[Samba] Upgrade from 4.6.7 to 4.17.4 fails with database problem (Failure during re-pack, so transaction must be aborted)

[Ricardo Esteves]

Yes, i agree, from now on i will definitely upgrade by adding new DCs.

Also now that i have samba running with 4.17.x, i will try to add a new 
DCs with Fedora own packages in order to get rid of compiled version.

Thank you once again.

On 18/05/2023 10:52, Rowland Penny via samba wrote:
>
>
> On 18/05/2023 09:41, Ricardo Esteves via samba wrote:
>> Hi,
>>
>> Thank you for help, probably updating using another host and use 
>> join, or demoting one of my DCs (i have 2, DC1, and DC2) upgrade and 
>> then join the demoted one, and then do the same for the other is the 
>> best way.
>>
>> But actually last night i was able to do the inplace upgrade 
>> successfully , basically the problem was on DC1 (probably had some 
>> bad old entries on the db), what i did was transfer all the fsmo 
>> roles to DC2, demote DC1, and then i did the inplace upgrade of DC2 
>> and went ok. Now i need to upgrade and join DC1 back.
>>
>> Btw, in order to avoid the issue you mentioned (I would never 
>> recommend upgrading a self compiled version of Samba in this way. 
>> There have been instances of files being removed from the version of 
>> Samba that you are upgrading to, that still exist on disk) what i 
>> usually do is remove
>> /usr/local/samba/bin|sbin|share|lib|lib64|libexec
>> i leave only  /usr/local/samba/etc|private|var
>> and then i do the ./configure / make / make install
>
> AD is so easy to upgrade between major versions by adding new DC's, 
> that I would never upgrade in place. Minor upgrades i.e. 4.x.5 to 
> 4.x.6 are probably okay, but I would never go from i.e. 4.15.x to 
> 4.16.x by upgrading in place, it is, in my opinion, just not worth the 
> risk.
>
> Also building Samba yourself isn't really required any more, unless 
> you need to test something. There are numerous repos that can supply 
> Samba packages now and this is further helped by the devs deciding 
> that perhaps MIT isn't experimental any more.
>
> Rowland
>
>