[SOLVED] writeback vs writeback (unsafe)

[adamb]

The wiki's seem to be missing anything related to the "writeback (unsafe)" cache options. Could someone explain the difference between just normal writeback and writeback (unsafe)? I am thinking there is a difference.

Am I also correct in thinking that cache=writeback is now safe so long your guest is using a file system which supports write barriers?

I appreciate the input!

 

[spirit]

>>The wiki's seem to be missing anything related to the "writeback (unsafe)" cache options. Could someone explain the difference between just normal writeback and writeback (unsafe)? I am thinking there is a difference.

unsafe : when the qemu vm is doing the fsync, to flush the datas to disk, the host acknowledge before datas is really writen. If you have a power loss, you'll have data corruption.


>>Am I also correct in thinking that cache=writeback is now safe so long your guest is using a file system which supports write barriers?

yes. barrier enabled , or > kernel 2.6.37 with fua support

http://monolight.cc/2011/06/barriers-caches-filesystems/

 

[TauriRed]

>> unsafe : when the qemu vm is doing the fsync, to flush the datas to disk, the host acknowledge before datas is really writen. If you have a power loss, you'll have data corruption.

So if host itself do have UPS - it's ok to use unsafe instead of writeback, right? (setup: regular lvm, no zfs)

 

[_gabriel]

it's ok to use unsafe instead of writeback, right? (setup: regular lvm, no zfs)

no, some precious data can be easy lost if there is system crash too. unsafe (even regular writecache) is only recommended for temporary VM, do not use writecache with 24/7 VM.
I already had corruption, Windows boot but error loading some apps or DB corrupt, with Lvmthin on Datacenter SSD.

 

[TauriRed]

no, some precious data can be easy lost if there is system crash too. unsafe (even regular writecache) is only recommended for temporary VM, do not use writecache with 24/7 VM.
I already had corruption, Windows boot but error loading some apps or DB corrupt, with Lvmthin on Datacenter SSD.

So you suggest using No cache in VM settings?(I do have daily backups but losing day is something I like to avoid). Did you host crash or VM crash?

 

[nikybiasion]

It depends on the storage you're using. Run tests with write-through, no cache, and write-back modes, and choose the one that provides the best performance. Ideally, one of the first two would be better for safe reasons.

 

[yangsm]

@_gabriel,My C drive cache mode is none, and I have placed the msSQL database file directory on the D drive. I have now set the D drive to write back (unsafe) mode. If my data can be lost and the virtual machine crashes, will my MS SQL application not start? Is the data damaged or only lost? When the backend storage is Ceph and your backend is THINLVM, is your database damaged? Is there any way to ensure that the data will not be damaged, but will be lost

 

[_gabriel]

Is there any way to ensure that the data will not be damaged, but will be lost

yes, keep close to baremetal.
so no another cache like writecache.
DB shouldn't be corrupt but I don't want to be fired in production.