[linux-lvm] Some ideas regarding snapshots (was: Reiserfs, 2.4.7, snapshots w/ 0.9)

[Wolfgang Weisselberg]

Iain Campbell (iain at sandon.ca) wrote 64 lines:
> Quote from Teigland and Mauelshagens's Volume Manager whitepaper:

> journalled file system. These operations (fsck or journal replay) require 
> writing to  the snapshot ligical volume, which is not allowed."

On a related subject: would it be possible to have a writable
snapshot (probably a 'reverse' snapshot)?

I remember a few situations where I would have done a lot for
a feature like that; e.g. building a system that's 'guaranteed'
the same after restarting/rebooting it.

The idea is along the lines of:
- you have the system in a known good, fscked state on the HD.
- you snapshot that (possibly huge) system
- you mount the snapshots r/w
- the original will probably stay unmounted or r/o (but not
  neccessarily, I am thinking of robust, networked stand-alone
  devices which could be upgraded in the background while
  running; a simple restart will have the upgrade applied.)
- The system can scribble onto the snapshots until they
  are full.
- To reset the complete system to the known good state, umount,
  unsnapshot, start over.  Or even -- horror -- reboot.

This should be *much* faster than a reintarnation or the use
of dd, not to mention easier, simpler, more elegant and using
less space.


And on another related idea: can we have the option of making
snapshots non-persistant again?  Or is that lots of work?
There sometimes are cases where snapshots lose all value over
a reboot, so they could/should automagically disappear.

-Wolfgang