Using consistency groups for Point in Time Copy
This topic provides a description about using consistency groups for point in time copy mechanism in IBM Spectrum Scale™.
The use of FlashCopy® consistency groups provides for the proper ordering of updates, but this method does not by itself suffice to guarantee the atomicity of updates as seen from the point of view of the user application. If the application process is actively writing data to GPFS™, the on-disk content of the file system may, at any point in time, contain some number of incomplete data record updates and possibly some number of in-progress updates to the GPFS metadata. These appear as partial updates in the FlashCopy image of the file system, which must be dealt before enabling the image for normal file system use. The use of metadata logging techniques enables GPFS to detect and recover from these partial updates to the file system's metadata. However, ensuring the atomicity of updates to the actual data remains the responsibility of the user application. Consequently, the use of FlashCopy consistency groups is suitable only for applications that implement proper mechanisms for the recovery from incomplete updates to their data.
The FlashCopy consistency group mechanism is used to freeze the source disk volumes at the logical instant at which their logical image appears on the target disk volumes. The appropriate storage system documentation should be consulted to determine how to invoke the Point in Time Copy with the consistency group option:
- Storage subsystems – 1
- LUN ids and disk volume names – lunS1 (hdisk11), lunS2
(hdisk12), lunT1, lunT2
lunS1 and lunS2 are the FlashCopy source volumes. These disks are SAN-connected and appear on the GPFS nodes as hdisk11 and hdisk12, respectively. A single GPFS file system fs0 has been defined over these two disks.
lunT1 and lunT2 are the FlashCopy target volumes. None of the GPFS nodes have direct connectivity to these disks.
lunS1 – lunT1 (source-target)
lunS2 – lunT2 (source-target)