Page Comparison

Requirements: memcached v1.4.14 or later

The memcached-based storage facility in IdPv3 is based on the spymemcached library, which has a number of compelling features for HA deployments:

• Optimized IO for high throughput

• Memcached failover facility

• Stable hashing algorithm supports memcached pool resizing

The failover facility merits further discussion. Failover is enabled by specifying multiple memcached hosts and failureMode="Redistribute". When the client encounters an unreachable host in redistribute mode, it will temporarily remove the unavailable host from the pool of available hosts. Any keys that hash to the unavailable host will be written or retrieved from a backup host. The high-level effect of this behavior on the IdP session management service is that a node failure will cause loss of IdP session, which would impact users as an unexpected authentication. The IdP session management service, however, would be fully functional during a host failure/recovery event. Also note that this behavior requires no state sharing (i.e. repcache) between memcached nodes.

Bear in mind that different storage use cases have different failover properties. While the replay cache would be similarly unimpacted, the artifact map failing to retrieve a previously stored artifact mapping would result in a failed login to the service to which the artifact was sent.

The following architecture is strongly recommended suggested for HA deployments that wish to use memcache:

Thus every IdP node runs a memcached service and the Java process running the IdP software connects to every memcached service. The following configuration example assumes the recommended architecture above and should be placed in conf/global.xml .

MemcachedStorageService Configuration

     <bean id="shibboleth.MemcachedStorageService"
          class="org.opensaml.storage.impl.memcached.MemcachedStorageService"
          c:timeout="2">
        <constructor-arg name="client">
            <bean class="net.spy.memcached.spring.MemcachedClientFactoryBean"
                  p:servers="idp-1.example.com:11211,idp-2.example.com:11211,idp-3.example.com:11211"
                  p:protocol="BINARY"
                  p:locatorType="CONSISTENT"
                  p:failureMode="Redistribute">
                <property name="hashAlg">
                    <util:constant static-field="net.spy.memcached.DefaultHashAlgorithm.FNV1_64_HASH" />
                </property>
                <property name="transcoder">
                    <!-- DO NOT MODIFY THIS PROPERTY -->
                    <bean class="org.opensaml.storage.impl.memcached.StorageRecordTranscoder" />
                </property>
            </bean>
        </constructor-arg>
    </bean>

Once a MemcachedStorageService bean has been defined as above, it can be used with subsystems that require a StorageService component. The following configuration snippet from conf/idp.properties indicates how to use memcached for session storage.

Memcached for IdP Sessions

idp.session.StorageService = shibboleth.MemcachedStorageService