WildFly Feature Development Process
by Brian Stansberry
Hey folks,
I'm overdue for a...
<wall-of-text>
The WildFly project has some pretty high standards around quality,
stability and backwards compatibility, and those heavily influence how we
go about bringing new features into WildFly. WildFly developers like me who
also work on Red Hat JBoss EAP are aware of EAP's policies and processes
for doing new feature development, and we bring those to how we do our work
on WildFly, but WildFly has its own needs and would benefit from processes
of …
[View More]its own.
This thread and other related discussions are intended to drive creation of
these kinds of policies, and to work through technical issues that stand in
the way. I’m hoping we can make good progress on this this month, and get
something finalized by the end of Q1 next year.
Skip ahead to the ‘Developing the Process’ section if you’re not interested
in background / context discussion.
Background
WildFly is the upstream project that Red Hat’s JBoss EAP product derives
from. Like most software products, EAP has processes that are meant to
govern how new functionality gets introduced into the product, in order to
ensure the high quality and long-term supportability of the product. EAP is
derived from WildFly, and a high percentage of the development work on
WildFly is performed by Red Hat engineers who also work on EAP, so it’s
natural that the EAP processes have affected how those engineers do their
work on WildFly.
This has brought a ton of benefit to WildFly, primarily because the EAP
requirements help ensure that people other than the developer coding up a
feature are involved in the identification of the use cases for and
requirements of the feature, and they also help ensure the feature is
properly documented and tested. People with quality engineering and
documentation expertise are involved in the development of the feature,
which is very much a good thing.
However, there are downsides to using EAP processes to guide a lot of
WildFly development:
* The EAP processes are meant for EAP and are not visible to the WildFly
community. An open source software community should be able to see and
understand how development work is done. A project with the scope and
complexity of WildFly needs processes to ensure quality, but those
processes should be WildFly’s own and should be publicly visible.
Quasi-borrowing EAP processes doesn’t meet this goal.
* The EAP processes can lead to an “all or nothing” effect. They require
the involvement of people from different groups within Red Hat before a
feature is merged. This is great in general but it can lead to situations
where people not being able to prioritize a particular feature blocks it,
even though the development of the feature has reached a point where it can
offer real value to some WildFly users. If the feature is blocked, those
users don’t benefit, and the WildFly developers don’t get the benefit of
feedback on the feature from those users or other WildFly developers who
would only encounter it once it's merged.
Developing the Process
So, what are we doing about this?
The goal is to define the various requirements for what constitutes a high
quality feature, and to define a path people working on that feature can
follow. The path should include a way for the feature to be made available
to some users before it reaches its final form, without exposing it to
users who only want finalized features.
Note that the main focus of this post is on coming up with a suitable
process, but there are technical tasks that need to be done to make it
practical to expose features to some but not all users and to allow them to
best control what they can use in their WildFly installation. A great
example of that kind of thing Paul Ferraro’s work on the management API:
https://lists.jboss.org/archives/list/wildfly-dev@lists.jboss.org/thread/...
For the process itself, back in September I created a working document to
try and capture some of the elements that would go into a possible process:
https://docs.google.com/document/d/15_yKhW74-X9s2zUhs_ZUuZ3h-RlMfH5xWmHHs...
The gist of that document is the notion from Paul’s work that features can
be merged with different initial quality/stability/maturity levels (we need
to pick a single term!), as represented by columns in the table, with a
given feature expected to be ‘promoted’ over time from columns to the left
toward columns to the right. The rows in the table represent different
aspects of ensuring a feature is at a particular quality/stability/maturity
level. The cells in the table then are quick notes on what requirements
would have to be met for a given aspect in order for the feature to be
merged at a given quality level.
What I’m hoping is discussion here and on Zulip about all this can help
flesh out what’s in that document, rework it as necessary, and eventually
lead to a more formal process document.
Some details on the quality levels:
* Experimental. This level would be for real bleeding edge stuff. No
WildFly distribution zip/tar would enable this level by default.
* Preview. This would be the level for features at a sufficient level to be
available by default[1] in WildFly Preview, but not in standard WildFly.
* Community. This would be the level for features at a sufficient level to
be available by default in standard WildFly.
* Default. Features at this level have gone through additional vetting to
ensure they are suitable for the long-term compatibility expectations of
the feature pack that provides them. (See ‘Relationship to Feature Packs’
for more on this.)
[1] Note that ‘available by default’ could but often wouldn’t mean ‘enabled
by default’, i.e. turned on in a standard OOTB configuration. It just means
a user could turn it on if they so choose using the current configuration
mechanisms.
Some details on the feature aspects:
* Feature Team -- this aspect relates to what people need to be involved
with the feature.
* Requirements Analysis -- this relates to ensuring the user stories and
technical requirements for the feature are thought through and written
down. We’ve been doing these for a long time now via the documents in the
wildfly-proposals repo.
* Implementation -- the production code aspect of the feature.
* Domain transformation -- a specialized production code aspect, related to
allowing a current version Domain Controller to manage Host Controllers and
servers running an earlier version.
* Test plan -- Thinking about and writing down what needs to be tested and
how.
* Test development -- writing the tests.
* Test verification -- test review, and verification that the test plan was
implemented and that the tests pass.
* Documentation -- proper documentation of the feature.
Note that it is likely that the set of rows in the table will need
expansion, e.g. to capture requirements for things like the HAL web
console, cloud images, tooling like WildFly Glow etc.
Relationship to Feature Packs
The main WildFly code base produces three different feature packs, and the
main reason there are three and not just one relates to some of the same
stability/long-term compatibility concepts that are a key part of the
feature development process I’m hoping we create. It’s important that we
think carefully about how the feature development process relates to the
feature packs, so to help with that I want to talk a bit about how feature
packs are meant to work.
Ideally for any feature pack, the documentation of that feature pack would
include somewhere ‘lifecycle’ information that can help users decide if the
feature pack is suitable for their needs. This would cover key elements
like:
* The basic scope of the feature pack.
* A likely release cadence for the feature pack.
* The expected long-term maintainability and compatibility for the feature
pack.
Different expectations for those key elements are prime reasons for
creating a new feature pack versus adding functionality to an existing one,
or for choosing one feature pack over another for a feature.
Honestly, the WildFly project does a poor job of documenting these things
for its feature packs, which is my fault, and is why I need to write some
of this in this already looooong post! Anyway…
WildFly produces three feature packs from its main repository. All share
the same expected release cadence (currently a new feature release roughly
quarterly and one bug fix release about a month after a feature release.)
The scope of all three is similar and broad -- they provide functionality
to run in, manage or act as a client to an application server process.
Their primary differences relate to long-term maintainability and
compatibility:
wildfly-ee -- We don’t talk about this feature pack a lot, often treating
it as an internal detail and not producing any downloadable zip/tar built
solely using it, but it’s an important piece of our ecosystem. The
defining characteristic of this feature pack is that it integrates
technologies where we have the highest confidence in our ability to provide
them in a largely compatible way for many years. We’re not perfect about
this, we made some mistakes in the early iterations of this feature pack,
and we can and will break compatibility if necessary. But we try not to and
try to give advance warning if we will. For example, this feature pack
provides the Elytron security layer, which was introduced as a replacement
for the Picketbox security layer many years before we removed support for
the Picketbox layer.
wildfly -- This feature pack depends upon wildfly-ee and adds functionality
in addition to what’s in wildfly-ee. The traditional standard WildFly
server zip is built using this feature pack. The primary reason we put
things in this feature pack instead of wildfly-ee is because what we’re
integrating is more likely to change in incompatible ways over a relatively
short time period. For example, MicroProfile specifications are comfortable
introducing breaking changes on an annual basis, making them not a great
fit for wildfly-ee. The observability space, particularly metrics and
tracing, is evolving rapidly, so our Micrometer and OpenTelemetry
extensions are not in wildfly-ee.
wildfly-preview -- This feature pack is all about the fact that it provides
no long term guarantees and can change significantly from release.
We need to think more about how feature quality levels relate to this, but
here are a few thoughts:
* Just because a user wants to use a particular preview or experimental
quality feature doesn’t mean they want to use an entire preview quality
feature pack like wildfly-preview. IOW there is a use case for preview or
experimental features in standard WildFly.
* OTOH we can use WildFly Preview to showcase functionality whose scope is
not tied to a particular reasonably scoped ‘feature’. Using it for EE 9 was
an obvious example. Not having an embedded messaging broker in the OOTB
configs is not a ‘feature’. IOW there is a use case for WildFly Preview
even if standard WildFly has preview features.
* What we mean by the ‘Community’ and ‘Default’ levels is “relative to” the
generally expected long-term maintainability and compatibility level of the
feature pack that provides it. IOW just because a feature provided by the
wildfly feature pack has been vetted as suitable for the ‘Default’ level
doesn’t mean it comes with higher expectations than the feature pack as a
whole. If we provide a specification, and we integrate it in a way that is
highly stable and very well tested, but then the spec changes in a year in
a significantly incompatible way, then we may have to as well.
Next Steps
I’d love to hear your thoughts and questions, either on this thread, in
Zulip (
https://wildfly.zulipchat.com/#narrow/stream/174184-wildfly-developers) or
in comments on the Google doc I linked above. Please have a look into
Paul’s work I mentioned above, as well as other technical work that will
likely be happening over the next six months.
As I noted at the start, I’m hoping we can have a solid process written and
published by the end of Q1 next year, and that we can nail down some of the
key concepts over the next few weeks well enough that we can integrate
Paul’s work.
I'm looking forward to hearing from you.
</wall-of-text>
Best regards,
--
Brian Stansberry
WildFly Project Lead
Principal Architect, Red Hat JBoss EAP
He/Him/His
[View Less]
1 year, 3 months
Processing an issue in a new Feature Development Process
by Tomasz Adamski
Hi,
This mail is a follow up to the discussion regarding processing WFLY-17180
in the new feature process. I would like to discuss what are the steps that
I have to take to follow the new process. I have a feature that is a model
change (adding query-time parameter to datasource validation) so it seems
to me that it would qualify as either preview or community feature: imho it
is neither complex nor risky, it is already implemented (the implementation
is partially done in IronJacamar library, …
[View More]so it would require
component upgrade), the tests are provided as part of implementation PR and
analysis as part of other PRs as well.
If we decide that this issue is suitable to be delivered using the new
process the initial questions that come to my mind are:
What is the maturity of the issue and how is it going to be decided?
How is it going to be traced in JIRA?
What steps do I need to take to bootstrap the process?
Regards,
Tomek
[View Less]
1 year, 3 months
Discovery of Galleon layers in the WildFly testsuite
by Jean Francois Denise
Hi,
an heads-up that we are changing the approach when provisioning WildFly
with Galleon layers in the WildFly testsuite.
Prior to https://github.com/wildfly/wildfly/pull/17153 (that should be
merged soon), a base layer (e.g.: cloud-server) + some decorator layers
(according to the test case) were explicitly set to provision servers
used by tests. So the actual set of required Galleon layers was
generally larger than what the test actually requires.
Thanks to the WildFly Glow <https:/…
[View More]/github.com/wildfly/wildfly-glow>
Arquillian Maven plugin, we are now scanning the Arquillian deployments
(prior to the test execution) to discover the required Galleon layers. A
provisioning.xml file containing the discovered layers is generated and
used by the WildFly Maven plugin to actually provision the test server.
In order to validate that what the scanning has discovered is expected,
you can configure the wildfly-glow-arquillian-plugin maven plugin to
contain the element <expected-discovery>. For example:
<expected-discovery>[cdi, ee-integration, ejb, ejb-lite,
elytron-oidc-client, naming,
servlet]==>ee-core-profile-server,ejb,elytron-oidc-client</expected-discovery>
The left part of the arrow contains the list of the discovered layers
according to the scanning. The right part is what will get provisioned.
Composed of a base layer (always ee-core-profile-server) + a list of the
discovered layers that has been cleaned-up to avoid to include dependencies.
BTW: The PR https://github.com/wildfly/wildfly/pull/17153 contains a lot
of examples of WildFly Glow scanning executions that you can use as
starting-point.
In case the test depends on some not discoverable features, you have the
ability to explicitly add add-ons to enrich the set of provisioned
layers (using the <add-ons> configuration element).
In addition, it can happen that some errors are identified by WildFly
Glow. Some need to be fixed, some could be fine in the context of the
tests. Errors that are fine in the context of the test can be ignored
using the <expected-errors> element.
Known WildFly Glow discovered errors:
* jakarta.naming.Context or InitialContext lookup: In case some lookup
is present in the test code that could hide the usage of a layer not
discovered. You can configure the plugin with <add-layers-for-jndi> to
add the layers that are required (if any) and hidden.
* ambiguous resource injection: An un-typed resource injection could
hide a required layer. You can configure the plugin with
<add-layers-for-jndi> to list the layers that are required (if any) and
hidden.
* an add-on of the messaging family is expected by the
messaging-activemq layer. One of embedded or remoting messaging add-on
is required.
* unbound datasources error: <name of ds>: A datasource is expected but
not found in the deployment. That is generally an error to ignore, the
test itself should add the datasource during the setup.
* no default datasource found error: It has been identified that a
default datasource is required. That can be fixed by adding the add-on
h2-database:default
You can find more information related to WildFly Glow in its
documentation <http://docs.wildfly.org/wildfly-glow/>.
Thank-you.
JF
[View Less]
1 year, 3 months
