1 |
On Tue, Jun 11, 2019 at 7:21 AM Davyd McColl <davydm@×××××.com> wrote: |
2 |
> |
3 |
> On Tue, 11 Jun 2019 at 12:34, Helmut Jarausch <jarausch@××××××.be> wrote: |
4 |
>> |
5 |
>> I had some trouble switching to the new profile 17.1. |
6 |
>> Following the advice in the news item didn't suffice. |
7 |
>> |
8 |
> |
9 |
> first off, `emerge -v1 /lib /lib32` didn't work out because I had an old library in there I |
10 |
> had to remove with `emerge --depclean` first. I also have an old install of sickbeard, which |
11 |
> I had to remove from world for the same reason: `emerge -v1 /lib /lib32` would just complain |
12 |
> about not being able to find an installable source (my words -- can't remember the original |
13 |
> terms), but it didn't really look like an error -- all green text. |
14 |
|
15 |
I've updated two hosts. One went very smoothly, but it is a fairly |
16 |
dedicated host. One had a few issues, and it has a LOT of history. |
17 |
|
18 |
I found that anything 32-bit tended to cause more trouble, and I had a |
19 |
few orphans as well. It wasn't a huge deal. |
20 |
|
21 |
I think a big part of that is that before I did ANYTHING I took a lot |
22 |
of steps to clean up. I ran depclean and revdep-rebuild as a start. |
23 |
I reviewed all the migration tool output and anything that looked |
24 |
non-essential was depcleaned. When I did the 32-bit rebuild anything |
25 |
that was giving me trouble was traced back to whatever pulled it in |
26 |
and depcleaned (I forget if I did that up-front or if I just deleted |
27 |
the offending library and depcleaned the rev dep later - obviously |
28 |
don't do that for anything you care about). |
29 |
|
30 |
On a more dedicated host/container/etc I suspect you won't have many |
31 |
issues, because you're not going to have a huge pile of legacy stuff |
32 |
lying around with complicated dependency relationships. |
33 |
|
34 |
Some of my rebuild and depclean issues were resolved with --backtrack |
35 |
and --with-bdeps=y. |
36 |
|
37 |
In general a good principle is that anytime you want to change |
38 |
profiles take some time to do some housekeeping. The less junk you |
39 |
have on your system, the less there is that can go wrong. |
40 |
|
41 |
On my one host I also took the opportunity to decide whether I REALLY |
42 |
needed wine. That is a TON of 32-bit stuff you otherwise probably |
43 |
don't need. After removing it you need to clean out package.use |
44 |
because we don't have soft USE dependencies yet. |
45 |
|
46 |
And of course before I did anything I took a zfs snapshot of my root |
47 |
filesystem which only contains the OS for the most part. So, if I ran |
48 |
into serious issues a rollback would probably have been a one-liner |
49 |
(I'm guessing that I'd do that from a rescue disk just to keep daemons |
50 |
with stuff in /var from going nuts). |
51 |
|
52 |
Overall it went better than I was anticipating actually. We haven't |
53 |
had a migration like this one in a while, but I do think that the |
54 |
risk-level of this one was a bit undersold. Restructuring all your |
55 |
libraries is obviously a risky task and while you shouldn't be |
56 |
alarmist it is something that has a lot of potential to go wrong. To |
57 |
be fair, the news item does say that you should do a backup. |
58 |
|
59 |
-- |
60 |
Rich |