Authentication cache
To improve server performance during single sign-on (SSO) authentication, Open Liberty provides an authentication cache to store a subject after a user authentication is successful. You can also use JCache to configure a distributed authentication cache that is shared among a group of Open Liberty servers that use the same set of users and groups.
Configure an in-memory authentication cache
The authentication cache is available by default when any version of the Application Security (Jakarta Security) feature is enabled in your server.xml
file. The cache is initialized with a set number of entries, which can be configured with the initialSize
attribute. It has a maximum number of entries, which can be configured with the maxSize
attribute. If the maximum size is reached, the earliest entries are removed from the cache. If a user is inactive for longer than the period that is specified by the timeout
attribute, the entry for that user is removed from the cache.
By default, the cache size is initialized to 50 entries and a maximum of 25000 entries, with a timeout of 600 seconds. To change the values for the authentication cache, configure the authCache element, as shown in the following example.
<authCache initialSize="100" maxSize="50000" timeout="15m"/>
Consider the following points if you adjust the authentication cache settings.
Larger authentication cache timeout values can increase security risks. For example, a user who is revoked can still log in by using a credential that is cached in the authentication cache until the cache is refreshed.
Smaller authentication cache timeout values can affect performance. When this value is smaller, the Liberty server accesses the user registry or repository more frequently.
Large numbers of entries in the authentication cache, which are caused by an increased number of users, increase the memory usage of the authentication cache, which can affect performance.
You can disable the authentication cache by setting the cachEnabled
attribute to false
in the authentication
element. The following example demonstrates how to disable the authentication cache for an authentication configuration that is called Basic
.
<authentication id="Basic" cacheEnabled="false" />
Any change to the user registry configuration in the server.xml
file clears the authentication cache. However, if you make changes to an external user registry, such as an LDAP user registry, the authentication cache is unaffected.
Configure a distributed authentication cache
When a group of servers uses the same set of users and groups, you can configure a distributed authentication cache to be shared across the servers.
When the distributed cache is shared among Open Liberty servers that use the same user registries, the work of populating the cache is not replicated on each server. In cloud environments, where Open Liberty servers might need to be scaled up or down, each server can start with an already populated cache, which can improve overall server performance.
Supported SSO mechanisms
The Open Liberty distributed authentication cache supports caching the subject credentials for the following SSO mechanisms and configurations.
LTPA
JWT, when the JSON Web Token Single Sign-On feature is enabled.
SPNEGO, when the Simple and Protected GSSAPI Negotiation Mechanism feature is enabled and the spnego:
includeClientGSSCredentialInSubject
attribute is set tofalse
.SAML, when the SAML Web Single Sign-On feature is enabled. The samlWebSso20:
spCookieName
attribute must be explicitly set and must match on all servers that share the authentication cache.OAuth, when the Social Media Login feature is enabled and the oauth2Login element is configured.
OIDC, when the Social Media Login feature is enabled and the oidcLogin element is configured.
OIDC, when the OpenID Connect Client feature is enabled. The webAppSecurity:
ssoCookieName
must be explicitly set and must match on all servers that share the authentication cache.
Subjects that are authenticated by unsupported mechanisms might be stored in the local cache instead.
Configure JCache for the distributed authentication cache
Open Liberty supports distributed authentication caching by using a JCache cache. To configure it, you must first configure your JCache caching provider, cache manager, and cache. For more information about using JCache with Open Liberty, see Distributed caching with JCache.
You must enable any version of the Application Security (Jakarta Security) feature in your server.xml
file. Configure the authCache: cacheRef
attribute to refer to the ID of the cache
instance that you create, as shown in the following example.
<featureManager>
<feature>appSecurity-4.0</feature>
</featureManager>
<library id="JCacheProviderLib">
<file name="${shared.resource.dir}/libs/jcacheprovider.jar" />
</library>
<cacheManager id="CacheManager" uri="<PROVIDER_SPECIFIC_URI>">
<properties property1="value1" property2="value2" />
<cachingProvider providerClass="org.acme.CachingProvider"
jCacheLibraryRef="JCacheProviderLib" />
</cacheManager>
<cache id="io.openliberty.cache.authentication" name="io.openliberty.cache.authentication" cacheManagerRef="CacheManager" />
<authCache cacheRef="io.openliberty.cache.authentication" maxSize="25000" timeout="10m" />
In this example, the library element specifies the location of the JCache provider JAR file. The cacheManager element references the library to configure the CachingProvider
and CacheManager
instances. The cache element references the cacheManager
element to configure the JCache instance. The authCache element references that JCache instance to specify it as the authentication cache for the server. This element also specifies maximum size and timeout values for the authentication cache that replicate the behavior of the local in-memory cache.
Caching custom classes in your subject
When you use the distributed authentication cache, you might need to store subjects that contain custom classes. Such classes include principals or credentials that were inserted into the subject in a custom JAAS LoginModule
instance or in a Trust Association Interceptor (TAI). In such cases, all classes that are inserted into the subject must be serializable. If any of the object classes that are stored in the subject are not serializable, the subject is not stored into the distributed authentication cache. In this case, the subject is stored in the local in-memory authentication cache instead. These subjects cannot be stored in the JCache cache and are not accessible by other Open Liberty servers. A warning is output to the server logs the first time that a class is encountered that prevents the subject from being serialized and stored to the distributed authentication cache.
Additionally, the same library reference must be used for the caching provider, JAAS LoginModule, the TAI, and any applications that access the class. Otherwise, you might encounter ClassCastException
exceptions when the custom classes are accessed in an application.
In the following example configuration, the JCacheProviderLib
library contains the JCache provider. The CustomLib
library contains some custom principal or credential classes that are set on the subject. The class is loaded by the same ClassLoader
instance in all instances of the class. The same library reference (commonLibraryRef
) is used for the caching provider, the application, the JAAS LoginModule, and the TAI.
<featureManager>
<feature>appSecurity-4.0</feature>
</featureManager>
<library id="JCacheProviderLib">
<file name="${shared.resource.dir}/libs/jcacheprovider.jar" />
</library>
<library id="CustomLib">
<file name="${shared.resource.dir}/libs/customlib.jar" />
</library>
<cachingProvider id="CachingProvider"
providerClass="org.acme.CachingProvider"
jCacheLibraryRef="JCacheProviderLib”
commonLibraryRef="CustomLib" />
<cacheManager id="CacheManager" uri="<PROVIDER_SPECIFIC_URI>" />
<cache id="io.openliberty.cache.authentication" name="io.openliberty.cache.authentication" cacheManagerRef="CacheManager" />
<application>
<classloader commonLibraryRef="CustomLib" />
</application>
<jaasLoginModule libraryRef="CustomLib" />
<trustAssociation>
<interceptors libraryRef="CustomLib" />
</trustAssociation>
This example specifies the appSecurity-4.0
feature but you can use any version of the Jakarta Security (Application Security) feature.
Cache sizing, eviction, and performance
To replicate the default behavior of the local in-memory authentication cache, configure a maximum size of 25,000 entries, with an access-based expiration policy of 10 minutes from the last access time. The previous example demonstrates how to configure these values in the authCache
element.
Distributed cache size limits are only approximate because partitioning of the data across servers can result in decreased realized capacity. To verify the maximum capacity, thoroughly test your configuration.
You can realize further performance gains by configuring a near cache for your JCache provider, if your provider supports it. A near cache allows the JCache provider to store and access selected cache entries on Open Liberty instead of needing to go out to the remote cache to retrieve those entries. Enabling the JCache provider near cache might increase the amount of memory that is required by the Open Liberty server as those cache entries are stored within the JVM for that server.
Securing the cache
Because the authentication cache contains security-sensitive information, the JCache cache contents must be secured. Treat the cache contents like any credentials that are used to access the server. Proper security provisions include, but are not limited to the following precautions. For more information, see the documentation for your JCache provider.
The security configuration must be compatible between all Open Liberty servers that share a distributed authentication cache. Specifically, any configuration that affects the subject must be compatible. This configuration includes, but is not limited to, user registries and any applicable single sign-on (SSO) configuration, such as LTPA, JWT, or SPNEGO.
Any subject that is present in the distributed authentication cache on one Open Liberty server can be used to access any other server that uses the same distributed authentication cache instance. If a subject’s user is not intended to be accessible from one of the other servers, including that subject in the distributed authentication cache might allow unintended access to protected resources.
Consider the following points when you secure your distributed authentication cache.
Enable security for data in motion. Enable TLS for transactions between the Liberty server and the JCache server.
Enable security for data at rest. Enable encryption for contents that are stored in the JCache server.
Enable authentication and authorization on the JCache server.
Follow any JCache provider security recommendations.
Clearing the cache
Unlike the local authentication cache, changes to security and user registry configuration do not clear the contents of the distributed authentication cache. This constraint ensures the integrity of the cache across all servers. Clear the distributed authentication cache whenever all servers are updated with configuration changes that might affect what is stored in the distributed authentication cache.
To clear the cache, use the Liberty DeleteAuthCache
MBean and call the removeAllEntries()
method on all Liberty servers to clear out the distributed and in-memory cache of all contents. Clearing the distributed cache by using any JCache provider utility does not clear the contents of the in-memory cache on any of the Liberty servers.
JCache configuration examples for the distributed authentication cache
When your Open Liberty servers use a JCache distributed authentication cache, you can configure the behavior of that JCache cache by using the relevant configuration files for your provider. In the following examples for Infinispan and Hazelcast authentication caches, the cache size and expiration values are set to replicate the behavior of the Open Liberty in-memory cache defaults. However, you must test your configuration in the target environment to determine the optimal sizing. For more information, see the Cache sizing, eviction, and performance section.
- Sample Infinispan cache configuration
The following
infinispan.xml
file example shows an Infinispan authentication cache configuration, with the cache size (max-count
) and expiration (max-idle
) values set to replicate the behavior of the Open Liberty in-memory cache defaults. The cache does not have to specify thedistributed-cache
mode. It can be whatever mode that best supports your needs. For more information, see Configuring Infinispan caches in the Infinispan documentation. This example was tested with Infinispan13.0.10.Final
.
<infinispan>
<distributed-cache name="io.openliberty.cache.authentication">
<memory max-count="25000" when-full="REMOVE" />
<expiration max-idle="600000" lifespan="-1" />
<encoding media-type="application/x-java-serialized-object" />
</distributed-cache>
</infinispan>
- Sample Hazelcast cache configuration
The following
hazelcast.xml
file example shows a Hazelcast authentication cache configuration. The cache size (eviction size
) and expiration (duration-amount
) values are set to replicate the behavior of the Open Liberty in-memory cache defaults. For more information, see JCache Setup and Configuration in the Hazelcast documentation. This example was tested with Hazelcast5.1.3
.
<hazelcast>
<cache name="io.openliberty.cache.authentication">
<key-type class-name="java.lang.Object" />
<value-type class-name="java.lang.Object" />
<eviction size="25000"
max-size-policy="ENTRY_COUNT" eviction-policy="LRU" />
<expiry-policy-factory>
<timed-expiry-policy-factory
expiry-policy-type="ACCESSED"
duration-amount="600"
time-unit="SECONDS" />
</expiry-policy-factory>
</cache>
</hazelcast>