On Tue, Apr 12, 2016 at 2:25 PM Lance Ball <lball@redhat.com> wrote:

Hi all

I've been emailing with Stian about versioning in the "Node.js adapter releases" thread, but as he pointed out, some of my concerns are broader than just Node.js and so I am broadening the conversation a bit.

Background: I am working as part of a Node.js RHT "middleware" team to, among other things, contribute towards the development of Node.js modules for our existing JBoss technologies. As part of this, we're doing what we can for the Keycloak Node.js pieces.

As it stands now, I understand that versioning of adapters must remain in lockstep with Keycloak core. I understand the motivation for this, but want to push back on this just a little bit, and open it up for discussion.

I see a couple of scenarios where this is potentially problematic. I am using Node.js and NPM here, but I think the concerns should apply to any adapter that is part of an ecosystem outside of Java.

1) There is a security flaw in some 3rd party dependency of the adapter, discovered the day of a Keycloak core release. This renders the "latest" version of an adapter useless until a new Keycloak server is released. I understand that the release cadence is anticipated to be approximately every 6 weeks (which is laudable), but still that's > 1 month that users have to wait for a security fix.

In my opinion it depends how critical and exploitable is the vulnerability. For critical updates, I'd say that the adapter should be release ASAP and the patch number increased and released. For example: vulnerability found at 0.0.16 we just bump it up to 0.0.17 and release.

2) There is no change in the adapter between releases of Keycloak server. In this case, it's not necessarily a problem to release a new adapter version, but it seems noisy and pointless if the bits are exactly the same.

When we look at version numbers, they are typically MAJOR.MINOR.PATCH with possibly a pre-release suffix like -Alpha1. I would like to discuss the possibility for adapters to issue patch level releases independent of a server release. This would allow for MAJOR.MINOR versions to remain consistent so to communicate compatibility with a given Keycloak server version. But would provide flexibility for adapters to deal with both issues noted above.

I'm +1 on keeping independent version numbers from the server and state something like connect-keycloak v0.0.17 is compatible with Keycloak server 1.9.2.Final for example. But I'm -1 on independent release dates — unless we have a critical security fix.

If we release whenever we want the adapters, would be hard to test, because we don't have bandwith for it. I see as something which could become out of control like v0.0.15, v0.0.16, v0.0.17 is compatible with KC 1.9.2, but v0.0.18 only works with KC 2.0.x.

And just for the sake of argument, let's look at a hypothetical situation where Keycloak is baptized a Product, and the release cadence slows down significantly to every 12-18 months. What if a major security flaw is discovered in an adapter? Should this trigger a new release of Keycloak server itself? Would it not be better to allow the adapter to issue a patch level release instead?

+1 for just the adapter issue a patch level release.

tl;dr my personal opinion is (anything that I'm going to say here can be innacurate, because I don't have the whole context about the conversations during the F2F):

- Adapters should be released on exactly the same dates as the server (with one single exception, critical security fixes)

- The version number should be independent, in this way we state something like Node.js adapters version 0.0.17 are fully compatible with KC 1.9.x.

If you've read this far - thanks! Looking forward to your thoughts.

Lance

_______________________________________________
keycloak-dev mailing list
keycloak-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/keycloak-dev