Nexenta Core Platform 2 Released

Posted May 26, 2009 19:31 UTC (Tue) by hppnq (guest, #14462)
In reply to: Nexenta Core Platform 2 Released by zooko
Parent article: Nexenta Core Platform 2 Released

Do tell -- when has it been possible to do something like this before?

AIX has had this for years. Solaris with Veritas as well, and I am sure there are more. The basic idea is the same as the ZFS one: you split a mirror or make a snapshot, keep one half untouched and upgrade the other one.

I think it is really cool that ZFS has made this available for home users, but it has existed for a long time in the proprietary Unix datacenter -- I used it. Technically, the possibility has or should have been there for years for Linux systems as well.

But I have never actually missed it. Especially with proprietary systems it is only *after* the reboot that you will find out whether an upgrade actually was successful (or rather, unsuccessful), because you are not upgrading a well-tested distribution, but a system typically consisting of a number of binary components not at all guaranteed to keep playing nicely together. It is mandatory that you have a way to actually go back to what worked before, and nothing else.

It's fine with me if you prefer reinstalling your operating system instead of having a transactional upgrade tool. But what does this have to do with proprietary vs. open source?

No problem, but that is not what I said. I prefer reinstalling the base system over making snapshots, if the very unlikely event of a completely screwed up upgrade occurs. Sorry if I was not clear, but all this has nothing to do with proprietary software, but with systems management.

Nexenta Core Platform 2 Released

Posted May 26, 2009 22:28 UTC (Tue) by hppnq (guest, #14462) [Link] (5 responses)

If you're interested, here's a 2004 article about using rpm's rollback feature when doing system upgrades. Not so long ago it was removed, because it was considered to be unreliable: think of scripts changing the system irreversibly. To be honest, I can't immediately see how ZFS would solve that. But I am no expert.

However, this kind of transactional package management has been an AIX feature since last century. In most serious environments, having a full, pristine copy of your operating system is a requirement anyway, so then having separate inline snapshots is simply overhead -- and note the obvious risk of not being able, for whatever reason, to get your hands on the snapshots, before or after boot.

Really, a console is all you need. ;-)

Nexenta Core Platform 2 Released

Posted May 26, 2009 22:32 UTC (Tue) by dlang (guest, #313) [Link] (4 responses)

you could do the same thing on linux with LVM snapshots and any filesystem you wanted to use.

Nexenta Core Platform 2 Released

Posted May 26, 2009 23:09 UTC (Tue) by hppnq (guest, #14462) [Link] (1 responses)

Indeed. Or check out the Nix package manager and NixOS!

(It seems to have nothing to do with nix of LWN fame. ;-)

Nexenta Core Platform 2 Released

Posted May 27, 2009 17:51 UTC (Wed) by nix (subscriber, #2304) [Link]

It is indeed entirely unrelated to me, although I was tickled by the
name :)

It seems like the sort of thing I'd like to come up with if I was better
at functional languages, though :) A very nifty piece of work indeed.

Nexenta Core Platform 2 Released

Posted May 27, 2009 10:23 UTC (Wed) by paulj (subscriber, #341) [Link] (1 responses)

Can LVM snapshots be promoted to full, R/W devices?

Nexenta Core Platform 2 Released

Posted May 27, 2009 15:05 UTC (Wed) by dtlin (subscriber, #36537) [Link]

Unlike LVM1, LVM2's snapshots are writable by default. It's very, very cool. With sbuild, I use r/w snapshots all the time: it can be set up to take a snapshot of a minimal base system, add minimal dependencies to run a package compile in this environment, and remove the snapshot when all is done.

The changes made to a snapshot go to the bit bucket when the snapshot is removed. There's patches which allow for merging a snapshot back into the origin device. They don't seem to be upstream yet, though.