Re: piece of mind (Re: Moderated posts?)
Scott Ferguson wrote:
On 15/10/14 03:33, Miles Fidelman wrote:
Scott Ferguson wrote:
On 15/10/14 01:54, Miles Fidelman wrote:
Scott Ferguson wrote:
On 14/10/14 23:54, Miles Fidelman wrote:
Andrei POPESCU wrote:
On Lu, 13 oct 14, 18:30:41, Miles Fidelman wrote:
Gee.... assuming that you don't run anything that has
systemd dependencies and/or systemd-shim is actually
maintained and kept up-to-date.
Have you actually looked into what depends on systemd?
---------------8<------------------->8---------------------------------
the majority of development. Embracing diversity and
conservatism (aversion to change) can be "a bit of a stretch".
How do you come to that conclusion.
Which conclusion?
That this is the selection criteria for LFS. (Mind you, building
from scratch is looking better and better - though Gentoo makes it a
lot easier.)
But, addressing (some of) your other points:
They weren't points (it's not a football match) :)
I simply didn't know which "conclusion" you meant (I made several) - so
to save time I covered them all.
What, it's NOT a football match?! Anyway - points as in bullet points :-)
That Debian is a progressive "Universal" OS? It changes as a result
of developers seeing a need to improve - I'd call that progressive
(rather than static or regressive).
Stability is an awfully nice virtue, one that Debian used to
subscribe to.
Stable is still as stable now as it was 20 years ago - no crashes. Do
you have different experiences?
There's also stable as in interfaces and behavior - particularly
important in a platform.
As to "no crashes" - well, not exactly, but that has more to do with Xen
than Debian.
That your own OS might suit your needs (and some others) better?
(did you take that as an offence??)
That Debian is NO LONGER a suitable operating system for my needs,
after more than a decade - yeah, that I kind of take offense to - or
at least I take offense to the inordinate amount of time that I
expect to waste on migration - be it to systemd, another distro, or
another o/s entirely.
OK - as long as you didn't take what I wrote as an offence.
Not a bit.
Parenthetically, I do notice that some here (not you) seem to take
offense at mentions of and/or comparisons with other distros and O/Ss.
Personally, I think sharing experiences about alternatives is an
important topic.
Based on the large number of posts you've made complaining that
Debian's plans don't match your needs.
---------------8<------------------->8---------------------------------e).
Some of us actually have to plan for things like transitions - and
the lack of clarity regarding development plans makes that rather
difficult.
In that sense your experience is little different than mine. I expect
hardware and user needs to change - and we anticipate it. For the last
decade change control has changed little - 6 months testing and
documentation, 2 years support. Except that the last stable has
(tenuous) LTS - which just makes our work easier.
Jessie is something we won't begin seriously testing until it actually
becomes stable.
The major difference is that I work with server, embedded devices and
desktop - if not for the last two categories I probably would work on a
derivative. Instead I rely on pre-seed and post-seed. My aim, probably
similar to yours, is not to get locked into hardware or software
(packages or distro). Legacy support can be a nice niche but it has it's
limits (is it worth it to the client?).
Similar situation. Out of more than curiousity, if not for embedded
devices and desktop, what derivatives appeal to you?
Re. pre-seeding and post-seeding: So far, my experience with
pre-seeding has been largely to automate the Q&A that's built into the
standard installer, and to install a few extra packages. I'm wondering
how granular one can get in terms of defining the install of the base
system -- i.e., to what extent one could define a pre-seed file that
installs sysvinit-core. Any thoughts on this?
Ok... so multiple init systems are going to be supported in Jessie,
and maybe beyond - ok. What seems to be very much up in the air is
whether that choice will be (well) supported at upgrade or install
time. That makes a rather big difference.
I'll wait and see - planning is only possible when the task is not
nailing smoke to the wall. I allow a two year gap between testing
wholesale changes and deployment with SLA - which is more than
sufficient (I have a much easier life than those that support Windoof!)
I don't want to learn multiple inits - I'm lazy (pick one).
Me to. Since I've already learned sysvinit, customized some of our
scripts, and it all just works - I REALLY don't want to either:
a. learn a new init system
b. deal with the (probably subtle and hard to find) ways that systemd's
claimed support for legacy systemv init scripts is less than 100%
<snip>
I've only had major upgrade problems with userland apps on webserver.
Given the huge number of packages that need to "just work together" I
can't think of another OS/Distro that even comes close to that degree of
upgrade stability (and I've plenty of experience with
Oracle/Solaris/RedHat).
For me, it's mostly mail related stuff - postfix, anti-spam, anti-virus,
list server -- I have yet to be able to build this stuff from packages.
It's always a build from source, and a lot of work to get all the
"wiring together" and configuration working properly. (Some of the
supporting stuff - Apache, MySQL, tool chains install just fine from
packages.)
I HATE major upgrades.
[*1] I suspect enough to support a tightly-focussed server OS
(if you can herd cats?) - maybe a Debian derivative? Strip out
all the DE packages and it might be do-able...
That used to be Debian.
Not in the 20+ years I've been deploying it. It was always 'better'
at server than desktop - now it's not so bad as a 'desktop'. But
different perspectives and different requirements - I *like* to
modify systems. Debian enable me to tailor it to suit a given
purpose - it I wanted someone else to tailor it for me I'd pay them
(and treat them nice).
hmmm.... I think we agree on this one --- the thing is, I've already
invested a lot of time and effort in tailoring our current
installation
Likewise - and the one before. Rinse and repeat until spud - though
between dokuwiki, OTRS, and well-documented scripts that's no longer a
real problem (compared to - well, anything. e.g SOE and SLAs for other
Distros, Windoof etc).
-- systemd is going to blow that all ways, or at least require: -
time and effort to build a system that doesn't install systemd by
default
Be grateful? Otherwise you may be replace with a bread cube and a
chicken. Glue bread cube to Enter key, glue chicken to desktop, install
:D
Well... I don't know. I'm really thinking hard about a really lean
platform, with each app. in its own container - there's a lot of appeal
to that model, and it seems to be working for a lot of people. Have to
rethink a lot of the platform issues, though - particularly around
storage. Linux RAID10 and DRBD make for a rock solid (and cheap)
storage platform that's hard to beat, coupled with Xen and automatic
failover for high availability.
(unless the installer folks include init-select as part of
installation), or, - time and effort to test anything and everything
that touches systemd - particularly customized initializations and
such (I'd sure like to see some regression testing on systemd's
support for legacy sysvinit) and - an expectation that somewhere, I'm
going to get bit by some big problem induced by systemd, that's going
to take a long time to figure out (and having looked at the
documentation, that scares me
I don't know what you support so I can't comment. I used to support OS/2
(only dropped support last year - the client still has several dozen
service stations that only run OS/2 - for everything). I no longer
support IBM PS/2s (sob). Your experience will differ (I won't stoop to
"all things must end" - but they do).
Fairly simple really -- mail, mailing lists, some simple web servers,
and our development sandbox. All but the last just needs to stay up and
running 24x7, with minimal care and feeding.
Beyond that: - the whole thing with systemd, it's monolithic nature,
and it's intrusiveness is that we lose a lot of that ability to
tailor things (or at least it makes that much harder) - as well as
being a major shift in design and architectural philosophy, away from
modularity toward monolithicness (if that's a word) - lots and lots
of central, unaudited code
Sufficient comment has been made on the bulk of that paragraph. As for
unaudited.... I'd be breaching the Coc to use the appropriate adjective
for how *little* GNU/Linux *has* been audited.
There is that. Then again, DHS has been funding Coverity to do
automated code analysis of lots of open-source stuff. Systemd is
listed, I'm waiting for approval to view the scan.
CentOS would fit the server focussed distro definition better
(well, limited architectures, on recent hardware...). Some people
swear by it i.e. "I've never used anything else - the others are
all rubbish" ;p
Well if I wanted Red Hat family, I'd go with CentOS or RHEL; though
when it comes to commercial distros, SUSE has always struck me as a
bit "cleaner" from a sysadmin perspective.
I've had a bit to do with Novell when I was at Telstra - we can pursue
that off-list perhaps. Doesn't seem appropriate here.
Yeah... we've all had to deal with Novell at one point or another -
might be worth following up off-list.
Miles Fidelman
Kind regards and sincerest wishes that all your fears become faint
memories. Until then - plan for the worst, hope for the best.
Isn't that "plan for the worst you can imagine, expect even worse?" Sigh...
Cheers,
Miles
--
In theory, there is no difference between theory and practice.
In practice, there is. .... Yogi Berra
Reply to: