[oi-dev] Updating Hipster and GUI log-in breakage with mounted ZFS datasets

[Nikola M]

On 02/18/15 08:24 AM, Alexander Pyhalov wrote:
> Hello.
> 1) Configurations where /opt is on separate pool are not supported. 
Who decided that? When and how?
Where are release notes noticing that? Where is discussion about 
architectural changes about that?
If I can't use /opt for all BE's that I need have to test OI, 
(Oracle-contriubted Firefox and Thunderbird lives there, etc)
I will not have enough disk space to test OI and I will need to say OI 
(At least OI Hipster goodbye.)

If you are one who decided that, please explain how you think to disable 
such important and valuable option of Unix-like OS, just like that?

> Imagine, IPS wants to update the system. How does it do it? 
It worked till now, then some package updates
(what updates, where, how, when in what release - noone knows because 
Hipster will not tell you that.)
It worked through all lifetime of Hipster till now.

I don't quite understand why that should be supported by community to 
remove obviously useful system functions like merely mounting system dir 
wherever I like.

/opt should not generally be used by packages that come with the system.
And if not explicitly installed by user /opt should be generally 
file-free from IPS
OR IPS and OI should not expect it to be inside boot archive.
> Asks libbe to create new boot environment and mount it and then 
> updates packages in it. When /opt is in separate pool libbe doesn't 
> know about this filesystem and creates empty /opt in new BE. IPS 
> installs updated package files to this new /opt and you receive 
> non-empty /opt, which can't be used as FS mountpoint on system boot.
I don't need know what is libbe is, why it is sudenly messing with my 
/opt, why I don't have GUI because of it
and user should simply not care. User cares for it's update to work and 
not to end up with non-working system because of newly introduced bugs.

It worked till now, it should work like explained, it's a bug, revert 
changes that caused it in 2014.1 so people can update their installs.
I can't work with OI without GUI and mounted dir in separate dataset, 
period.

If non-empty Dir can not be mountpoint at boot, IPS should not install 
anything there.
Or it's time to update IPS or update libbe , or system configuration then,
it worked for me till now.

> Now, how do you propose libbe to manage such configurations? Create 
> snapshot of each dataset in separate pool and mount cloned FS? At 
> least, they can be inconsistent, as they are in separate pools and we 
> can't create consistent snapshot. However, I understand, that 
> consistency sometimes can be sacrificed. In any way, libbe doesn't 
> behave this way, but it can be a valuable RFE. Feel free to fill it on 
> bug tracker under illumos project.
It worked before. Why it does not work now?
Libbe should not be dealing with what is not in it's way (and /opt is 
not and it's subvolumes and subvolumes of user's dir volumes.
Something changed and you are master of Hipster that thinks Hipster does 
not need versioned numbers on updates so figure out what changed that 
between August 2014 and 20141010.

I am unable to incrementally update from August to 20141010, so I am 
unable to test it when it started to happen. It wasn't happening till my 
20140824 and 20140826 BEs. It's a bug introduced after it.
>
> 2) As for your system freezing after logging in - it's more interesting.
> Could you get you ~/.xsession-erros log ?
> What if you log in to terminal session and start gnome-session by hands?
I also have mounted datasets under my home dir. Maybe that is also 
Hipster's problem too...

Most intersting thing is that since 151a7 Openindiana is getting more 
and more bugs and user problems, one by one things are stopping to work 
and noone even cares about cries of users.

Things do not get tested before releases, releases are not being 
discussed , made and polished for updates, priorities have not being set,
all we actually have is "updating packages" without idea where OI is 
heading and why and who actually needs a system that does have only part 
of functionality it had in 2009.
At least till 151a7 everything worked, after that it started falling a part.

Hipster was fun when I was thinking it leads to the next /dev.
It's not fun anymore for a very long time.
Rolling releases are just lunacy for general use and could exist only 
between /dev releases,
And not every random package change should survive to next release).

Things should be discussed , announced, explained and problems should be 
worked on instead of putting them under the carpet.

"New morning, new update from the clear sky" is not kind of my distribution.
Testing and releases and support and requests for opinions Before 
something change - is.