2015-03-03 06:54 AM - edited 2015-12-18 01:05 AM
How can I guarantee snapshot consistency across controllers and volumes? I thought listing luns in the snapdrive snap create file_spec creates a consistency group that freezes I/O across controllers and FlexVols. It doesn't seem to work that way.
We have an Oracle Linux 6.5 database server, 11gR2, two Oracle ASM volumes(DataASM and FlashASM), FCP, with LUNs mapped from two controllers and seven volumes(five data volumes and two archive volumes). The redo logs are kept in the data ASM DG.
A shell script puts the database in hot backup mode, and runs a snapdrive snap create command, similar to this:
snapdrive snap create -lun netapp1:/vol/dataA/lun1 lun2 lun3 netapp2:/vol/dataB/lun1 lun2 lun3 ... netapp1:/vol/flashA/lun1 lun2 lun3 netapp2:/vol/FlashA/lun1 lun2 lun3 ... -snapname snap.1
Approximately 1-out-of-4 runs fail with this error:
SnapDrive error 0001-559, Detected I/Os while taking snapshot. Please quiesce your application. See Snapdrive Admin. Guide for more information.
We modified the snapdrive.conf snapcreate-consistency-retries to 5, to increase the # of attempts, but it doesn't fix anything.
If I can't guarantee a consistent snapshot every time, how do I improve the odds? I want to move redo logs to dedicated ASM DG, which I think will improve the snapshot odds for the data ASM DG.
2015-03-30 12:07 AM
Do you have LVM snapshots on your server? LVM snapshots and Snapdrive snapshots do not go along so well. If so, the only workaround that I can see is to use 'fsfreeze' command to quiesce the filesystem before taking a snapdrive snapshot.