freebsd guest on azure

[Frank Leonhardt]

On 2019-03-23 18:57, tech-lists wrote:
> On Sat, Mar 23, 2019 at 09:31:17PM +0800, Frank Leonhardt wrote:
> 
>> As to your question, what I've done is run a ZFS pool and simply send 
>> a snapshot.
> 
> That's a good idea. I presumed I'd be stuck just with UFS. My most
> favoured way of doing this is having a zfs-vol backed vm on a freebsd
> host but it's not my choice.
> 
>> I can't remember now if I made it boot from ZFS as I prefer to boot 
>> from UFS
> 
> this is me. I like to boot off something fast and simple.
> 
>> and use ZFS for the rest. But I think I did this with Azure exactly 
>> because  I wanted to copy  the disk around. It was early last year so 
>> you can't quite me on it.
>> 
>> If you're not sure about sending datasets on ZFS then ask.
> 
> Yes please how are you invoking sending a dataset in this context? For
> example, are you using compression for the data stream? If so, what
> sort? Is there anything else pertinent to the azure context when using
> zfs send?

I have a nagging feeling that Azure insisted on its version of FreeBSD 
and did not install from a normal CD image. But even if this was the 
case, it could be different now. If it did I would have attached another 
vdisk for the zpool, but it means you can't transfer the OS partition. 
However, I don't want to do this anyway...

Now I run everything in a jail, which means that the OS partition 
contains nothing but the base OS. I just move the jail to another host 
and run it there. This is also a great way of getting several VMs whilst 
only paying for one. So I'm only moving the dataset with the jail on it.

Your friend here is the "zfs send" command. For obvious reasons (on a 
running system) you need to take and send a snapshot. The send command 
serializes the dataset and sends it to stdout (but not if it's a tty, as 
that would be silly). zfs recv does the opposite. You can copy datasets 
with something like zfs send z/test at today | zfs recv junk/foo

if you want to send it to another host, use ssh. To be precise:

zfs send z/test at today | ssh anotherhost zfs recv junk/foo

One annoying feature was that data used to be decompressed and 
re-compressed on a send. You could then pipe it through compress if you 
wanted to save WAN bandwidth. However, since 11.2 (and 12.0) a 
--compressed option has appeared, which leaves the data in the state it 
was stored. This relies on the zpool on the other end supporting the 
same compression method, so I've not used this in anger yet. Using lz4 
compression is normally no-brainer.

Regards, Frank.