed required.

Getting the data synchronized between two systems is only a part of the task. You need to change the network connections since the primary system is not working. That involves some method to disable the primary system's LAN cards to prevent duplicate IP problems. Probably someone will have to move the LAN cables (not very convenient at all with a blade box). The standby box must have it's IP address(es) changed and other systems on the network would have to wait for their arp tables to expire (minutes) or you must change the MAC address of the common IP address to get remote systems quickly connected.

Then there is the problem in detecting that the "server is gone". Does that mean that the OS has crashed? Would a failure also be a part of the application failing? If so, how is that detected and someone notified? What about a full filesystem that hangs the primary system? Or a kernel parameter such as nproc or nfile has been exceeded and several random failures start occurring?

All of these concerns can be handled by a senior system administrator although it might take a long time to get the second system operational since there are so many steps that require decisions to be made before making changes.