Virtuoso "AnalyzeFixQuadStore" parameter for fixing RDF Quad Store Index corruption

In Virtuoso versions before 6.3, some RDF data in the quad store could be stored in a way that could break the sequence of an index, causing wrong results to be passed back.

To fix this, we have added an automated check into 6.3+ builds to detect whether this condition is present, and to fix the index when needed.

When a DBA starts the database with this newer Virtuoso binary, the following text will appear in the virtuoso session log:

  21:05:36 PL LOG: This database may contain RDF data that could cause indexing problems on previous versions of the server.
  21:05:36 PL LOG: The content of the DB.DBA.RDF_QUAD table will be checked and an update may automatically be performed if
  21:05:36 PL LOG: such data is found.
  21:05:36 PL LOG: This check will take some time but is made only once.

As noted, this check may take some time, depending on the number of stored quads.

If the check succeeds, the following message is entered in the log, and Virtuoso will flag the check as "done," so it will not affect subsequent restarts of the database server.

21:05:36 PL LOG: No need to update DB.DBA.RDF_QUAD

The database will continue to perform the startup routines and go into an online state.

If the condition is detected, however, the following message will appear in the log, and the Virtuoso server will refuse to start --

  21:05:36 PL LOG: An update is required.
  21:05:36 PL LOG:
  21:05:36 PL LOG: NOTICE: Before Virtuoso can continue fixing the DB.DBA.RDF_QUAD table and its indexes
  21:05:36 PL LOG:         the DB Administrator should check make sure that:
  21:05:36 PL LOG:
  21:05:36 PL LOG:          * there is a recent backup of the database
  21:05:36 PL LOG:          * there is enough free disk space available to complete this conversion
  21:05:36 PL LOG:          * the database can be offline for the duration of this conversion
  21:05:36 PL LOG:
  21:05:36 PL LOG:         Since the update can take a considerable amount of time on large databases
  21:05:36 PL LOG:         it is advisable to schedule this at an appropriate time.
  21:05:36 PL LOG:
  21:05:36 PL LOG: To continue the DBA must change the virtuoso.ini file and add the following flag:
  21:05:36 PL LOG:
  21:05:36 PL LOG:     [Parameters]
  21:05:36 PL LOG:     AnalyzeFixQuadStore = 1
  21:05:36 PL LOG:
  21:05:36 PL LOG: For additional information please contact OpenLink Support <support@openlinksw.com>
  21:05:36 PL LOG: This process will now exit.

Since the update will take a substantial amount of time and additional disk space depending on the size of the quad store, we have decided not to automatically start the update process. Instead, we hand control back to the DBA and let him decide when to perform this update. If the DBA wants to delay the update until a more appropriate time, he should restart with the previous binary as this latest binary will never start the database.

Once the DBA has checked the backups and disk space, and found an appropriate time-slot to run this update, he should edit the virtuoso.ini file and add the following flag:

[Parameters]
AnalyzeFixQuadStore = 1

Upon starting the Virtuoso server, the following messages will appear in the virtuoso.log file:

  21:05:57 PL LOG: This database may contain RDF data that could cause indexing problems on previous versions of the server.
  21:05:57 PL LOG: The content of the DB.DBA.RDF_QUAD table will be checked and an update may automatically be performed if
  21:05:57 PL LOG: such data is found.
  21:05:57 PL LOG: This check will take some time but is made only once.
  21:05:57 PL LOG:
  21:05:57 PL LOG: An update is required.
  21:05:57 PL LOG: Please be patient.
  21:05:57 PL LOG: The table DB.DBA.RDF_QUAD and two of its additional indexes will now be patched.
  21:05:57 PL LOG: In case of an error during the operation, delete the transaction log before restarting the server.
  21:05:57 Checkpoint started
  21:05:57 Checkpoint finished, log off
  21:05:57 PL LOG: Phase 1 of 9: Gathering statistics ...
  21:05:58 PL LOG:  * Index sizes before the processing: 002565531 RDF_QUAD, 002565531 POGS, 001171100 OP
  21:05:58 PL LOG: Phase 2 of 9: Copying all quads to a temporary table ...
  21:07:26 PL LOG: * Index sizes of temporary table: 001171100 OP
  21:07:26 PL LOG: Phase 3 of 9: Cleaning the quad storage ...
  21:07:51 PL LOG: Phase 4 of 9: Refilling the quad storage from the temporary table...
  21:09:17 PL LOG: Phase 5 of 9: Cleaning the temporary table ...
  21:09:41 PL LOG: Phase 6 of 9: Gathering statistics again ...
  21:09:41 PL LOG: * Index sizes after the processing: 002565531 RDF_QUAD, 002565531 POGS, 001171100 OP
  21:09:41 PL LOG: Phase 7 of 9: integrity check (completeness of index RDF_QUAD_POGS of DB.DBA.RDF_QUAD) ...
  21:10:00 PL LOG: Phase 8 of 9: integrity check (completeness of primary key of DB.DBA.RDF_QUAD) ...
  21:10:17 PL LOG: Phase 9 of 9: final checkpoint...
  21:10:20 Checkpoint started
  21:10:22 Checkpoint finished, log off
  21:10:22 PL LOG: Update complete.

If the update process detects any problem, it will put some debug output into the virtuoso.log and exit. At this point the DBA is advised to remove the virtuoso.trx file and contact OpenLink Support.

After the update process has completed, the database is left in an online state.