2009-04-22 07:48 AM
We currently are using SnapManager for SQL, snapmirroring the data to another data center, and we currently have a script setup to automate the restore process. The machine is a Windows Guest running on ESX 3.5 over FCP RDM luns. Currently we are using the sdcli disk connect command to attach these LUNS to are DR system when we want to mount the databases, but this process is very slow in ESX with FCP luns as it has to rescan the HBA each time it attaches a drive to the VM.
Initially we left the drives connected to the VM, but did not assign drive letters in Windows. The annoyance we ran into in that scenario was after we detached the databases and removed the drive letters from the system we would sometimes see Delayed Write Fail messages. We built pauses into the script to wait after detaching the databases, but it seemed like we always ran into that problem. Does anyone have any recomendations in this regard? What are other users doing? Is it best to just disconnect and reconnect the drives each time, or should I just build in longer pauses to try and avoid the delayed write fail messages?
2009-04-22 08:17 AM
Welcome to the community
I was told about this issue on the VMWare course and think it can be solved with LUN masking. Not sure how off the top of my head however.
Have you seen this http://communities.netapp.com/community/events/ask_the_expert_vdi as it is full of great tips and links?
Think you need to look at http://communities.netapp.com/message/8833#8833 and the links on it for the answer.
Hope it helps
2009-04-22 08:46 AM
Yeah we don't have an issue with our backups. Those work fine, the restore works fine, everything is all good. I am just looking to see what other users in a setup like mine are doing. My restore script works like gangbusters, it is just slower than I would like due to the fact that Snapdrive 6 has to rescan the HBA in ESX in order for it to see the new drives that are attached. I can leave the drive attached, and not assign a drive letter, but I get those annoying delayed write failed messages since Windows is trying to do something to the RO Luns when the Luns are in their snapmirrored state. Lun masking won't work because we are talking about ESX here. If I hide the WWN from the initator group it is removed for all the VM's on that HBA.
2009-04-23 07:08 AM
Just to note, it takes 15 minutes for us to do our restore process. About 95% of that time is due to ESX rescanning the HBA for each drive we attach to the SQL server (User Databases -D, Database Logs -L, System Databases -S, Snapinfo -I). While 15 minutes is acceptable for now we would like to cut down that RTO.