Check and Fix Search Index Inconsistencies
Fix inconsistencies between the search index and the content repository.
In Hippo Repository nodes and properties are indexed by Lucene. The index allows efficient query execution. Whenever a change is made to the repository the index is updated. These indexes are not shared between the different nodes of a cluster, but are stored on the local filesystem of the cluster node for optimal performance.
There are circumstances under which the index can get out of sync with the contents of the repository. This can cause problems when queries and searches fail to find content that is in the repository or return hits of content that is no longer in the repository. We have been able to establish that index synchronization problems can especially happen when the repository is not shut down correctly or the corresponding Java process is killed unexpectedly.
Enable the Search Index Consistency Check
To enable the search index consistency check on repository startup specify the following parameters in the search index section of the repository configuration:
<param name="enableConsistencyCheck" value="true" /> <param name="autoRepair" value="true" />
Verify (and, if needed, add) the following entries to conf/log4j.xml. The provided log levels are the recommended ones for these two categories - adjust them as experience dictates:
<logger name="org.apache.jackrabbit.core.query.lucene.ConsistencyCheck"> <level value="info"/> </logger> <logger name="org.hippoecm.repository.jackrabbit.RepositoryImpl"> <level value="warn"/> </logger>
Efficient, reliable, and complete search index consistency checks and repairs can now be executed during startup of the repository thanks to double checks for false positives. Even in live and busy clustered setups.
Impact and Recommendations
The search index consistency check is disabled by default. Enabling the check causes application startup to take considerably longer. This is the only main disadvantage. Otherwise running the check is safe.
Normally there is no reason to enable the check. We recommend to enable the check only in cases where an inconsistency in the search index is suspected. Once the check has finished and inconsistencies has been fixed, we recommend to disable the check again.