RHEV 3.3 and ext4/LVM discards
Does anybody know whether 3.3 will format iscsi block devices ext4? If so, will the 'discard' option be in use? We're looking at purchasing a flash-based storage array, but being able to reclaim unused blocks is important. Thanks,
Doug
Responses
Red Hat
Guru
4649 points
Are you asking whether a block storage domain created using iSCSI lun be formatted using ext4? If yes, then we don't create any filesystem on such an iSCSI lun. We create storage domain by,
- Making a PV on the iSCSI lun.
- Creating a Volume Group using the PV.
- Creating lvs for each disk requested by user from that Volume Group.
Then the user see the lv as disks inside the vm and they create filesystem on it. They are free to create ext3/ext4 or whatever filesystem they want depending upon the the OS they use.
Guru
6863 points
Great question Doug.
Sadique - do you know whether storage Thin Reclamation will be possible? I.e. if inside the VM a significant amount of data is removed/deleted (i.e. someone truncates a 2GB log file), will that initiate those blocks to returned to the storage pool? I have seen the term "UNMAP" used in this context, but I don't know if that is slang for another technical term.
So, specifically:
Would the storage array indicate that 2GB was once again available?
Would the storage domain in RHEV also indicate that the Virtual Disk was again 2GB smaller?
Again - great question Doug. I am hoping that Thin Storage Reclamation (or whatever it is called) is either available now, or will be in a future release!
Red Hat
Guru
4649 points
James,
There are two areas to explore when it comes to discard support.
1 - When a disk is removed from RHEV-M (this does lvremove to remove the underlying lv associated with the disk), whether the hypervisor will issue a discard to the storage or not to reclaim the space used by that lv/disk.
Answer: RHEL6 has support for this, but this is by default disabled in RHEV-H hypervisors and RHEL based hypervisors. If you are using RHEL6 based hypervisors, you can change this in /etc/lvm/lvm.conf
# 1 enables; 0 disables.
issue_discards = 0
An RFE is already filed to make issue_discards = 1 the default which may end up in a future rhev-h.
2 - If a file is removed from the vm filesystem, whether a discard will be issued to the storage or not.
Answer: This requires discard support in filesystem layer and qemu layer. The filesystem layer support is there in ext4, but support in qemu layer is still a work in progress for RHEL7. So this is highly unlikely to get implemented before rhel7 becomes the base for hypervisors.
Welcome! Check out the Getting Started with Red Hat page for quick tours and guides for common tasks.