[syslog-ng] Redhat repo

Fri Mar 29 17:04:34 UTC 2019

Hi,

[ personal opinion, just trying to share another point of view(developers) ]

LTS vs. rolling releases ... not an easy question... until you have enough time for everything.

## LTS
* requires much more effort
  * backward/forward port vs. number of bugfixes (most of the time fixing the same/very similar issue is not straightforward between different branches)
  * much more complex CI/CD pipelines
  * need to support more complex deployments (we have Docker images... we should have to support build/debug environment, etc...)
  * a complex bugfix could also break the LTS... (and maybe in a different way in all the supported versions... :  a nightmare)

* requires a better cooperation with distro package maintainers (when we release a new maintenance version, would be nice if package maintainers would integrate it into the distros) [note: this better cooperation is not only depend on us, I think]

* if we are talking about unofficial repositories (don't forget, Balabit does not offer official packages for OSE), then in my case, I simply could not allocate enough time for support 5 different versions of syslog-ng

* supporting upgrade between LTSs also not an easy task [especially when we are talking about 4-7 years EOLs and 5 different versions]

## Rolling release
* smaller changeset, smaller risk (compared to LTS upgrades...)

* more bugs we can fix on a single track

* have time for other activities (like Google Summer Of Code, stabilize main track)

* negative effects:
  * using a rolling release in production system requires a test environment
  * users don't have the possibility to upgrade to the latest version

Maybe I'm wrong, but since when we are focusing on a single track (not counting other editions...), syslog-ng is more stable than before. In the good old days we just run some unit tests, some functional tests and that was all.

On today, we have a unified unit test framework, run internal tests for OSE (on dedicated AWS machines), run internal performance tests and currently we are also working on replacing legacy functional tests with a new, internally developed Python test framework(based on PyTest).
All of these work would be not possible if we would have to focus on backport/forward ports.

But I understand your point of view and your pain: we can do whatever we want if you, the users don't want to, or cannot upgrade to the latest versions ...

L.

________________________________
From: syslog-ng <syslog-ng-bounces at lists.balabit.hu> on behalf of Evan Rempel <erempel at uvic.ca>
Sent: Friday, March 29, 2019 4:13 PM
To: syslog-ng at lists.balabit.hu
Subject: Re: [syslog-ng] Redhat repo

I don't have a versioning issue with versioned repos. After all, that is what all of the
OS repos are: Redhat 6, Sientific Linux 7, Centos 8, Debian 9, Ubunto 16 etc.

The issue I have with syslog-ng in specific is that once a new version comes out,
the previous version is abandoned. No bug fixes. No security fixes etc.
This is NOT the same as the OS versions repos where bug fixes and security fixes
are continued within the repo.

What is being suggested in this thread is for major version repo to be created and
maintained in the long term support (LTS) role. Given the history of version numbers for
syslog-ng there is no clear major version transition. Syslog-ng3 (3.0 thorught 3.19)
has been changing, breaking and adding new features for nearly 6 years. That's
about 3 releases per year. Without some automated update process this is impossible to
keep up with for anything more that a handful of systems.

Some thought as to designing a policy around when a new major version should be created
and supported the long term support and then automatic updates (not upgrades) would be possible
within that major version and its associated repo. A balance needs to be found between frequency
of releases and how many versions need to be supported. I would recommend only 1 release per
year that is supported. That would limit the number of supported versions to no more than 5.
Possibly include an unsupported head for folks that absolutely need a latest feature before the
year it is released. That group can transition to the released version once it is released as a
supported version.

With regards to the duration of the long term support, I think that 3 years is a little short.
With a 3 year LTS cycle organizations that use syslog-ng would have to start the planning, testing
and approval process perhaps as frequently as every 2 years. Redhat basic support is 5 years
with extended support pushiing that out to 7 years. That cycle might be a little long and costly for the
vendor/community to support that long.

I would propose nothing shorter than 4 years, so that customers can do the planning, testing, approval
and upgrade process every 3 years at a minimum.

On 3/29/19 6:23 AM, Nik Ambrosch wrote:
Having a syslog-ng-latest repo to compliment the existing syslog-ng-317 versioned repos might be a good idea for people who use these builds for desktops and non-mission-critical applications.

I don't like surprises so I use the versioned repos and only change when there's a *need* to upgrade (new feature/fix listed in the changelog).

On Fri, Mar 29, 2019 at 4:24 AM Czanik, Péter <peter.czanik at balabit.com<mailto:peter.czanik at balabit.com>> wrote:
Hi,

Here I offer my personal opinion about rolling releases, which differs
from the official Balabit opinion...

When I first started my unofficial syslog-ng rpm packages a long time
ago in the galaxy, I had a single repository. When a new release came
out I packaged it into this repo. The feedback I received -- and I
fully agreed as a practicing sysadmin -- that "don't do automagic
version updates for us, as it might break things". That's how
versioned repositories started. From the admin point of view an
automatic update is a risk, as even with thousands of automatic tests,
configurations can break. I recall quite a few breaking changes, some
intentional, some accidental. From this point of view things got just
worse when from yearly releases we changed to two months. Quite a few
places test new releases for more than two months before adding to
production...

Here I fully agree with Evan, and this was also a feedback I often
received in private (unfortunately not on a public mailing list, blog
post or whatever), that continuous rolling is a major pain. To best
accommodate such users, there could be an LTS releases next to the
rolling, which receives fixes for critical problems. To avoid having
many parallel supported versions, we could align this with a major
release of Redhat, and keep that version supported for ~3 years.

In this case there could be a versioned repo for LTS releases:
syslog-ng4
syslog-ng5
etc.
And for those needing the latest features:
syslog-ng-rolling
Where people receive new feature releases automatically as they appear.

Bye,

Peter Czanik (CzP) <peter.czanik at balabit.com<mailto:peter.czanik at balabit.com>>
Balabit (a OneIdentity company) / syslog-ng upstream
https://syslog-ng.com/community/
https://twitter.com/PCzanik

On Thu, Mar 28, 2019 at 3:57 PM Evan Rempel <erempel at uvic.ca<mailto:erempel at uvic.ca>> wrote:
>
> The syslog-ng package maintainers and those deciding on version numbering need to decide on when backwards compatibility breaking occurs. IMO there is no reason to not upgrade to newer versions automatically (whatever schedule the client performs yum updates), with the one exception being updates that no longer work with a configuration file. Having an exact version number in the configuration file (@version 3.17) is not useful if the configuration works with multiple releases. I would think
> that having a configuration version that remains the same for the same configuration syntax and semantcs would be more meaningful. Just thinking in public/out loud, perhaps all syslog-ng version 3.x use the same configuration syntax, then the @version would be 3, and when the config syntax is no longer backwards compatible, a new package name and @version is produced. This means that package NAMES would need to contain the major version number "syslog-ng-3" so that you do NOT get automatic
> updates to "syslog-ng-4".
>
> The down side for this is that syslog-ng-3 would need to be security fixed along side the syslog-ng-4 package for some support period and then finally deprecated in favour of the newer syslog-ng-4 package.
>
> I find that the postgresql community has done this very well. They make a major version release once per year which is how they introduce new features. They provide security fixes (and most bug fixes) for a period of 5 years. That means that at any given time they are supporting 5 major releases. Because this is a published schedule, installations are guaranteed to be supported/patched for 5 years, which falls into line with vendor OS support models quite well and provides the postgresql
> community the ability to make backwards breaking changes every year.
>
> For enterprise environments similar to what Mark describes below, it is critical that installed packages don't become abandoned with regards to patches. New features are not required to be added automatically, but security and bug fixes are. Some way to get security and bug fixes automatically without breaking compatibility over a long time preriod (3-5 years) is what is needed to ensure stability, security and maintainability in an enterprise environment.
>
> Our company suffers from these requirements which are made that much more difficult by packaging decisions like what syslog-ng has decided on. Don't think  I am picking on syslog-ng. elasticsearch and oracle java suffer from this too. I am just trying to make the syslog-ng package as awesome as the syslong-ng program!
>
> I am free for further discussing if the package maintainers want to reach out.
>
> Evan Rempel
>
> On 3/28/19 6:43 AM, Faine, Mark R. (MSFC-IS40)[NICS] wrote:
> > You can easily version lock a package in Redhat (don’t know about SUSE, but I’d imagine so)  There is no need for a separate repository.  It makes more sense for the default to be the latest release and for locking to a specific release to be the exception.  It would probably even be easier for him to manage.
> >
> > The Redhat version is very old (3.2.5), so I’d rather not use it but not having a repo that tracks with the latest release breaks my installation process and means that I need a few separate tasks just for syslog-ng.
> >
> > If there is an update, I’ll have to remove the repo, add the new repo, then upgrade.  Instead of just doing a yum update.  I usually patch secondary servers first and then a few weeks later the primary or production servers, I think that would be enough warning if there were any breaking changes.
> >
> > Hopefully, if he is going to choose not to provide a repo with the latest release, I hope he is at least standardizing the naming convention and URLs so that I can automate the process, even so my problem is that it would still be a manual process, at least partially, since I’ll have to keep track of when a new version is released and update my configuration accordingly.  Manual processes such as this increase the potential for oversight which could allow a security vulnerability to slip through.
> >
> > I don’t know how many security vulnerabilities typically occur with syslog-ng but I know that it’s a big deal here and we are constantly pushed to keep systems up-to-date.  If there is some random high profile security vulnerability found on my system, I don’t want to be called out for having out-of-date software from a third-party repo.  I know it’s not fair but that’s the way they’ll frame it.
> >
> > In the end, I might be forced to use the Redhat version in the optional rpms repo, though since I’ve been working with 3.18, it’s possible that there may be things in my configuration that aren’t even compatible with that version. The thing is, they’d rather I use the old version that is, at least allegedly, supported by Redhat, than a newer, arguably more secure, and more performant version. Their reasoning breaks down to nothing more than CYA.
> >
> > I guess my point is that I’d hoped to argue for getting to use the czanik packages for the reasons I’ve mentioned and knowing already it would be an uphill battle but if they get a whiff of how the repos are structured it will make it even harder to justify, perhaps impossible.
> >
> > Thanks,
> > -Mark
> >
> > From: syslog-ng <syslog-ng-bounces at lists.balabit.hu<mailto:syslog-ng-bounces at lists.balabit.hu>> On Behalf Of Péter, Kókai
> > Sent: Thursday, March 28, 2019 4:16 AM
> > To: Syslog-ng users' and developers' mailing list <syslog-ng at lists.balabit.hu<mailto:syslog-ng at lists.balabit.hu>>
> > Subject: Re: [syslog-ng] Redhat repo
> >
> > Hello,
> >
> > I've just spoke with Czanik Peter - the maintainer of those repositories.
> >
> > He said that initially there was one repo having the latest released, but majority of the user want to control when to upgrade to a newer version from both RHEL and SUSE side.
> > (I guess it is an *easy* way to prevent accidental updates.)
> > Also he said that there was not much demand for such structured repository since the shift (only 1-2 person).
> >
> >
> >
> > --
> > Kokan
> >
> > On Wed, Mar 27, 2019 at 3:40 PM Faine, Mark R. (MSFC-IS40)[NICS] <mailto:mark.faine at nasa.gov<mailto:mark.faine at nasa.gov>> wrote:
> > I have a question about the Redhat repo for the unofficial OSE builds.  It seems as though they are tied to a specific version.  For example, I'm currently using czanik-syslog-ng318epel6, however, I see that 3.19 has a different repo specifically for that version.  This seems to be counter to the point of yum repositories and it imposes an additional maintenance burden.  Am I maybe not properly understanding how it works, and if not, is there a repo that tracks with the latest stable version?
> >
> > -Mark
> >

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.balabit.hu/pipermail/syslog-ng/attachments/20190329/6c36e762/attachment-0001.html>