[Samba] Accidentially upgraded production DC to git master branch

[Rowland Penny]

On 10/09/15 15:24, Pinja-Liina Jalkanen wrote:
> On 10/09/15 16:31, Rowland Penny wrote:
>> On 10/09/15 13:06, Pinja-Liina Jalkanen wrote:
>>> Hi all,
>>>
>>> So I've accidentially installed a development version of Samba on a
>>> production machine, and not even realised it for a month, because there
>>> hasn't been any problems.
>>>
>>> I want to return to a supported version of Samba, whatever it is (4.2 or
>>> 4.3). What is the right way to remedy this? Do I have to demote Primary
>>> and re-join it, or would it be safe to just downgrade Primary to, say,
>>> 4.3.0?
>>>
>>> As far as I know I currently have no replication or other problems, but
>>> with Primary running whatever was bleeding edge a month ago, my current
>>> situation feels precarious at best.
>>>
>>> Any suggestions?
>>>
>> I fail to see how you can accidentally install a development release,
>> you would have had to do a git pull then ./configure, make, make install.
> By being a little carefree here. And yes, I acknowledge that this is a
> silly mistake entirely of my own making that shouldn't really have ever
> been happened. Just that now it has.
>
> I've always installed by checking out master, doing a git pull, checking
> out the release tag I want (which puts Git to a "detached head" state),
> and then ./configure && make && sudo make install.
>
> Now, I've clearly forgotten to check out the right release tag before
> running configure and then furthermore also forgotten to check what
> version actually got compiled. It's not like configure or make is
> actually shouting out loudly the version being compiled.
>
> And because you can't do git pull in a detached head state, you'll have
> to temporarily check out some branch for that anyway. But in retrospect,
> it would've made far more sense to use one of the stable branches
> instead of the master branch; this mistake would still have been
> possible, but the resulting version wouldn't have been that far away
> from a tag release, causing less trouble.
>
>> Anyway, that is what you are going to have to do, but instead of doing a
>> git pull, download the tarball for 4.3.0 and unpack this, stop samba,
>> back it up and then compile samba using the same settings as when you
>> installed the wrong version. Once installed, restart samba again and you
>> should now be using version 4.3.0
> So you think that the DB should still be OK after downgrading to 4.3.0,
> without any other action? OK, I'll give it a try (but I'll still use Git).
>
> After reading the Git logs between
> 8eb57316f040a445d47cff8de0ca9bf4e9280acb and
> 1fcad53d7ca1be619bbc9572100b38886114dee7 and not finding anything that
> looked like a DB change I would have most likely done that anyway, but I
> just wanted a secondary opinion. Thanks!
>
>

The wiki used to have a warning about using samba from git in production 
(why did you remove it Marc ?). You cannot be sure that it isn't broken, 
it shouldn't be, but it could be. What is in samba-master gets branched 
off for the next release (or so I understand), this is then released as 
'rc' candidates to find and fix any bugs, you are therefore getting any 
possible unfixed bugs by using samba-master. I would suggest you use the 
latest stable version (4.3.0) instead, but of course, I cannot force you 
to do so.

Rowland