[OpenIndiana-discuss] Proposal: OpenIndiana Stable Branch

Sat Jan 15 01:20:24 UTC 2011

Hi,
Sounds like an good move, however I don't think that you mentioned or
proposed how we tackle one issue of taking OI into production server (which
is possible, I go live with 3 OI servers on Monday <gulp> ;) ).

Currently the dependency chain of the packages, is erm to be polite utterly
broken... Before we can consider a stable build, we have to fix that a text
install takes 2.6GiB and includes so much stuff that doesn't belong. Trying
to remove via pkg gets you nowhere as it horrible chained together
incorrectly. It's a security nightmare, the only way I currently feel safe
is that I have a zone that faces the world because zones (for some reason)
install much more striped down installs.

IMHO First thing should be making a minimal server install, debootstrap
minimal. Get a unix base system, IPS package manager and wget and the rest
can come later. TBH the install a new zone gets is more like the default
install should be imho.

Then have a number of repositories with different classes of supported apps,
Primary being your list and with critical fixes etc. and then secondary
being less supported apps.
Your proposal to focus on a small set of apps is correct imho, new users to
OI stable will be early adopters almost by definition, so by being honest
and saying OS is stable and great and so are these major programs, but not
everything out there is to the same level, we encourage champions to take
their favorite program and get it on the major supported list.

Also a smaller core will make the illumos switch faster, I'm personally not
sure if stable should become before illumos integration. OI on illumos works
now, with locales being the major issues (being worked on), it doesn't feel
right to call OI stable without it using (even a WIP) the base that it
requires going forward (OI on ON isn't really stable as it's a EOL, which
implies an unstable future).

As you're worried about missing the window, as OSol users migrate to linux
or FreeBSD IMHO that is more a perception issue. OI website appears very
slow moving, even dead. Bringing some life there may help that issue, call
the WIP stable build, Early Adopter build or something like that, post EA
new builds once a week on the front page. Get silly screen shots of shell
doing zfs, or apache configuration files, all completely useless BUT
highlights that look this thing is real and running apps you, as a IT geek
are wanting to run...

As a production OI deployer, I really care about 
1) Minimal install with just the programs I want, 
2) Critical fixes for the OS and those apps if I use the package system.
3) A safe build environment, as there is a fair chance I'll be building app
myself at this stage (I use a separate machine for this as the safest way :)
)
4) Something that will upgrade nicely for the say 3 years. For OI that
scream illumos IMHO
5) A community with nice central info pool, currently the OI wiki and
webpages doesn't feel like a community, wiki access is restricted, so not
encouraging writing up notes and most of the useful information  isn't on
there anyway. Half the time you end up on Oracle web pages, which makes you
wonder if this is a real OS.
6) Security info and concerns, from articles to hardening the OS to using
VMs (Xen, Zones, Virtual box?) to isolate components. Probably just an
extension of the wiki and/or blogs but I'm sure some of the in the trenches
guys would be happy to write a few articles on how we got OI onto the front
line and in use.

Hope this doesn't sound negative, as mostly I agree with your proposal (only
thing I really disagree on is non illumos). At the moment OI is very much a
shadow of OSol choices, which I don't think apply here, for it to go stable
it needs to shake of its old masters clothes and choose its own route. 
Starting with a small server distro that just happens to have a huge repo of
other apps including desktop, allows it to find a niche and then expand out
from there. As a server (especially a storage server) OS imho its second to
none :)

Bye,
Deano
Deano at cloudpixies.com

-----Original Message-----
From: Alasdair Lumsden [mailto:alasdairrr at gmail.com] 
Sent: 14 January 2011 20:36
To: Discussion list for OpenIndiana; OpenIndiana Developer mailing list
Subject: [OpenIndiana-discuss] Proposal: OpenIndiana Stable Branch

Hi All,

I believe now would be a really good time for us to create our first stable
branch of OpenIndiana, given the timing of some developments within the
project.

Below I've outlined my proposal and I'd love feedback from the community and
from OI developers!

Obviously as a new project with a small (but growing) developer base,
providing support for the whole release isn't feasible - there are literally
thousands of packages in the distribution. But we have to start somewhere,
so I'm proposing we provide limited support (outlined below) for a set of
core packages.

********
* Why? *
********

Prior to the Oracle takeover, Solaris 10 was free to use in production, and
for a long time, security updates were provided free of charge. OpenSolaris
was also free to use, and updates were available by living on the bleeding
/dev edge. People were (mostly) happy.

Then Sun hit financial difficulties and discontinued free security updates
for Solaris 10. Then Oracle happened, ending the free use of Solaris in
production.

This has left people wishing to use Solaris technologies on their production
servers in a difficult position. They have to pay Oracle, or use
distributions that don't provide security updates. Or switch to Linux.

There are a great many people who would jump at the chance to use Solaris if
there were a production ready version with security and bug fixes provided
for free.

Indeed, this is what people have come to expect from mainstream UNIX
platforms - Linux distributions such as Debian, CentOS, Ubuntu, etc, provide
updates free of charge - and this is one of the reasons they have become so
popular.

We have a real opportunity to capitalise on the situation left by Oracle, to
capture server market share away from OpenSolaris, Solaris 10, and give
users a migration path other than switching to Linux (which a lot of people
are doing).

There are a lot of people out there who *really really* want a stable build
of OpenIndiana - myself included, and I believe OpenIndiana's best chance of
gaining acceptance, market share, and building a thriving development
community is by capturing the server market.

There is also a risk that if we *don't* do this, we'll become an obscure
fringe distribution, like DragonflyBSD.

The goal here is to be the *mainstream* accepted de-facto Solaris
distribution. Something people talk about and seriously consider using.

Solaris contains killer technologies not seen on other platforms;
technologies like ZFS, Zones, SMF, DTrace, COMSTAR, Crossbow - I couldn't
live without any one of these, and we should capitalise on this while we
can.

It's also worth keeping in mind that despite warning users that oi_147 and
oi_148 were development releases, people are already using it in production
environments, myself included, due to a lack of alternatives. The great news
is that it has proven to be exceedingly reliable, and I have no hesitation
in recommending it for busy workloads. All we need to do is add security
updates and critical bug fixes on top and we'll be in a great position. No
small feat I grant you, but we can start off small and work our way up.

Now is also an opportune time to do this - our next release will be based on
Illumos, which has seen rapid development and will involve some integration
pain. Some have called for a stable branch after Illumos is integrated, but
it could be many months until we have an Illumos dev build suitable for
respinning as a stable branch. That's months of lost opportunity.

So I say we do it now.

/dev builds will continue as normal, the next one will be Illumos based -
Desktop users can continue to use our /dev builds, and internet facing
servers can use the stable branch.

*********************
* What we'd provide *
*********************

The release would be aimed for February, and titled "2011.02". It would be
based
on oi_148. We would only provide the Text Installer and Automated Installer
ISOs.

We would provide security and critical bug fixes only for:

1. OS/Net (The core OS consolidation)
2. A limited set of server oriented packages that have the greatest usage
and
attack "surface area". The initial list I can think of includes:

 - OpenSSL
 - Sendmail
 - Perl 5.8.4
 - Python 2.6
 - Ruby
 - zip, bzip2, gzip
 - Apache HTTPD 2.2
 - PHP 5.X
 - MySQL 5.X.X
 - Postgresql 8.4
 - Java
 - Tomcat
 - GNU Coreutils
 - GCC
 - RSync
 - ISC BIND
 - Bash
 - Curl
 - wget

We should also aim to provide security fixes for any bit of software in the
repo that allows an easily exploitable remote access vulnerability or root
privilege escalation, although we cannot guarantee to do so as monitoring
security updates for over 1000 software packages is unfeasible. An example
would be the recent Exim vulnerability on CentOS that allowed remote root
access by sending appropriately formatted emails. This area is something
where we will depend on users, not OI developers, alerting the project to
the issue so that a judgement call can be made on whether we have the
resources to fix the issue.

Security updates would be provided from 6 months of the release date, or
until the next stable release is released. Potentially we have the option as
a project of providing commercial support past the 6 month date if
enterprises desired this. I feel this could be a good way of generating
revenue for the project to fund development if there was a market for it.

If external contributors were able and willing to commit patches/fixes
beyond the supported list, we'd accept them with open arms, and this could
be a great way to extend the contributor list and get more people involved.

******************
* How we'd do it *
******************

1. We do a re-spin of oi_148 fixing any of the major bugs that we can (Eg
things like the Broadcom driver issue introduced in oi_148)

2. This gets pushed into pkg.openindiana.org/stable (or /release - tbc)

3. Security fixes and critical bug fixes for the supported packages get
pushed into the repo. People doing an image-update would then receive the
latest packages.

4. Security fixes and bug fixes would be backports to the version we
currently provide.

People should be able to update from oi_148 to 2011.02. And people should be
able to update from 2011.02 to oi_150. But people should not be able to
downgrade from oi_150 or later to 2010.02. This is the same as the situation
was with OpenSolaris releases.

To make the above easier to manage, one proposal I have is to match the
versions of Apache, PHP, MySQL, Tomcat etc to the same versions shipped in
RHEL 6/CentOS 6. This way we can monitor their repositories for security
updates against these packages, and share the same backports. This will make
life a lot easier for us as a project.

The main thing will then be doing rebuilds of the packages involved. I would
suggest we keep a set of Zones on infra01.uk.openindiana.org around for
doing this, so that doing a rebuild is very easy to do, and well documented.
Just a case of logging in, patching the appropriate files, running a build,
pushing to a test repo, testing it, and then pushing into the public repo.

**********************
* Concluding Remarks *
**********************

I believe this is a great opportunity for us and I think it's the right time
to do it.

Although we're starting on the server only front, there's no reason why we
can't at a later date add support for the desktop if sufficient contributors
are able to make it happen. 

I'm confident that with a stable branch, we can really increase our userbase
on servers, which will bring commercial opportunities from the enterprise,
and accelerate development of our favourite operating system :-)

Looking forward to feedback!

Cheers,

Alasdair.
_______________________________________________
OpenIndiana-discuss mailing list
OpenIndiana-discuss at openindiana.org
http://openindiana.org/mailman/listinfo/openindiana-discuss