Re: Linux Distribution Preferences? - Mailing list pgsql-general

On Sun, Jan 13, 2013 at 4:06 PM, SUNDAY A. OLUTAYO <olutayo@sadeeb.com> wrote:
> 4 reasons:
>
>  1. One place where I worked Ubuntu was standard, I tried it and found
>     that it lacked at least a couple of desktop features in GNOME 2 that
>     I found very useful into Fedora. Fortunately, I was allowed to
>     revert back to Fedora. Prior to that, I was using Fedora mainly by
>     default.
>
>  2. Twice I came across features that I liked and Ubuntu seemed to imply
>     they had done them, later I found the projects been initiated and
>     sponsored largely by Red Hat. Especially as Red Hat is in the top
>     ten contributors to the kernel, and the contribution of Ubuntu is
>     not significant.
>
>  3. Ubuntu distributions are now starting to be filled with crapware and
>     ant-privacy features features.
>
>  4. Ubuntu seems very good at collecting fanbois.

Not one of those is a good reason to avoid Ubuntu server for pgsql.
There are reasons to not use it, but those are not them.  I've run
PostgreSQL servers on Redhat (before RHEL existed and there was JUST
Redhat) 5.1, RHEL 4, 5 and 6, Debian Lenny and Squeeze, just one on an
old version of Suse, and on Ubuntu server 8.04LTS and 10.04LTS and
12.04LTS.

My preference personally is for debian based distros since they
support the rather more elegant pg wrappers that allow you to run
multiple versions and multiple clusters of those versions with very
easy commands.  RHEL is great for building a stable but not
necessarily ultra faster server, and if you can afford their
commercial support it IS top notch.  Debian and Ubuntu feel much the
same to me, from the command line, on a server.

The reasons to NOT use ubuntu under PostgreSQL are primarily that 1:
they often choose a pretty meh grade kernel with performance
regressions for their initial LTS release.  I.e. they'll choose a
3.4.0 kernel over a very stable 3.2.latest kernel, and then patch away
til the LTS becomes stable.  This is especially problematic the first
6 to 12 months after an LTS release.  Ubuntu support is a pitiful
thing compared to RHEL support.  I've reported bugs for RHEL that were
fixed within weeks, or at least a workaround came out pretty quick.
I've reported LTS bugs that are now YEARS old and Canonical has done
NOTHING to fix them.  There's a bug in 10.04LTS workstation for
instance that meant you couldn't have > 1 profile for a given WAP.
Never fixed.  Only recommendation was to upgrade.  From an LTS.  sigh.

There are reasons TO use Ubuntu as well.  Of if you are running very
late model hardware you can't get good support from an older release,
and using a more recent, possibly not LTS release is a good way to get
best performance.  I have often installed a late model release like
11.10, to get support for odd / new / interesting / high performance
hardware, and then at a later date could update that platform to an
LTS release for stability.  Note that I often waited til a good 3 or 4
months after the next release before I even started testing it, let
alone upgrading to it.  Ubuntu often has fairly late model versions of
many packages like pgsql or php or whatever that more RHEL like
distros will not get due to their longer release cycles.  It's easier
to add a ppa: repo to debian or ubuntu than to add an RPM repo to RHEL
and I've found they're usually better maintained and / or more up to
date.

Simple answer of course is that there is no simple answer.

Frequently released / updated distros (fedora, ubuntu non-LTS, debian
beta and so on) are GREAT for doing initial development on, as once
the stable branch based on it comes out you'll be deploying against
something with a long stable release branch.  So the latest version of
Ruby, Perl, PHP, Python and so on are on the server, as are the
latest, or nearly so, versions of pgsql and slony and other packages.

Long term distros (debian stable, Ubuntu LTS, RHEL) are all good for
deploying things on you don't need the latest and greatest hardware
support nor the absolute fastest performance but instead stability are
paramount.  When downtime costs you $10k a minute, using the latest
code is not always the best idea.

Most importantly, if you've got LOTS of talent for one distro or
another, you're probably best off exploiting it.  If 95% of all the
developers and ops crew run Ubuntu or Debian, stick to one of them.
If they favor Fedora / RHEL stick to that.  If they work on windows,
find a new job if at all possible.