diff options
author | Dieter Plaetinck <dieter@plaetinck.be> | 2008-12-14 18:27:19 +0100 |
---|---|---|
committer | Dieter Plaetinck <dieter@plaetinck.be> | 2008-12-14 18:27:19 +0100 |
commit | 801b3d4a7f925e7b9ad63f795f7a7bdc92fd75a7 (patch) | |
tree | 6e3266792d5abb99d134a530e47750bbaf1208a2 /NOTES | |
parent | 1b5b6ed697fdce45da8657bd580211f682f2bee8 (diff) |
switch from dead-end 'try to be smart and clean environment first' approach to 'rollback in case of failures' system
Diffstat (limited to 'NOTES')
-rw-r--r-- | NOTES | 13 |
1 files changed, 13 insertions, 0 deletions
@@ -0,0 +1,13 @@ +* umounting/destructing blockdevice/filesystem stuff in the blockdevice library. +** Goals +not break while trying to build the setup like the user requested (breakage could happen if a device mapper volume is still active or a filesystem is still mounted) +still allow user to mount stuff himself behind the installers back. he is smarter then us. just do what we're told. +** Options +*** umount/deconstruct before trying to build +problems: - it's hard to know what we should delete, our 'build' plan might be different then the current environment (eg devices with same name but other function), + usually because of a previous run with the wrong settings, or which failed + - we can't base ourselves on things like "we should only have / and /dev/shm". The user can mount things himself + - quite complicated code if want to make it smart, but it's a dead end anyway. +*** if buildup fails, ask user to rollback -> implemented approach +- user should not ctrl-c and installer should not crash. this is doable. a 'wrong' state can be an acceptable exception. +- right now we can start repartitioning a disks that has filesystems mounted. is this harmfull? this only happens if unclean rollback or user did it, so NP i think |