[lxc-users] lxd: "-B backingstore" equivalent?

[Mark Constable]

On Fri, 5 Jun 2015 05:05:21 PM Serge Hallyn wrote:
> > > > Does this mean that btrfs is considered a second class option
> > > 
> > > It is, for a few reasons.
> > 
> > Sorry to persist with this but would you mind elaborating briefly on
> > some of those reasons or point me to further discussion please?
> 
> We didn't want to depend on a single fs.  Also, btrfs still has some
> performance issues (esp at fsync, which kills apt-get),

I suspect a lot of performance issues revolve around unbalanced systems.

> and people still seem to hit corruption with it (though other people
> seem to run it rock-solid with no issues).

Older war stories mostly revolve around folks letting their btrfs systems
get to 100% full and/or involve earlier series 3 kernels and those earlier
bad experiences are still being used as a reason why btrfs is "not ready".

> > I have invested heavily in btrfs so I am a little "shocked" at this
> > news. If I want to stick to btrfs then would I be better off relying
> > on legacy lxc?
> 
> I don't think we'll be dropping the support we have.

Sure, I wouldn't expect that, but it means that most future devel, testing,
tutorials and example setups will be based on LVM instead of btrfs and
that concerns me (not that my concerns matter in the real world.)

> We definately won't be adding support for zfs, overlayfs, etc.

Good.

> Can you say a bit more about how your usage depends on btrfs?

I can't compare btrfs to LVM because I've been using btrfs for so long
now that I have forgotten all I knew about LVM... and very glad of that
because btrfs is so much simpler and more flexible.

I have a couple of dozen personal and professional systems and all run
utopic and btrfs. The busiest server with 1000s of clients and 100's
of vhost domains has been up for 6 months without any problems other
than initial performance issues because the fs needed to be rebalanced.
Once that was done, and once a month, it's been perfectly satisfactory.
I also got caught out with sparse sqlite3 databases from Dspam but once
they were regularly vacuumed that problem disappeared. I didn't notice
that particular problem on the previous ext4/dell-raid system.

Personally, my own pair of HP microservers for local backup were
renovated from zfs to btrfs 3 months ago and have been working perfectly.
Again, particularly so since being rebalanced. The ease of management
and flexibility, especially being able to use send/receive to sync
them, is just not (so easily) available without btrfs.

The key points over LVM is being able to use disks of any size, online
transition of raid personalities, file system (not hardware) level
checksumming and... subvolumes.

I guess my "usage depends on btrfs" is because of it's ease of use and
flexibility to cover everything from a single laptop SSD through to
various RAID configurations but short of enterprise level openstack-like
systems. There the extra stability and performance of LVM is justified
in 2015 (maybe 2016) but short of that fairly lofty niche enterprise
level of need, this year, I believe btrfs is an overall superior fs
solution and a perfect fit for lxc/lxd.

Obviously IMHO.