Create CacheEmbeddedConfigProvider #39126

pruivo · 2025-04-22T14:18:52Z

Closes #38497

Notes

Removed DefaultInfinispanConnectionProviderFactory#initEmbedded() -> it was used by the store model tests.
A Infinispan configuration file is always required -> create a basic configuration file for the store model tests
Most of the logic is present in DefaultCacheEmbeddedConfigProviderFactory
CacheConfigurator has the logic to configure caches based on CLI options; JGroupsConfigurator is the same for the JGroups configuration. Trying to keep the class DefaultCacheEmbeddedConfigProviderFactory small and easy to read.

Tasks

Update changes documentation
Review documentation
Javadoc
Statis analysis: remove or deprecate unused code

ahus1 · 2025-04-23T20:06:54Z

Failing FIPS tests have been fixed in #39125

ahus1

Looks good to me, see pruivo#8 for two changes:

Removal of some files which are IMHO not part of the public API, and I also don't see how they are being used.
I saw you added an option to disable the ISPN metrics. IMHO this wasn't available earlier. When we add it, I want to make this option explicit and documented in the CLI. I moved it to the other metrics open that we have for the embedded caches, so they are in a similar place.

See pruivo#8 for the two commits.

Once these items are resolve, I'm happy to merge it.

docs/documentation/upgrading/topics/changes/changes-26_3_0.adoc

pruivo · 2025-04-28T08:20:30Z

CLI option

It is not a new option, it is a bridge to map the --metrics-enabled option to the SPI option. It should not be exposed.

The current code is reacting to this option to enable the metrics in Infinispan

keycloak/quarkus/runtime/src/main/java/org/keycloak/quarkus/runtime/storage/infinispan/CacheManagerFactory.java

Line 145 in c97a956

if (Configuration.isTrue(MetricsOptions.METRICS_ENABLED)) {

But the option is already mapped to a quarkus option here

keycloak/quarkus/runtime/src/main/java/org/keycloak/quarkus/runtime/configuration/mappers/MetricsPropertyMappers.java

Lines 17 to 19 in 3839f8e

    
           fromOption(MetricsOptions.METRICS_ENABLED) 
        
                   .to("quarkus.micrometer.enabled") 
        
                   .build(),

Steven may implement another alternative to get rid of the mapping:

#39123

Site Name

jboss.site.name has been removed but we never documented it and I added that piece of information to the changelog.

keycloak/model/infinispan/src/main/java/org/keycloak/connections/infinispan/TopologyInfo.java

Lines 61 to 65 in 4dc4de7

    
           if (System.getProperty(InfinispanConnectionProvider.JBOSS_SITE_NAME) != null) { 
        
               throw new IllegalArgumentException( 
        
                       String.format("System property %s is in use. Use --spi-connections-infinispan-%s-site-name config option instead", 
        
                               InfinispanConnectionProvider.JBOSS_SITE_NAME, providerId)); 
        
           }

Multi-site has never used the site's name value. There is no clustering and all the events use DCNotify.ALL_DCS. It is ignored here:

keycloak/model/infinispan/src/main/java/org/keycloak/cluster/infinispan/WrapperClusterEvent.java

Line 65 in bf26d33

senderSite = dcNotify == ClusterProvider.DCNotify.ALL_DCS ? null : senderSite;

It was used by the code that Michal removed and, at the moment has no use. It will become relevant for stretched clusters.

This PR changes --spi-connections-infinispan-quarkus-site-name (still reads this SPI options) to --spi-cache-embedded-default-site-name

Node Name

jboss.node.name is not documented anywhere. It is not used at all in production as it uses the address generated by JGroups.

keycloak/model/infinispan/src/main/java/org/keycloak/connections/infinispan/TopologyInfo.java

Lines 68 to 75 in 4dc4de7

    
           var addr = cacheManager.getAddress(); 
        
           if (addr != null) { 
        
               nodeName = addr.toString(); 
        
               siteName = cacheManager.getCacheManagerConfiguration().transport().siteId(); 
        
               if (siteName == null) { 
        
                   siteName = config.get("siteName"); 
        
               } 
        
           } else {

I'm changing quite a bit in the PR

node name is taken into consideration and used to configure JGroups. Worth mentioning that it is only used for logging purposes (a friendly name than using internal uuid)
Keycloak fails to start if jboss.node.name is set.
A new SPI option to set the node's name.

@ahus1 I hope this clarifies my decision in this PR.

Closes keycloak#38497 Signed-off-by: Pedro Ruivo <[email protected]>

Signed-off-by: Pedro Ruivo <[email protected]>

ahus1 · 2025-04-28T08:38:25Z

It is not a new option, it is a bridge to map the --metrics-enabled option to the SPI option. It should not be exposed.

Thanks, this explains it. A comment would have been helpful to state that "why" it was added. If you think exposing it adds value and would be a feature, consider picking my changes from the other PR. I leave that up to you.

Signed-off-by: Alexander Schwartz <[email protected]>

ryanemerson · 2025-04-28T09:06:18Z

model/infinispan/src/main/java/org/keycloak/connections/infinispan/InfinispanUtil.java

+    /**
+     * @deprecated to be removed without replacement.
+     */
+    @Deprecated(since = "26.3", forRemoval = true)


The deprecated methods in this class (and others) are not used internally anymore. @ahus1 @pruivo What's the policy for deprecating/removing classes in the model modules. Should all code in this module be considered public API or just interfaces? I don't see a package-info.java so it's not clear to me what the expectations are here.

Technically we are allowed to change them in any release, as they are not considered public. Sometimes it is still nice to keep them around for a while if we think people might use them anyway and to allow them a smoother migration.

I assume keeping this methods comes only at the cost that we would need to have an extra PR later to actually remove the code.

I'll leave that final call with @pruivo.

It is not clear what is public API and what users are using (I'm aware some users are creating caches for their providers). I'm playing safe and keeping them around and removing them in 26.4 or 26.5.

That's fine by me. I see we're now making better use of impl.* packages, so in the future hopefully what is/isn't supported will be clearer.

ryanemerson · 2025-04-28T09:26:34Z

...ava/org/keycloak/spi/infinispan/impl/embedded/DefaultCacheEmbeddedConfigProviderFactory.java

+
+    @Override
+    public Set<Class<? extends Provider>> dependsOn() {
+        return Set.of(JGroupsCertificateProvider.class);


Not related to these changes per se, but I noticed that RemoteLoadBalancerCheckProviderFactory depends on InfinispanConnectionProvider but does not have a dependsOn implementation. Is there a reason for this, or should I create an issue to address this?

I assume it is because dependsOn() method was added after RemoteLoadBalancerCheckProviderFactory.
This is a "lazy provider" and is only instantiated when the REST endpoint is invoked.

pruivo force-pushed the t_37912_config_spi branch 4 times, most recently from 650cd5e to 9ba029c Compare April 23, 2025 17:30

pruivo force-pushed the t_37912_config_spi branch from 9ba029c to 42075c3 Compare April 24, 2025 08:12

ahus1 reviewed Apr 25, 2025

View reviewed changes

docs/documentation/upgrading/topics/changes/changes-26_3_0.adoc Outdated Show resolved Hide resolved

docs/documentation/upgrading/topics/changes/changes-26_3_0.adoc Outdated Show resolved Hide resolved

ahus1 self-assigned this Apr 25, 2025

ryanemerson self-requested a review April 28, 2025 07:38

pruivo added 2 commits April 28, 2025 09:21

Create CacheEmbeddedConfigProvider

13070be

Closes keycloak#38497 Signed-off-by: Pedro Ruivo <[email protected]>

remove ManagedCacheManagerProvider

515a858

Signed-off-by: Pedro Ruivo <[email protected]>

pruivo force-pushed the t_37912_config_spi branch from 42075c3 to 515a858 Compare April 28, 2025 08:23

Review

dcb27fd

Signed-off-by: Alexander Schwartz <[email protected]>

ryanemerson reviewed Apr 28, 2025

View reviewed changes

Merge branch 'main' into t_37912_config_spi

a1c5920

ahus1 marked this pull request as ready for review April 28, 2025 10:04

ahus1 requested review from a team as code owners April 28, 2025 10:05

keycloak-github-bot bot added team/cloud-native team/core-clients team/core-iam labels Apr 28, 2025

ahus1 enabled auto-merge (squash) April 28, 2025 10:06

ahus1 approved these changes Apr 28, 2025

View reviewed changes

ahus1 merged commit eafe08a into keycloak:main Apr 28, 2025
80 checks passed

pruivo deleted the t_37912_config_spi branch April 28, 2025 12:22

ryanemerson mentioned this pull request Jun 12, 2025

39429/OfflineSessionPersistenceTest testPersistenceMultipleNodesClientSessionsAtRandomNode #40459

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Create CacheEmbeddedConfigProvider #39126

Create CacheEmbeddedConfigProvider #39126

Uh oh!

pruivo commented Apr 22, 2025 •

edited

Loading

Uh oh!

ahus1 commented Apr 23, 2025

Uh oh!

ahus1 left a comment

Uh oh!

Uh oh!

Uh oh!

pruivo commented Apr 28, 2025

Uh oh!

ahus1 commented Apr 28, 2025

Uh oh!

ryanemerson Apr 28, 2025

Uh oh!

ahus1 Apr 28, 2025

Uh oh!

pruivo Apr 28, 2025

Uh oh!

ryanemerson Apr 28, 2025

Uh oh!

ryanemerson Apr 28, 2025

Uh oh!

pruivo Apr 28, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Create CacheEmbeddedConfigProvider #39126

Create CacheEmbeddedConfigProvider #39126

Uh oh!

Conversation

pruivo commented Apr 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Notes

Tasks

Uh oh!

ahus1 commented Apr 23, 2025

Uh oh!

ahus1 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

pruivo commented Apr 28, 2025

Uh oh!

ahus1 commented Apr 28, 2025

Uh oh!

ryanemerson Apr 28, 2025

Choose a reason for hiding this comment

Uh oh!

ahus1 Apr 28, 2025

Choose a reason for hiding this comment

Uh oh!

pruivo Apr 28, 2025

Choose a reason for hiding this comment

Uh oh!

ryanemerson Apr 28, 2025

Choose a reason for hiding this comment

Uh oh!

ryanemerson Apr 28, 2025

Choose a reason for hiding this comment

Uh oh!

pruivo Apr 28, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

pruivo commented Apr 22, 2025 •

edited

Loading