Turn on dependabot
by Brian Stansberry
Occasionally we've thought about turning on dependabot for the main WildFly
repo, and a couple current discussions (see [1] and [2]) relate to that, so
it seems a good time to discuss further and perhaps take action.
My main concern with dependabot is it doesn't integrate with JIRA. JIRA is
really important to how we're able to keep a handle on a project as complex
as WildFly. And I think it's important to track component upgrades in JIRA
so our users can keep an eye on what we're providing. Particularly
important in the world of ubiquitous CVE scanners.
But James Perkins has pointed out that such JIRA tracking is kind of
overkill for non-production dependencies (e.g. test and build deps) and I
agree.
So, how about we turn on dependabot and require a JIRA to be filed and
linked to the PR if the proposed upgrade is production code dep? For
non-production deps a JIRA would be optional.
The other thing I care about a lot is being able to grep the git log for
commits related to a JIRA. That would of course be lost for non-production
upgrades with no JIRA. Oh well. Also though dependabot wouldn't put our
JIRA in its commit messages. But for PRs where we file a JIRA we can
require human edit of the dependabot PR title to reference the JIRA. That
will result in the JIRA appearing in the log via the merge commit Github
generates. That solves the git log use case adequately enough IMO.
Thoughts?
[1]
https://lists.jboss.org/archives/list/wildfly-dev@lists.jboss.org/thread/...
[2]
https://lists.jboss.org/archives/list/wildfly-dev@lists.jboss.org/thread/...
Best regards,
Brian
2 years, 3 months
Changes to WildFly Preview test profile activation
by Brian Stansberry
I've filed https://github.com/wildfly/wildfly/pull/17049 which, if merged,
will change what developers need to do to run the testsuite against WildFly
Preview.
Instead of -Dts.ee9 you would use -Dts.preview.
For bootable jar testing instead of -Dts.bootable.ee9 you would use
-Dts.bootable.preview.
I configured the ci.wildfly.org jobs that test main to use both sets of
properties, so they can test the PR. Once it's merged I'll remove the ee9
variants from the job configs.
Best regards,
Brian Stansberry
He/Him/His
2 years, 3 months
Future of management console / HAL codebase
by Harald Pehl
With the latest release of PatternFly 5 [1], I'd like to open a discussion about the future of the HAL codebase and possible major updates.
HAL currently depends on PatternFly 3.x, released on Feb 1st, 2016 [2]. PatternFly 3.x itself depends on many old JS dependencies like jQuery 3.x and Bootstrap 3.x, which are either no longer supported [3] or don't receive new updates.
While HAL itself is pretty stable, this brings the danger that we run into CVEs that can no longer be fixed simply by updating the dependency. One possible solution would be to update to the latest PatternFly version. This would have more advantages:
- Fresh, new look and feel
- Enhance user experience
- Align with other consoles like OpenShift & Keycloak
I see two basic approaches when updating to PatternFly 5:
1. Switch the technical stack to TypeScript & React
Pros:
- Recommended way how to use PatternFly
- Well documented
- Active community
- State of the art
Cons:
- Rewrite HAL from scratch
- Hard to re-use existing codebase
- Technology-independent building blocks like DMR-related code, RBAC security checks, and model-based UI would be lost
- Weak expertise
- Contributors need to adapt
2. Stick with the current architecture (Java & GWT) and migrate the codebase
Pros:
- Technology-independent building blocks can be re-used
- Could also be the foundation of halOS
- Well-known and rock-solid architecture
- Strong expertise
- Some work has already been done [4]
Cons:
- GWT community does still exist, but it is small and not very active
- Rewrite a lot of UI-related code
Finally, we should consider the following issues:
- With the focus on running WildFly in the cloud, do we really need to make this effort? Is it worth it?
- When updating HAL, we should keep in mind that the test suite can be adapted with minimal effort.
Before going into any direction, I'm keen to hear what you think about it. Suggestions, ideas, and critics are welcome!
---
Harald Pehl
JBoss by Red Hat
[1] https://www.patternfly.org/get-started/upgrade
[2] https://github.com/patternfly/patternfly-3/releases/tag/v3.0.0
[3] https://blog.getbootstrap.com/2019/07/24/lts-plan/
[4] https://github.com/patternfly-java
2 years, 3 months
WildFly 30 Schedule
by Brian Stansberry
In May I published a rough roadmap for WildFly feature release dates in
2023-2023. That had October 12 as a likely date for WildFly 30. That date
still looks reasonable so I'd like to stick to it.
That would result in these detailed dates for getting the release out:
30 Beta
Wed Sep 20 -- PR submission deadline
Fri Sep 22 -- WildFly Core ready to tag
Mon Sep 25 -- Final merges and checks
Tue Sep 26 -- Buffer day
Wed Sep 27 -- Tag and deploy to repository.jboss.org
Thu Sep 28 -- Release available on wildfly.org, public announcement.
Fri Sep 29 -- Other release deliverables not ready on release day.
30 Final
Wed Oct 4 -- PR submission deadline
Fri Oct 6 -- WildFly Core ready to tag
Mon Oct 9 -- Final merges and checks
Tue Oct 10 -- Buffer day
Wed Oct 11 -- Tag and deploy to repository.jboss.org
Thu Oct 12 -- Release available on wildfly.org, public announcement.
Fri Oct 13 -- Other release deliverables not ready on release day.
[1] https://www.wildfly.org/news/2023/05/11/WildFly-Roadmap/
Best regards,
Brian Stansberry
WildFly Project Lead
He/Him/His
2 years, 3 months
WFCORE-6221 Incorporating preview/experimental features in WildFly
by Paul Ferraro
We have long wanted the ability to easily give users opt-in access to
less-than-stable features within WildFly.
To this end, WildFly currently includes a preview feature pack that
facilitates the delivery of preview features to WildFly users.
This gives us the ability to include new or alternate versions of modules
not included in our default feature pack.
However, this mechanism is not well utilized, as evident by the large
number of feature proposals sitting the pull request queue, largely because
the vast majority of "features" do not naturally arrive via a new module
(or different version of an existing module), but rather via changes to
existing modules usually residing within the WildFly codebase.
For the purpose of this proposal, I am mostly concerned with "features" as
defined as new runtime behavior enabled via configuration within a new or
existing subsystem.
Usually, development of a new feature not only involves the feature code
itself, which may be bundled with the wildfly codebase, or via an external
component; but also changes to the management model of the corresponding
subsystem otherwise required to enable the feature. This might be a new
subsystem, but more typically, a new resource within an existing subsystem,
or a new attribute of an existing resource, etc.
Rather than only being able to control the set of available features via
controlling the modules of a given feature pack, it would be more useful, I
think, to allow existing modules to enable features by filtering a
subsystem's management model, thus exposing/restricting the configuration
needed to enable feature's runtime behavior.
Several months ago, I created https://issues.redhat.com/browse/WFCORE-6221
which proposes to formalize the concept of a "feature stream" within the
WildFly kernel.
We currently only support the inclusion of stable, well-tested features,
which generally requires a new subsystem management model version. Let's
call this the STABLE feature stream, where a "feature stream" is a set of
features with specific stability guarantees, e.g. STABLE, PREVIEW,
EXPERIMENTAL, etc.
By associating incoming features with a non-STABLE "feature stream", e.g.
PREVIEW, EXPERIMENTAL, we can more quickly include new features into
WildFly, allowing users access to them via a simple opt-in mechanism. This
way we can more quickly evolve WildFly while still retaining the same
testing standards required for a feature to be deemed STABLE.
While we can complicate things later, let's assume for now that feature
streams are a nested hierarchy.
i.e.
- a server configured with the STABLE feature stream will only contain
STABLE features, not PREVIEW nor EXPERIMENTAL features
- a server configured with the PREVIEW feature stream will contain
STABLE and PREVIEW features, but not EXPERIMENTAL features.
- a server configured with the EXPERIMENTAL feature stream will contain
all features.
WFCORE-6221 proposes that the features exposed by a given subsystem are
defined, not just by its management model, but also the feature stream of
the server.
To achieve this, WFCORE-6221 proposes the following changes to WildFly core:
- Add the ability to start WildFly with a specific "feature stream"
- This takes inspiration from JEP 12, which introduced "preview
features" to OpenJDK (https://openjdk.org/jeps/12)
- e.g.
- ./standalone.sh --feature-stream=experimental
- ./domain.sh --feature-stream=experimental
- Add the ability to manipulate management model registration based
on the "feature stream" of the server
- Add support for "feature stream"-specific subsystem XML namespaces
A WildFly server instance is assigned a "feature stream" at startup, either
via the command line (for the standalone use case), or via its host
controller (for the managed domain use case). By default, a server will
use the STABLE feature stream.
Let's look at a few different use cases, and explore how each might be
handled. Forgive me in advance if all of my examples are
clustering-related... :)
In general, I will show 2 approaches: one using programmatic filtering, and
the other using auto-filtering.
I expect most users would use the auto-filtering approach.
1. Introducing an experimental feature enabled via a new subsystem
e.g. https://issues.redhat.com/browse/WFLY-14953
The module containing the extension for an experimental subsystem needs to
be made available within the target feature pack.
However, an experimental subsystem simply skips registration if the current
feature stream does not support EXPERIMENTAL features.
e.g.
public class FooExtension implements Extension {
@Override
public void initialize(ExtensionContext context) {
if (context.enables(FeatureStream.EXPERIMENTAL)) {
SubsystemRegistration subsystem =
context.registerSubsystem("foo",
FooSubsystemModel.VERSION_1_0.getVersion());
// ...
}
}
// ...
}
To promote this feature to the PREVIEW stream, we simply change our logic
accordingly:
Promotion to the STABLE stream can remove the condition entirely, since
context.enables(FeatureStream.STABLE) always return true.
However, promotion to STABLE will likely involve incrementing the
management model version, so existing processes for stable features will
apply.
Alternatively, if the subsystem is self-contained within its own extension
(as opposed to an existing extension), we can simply associate the
extension with a specific feature stream.
e.g.
public class FooExtension implements Extension {
// ...
@Override
public FeatureStream getFeatureStream() {
return FeatureStream.EXPERIMENTAL;
}
}
When the server loads the extension, it will automatically skip
initialization of any extensions not enabled by the current feature stream
of the server.
2. Introducing an experimental feature enabled via a new resource of an
existing subsystem
e.g. https://issues.redhat.com/browse/WFLY-16345
Similar to the above, we need to skip registration of the experimental
resource definition if the current feature stream does not support
EXPERIMENTAL features.
If the experimental resource is never registered, it never installs the
services required to enable the experimental feature.
e.g.
@Override
public void registerChildren(ManagementResourceRegistration parent) {
if (parent.enables(FeatureStream.EXPERIMENTAL)) {
parent.registerSubModel(new FooResourceDefinition(...));
}
}
Alternatively, we can simply associate the ResourceDefinition with a
specific feature stream.
e.g.
class FooResourceDefinition extends SimpleResourceDefinition {
// ...
@Override
public FeatureStream getFeatureStream() {
return FeatureStream.EXPERIMENTAL;
}
}
When registering this resource via
ManagementResourceRegistration.registerSubModel(new
FooResourceDefinition(...)), the server will omit registration if the
feature stream associated with the ResourceDefinition is not enabled by the
server.
N.B. Care must be taken when using this approach, as the
registerSubModel(...) method will return null if registration was skipped.
3. Introducing an experimental feature enabled via a new attribute of an
existing subsystem resource
https://issues.redhat.com/browse/WFLY-18000
Similar to the above, we need to skip registration of the experimental
attribute if the current feature stream does not support EXPERIMENTAL
features.
e.g.
class FooResourceDefinition extends SimpleResourceDefinition {
static final AttributeDefinition BAR = ...; // Our new attribute that
enables the new experimental feature
// ...
@Override
public void registerAttributes(ManagementResourceRegistration
registration) {
if (registration.enables(FeatureStream.EXPERIMENTAL)) {
registration.registerReadWriteAttribute(BAR, null, new
ReloadRequiredWriteAttributeHandler(FOO);
}
}
}
Unfortunately, the current registration mechanism available in
wildfly-core, which registers the OperationDefinition parameters of the add
operation parameters independently from resource attributes (via different
ResourceDefinition.registerXXX(...) methods), makes this awkward.
Additionally, resource add operation handlers and write-attribute operation
handlers are constructed with a separately defined set of parameters
(rather than using the parameters of the corresponding OperationDefinition).
For this reason, I submitted https://issues.redhat.com/browse/WFCORE-6407
(WIP https://github.com/wildfly/wildfly-core/pull/5563) which eliminates
the need to construct add resource operation handlers or write-attribute
operation handlers using a set of attributes.
Until that change is in place, most resource definitions for most
subsystems (i.e. those not using the registration mechanics from
wildfly-clustering-common) will require separate logic to exclude the
EXPERIMENTAL attributes from its add operation handler independently from
the resource's attributes. Consequently, until WFCORE-6407 is complete,
add operation parameter handling will be very awkward:
e.g.
class FooResourceDefinition extends SimpleResourceDefinition {
static final AttributeDefinition ATTRIBUTE = //... an existing attribute
// Our new experimental attribute
static final AttributeDefinition BAR = new
SimpleAttributeDefinitionBuilder("bar", ModelType.STRING);
// N.B. FeatureStream.complete(...) is a convenience method that
returns a full map of feature-per stream
// e.g. will auto-map FeatureStream.PREVIEW to the FeatureStream.STABLE
value
// In this way, the addition of a new feature stream will not affect
existing usage
static final Map<FeatureStream, Collection<AttributeDefinition>>
ATTRIBUTES = FeatureStream.complete(Map.of(FeatureStream.STABLE,
List.of(ATTRIBUTE), FeatureStream.EXPERIMENTAL, List.of(List.of(ATTRIBUTE,
BAR)));
// ...
public FooResourceDefinition(ManagementResourceRegistration parent) {
super(new Parameters(PATH, DESCRIPTION_RESOLVER).setAddHandler(new
ReloadRequiredAddStepHandler(ATTRIBUTES.get(parent.getFeatureStream()))));
}
// ...
}
W.R.T. runtime, if the experimental attribute is never registered, it will
not be allowed within our resource's add operation, and thus will always
resolve to its default value.
Alternatively, once WFCORE-6407 is complete, we can associate an
AttributeDefinition with a FeatureStream and perform the conditional
registration automatically.
e.g.
static final AttributeDefinition BAR = new
SimpleAttributeDefinitionBuilder("bar", ModelType.STRING)
.setRequired(false)
.setValidator(new EnumValidator<>(EnumSet.allOf(Baz.class))
.setFeatureStream(FeatureStream.EXPERIMENTAL)
.build();
The attribute registration methods of ManagementResourceRegistration will
omit registration of an attribute its associated feature stream is not
enabled by the server.
Similarly, the OperationDefinition of the add operation of the containing
ResourceDefinition will omit this attribute from its allowed parameters if
the feature stream associated with the AttributeDefinition is not enabled
by the server.
4. Introducing an experimental feature enabled via a new value of an
existing subsystem resource attribute.
e.g. https://issues.redhat.com/browse/WFLY-13904
Typically, this would involve adding a new value to an existing enum.
Here we need to conditionally register a ParameterValidator specific to the
current FeatureStream.
As with the previous example, selecting the appropriate validator for a
given "feature stream" is also awkward due to the way that resource
attributes vs resource add operation parameters are handled.
With the existing limitations, a "feature stream"-specific validator can be
registered using logic such as:
e.g.
Using our AttributeDefinition BAR from the above example, which specifies a
value enumerated by the enum Baz.
Our experimental feature involves a newly added QUX value to our Baz enum.
static final Map<FeatureStream, Set<Baz>> BAZ_VALUES =
FeatureStream.complete(Map.of(FeatureStream.STABLE,
Enum.complementOf(EnumSet.of(Baz.QUX)), FeatureStream.EXPERIMENTAL,
EnumSet.allOf(FeatureStream.class)));
During attribute registration, we specify the validator specific to the
current stream.
e.g.
@Override
public void registerAttributes(ManagementResourceRegistration registration)
{
ParameterValidator bazValidator = new
EnumValidator<>(BAZ_VALUES.get(registration.getFeatureStream()));
// Copy attribute and apply correct validator
AttributeDefinition attribute
= SimpleAttributeDefinitionBuilder.create(BAR).setValidator(bazValidator).build();
registration.registerReadWriteAttribute(attribute, null, new
ReloadRequiredWriteAttributeHandler(attribute));
}
Not so pleasant...
Due to the same limitation of the current registration mechanics as
described previously, a similar hack will be needed to ensure that the
AttributeDefinition provided to the constructor of the add
OperationStepHandler has the correct validator applied. Again, this
limitation will be addressed via WFCORE-6407.
Alternatively, with some minor changes to the ParameterValidator interface,
and once WFCORE-6407 is complete, we can associate a ParameterValidator
with an AttributeDefinition per feature stream and perform the selection
automatically wherever necessary, e.g. via the base OperationStepHandler
implementations. I have not completely thought this through, but my
current thinking is something like:
e.g.
static final AttributeDefinition BAR = new
SimpleAttributeDefinitionBuilder("bar", ModelType.STRING)
.setRequired(false)
.setValidator(new
FeatureStreamValidator(Map.of(FeatureStream.STABLE, new
EnumValidator<>(Enum.complementOf(EnumSet.of(Baz.QUX))),
FeatureStream.EXPERIMENTAL, new EnumValidator<>(Enum.allOf(Baz.class)))))
.build();
... where FeatureStreamValidator is a composite ParameterValidator
implementation that delegates to a specific ParameterValidator depending on
the feature-stream of the server.
5. Subsystem XML parsing
Just as the feature stream is a new dimension to a subsystem's management
model version - so too is the feature stream an optional dimension of a
subsystem configuration XML namespace.
Say the current version of an existing subsystem uses the XML namespace
"urn:wildfly:foo:2.1"
Implementing a new experimental feature would involve a new XML namespace
"urn:wildfly:foo:experimental:2.1"
If/when this feature is promoted to STABLE, we would need to increment the
schema version itself, e.g. "urn:wildfly:foo:2.2"
If instead, a new stable feature is added, and the experimental feature
remains experimental, we would increment the version for both the stable
and experimental schemas.
e.g. "urn:wildfly:foo:2.2", "urn:wildfly:foo:experimental:2.2"
W.R.T. XML parsing, filtering attributes/resource by stream must be done
inline with existing filtering by version.
e.g.
Consider the following set of subsystem namespaces:
public enum FooSubsystemSchema implements
PersistentSubsystemSchema<FooSubsystemSchema> {
VERSION_1_0(1),
VERSION_2_0(2),
VERSION_2_0_EXPERIMENTAL(2, FeatureStream.EXPERIMENTAL), // We added a
new experimental attribute
;
private final VersionedNamespace<IntVersion,
ExperimentalSubsystemSchema> namespace;
ExperimentalSubsystemSchema(int major) {
this(major, FeatureStream.DEFAULT);
}
ExperimentalSubsystemSchema(int major, FeatureStream stream) {
this.namespace =
SubsystemSchema.createSubsystemURN(FooSubsystemResourceDefinition.SUBSYSTEM_NAME,
new IntVersion(major), stream);
}
@Override
public VersionedNamespace<IntVersion, ExperimentalSubsystemSchema>
getNamespace() {
return this.namespace;
}
@Override
public PersistentResourceXMLDescription getXMLDescription() {
PersistentResourceXMLBuilder builder =
builder(FooSubsystemResourceDefinition.PATH, this.namespace);
if (this.namespace.since(VERSION_2_0)) {
// BAR is new since version 2.0, but only for specific feature
streams
builder.addAttributes(FooSubsystemResourceDefinition.ATTRIBUTES.stream().filter(this::enables));
} else {
// BAR does not exist prior to version 2.0
builder.addAttributes(FooSubsystemResourceDefinition.ATTRIBUTES.stream().filter(Predicates.not(BAR)));
}
return builder.build();
}
}
Registering subsystem parsers should generally look the same as it does
now, since the server can skip registration of schemas associated with a
feature stream not supported by the server.
e.g.
@Override
public void initializeParsers(ExtensionParsingContext context) {
// This will skip registration of
FooSubsystemSchema.VERSION_2_0_EXPERIMENTAL if the server does not support
it
context.setSubsystemXmlMappings(FooSubsystemResourceDefinition.SUBSYSTEM_NAME,
EnumSet.allOf(FooSubsystemSchema.class));
}
Subsystem extensions will also need to register the appropriate writer
based on the feature stream of the server.
// The "current" schema will depend on the feature stream of the server
static final Map<FeatureStream, FooSubsystemSchema> CURRENT_SCHEMAS =
FeatureStream.complete(Map.of(FeatureStream.STABLE, VERSION_2_0,
FeatureStream.EXPERIMENTAL, VERSION_2_0_EXPERIMENTAL));
@Override
public void initialize(ExtensionContext context) {
SubsystemRegistration subsystem =
context.registerSubsystem(FooSubsystemResourceDefinition.SUBSYSTEM_NAME,
FooSubsystemModel.VERSION_2_0.getVersion());
// ...
subsystem.registerXMLElementWriter(new
PersistentResourceXMLDescriptionWriter(CURRENT_SCHEMAS.get(context.getFeatureStream())));
}
6. Misc concerns
- Subsystem model transformers for mixed-domains
- I anticipate that we would restrict the use of mixed-domains to the
STABLE feature stream. That means that only STABLE features need to be
concerned with subsystem model transformations.
- Experimental/preview wildfly kernel features
- The above mechanisms should work for any features configured by a
ResourceDefinition/AttributeDefinition, even if they have no
corresponding
subsystem
- Anything else would need to conditionally enable based on the
feature stream of the controller
That's about all I have for now.
Again, I think this approach should cover the bulk of feature development
use cases in WildFly.
Let me know if anything was particularly unclear, confusing, or requires
elaboration; or if there are any major use cases that I have missed.
STATUS:
I have a pull request open for WFCORE-6221 [1] that implements most of the
above. It is still a work in progress - and needs to be rebased on my
WFCORE-6407 branch (once that is complete).
Please browse my topic branch [2], and leave any comments on the PR [3]. A
good place to start is the integration tests [4], which validates this
against a sample subsystem demonstrating several of the above use cases.
For any design-related discussion, either reply to this thread or to the
WFCORE-6221 jira itself.
Paul Ferraro
[1] https://issues.redhat.com/browse/WFCORE-6221
[2] https://github.com/pferraro/wildfly-core/tree/
[3] https://github.com/wildfly/wildfly-core/pull/5413
[4]
https://github.com/pferraro/wildfly-core/tree/WFCORE-6221/subsystem-test/...
2 years, 3 months
