Copyright (C) 1998-2018 OpenLink Software [email protected]
Always make sure the database has been properly shutdown and the transaction log (virtuoso.trx) is empty before performing any of the following updates/upgrades.
Before upgrading any database, it is always a wise precaution to make a proper backup.
The database format has not changed between various versions of Virtuoso 5.0.X, so from a database standpoint no particular steps need to be performed before upgrading to the latest version of Virtuoso 5.
The only requirement is that you have properly shutdown the database prior to installing the latest binaries, as the transaction logs can have a different version tag. In this case the virtuoso server will print the following message and refuses to start the database:
The transaction log file has been produced by server version
'05.00.XXXX'. The version of this server is '05.00.YYYY'. If the
transaction log is empty or you do not want to replay it then delete
it and start the server again. Otherwise replay the log using the
server of version '05.00.XXXX' and make checkpoint and shutdown
to ensure that the log is empty, then delete it and start using
new version.
The database format has substantially changed between these two versions of Virtuoso. To upgrade your database, you must dump all your data from your VOS 5.0 database and re-load it into VOS 6.0.
Regular VOS 5.0 RDBMS tables can be dumped with the dbdump tool into scripts that can be replayed using the isql tool into the VOS 6.0 database.
For VOS 5.0 RDF_QUAD table, we have a set of dump/load stored procedures to dump graphs into a set of backup files, which can then be reloaded into the VOS 6.0 database. Contact us at [email protected] for more info.
If you attempt to start a VOS 5.0 database with a VOS 6.0 server, the server will print the following message ans refuses to start the database:
The database you are opening was last closed with a server of
version 3016. The present server is of version 3126. This server
does not read this pre 6.0 format.
The database disk format has not changed, but the introduction of a newer RDF index requires you run a script to upgrade the RDF_QUAD table. Since this can be a lengthy task and take extra disk space (upto 2x space of original RDF_QUAD table during conversion) this is not done automatically on startup.
After upgrading the binary you cannot perform any SPARQL queries until the new RDF_QUAD table is installed. The steps for this are:
-
Shutdown the database and verify .trx file is empty
-
Check to make sure you have enough disk space to perform operation
-
Check to make sure you have a proper backup of the database
-
Edit virtuoso.ini and disable VADInstallDir and possibly the HTTPServer section for duration of the upgrade.
-
Start up the database
-
Connect with isql to your database and run the libsrc/Wi/clrdf23.sql script. Depending on the number of quad records this can take several hours. Once the conversion is complete, the database will shutdown itself.
-
Edit virtuoso.ini and enable VADInstallDir and possibly HTTPServer section
-
Start up database
The database format has not changed between various versions of Virtuoso 6.1.X, so from a database standpoint no particular steps need to be performed before upgrading to the latest version of Virtuoso 6.1.x.
The only requirement is that you have properly shutdown the database prior to installing the latest binaries, as the transaction logs can have a different version tag. In this case the virtuoso server will print the following message and refuses to start the database:
The transaction log file has been produced by server version
'06.01.XXXX'. The version of this server is '06.01.YYYY'. If the
transaction log is empty or you do not want to replay it then delete
it and start the server again. Otherwise replay the log using the
server of version '06.01.XXXX' and make checkpoint and shutdown
to ensure that the log is empty, then delete it and start using
new version.
In Virtuoso versions before 6.1.4 some XML data was stored in the QUAD store in such a way it could break the sequence in an index causing wrong results to be passed back.
To fix this we've added an automated check into 6.1.4 to detect if this condition occurs 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.
This check will take some time depending on the number of stored QUADS, but if the check succeeds the following message is entered in the log and virtuoso will flag this check is 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.
However if the condition is detected, the following message will appear in the log and the virtuoso server will refuse to start up:
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 [email protected] 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, OpenLink has decided not to automatically start the update process but hand control back to the DBA and let him decide when to perform this update. In case the DBA wants to delay the update until a more appropriate time, he should restart with the previous binary as this latest binary will not allow him to startup.
Once the DBA has checked the backups, 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 [email protected].
After the update process has completed, the database is left in an online state.