Virtuoso Elastic Cluster Configuration

What

Virtuoso 7 and later versions can be configured in Elastic scale out cluster mode. The data is sharded in a large number of self-contained partitions. These partitions are divided among a number of database server processes and can migrate between them.

Why

An Elastic cluster can be dynamically resized with new partitions or shards dynamically added to the same or new hardware resources as the need to increase the size of the scale cluster or relocation of partitions is required.

How

This documentation details the steps for the installation and configuration of a Virtuoso Elastic Cluster on Unix.

Perform Virtuoso 7 Unix installation:


$ tar xvf virtuoso-universal-server-7.x.tar
x install.sh
x universal-server.taz
$ sh install.sh

- Extracting Virtuoso Universal Server v7.x

- Creating default environment settings

- Creating default database settings
Configuring: database
Creating directory $VIRTUOSO_HOME/database/backup
Installing new virtuoso.ini in $VIRTUOSO_HOME/database
Installing new php.ini in $VIRTUOSO_HOME/database
Creating symlink to $VIRTUOSO_HOME/bin/virtuoso-iodbc-t

- Registering ODBC drivers

- Registering .NET provider for Mono

- Installing VAD packages in database (this can take some time)

- Checking where license should be stored

- Starting OpenLink License Manager

- Checking for initial Virtuoso license

- Starting Virtuoso server instance

- Finalizing installation


This concludes the first part of the installation.


- Attempting to start browser

Please start a browser manually and open the following URL to finalize
the installation process:

     http://cname:8890/install/


Installation completed
$

Determine Elastic Cluster size

  1. Determine how many nodes you want to start the elastic cluster with: 2, 3, 4, 5, 6, 7, 8 etc. Note: the number of the nodes must be >= 2
  2. Setup Virtuoso environment and stop default database: 

    $ . ./virtuoso-environment.sh
$ virtuoso-stop.sh
Shutting down Virtuoso instance in [database]
$

  3. Optionally remove the default database such that is does not get started with the cluster: 

    rm  -rf database

  4. Full list of supported options for the script "virtuoso-mkcluster.sh": 

      -cluster-size       Number of nodes in cluster
  -cluster-node       Node number in a cluster
  -cluster-port       Base portnumer for cluster
  -cluster-ipc-port   Base portnumer for cluster IPC
  -cluster_size       Size of the cluster
  -virtuoso_home      Virtuoso home path

  5. Run the virtuoso-mkcluster.sh script to create the cluster, note the default number of nodes is 4, but this can be changed by setting the environment variable CLUSTER_SIZE to the required number of nodes to be created before running the script: 

    $ virtuoso-mkcluster.sh
Configuring cluster node: 1/4

Creating directory $VIRTUOSO_HOME/cluster_01
Creating directory $VIRTUOSO_HOME/cluster_01/backup
Creating directory $VIRTUOSO_HOME/cluster_01/logs
Installing new virtuoso.ini in $VIRTUOSO_HOME/cluster_01
Installing new php.ini in $VIRTUOSO_HOME/cluster_01
Creating symlink to $VIRTUOSO_HOME/bin/virtuoso-iodbc-t
Configuring cluster node: 2/4

Creating directory $VIRTUOSO_HOME/cluster_02
Creating directory $VIRTUOSO_HOME/cluster_02/backup
Installing new virtuoso.ini in $VIRTUOSO_HOME/cluster_02
Removing unneeded sections from virtuoso.ini
Creating symlink to $VIRTUOSO_HOME/bin/virtuoso-iodbc-t
Configuring cluster node: 3/4

Creating directory $VIRTUOSO_HOME/cluster_03
Creating directory $VIRTUOSO_HOME/cluster_03/backup
Installing new virtuoso.ini in $VIRTUOSO_HOME/cluster_03
Removing unneeded sections from virtuoso.ini
Creating symlink to $VIRTUOSO_HOME/bin/virtuoso-iodbc-t
Configuring cluster node: 4/4

Creating directory $VIRTUOSO_HOME/cluster_04
Creating directory $VIRTUOSO_HOME/cluster_04/backup
Installing new virtuoso.ini in $VIRTUOSO_HOME/cluster_04
Removing unneeded sections from virtuoso.ini
Creating symlink to $VIRTUOSO_HOME/bin/virtuoso-iodbc-t
$

  6. In each cluster_XX directory the following the cluster.ini files are created for cluster internal communication 

    $ more cluster_01/cluster.ini
[Cluster]
Threads             = 20
Master              = Host1
ThisHost            = Host1
ReqBatchSize        = 10000
BatchesPerRPC       = 4
BatchBufferBytes    = 20000
LocalOnly           = 2
MaxKeepAlivesMissed = 1000
Host1               = localhost:22201
Host2               = localhost:22202
Host3               = localhost:22203
Host4               = localhost:22204
MaxHosts            = 5
$ more cluster_02/cluster.ini
[Cluster]
Threads             = 20
Master              = Host1
ThisHost            = Host2
ReqBatchSize        = 10000
BatchesPerRPC       = 4
BatchBufferBytes    = 20000
LocalOnly           = 2
MaxKeepAlivesMissed = 1000
Host1               = localhost:22201
Host2               = localhost:22202
Host3               = localhost:22203
Host4               = localhost:22204
MaxHosts            = 5
$ more cluster_03/cluster.ini
[Cluster]
Threads             = 20
Master              = Host1
ThisHost            = Host3
ReqBatchSize        = 10000
BatchesPerRPC       = 4
BatchBufferBytes    = 20000
LocalOnly           = 2
MaxKeepAlivesMissed = 1000
Host1               = localhost:22201
Host2               = localhost:22202
Host3               = localhost:22203
Host4               = localhost:22204
MaxHosts            = 5
$ more cluster_04/cluster.ini
[Cluster]
Threads             = 20
Master              = Host1
ThisHost            = Host4
ReqBatchSize        = 10000
BatchesPerRPC       = 4
BatchBufferBytes    = 20000
LocalOnly           = 2
MaxKeepAlivesMissed = 1000
Host1               = localhost:22201
Host2               = localhost:22202
Host3               = localhost:22203
Host4               = localhost:22204
MaxHosts            = 5
$

Enable Elastic Cluster mode

  1. The cluster.ini files need to be reconfigured as detailed below for elastic cluster and file slicing/sharding to be enabled.
  2. A common file called cluster.global.ini can be created and placed in the home directory of the Virtuoso installation: 

    [Cluster]
Threads             = 300
ReqBatchSize        = 10000
BatchesPerRPC       = 4
BatchBufferBytes    = 20000
LocalOnly           = 2
MaxKeepAlivesMissed = 100
RDFLoadBytes        = 52428800
Host1               = localhost:22201
Host2               = localhost:22202
Host3               = localhost:22203
Host4               = localhost:22204

  3. This file can then be sym-linked to each of the cluster_XX directories of the cluster, and its content will be merged with the cluster.ini in the respective directory of each node when starting the cluster. 

    ln -s cluster.global.ini cluster_01/cluster.global.ini
ln -s cluster.global.ini cluster_02/cluster.global.ini
ln -s cluster.global.ini cluster_03/cluster.global.ini
ln -s cluster.global.ini cluster_04/cluster.global.ini

  4. Edit each of the the cluster.ini files in the cluster_XX directories as follows to enable, elastic cluster mode: 

    $ more cluster_01/cluster.ini
[Cluster]
Master              = Host1
ThisHost            = Host1

[ELASTIC]
Segment1 = 4G, database.db = q1
Slices = 6
MaxSlices = 2048

$ more cluster_02/cluster.ini

[Cluster]
Master              = Host1
ThisHost            = Host2

[ELASTIC]
Segment1 = 4G, database.db = q1
Slices = 6
MaxSlices = 2048

$ more cluster_03/cluster.ini

[Cluster]
Master              = Host1
ThisHost            = Host3

[ELASTIC]
Segment1 = 4G, database.db = q1
Slices = 6
MaxSlices = 2048

$ more cluster_04/cluster.ini

[Cluster]
Master              = Host1
ThisHost            = Host4

[ELASTIC]
Segment1 = 4G, database.db = q1
Slices = 6
MaxSlices = 2048

  5. The [ELASTIC] section in the cluster.ini files above, enables the elastic cluster mode, where multiple segments and stripes as detailed in the standard Virtuoso documentation database striping.
  6. The Slices parameter above should be set to the number of hardware threads on the CPUs running on. Thus in the example above where it is set to 6, this assumes all nodes are running on the same physical machine with 12 cores with hyper-threading enabled i.e. 24 threads, thus 6 per cluster node.
  7. The MaxSlices parameter above sets the the maximum number of physical slices in the cluster.

Start Elastic Cluster

  1. Start the Elastic cluster using the standard virtuoso-start.sh script run from the home directory of the Virtuoso installation,which automatically detects the number for nodes to be started: 

    $ virtuoso-start.sh
Starting Virtuoso instance in [cluster_01]
Starting Virtuoso instance in [cluster_02]
Starting Virtuoso instance in [cluster_03]
Starting Virtuoso instance in [cluster_04]
$

  2. The default SQL port of the master node is 12201, as indicated in the virtuoso.ini file of the cluster_01 directory, and can then be used for connecting to the newly created cluster and check its status to ensure all nodes are online: 

    $ isql 12201
Connected to OpenLink Virtuoso
Driver: 07.10.3211 OpenLink Virtuoso ODBC Driver
OpenLink Interactive SQL (Virtuoso), version 0.9849b.
Type HELP; for help and EXIT; to exit.
SQL> status('cluster_d');
REPORT
VARCHAR
_______________________________________________________________________________

Cluster: No samples, Please refresh

1 Rows. -- 22 msec.
SQL> status('cluster_d');
REPORT
VARCHAR
_______________________________________________________________________________

Cluster 4 nodes, 2 s. 5 m/s 0 KB/s  5% cpu 0%  read 0% clw threads 1r 0w 0i buffers 349144 2 d 0 w 0 pfs
cl 1: 2 m/s 0 KB/s  0% cpu 0%  read 0% clw threads 1r 0w 0i buffers 45391 2 d 0 w 0 pfs
cl 2: 0 m/s 0 KB/s  0% cpu 0%  read 0% clw threads 0r 0w 0i buffers 43367 0 d 0 w 0 pfs
cl 3: 0 m/s 0 KB/s  0% cpu 0%  read 0% clw threads 0r 0w 0i buffers 50129 0 d 0 w 0 pfs
cl 4: 0 m/s 0 KB/s  0% cpu 0%  read 0% clw threads 0r 0w 0i buffers 39383 0 d 0 w 0 pfs

5 Rows. -- 22 msec.
SQL> 
SQL> cl_ping(1,500, 1000);

Done. -- 7 msec.
SQL> cl_ping(2,500, 1000);

Done. -- 52 msec.
SQL> cl_ping(3,500, 1000);

Done. -- 60 msec.
SQL> cl_ping(4,500, 1000);

Done. -- 51 msec.
SQL>

  3. The database.log of the master node can be checked to verify the Elastic cluster mode has been enable on server startup, which will contain the entry PL LOG: Elastic cluster setup: 

    $ more cluster_01/database.log

                Thu Oct 09 2014
05:11:49 { Loading plugin 1: Type `plain', file `wikiv' in `../hosting'
05:11:49   WikiV version 0.6 from OpenLink Software
05:11:49   Support functions for WikiV collaboration tool
05:11:49   SUCCESS plugin 1: loaded from ../hosting/wikiv.so }
05:11:49 { Loading plugin 2: Type `plain', file `mediawiki' in `../hosting'
05:11:49   MediaWiki version 0.1 from OpenLink Software
05:11:49   Support functions for MediaWiki collaboration tool
05:11:49   SUCCESS plugin 2: loaded from ../hosting/mediawiki.so }
05:11:49 { Loading plugin 3: Type `plain', file `creolewiki' in `../hosting'
05:11:49   CreoleWiki version 0.1 from OpenLink Software
05:11:49   Support functions for CreoleWiki collaboration tool
05:11:49   SUCCESS plugin 3: loaded from ../hosting/creolewiki.so }
05:11:49 { Loading plugin 4: Type `plain', file `im' in `../hosting'
05:11:49   IM version 0.61 from OpenLink Software
05:11:49   Support functions for Image Magick 6.8.1
05:11:49   SUCCESS plugin 4: loaded from ../hosting/im.so }
05:11:49 { Loading plugin 5: Type `plain', file `wbxml2' in `../hosting'
05:11:49   WBXML2 version 0.9 from OpenLink Software
05:11:49   Support functions for WBXML2 0.9.2 Library
05:11:49   SUCCESS plugin 5: loaded from ../hosting/wbxml2.so }
05:11:49 { Loading plugin 6: Type `attach', file `libphp5.so' in `../hosting'
05:11:49   SUCCESS plugin 6: loaded from ../hosting/libphp5.so }
05:11:49 { Loading plugin 7: Type `Hosting', file `hosting_php.so' in `../hosting'
05:11:49   Hosting version 3208 from OpenLink Software
05:11:49   PHP engine version 5.3.21
05:11:49   SUCCESS plugin 7: loaded from ../hosting/hosting_php.so }
05:11:49 { Loading plugin 8: Type `plain', file `qrcode' in `../hosting'
05:11:49   QRcode version 0.1 from OpenLink Software
05:11:49   Support functions for ISO/IEC 18004:2006, using QR Code encoder (C) 2006 Kentaro Fukuchi <fukichi@megaui.net>
05:11:49   SUCCESS plugin 8: loaded from ../hosting/qrcode.so }
05:11:49 OpenLink Virtuoso Universal Server
05:11:49 Version 07.10.3211-pthreads for Linux as of Oct  6 2014
05:11:49 uses parts of OpenSSL, PCRE, Html Tidy
05:11:51 SQL Optimizer enabled (max 1000 layouts)
05:11:52 Compiler unit is timed at 0.000403 msec
05:12:03 Checkpoint started
05:12:04 Roll forward started
05:12:04 Roll forward complete
05:12:07 PL LOG: Elastic cluster setup
05:12:08 Checkpoint started
05:12:09 Checkpoint finished, log reused
05:12:11 Checkpoint started
05:12:12 Checkpoint finished, log reused
05:12:12 PL LOG: new clustered database:Init of RDF
05:12:23 Checkpoint started
05:12:25 Checkpoint finished, log reused
05:12:50 PL LOG: Installing Virtuoso Conductor version 1.00.8727 (DAV)
05:12:51 Checkpoint started
05:12:53 Checkpoint finished, log reused
05:13:23 Checkpoint started
05:13:25 Checkpoint finished, log reused
05:13:26 Checkpoint started
05:13:26 Checkpoint finished, log reused
05:13:28 HTTP/WebDAV server online at 8890
05:13:28 Server online at 12201 (pid 15211)
05:13:29 ZeroConfig registration CLUSTER (MASALA)

  4. The cluster node directories can also be checked, where the database slice/shard files i.e. database.db.X, database.db.Y, database.db.Z can be seen: 

    $ ls -ltr cluster_01
total 2322804
drwxr-xr-x. 2 virtuoso virtuoso      4096 Oct  9 04:26 backup
drwxr-xr-x. 2 virtuoso virtuoso      4096 Oct  9 04:26 logs
-rwxr-xr-x. 1 virtuoso virtuoso     70607 Oct  9 04:26 php.ini
lrwxrwxrwx. 1 virtuoso virtuoso        24 Oct  9 04:26 virtuoso -> ..//bin/virtuoso-iodbc-t
-rw-r--r--. 1 virtuoso virtuoso         0 Oct  9 04:26 database.pxa
-rwxr-xr-x. 1 virtuoso virtuoso      6594 Oct  9 04:33 virtuoso.ini
-rw-r--r--. 1 virtuoso virtuoso       137 Oct  9 05:03 cluster.ini
lrwxrwxrwx. 1 virtuoso virtuoso        21 Oct  9 05:04 cluster.global.ini -> ../cluster.global.ini
drwxr-xr-x. 2 virtuoso virtuoso    131072 Oct  9 05:37 dump
-rw-r--r--. 1 virtuoso virtuoso     32915 Oct  9 06:55 database.2pc
-rw-r--r--. 1 virtuoso virtuoso        25 Oct 10 02:24 database.map
-rw-r--r--. 1 virtuoso virtuoso  56623104 Oct 10 03:23 database-temp.db
-rw-r--r--. 1 virtuoso virtuoso 824180736 Oct 10 05:10 database.db.16
-rw-r--r--. 1 virtuoso virtuoso 723517440 Oct 10 05:10 database.db.8
-rw-r--r--. 1 virtuoso virtuoso 740294656 Oct 10 05:10 database.db.0
-rw-r--r--. 1 virtuoso virtuoso  33554432 Oct 10 05:10 database.db
-rw-r--r--. 1 virtuoso virtuoso         0 Oct 10 05:10 database.trx
-rw-r--r--. 1 virtuoso virtuoso     34234 Oct 10 05:10 database.log
$

Splitting Cluster nodes across different machines

  1. To split the node across across physical machines for better scale out performance, scalability and growth, simply perform a parallel Virtuoso installation on the additional physical machines and move the cluster nodes required to the designated machine. Example, for the default 4 node cluster to be split across two identical machines it would make sense to split 2 nodes across each machine, thus you would move say the cluster_03 and cluster_04 directory nodes to the new machine (removing them from the original). The cluster.global.ini file on each node would then need to be updated to set the HostXX parameters to point to the new locations for nodes 03 and 04: 

    
$ more cluster.global.ini

[Cluster]
Threads             = 300
ReqBatchSize        = 10000
BatchesPerRPC       = 4
BatchBufferBytes    = 20000
LocalOnly           = 2
MaxKeepAlivesMissed = 100
RDFLoadBytes        = 52428800
Host1               = hostname1:22201
Host2               = hostname1:22202
Host3               = hostname2:22203
Host4               = hostname2:22204

Machine 1 (hostname1)

$ more cluster_01/cluster.ini
[Cluster]
Master              = Host1
ThisHost            = Host1

[ELASTIC]
Segment1 = 4G, database.db = q1
Slices = 12
MaxSlices = 2048

$ more cluster_02/cluster.ini

[Cluster]
Master              = Host1
ThisHost            = Host2

[ELASTIC]
Segment1 = 4G, database.db = q1
Slices = 12
MaxSlices = 2048

Machine 2 (hostname2)

$ more cluster_03/cluster.ini

[Cluster]
Master              = Host1
ThisHost            = Host3

[ELASTIC]
Segment1 = 4G, database.db = q1
Slices = 12
MaxSlices = 2048

$ more cluster_04/cluster.ini

[Cluster]
Master              = Host1
ThisHost            = Host4

[ELASTIC]
Segment1 = 4G, database.db = q1
Slices = 12
MaxSlices = 2048

  2. Note assuming the same number of CPU threads on each machine i.e. 24, then the Slices param can be doubled to 12 for each node in cluster.ini, as above.
  3. The cluster nodes can then be started on each machine, with 2 nodes being started on each in this case to form the cluster: 

    Machine 1 (hostname1)

$ virtuoso-start.sh
Starting Virtuoso instance in [cluster_01]
Starting Virtuoso instance in [cluster_02]
$

Machine 2 (hostname2)

$ virtuoso-start.sh
Starting Virtuoso instance in [cluster_03]
Starting Virtuoso instance in [cluster_04]
$

  4. To stop the cluster use the standard virtuoso-stop.sh script on each machine, which automatically detects the number for nodes to be stopped: 

    Machine 1 (hostname1)

$ virtuoso-stop.sh
Stopping Virtuoso instance in [cluster_01]
Stopping Virtuoso instance in [cluster_02]

Machine 2 (hostname2)

$ virtuoso-stop.sh
Stopping Virtuoso instance in [cluster_03]
Stopping Virtuoso instance in [cluster_04]
$

  5. The cl_exec('shutdown') command can also be run from isql on any node of the cluster to shutdown all nodes at once: 

    $ isql 12201
Connected to OpenLink Virtuoso
Driver: 07.10.3211 OpenLink Virtuoso ODBC Driver
OpenLink Interactive SQL (Virtuoso), version 0.9849b.
Type HELP; for help and EXIT; to exit.
SQL> cl_exec ('checkpoint');

Done. -- 2487 msec.
SQL> cl_exec ('shutdown');

*** Error 08S01: VD CL065: Lost connection to server
at line 2 of Top-Level:
cl_exec ('shutdown')
$

Related