Sample Platform X3 implementation for Transactional and Analytical Workloads
MariaDB Platform X3 responds to business design challenges by integrating:
- MariaDB Server, the leading enterprise open source database;
- MariaDB ColumnStore, a columnar database for on-demand analytical processing; and
- MariaDB MaxScale, the world's most advanced database proxy.
MariaDB Server powers website and application workloads as a modern RDBMS. MariaDB Platform X3 extends MariaDB Server's capabilities to include:
- Transactional workloads (OLTP);
- Analytical workloads (OLAP); and
- the combination of the two as Hybrid Transactional and Analytical Processing (HTAP) queries.
This guide has been written for the DBA, developer and operator to help you stand up Platform X3 for HTAP queries, unleashing the ability to perform analysis across events as they are happening. It is also a deployment that can scale from the small cluster of the examples below to accommodate more transactions, larger analytical processing and high availability.
Platform X3 Query Routing and Data Streaming
When MariaDB Platform X3 is deployed for HTAP, web and mobile services send queries to MariaDB MaxScale. In turn, MaxScale distributes these queries according to their purpose, transactional queries are sent to MariaDB Servers for OLTP workloads, and analytical queries are sent to MariaDB ColumnStore for OLAP operations.
On the back-end, changes made to the MariaDB Servers are sent through MaxScale streaming data adapters to ColumnStore, ensuring that ColumnStore remains up-to-date.
Platform X3 Scaleout
MariaDB Platform X3 can operate from individual servers, but as your application grows more complicated and your database workload increases, each component can scale out to suit your particular infrastructure needs.
For OLTP operations, our sample Platform X3 deployment we start with four MariaDB Servers, configured to run as one master and three slaves synchronized with each other in a MariaDB Replication cluster. In scaling OLTP, you can increases the number of MariaDB Servers, allowing for high availability, replication backups and failover.
For OLAP operations, our sample deployment uses five MariaDB ColumnStore nodes, two of which are configured as User Modules (UM's) and three as Performance Modules (PM's). In scaling OLAP, you can increase the number of UM's to handle more incoming queries or increase the number of PM's to better handle the processing of those queries.
To manage network communication between the application and our deployment and between the database servers we use two MaxScale servers, one to handle the back-end data streaming and the other to handle selective query proxying from your application. In scaling for the network load, you can add MaxScale servers to the first to handle a larger database write load or to the second to manage a greater number of queries from your application.
Sample Use Cases
Retail Store Example
A retail store wants to amplify sales by providing a personalized shopping experience. When a customer checks out at a cash register or online, the customer is presented promotions tailored to the customer's interests. These offers may also be integrated across channels, available when the known customer visits the website, shops by mobile phone app, or be included in the next personalized email sent to the customer.
At a technical level, when an OLTP query is performed to process the customer's purchase, the customer's past and current purchase history is analyzed with an OLAP query to provide promotions tailored to the customer's buying history.
Retail Bank Example
A retail bank maintains a customer database that includes their account information and a transaction activities ledger. When a customer makes a deposit or a withdrawal, the application updates the transaction activities ledger of their account. The customer can access their account information and activity through the bank's online portal.
Additionally, the application generates reports analyzing transaction activities. These reports are adapted for categories of customers (business, student, regular checking, savings) or for types of transactions (cash deposits, checks, ATM deposits, in-branch deposits, transfers, withdrawals). These reports can be run by the customers on their individual accounts or by the bank's back office on all customer activities.
In this scenario, queries listing account information and general transaction activities are OLTP operations. Reports analyzing transaction activities run by the customer for individual accounts or by the bank on all customers are OLAP operations.
Internet of Things Example
A chain of convenience stores maintains an IoT (Internet of Things) network in which each store records data on its milk inventory levels and sensor data such as refrigerator temperature. The central office continuously monitors inventory levels to trigger replenishment on an as-needed basis. The maintenance teams and store also receive real-time alerts if issues arise with the cooling system, speeding repair and reducing product losses. A holistic, whole-picture view of supply levels and status allows the chain to keep costs low and the customer experience consistent.
At a technical level, purchasing of a milk carton or container triggers an OLTP query, and inventory reporting is an OLAP query. OLTP data is used for logging, and analysis of OLAP data drives understanding of product losses, replenishment patterns, and equipment failures.
Sample Deployment
The following sections detail how to implement a sample deployment of Platform X3 for HTAP. The first steps cover server installation and deployment; the next cover configuration for Replication, Data Streaming, and Application Traffic, and lastly Testing with OLTP and OLAP queries and with DML statements.
Deploy MariaDB Servers
Our sample deployment calls for four servers running MariaDB Server to handle OLTP workloads, which we've named Server-1 to Server-4. These are shown at the left of our sample deployment diagram, in orange. We use InnoDB as the storage engine on these servers.
These four servers run MariaDB Server 10.3, installed per the instructions at: Getting and Upgrading MariaDB.
Deploy MariaDB ColumnStore Servers
Our sample deployment calls for five servers to run MariaDB ColumnStore to handle OLAP workloads. Two of these servers operate as User Module servers, named UM-1 and UM-2, and receive application traffic from MaxScale. The other three operate as Performance Module servers, named PM-1 through PM-3, and perform distributed query processing.
These servers run MariaDB ColumnStore 1.2.2 and are installed using the non-root and non-distributed installation method, as per the instructions at: Preparing and Installing MariaDB ColumnStore.
Deploy MariaDB MaxScale Servers
Our sample deployment calls for two servers to run MariaDB MaxScale. The first server, named MaxScale-1, handles data streaming from the MariaDB Servers to the MariaDB ColumnStore servers. The second, named MaxScale-2, selectively proxies application traffic to the respective servers for OLTP and OLAP workloads.
These servers run MariaDb MaxScale 2.3.1, installed per the instructions at: Installing MariaDB MaxScale.
Configure for Replication
Once we have the server software installed on the respective hosts, we can begin configuring them for use. To start, our sample deployment calls for the four MariaDB Servers to synchronize data using MariaDB Replication. This allows for high availability on OLTP operations, replication backup and failover.
In MariaDB Replication, one server operates as the master receiving all writes from the application and replicating changes to the cluster. The other servers operate as slaves, receiving reads from the application and only accepting writes from the master server.
For our sample deployment, Server-1 operates as the replication master while Server-2 through Server-4 operates as the replication slaves.
Configure Server-1 (master)
Add the following lines to the [mysqld] section of /etc/my.cnf.d/server.cnf:
[mysqld] server_id = 1 log_bin = mariadb-bin binlog_format = ROW gtid_strict_mode = 1 log_error log-slave-updates
In streaming data from MariaDB Server to ColumnStore for analysis, MaxScale requires that the Servers format the binary log events by each row modified by a statement, rather than by operation. So, when deploying a cluster for HTAP, ensure that the binlog_format system variable on the MariaDB Servers is always set to the ROW value.
For more information on these server system variables and others, see Server System Variables.
Configure Server-2 (slave)
Add the following lines to the [mysqld] section of /etc/my.cnf.d/server.cnf:
[mysqld] server_id = 2 log-bin = mariadb-bin binlog-format = ROW gtid_strict_mode = 1 log_error log-slave-updates
Configure Server-3 (slave)
Add the following lines to the [mysqld] section of /etc/my.cnf.d/server.cnf:
[mysqld] server_id = 3 log-bin = mariadb-bin binlog-format = ROW gtid_strict_mode = 1 log_error log-slave-updates
Configure Server-4 (slave)
Add the following lines to the [mysqld] section of /etc/my.cnf.d/server.cnf:
[mysqld] server_id = 4 log-bin = mariadb-bin binlog-format = ROW gtid_strict_mode = 1 log_error log-slave-updates
Restart Server-1 to Server-4
Restart all four MariaDB Servers. Log into each server (Server-1, Server-2, Server-3 and Server-4), and issue the restart command to systemctl on each server:
# systemctl restart mariadb.service
Create a replication user on Server-1
When MariaDB Servers run as replication slaves, they replicate data through client connections with the master server. In order for these servers to establish client connections, create a replication user on the master server, Server-1, and grant the user the relevant privileges to retrieve the data.
Connect to the master MariaDB Server through the client:
$ mysql -u root -p -h <Server-1-ip> -P 3306
Once connected, reset the master:
RESET MASTER;
Then, create a replication user for MaxScale and the slave MariaDB Servers and grant the relevant privileges to the user:
CREATE USER 'repl'@'%' IDENTIFIED BY 'pass'; GRANT REPLICATION SLAVE ON *.* TO 'repl'; GRANT SELECT ON mysql.user TO 'repl'; GRANT SELECT ON mysql.db TO 'repl'; GRANT SELECT ON mysql.tables_priv TO 'repl'; GRANT SELECT ON mysql.roles_mapping TO 'repl'; GRANT SHOW DATABASES ON *.* TO 'repl'; GRANT REPLICATION CLIENT ON *.* TO 'repl';
For more information on MariaDB Replication, see Performance Tuning with MariaDB Replication.
Configure and start replication on Server-2 to Server-4
With each slave MariaDB Server in your deployment, configure it to replicate data from the master server and start the replication process. Perform the following operations on each slave server, (that is, Server-2 through Server-4).
First, connect to the slave server:
$ mysql -u root -h <slave-server-ip> -P 3306 -p
If replication is currently running, reset the master so you can update its configuration:
RESET MASTER;
Issue a CHANGE MASTER TO statement to configure the slave to replicate from Server-1:
CHANGE MASTER TO
MASTER_HOST='Server-1-ip',
MASTER_USER='repl',
MASTER_PASSWORD='pass';
Then, start the replication slave:
START SLAVE;
You can verify that replication is working using a SHOW SLAVE STATUS statement.
SHOW SLAVE STATUS;
Configure for data streaming
In HTAP deployments the only queries issued to MariaDB ColumnStore are those specific to OLAP workloads, which does not include writes. In order to update ColumnStore with new data written to the MariaDB Servers, configure MaxScale on the back-end to stream writes to ColumnStore.
Configure MaxScale
Our sample deployment calls for a MaxScale server to handle data streaming between the MariaDB Servers and the MariaDB ColumnStore cluster. On the MaxScale-1 server, edit the /etc/maxscale.cnf configuration file, adding the following lines to set it up for data streaming to ColumnStore using the Avro Listener.
First, configure the replication router. Set it to use the particular server_id and the binary log:
[replication-router] type = service router = binlogrouter user = repl password = pass server_id = 5 master_id = 1 Binlogdir = /var/lib/maxscale Mariadb10-compatibility = 1 filestem = mariadb-bin
Then, configure the replication listener:
[replication-listener] type = listener service = replication-router protocol = MySQLClient port = 6603
MaxScale now listens for configuration connections on port 6603. This allows a MariaDB client to set up replication using commands similar to those that manage a replication slave server. It only uses the user and password to authentication the configuration connection, (the credentials for connecting to the Server are specified in the configuration below).
Next, configure the router for the Avro service.
[avro-router] type = service router = avrorouter source = replication-router avrodir = /var/lib/maxscale
This generates JSON files from the binary logs it receives from the master MariaDB Server and stores them in the avodir directory (that is, /var/lib/maxscale/) using the replication router.
Then, configure the listener for the Avro service to use a specific port:
[avro-listener] type = listener service = avro-router protocol = cdc port = 4001
Once this is done, save the file and restart MaxScale to apply the new configuration:
# sudo systemctl restart maxscale
Lastly, using the maxctrl utility, create a user for the Avro Router to capture data changes. This user handles streaming data MaxScale retrieves from the MariaDB Servers to ColumnStore.
# maxctrl call command cdc add_user avro-router cdcuser cdc
For more information on the Avro Router, see: Avro Router.
Configure MaxScale as a slave
When the MaxScale server streams data to MariaDB ColumnStore it retrieves it from the master server using the same process that the slaves use in MariaDB Replication. In effect it operates as a replication slave, only instead of writing data locally, it streams the writes to the ColumnStore User Modules.
Connect to MaxScale using the MariaDB Client. Unlike when connecting to MariaDB Servers previously, use port 6603, (which you configured above in the /etc/maxscale.cnf file as the replication listener port).
$ mysql -h <MaxScale-1-ip> -P 6603 -u repl -p
Issue a CHANGE MASTER TO statement to use the master MariaDB Server host (that is, the IP address to Server-1) and the port for client connections, (which defaults to 3306). Set the user and password as defined for the replication router in /etc/maxscale.cnf above.
CHANGE MASTER TO MASTER_HOST='<Server-1-ip>',
MASTER_PORT=3306,
MASTER_USER='repl',
MASTER_PASSWORD='pass',
MASTER_LOG_FILE='mariadb-bin.000001';
Then, start the slave:
START SLAVE;
Once you've started the replication slave process on MaxScale, you can check it using the SHOW SLAVE STATUS statement, just as you would when checking the status of a slave MariaDB Server.
SHOW SLAVE STATUS;
If there are no errors, MaxScale-1 is now running as a replication slave to Server-1.
Configure the CDC user
On the master MariaDB Server, (that is, Server-1), create a user for the CDC service. The CDC Data Adapter authenticates with this user when retrieving data from the MariaDB Servers.
Connect using the MariaDB Client to Server-1:
$ mysql -u root -p -h <Server-1-ip>
Then, issue the following statements to create the CDC user and grant it the necessary privileges:
CREATE USER 'cdcuser'@'%' IDENTIFIED BY 'cdc'; GRANT REPLICATION SLAVE ON *.* TO 'cdcuser'; GRANT SELECT ON mysql.user TO 'cdcuser'; GRANT SELECT ON mysql.db TO 'cdcuser'; GRANT SELECT ON mysql.tables_priv TO 'cdcuser'; GRANT SELECT ON mysql.roles_mapping TO 'cdcuser'; GRANT SHOW DATABASES ON *.* TO 'cdcuser'; GRANT REPLICATION CLIENT ON *.* TO 'cdcuser';
Install the CDC Streaming Data Adapter
The MaxScale CDC Streaming Data Adapter allows you to stream binary log events from MariaDB Servers to MariaDB ColumnStore clusters. In order to use it, install the ColumnStore Bulk Write SDK and the MaxScale CDC Adapter packages on a dedicated host or on any MaxScale server that you want to use for data streaming, (MaxScale-1 in our sample deployment).
Downloads are available at: Downloads.
Note these packages conflict with ColumnStore installations. Don't install them on any of your ColumnStore servers.
To install the ColumnStore Bulk Write SDK, download the RPM package from MariaDB, then install the EPEL releases and package dependencies using YUM:
$ wget https://downloads.mariadb.com/Data-Adapters/mariadb-columnstore-api/1.2.2/centos/x86_64/7/Mariadb-columnstore-api-1.2.2-1-x86_64-centos7-cpp.rpm $ sudo yum install epel-release $ sudo yum install -y libuv libxml2 snappy python34 $ sudo yum install -y mariadb-columnstore-api-1.2.2-1-x86_64-centos7-cpp.rpm
To install the CDC Data Adapter, download the RPM package from MariaDB, then install it using YUM:
$ wget https://downloads.mariadb.com/Data-Adapters/mariadb-streaming-data-adapters/cdc-data-adapter/1.2.2/centos-7/mariadb-columnstore-maxscale-cdc-adapters-1.2.2-1-x86_64-centos7.rpm $ sudo yum install -y mariadb-columnstore-maxscale-cdc-adapters-1.2.2-1-x86_64-centos7.rpm
For more information on installing the Data Adapter, see Data Adapters Installation.
Configure the CDC Data Adapter
With the CDC Data Adapter installed you can configure it to stream data to MariaDB ColumnStore. This is done by copying the Columnstore.xml configuration file from one of the ColumnStore nodes to the MaxScale-1 server, where the CDC Data Adapter can use it.
$ scp root@columnstore-host:/home/mysql/columnstore/etc/Columnstore.xml \
~/Columnstore.xml
$ sudo mv Columnstore.xml /etc
Note, installing MaxScale and the CDC Data Adapter as root creates the /var/lib/mxs_adapter/ directory. If you intend to run mxs_adapter as a non-root user, ensure that the user can read and write from this directory. If not, change the ownership to that of your user. For instance,
$ sudo chown ec2-user /var/lib/mxs_adapter
Lastly, check the firewall and SELinux. ColumnStore uses ports 8600 to 8630, as well as ports 8700 and 8800. The CDC Data Adapter uses the same ports to stream data from MaxScale-1 to ColumnStore. Check each server to ensure that there is no firewall blocking these ports. Additionally, ensure that SELinux has a policy allowing these connections or that it is running in permissive mode.
Verify data streaming
Once you've installed and configured the MaxScale and the CDC Data Adapter, you can run checks to verify that it is properly configured and able to communicate and stream data from the MariaDB Servers to the MariaDB ColumnStore cluster. Using the mxs_adapter utility, you can connect to MaxScale and test data streaming.
In order to test this feature, you first need to create tables to store the data the CDC Data Adapter transfers. First, connect to the master MariaDB Server, Server-1, and create an InnoDB table to use in storing test data:
CREATE TABLE test.t6(a INT, b INT) ENGINE=InnoDB;
Then, connect to one of the User Modules and create a ColumnStore table with the same name and schema:
CREATE TABLE test.t6(a INT, b INT) ENGINE=ColumnStore;
In order to stream data from the MariaDB Servers to ColumnStore,call the mxs_adapter utility. From the MaxScale-1 server, run the following command:
$ mxs_adapter -c /etc/ColumnStore.xml -u cdcuser -p cdc \ -h localhost -P 4001 -r 2 -d -n -z test t6
Use the username and password for the CDC user created in the previous section. In the MaxScale configuration, we set port 4001 for the listener service.
The last two arguments tell it to stream the t6 table on the test database. If you want to stream multiple tables, replace these arguments with a -f option that gives the path to a table list file. Format the file for one table per line, separating the database name and table name by a tab. For instance,
$ cat tbl.lst test t6 test t7 test t8
When you start streaming data, the mxs_adapter utility begins printing logging messages to stdout. As you add data to the MariaDB Servers, you can check this output to see binary events streaming over to ColumnStore.
To test this, begin inserting data onto the test.t6 table on the Server-1 server. This is the master server in the MariaDB Replication and the only one that accepts write operations.
INSERT INTO test.t6 VALUES (1,1); INSERT INTO test.t6 VALUES (1,2); INSERT INTO test.t6 VALUES (1,3); INSERT INTO test.t6 VALUES (1,4); INSERT INTO test.t6 VALUES (1,5);
Then, issue a SELECT statement to see what data is available on the MariaDB Servers for the OLTP operations:
SELECT * FROM test.t6; +----+----+ | a | b | +----+----+ | 1 | 1 | | 1 | 2 | | 1 | 3 | | 1 | 4 | | 1 | 5 | +----+----+ 5 rows in set (0.010 sec)
By checking the logging messages coming from the mxs_adapter utility, you can watch these INSERT statements as they stream from the MariaDB Servers through MaxScale to ColumnStore:
2018-11-23 18:39:09 [main] Started thread 0x176b470 2018-11-23 18:39:09 [main] Started 1 threads 2018-11-23 18:39:09 [test.t6] Requesting data for table: test.t6 2018-11-23 18:39:09 [test.t6] INSERT INTO `test`.`t6` (`a`,`b`) VALUES (1,1) 2018-11-23 18:39:09 [test.t6] DML average: 12ms 2018-11-23 18:39:19 [test.t6] Read timeout 2018-11-23 18:39:22 [test.t6] INSERT INTO `test`.`t6` (`a`,`b`) VALUES (1,2) 2018-11-23 18:39:22 [test.t6] DML average: 4ms 2018-11-23 18:39:24 [test.t6] INSERT INTO `test`.`t6` (`a`,`b`) VALUES (1,3) 2018-11-23 18:39:24 [test.t6] DML average: 4ms 2018-11-23 18:39:26 [test.t6] INSERT INTO `test`.`t6` (`a`,`b`) VALUES (1,4) 2018-11-23 18:39:26 [test.t6] DML average: 4ms 2018-11-23 18:39:28 [test.t6] INSERT INTO `test`.`t6` (`a`,`b`) VALUES (1,5) 2018-11-23 18:39:28 [test.t6] DML average: 9ms
Then, you can issue a similar SELECT statement to the test.t6 table on either of the User Modules for MariaDB ColumnStore to see that the data is now available for OLAP operations:
SELECT * FROM test.t6; +----+----+ | a | b | +----+----+ | 1 | 1 | | 1 | 2 | | 1 | 3 | | 1 | 4 | | 1 | 5 | +----+----+ 5 rows in set (0.010 sec)
Next, test out other write operations using either an UPDATE or DELETE statement on Server-1:
UPDATE test.t6 SET a = 2 WHERE b > 3; SELECT * FROM test.t6; +----+----+ | a | b | +----+----+ | 1 | 1 | | 1 | 2 | | 1 | 3 | | 2 | 4 | | 2 | 5 | +----+----+ 5 rows in set (0.000 sec)
Using the logging messages from the CDC Data Adapter, you can watch the binary events for this operation stream through MaxScale:
2018-11-23 18:44:18 [test.t6] Read timeout 2018-11-23 18:44:22 [test.t6] UPDATE `test`.`t6` SET `a` = 2, `b` = 4 WHERE `a` = 1 AND `b` = 4 2018-11-23 18:44:22 [test.t6] DML average: 12ms 2018-11-23 18:44:22 [test.t6] UPDATE `test`.`t6` SET `a` = 2, `b` = 5 WHERE `a` = 1 AND `b` = 5 2018-11-23 18:44:22 [test.t6] DML average: 13ms 2018-11-23 18:44:27 [test.t6] Read timeout
As you can see from the logging messages, MaxScale detected the UPDATE statement and streamed it through the CDC Data Adapter to ColumnStore. The CDC Data Adapter then begins logging Read timeout messages to indicate that it is done streaming and is waiting on additional binary events from the MariaDB Servers.
You can confirm that the data was successfully transferred by issuing a SELECT statement to one of the MariaDB ColumnStore User Modules:
SELECT * FROM test.t6; +----+----+ | a | b | +----+----+ | 1 | 1 | | 1 | 2 | | 1 | 3 | | 2 | 4 | | 2 | 5 | +----+----+ 5 rows in set (0.011 sec)
Notice that rows 4 and 5 now contain new values.
Configure for Application Traffic
When your application issues queries to Platform X3 for HTAP operations, it doesn't connect to either the MariaDB Servers or to the MariaDB ColumnStore User Modules directly. Instead, it connects to a MaxScale server configured to selectively routes queries, ensuring that OLTP operations execute on MariaDB Servers and OLAP operations execute on ColumnStore.
In order to better illustrate how MaxScale distributes queries between the servers, we are going to install a sample banking database and show how to process payments and analyze loan data.
Our sample database contains the following tables:
-
account- each record describes the static characteristics of an account -
client- each record describes the characteristics of a client -
client_accts- each record relates together a client with an account -
loan- each record describes a loan granted for a given account
With the configuration so far, our sample deployment will streams these new tables from the MariaDB Serves to ColumnStore through the CDC Data Adapter running on MaxScale-1. We'll then configure the second MaxScale server to selectively route your application traffic for HTAP operations.
Prepare MariaDB Server and MariaDB ColumnStore to receive traffic from MaxScale
Create a read user for MaxScale on the master MariaDB Server Server-1 and the ColumnStore User Modules. Grant it the necessary privileges to operate.
CREATE USER 'maxscale' IDENTIFIED BY 'pass'; GRANT SELECT ON mysql.user TO 'maxscale'; GRANT SELECT ON mysql.db TO 'maxscale'; GRANT SELECT ON mysql.tables_priv TO 'maxscale'; GRANT SHOW DATABASES ON *.* TO 'maxscale';
Then, create a write user for MaxScale on the same servers.
CREATE USER 'maxuser'@'%' IDENTIFIED BY 'maxpwd'; GRANT ALL ON *.* TO 'maxuser'@'%';
Create the schema
Download and unzip the sample dataset onto both Server-1 and a ColumnStore User Module from test-db. This file is test-db.zip, which unzips into the test-db/ directory . It contains following SQL and CSV files:
$ unzip test-db.zip $ ls test-db/ create-db-cs.sql create-db-innodb.sql account.csv client_accts.csv client.csv loan.csv
On the master MariaDB Server, Server-1, change into the unzipped directory and launch the client as the write user maxuser you created above for MaxScale:
$ cd test-db $ mysql -u 'maxuser' -p
Use the SOURCE command to load the create-db-innodb.sql file, initializing the base database:
SOURCE create-db-innodb.sql;
Then, modify the schema to add a balance column to the bank.loan table. This column will be used in the ColumnStore examples below.
SET sql_log_bin = 0; ALTER TABLE bank.loan ADD COLUMN balance decimal(10,2); SET sql_log_bin = 1;
On a ColumnStore User Module, connect from the same directory and with the same user:
$ cd test-db $ mysql -u 'maxuser' -p
Then, use the SOURCE command to load the create-db-cs.sql file:
SOURCE create-db-cs.sql;
At this point, we have created the bank database and tables, and have loaded the data into the MariaDB Servers, (though we only wrote to Server-1, as the master server it has replicated the data out to the slaves). We have also created the bank database and tables in MariaDB ColumnStore. The tables on MariaDB ColumnStore are empty at this point.
Streaming Data from MariaDB Servers to MariaDB ColumnStore
Next, we will start streaming data via the mxs_adapter utility, so that the data we've loaded on the MariaDB Servers can stream to MariaDB ColumnStore. On MaxScale-1, create a TSV (tab-separated) file named bank.lst with the bank database tables we want to stream:
$ cat bank.lst bank account bank client bank client_accts bank loan
Now, start the mxs_adapter utility designating this file with the -f option for this file:
$ mxs_adapter -c /etc/Columnstore.xml -u cdcuser -p cdc \
-h <maxscale-1-host> -P 4001 -r 50 -d -n -f bank.lst
When you run the mxs_adapter utility, it streams logging messages about the operations it's performing to stdout. You can monitor this information to see the binary events its streaming from the MariaDB Servers to MariaDB ColumnStore.
2018-11-28 03:56:49 [bank.client_accts] client_id: 13971 2018-11-28 03:56:49 [bank.client_accts] account_id: 11362 2018-11-28 03:56:49 [bank.client_accts] ca_type: OWNER 2018-11-28 03:56:49 [bank.client_accts] ca_id: 13690 2018-11-28 03:56:49 [bank.client_accts] client_id: 13998 2018-11-28 03:56:49 [bank.client_accts] account_id: 11382 2018-11-28 03:56:49 [bank.client_accts] ca_type: OWNER 2018-11-28 03:56:49 [bank.client_accts] ca_id: 0 2018-11-28 03:56:49 [bank.client_accts] client_id: 2018-11-28 03:56:49 [bank.client_accts] account_id: 2018-11-28 03:56:49 [bank.client_accts] ca_type: 2018-11-28 03:56:52 [bank.account] Read timeout 2018-11-28 03:56:57 [bank.loan] Read timeout 2018-11-28 03:56:58 [bank.client] Flushing batch 2018-11-28 03:56:58 [bank.client] 21 rows, 0 transactions inserted over 10.1008 seconds. GTID = 0-1-23:5371 2018-11-28 03:56:59 [bank.client_accts] Flushing batch 2018-11-28 03:56:59 [bank.client_accts] 21 rows, 0 transactions inserted over 10.1153 seconds. GTID = 0-1-27:5371 2018-11-28 03:57:02 [bank.account] Read timeout
When all the loaded data has been streamed from the MariaDB Servers to ColumnStore, you'll begin to see Read timeout messages in the output. This means that the mxs_adapter utility is now waiting on additional binary events to occur on the MariaDB Servers.
At this point, if you issue a SELECT COUNT(*) statements to MariaDB ColumnStore, you should get following result-sets:
SELECT COUNT(*) FROM bank.account; +----------+ | COUNT(*) | +----------+ | 4502 | +----------+ 1 row in set (0.057 sec) SELECT COUNT(*) FROM bank.client; +----------+ | COUNT(*) | +----------+ | 5371 | +----------+ 1 row in set (0.051 sec) SELECT COUNT(*) FROM bank.loan; +----------+ | COUNT(*) | +----------+ | 684 | +----------+ 1 row in set (0.056 sec) SELECT COUNT(*) FROM bank.client_accts; +----------+ | COUNT(*) | +----------+ | 5371 | +----------+ 1 row in set (0.027 sec)
The MariaDB Servers and ColumnStore now contain the same data.
Configure HTAP routing
Our sample deployment proxies application traffic through a second MariaDB MaxScale server for the selective routing of HTAP queries. Specifically, it involves configuring the MaxScale-2 server so that:
- All
INSERT,UPDATE,DELETEstatements always route toServer-1for (OLTP) - All
SELECTqueries on the loan table always route to MariaDB ColumnStore (OLAP) - All remaining read queries will route to any MariaDB Server.
Below is a diagram of the MaxScale-2 server configuration:
Here is the configuration file you should have in /etc/maxscale.cnf on MaxScale-2 to achieve the above.
[maxscale] threads = auto sql_mode = default ## Identify the Master MariaDB Server: Server-1 [sw-db1] type = server address = <MariaDB-Server-1-IP> port = 3306 protocol = MariaDBBackend ## Identify the Slave MariaDB Server: Server-2 [sw-db2] type = server address = <MariaDB-Server-2-IP> port = 3306 protocol = MariaDBBackend ## Identify the Slave MariaDB Server: Server-3 [sw-db3] type = server address = <MariaDB-Server-3-IP> port = 3306 protocol = MariaDBBackend ## Identify the Columnstore Server: UM-1 [sw-mcs-um1] type = server address = <MariaDB-UM-1-IP> port = 3306 protocol = MariaDBBackend ## Monitor all servers [MariaDB-Monitor] type = monitor module = mariadbmon servers = sw-db1,sw-db2,sw-db3,sw-mcs-um1 user = maxuser password = maxpwd monitor_interval = 10000 ## Service to talk to the servers. [MDB-Service] type = service router = readwritesplit servers = sw-db1,sw-db2 user = maxuser password = maxpwd ## Listener that clients use to access the MariaDB Servers. [MDB-Listener] type = listener service = MDB-Service protocol = MariaDBClient port = 4009 ## The MDB-Service abstracted as a server [MDB-Service-as-server] type = server address = 127.0.0.1 port = 4009 protocol = MariaDBBackend ## Service to talk to the ColumnStore UM [CS-Service] type = service router = readconnroute router_options = running servers = sw-mcs-um1 user = maxuser password = maxpwd ## Listener that clients use to access the CS-Service. [CS-Listener] type = listener service = CS-Service protocol = MariaDBClient port = 4010 ## CS-Service abstracted as a server [CS-Service-as-server] type = server address = 127.0.0.1 port = 4010 protocol = MariaDBBackend ## Filter the _datamart_ queries to the ColumnStore server and rest of the queries to the MariaDB Servers [target-selector] type = filter module = namedserverfilter match01 = (?i)SELECT.*loan target01 = CS-Service-as-server match02 = .* target02 = MDB-Service-as-server ## Filter to replace loan table name without database qualifier, with bank.loan, so that the connection to ColumnStore knows which database to use [loan-table-filter] type = filter module = regexfilter options = ignorecase match = \sloan\s replace = /* */ bank.loan /* */ log_trace = true log_file = /tmp/regexfilter.log ## Combines the two services as one [HTAP-Service] type = service Router = schemarouter ignore_databases_regex = .* servers = MDB-Service-as-server,CS-Service-as-server Preferred_server = MDB-Service-as-server user = maxuser password = maxpwd filters = target-selector ## Listener clients use to access the combined service [HTAP-Listener] type = listener service = HTAP-Service protocol = MariaDBClient port = 4011
Then, start MaxScale:
# systemctl start maxscale
Test HTAP Application Traffic
With the MariaDB Servers, ColumnStore and MaxScale servers configured and deployed, you can begin testing the sample HTAP deployment. The application connects to the second MaxScale server, MaxScale-2, on port 4011, where it performs selective query routing:
- All
INSERT,UPDATE,DELETEqueries always route toServer-1(OLTP) - All
SELECTqueries on loans table always route to MariaDB ColumnStore (OLAP) - All remaining read queries route to any MariaDB Server (OLTP)
Connect from application to MaxScale HTAP Service
From your application server use the MariaDB Client to connect to the MaxScale HTAP Service.
$ mysql -h <maxscale2-host> -P 4011 -u maxuser -p
These are the same command-line options as you would use to connect to a MariaDB Server, but instead of an individual server, you connect to MaxScale, which sends the queries to the Servers or to one of the ColumnStore UM's.
Use the password 'maxpwd' value that you previously set for this user.
Transactional queries
The MariaDB MaxScale server configuration above designates queries on tables other than bank.loan as transactional and routes them to the MariaDB Servers rather than ColumnStore. You can identify which server cluster the query executes on using the version_comment system variable.
SELECT *, @@version_comment FROM bank.account LIMIT 5; +------------+-------------+------------------+------------+-------------------+ | account_id | district_id | frequency | a_d | @@version_comment | +------------+-------------+------------------+------------+-------------------+ | 576 | 55 | POPLATEK MESICNE | 1993-01-01 | MariaDB Server | | 3818 | 74 | POPLATEK MESICNE | 1993-01-01 | MariaDB Server | | 704 | 55 | POPLATEK MESICNE | 1993-01-01 | MariaDB Server | | 2378 | 16 | POPLATEK MESICNE | 1993-01-01 | MariaDB Server | | 2632 | 24 | POPLATEK MESICNE | 1993-01-02 | MariaDB Server | +------------+-------------+------------------+------------+-------------------+ 5 rows in set (0.039 sec)
Since MaxScale routes this query as a transactional operation, the version_comment system variable returns MariaDB Server.
Analytical queries
The MariaDB MaxScale server configuration above designates queries on the bank.loans table as analytical queries and routes them to the MariaDB ColumnStore User Modules rather than the MariaDB Servers. You can identify which server cluster the query executes on using the version_comment system variable.
SELECT loan_id, account_id, amount, duration, payments, @@version_comment FROM loan LIMIT 5; +---------+------------+-----------+----------+----------+---------------------+ | loan_id | account_id | amount | duration | payments | @@version_comment | +---------+------------+-----------+----------+----------+---------------------+ | 0 | 0 | 0.00 | 0 | 0.00 | Columnstore 1.2.1-1 | | 5314 | 1787 | 96396.00 | 12 | 8033.00 | Columnstore 1.2.1-1 | | 5316 | 1801 | 165960.00 | 36 | 4610.00 | Columnstore 1.2.1-1 | | 6863 | 9188 | 127080.00 | 60 | 2118.00 | Columnstore 1.2.1-1 | | 5325 | 1843 | 105804.00 | 36 | 2939.00 | Columnstore 1.2.1-1 | +---------+------------+-----------+----------+----------+---------------------+
Since MaxScale routes the query as an analytical operation, the version_comment system variable indicates a ColumnStore server.
DML Statements
The MariaDB MaxScale server configuration above designates data manipulation statements such as INSERT, UPDATE and DELETE as transactional and routes these statements to the MariaDB Servers. The other MaxScale server then streams the changes over to ColumnStore.
Issue an UPDATE statement to set the initial balance for a loan:
UPDATE bank.loan SET balance = amount WHERE loan_id = 5314;
Then, query the updated row to see the changes on ColumnStore:
SELECT loan_id, account_id, amount, payments, @@version_comment FROM bank.loan WHERE loan_id = 5314; +---------+------------+----------+----------+-----------+---------------------+ | loan_id | account_id | amount | payments | balance | @@version_comment | +---------+------------+----------+----------+-----------+---------------------+ | 5314 | 1787 | 96396.00 | 8033.00 | 96369.00 | Columnstore 1.2.1-1 | +---------+------------+----------+----------+-----------+---------------------+ 1 row in set (0.113 sec)
The loan balance has been set to its initial value of the received funds.
As you can see, the account holder has made payments on the loan. This amount should be removed from the balance to reflect their paying it down. Issue another UPDATE statement to modify the balance, removing payments made on the account:
UPDATE bank.loan SET balance = balance - payments WHERE loan_id = 5314;
Then, query the loan to view the updated rows on ColumnStore:
SELECT loan_id, account_id, amount, payments, @@version_comment FROM bank.loan WHERE loan_id = 5314; +---------+------------+----------+----------+----------+---------------------+ | loan_id | account_id | amount | payments | balance | @@version_comment | +---------+------------+----------+----------+----------+---------------------+ | 5314 | 1787 | 96396.00 | 8033.00 | 88336.00 | Columnstore 1.2.1-1 | +---------+------------+----------+----------+----------+---------------------+ 1 row in set (0.094 sec)
The loan balance has now been reduced to reflect account payments. MaxScale has also streamed the data from the changes over to ColumnStore.
For More Information
Obtaining the Products
When you're ready to install MariaDB Platform X3, go to Downloads and select Platform X3. If you use an RPM or APT based distribution of Linux, you can configure your server repositories to install it through the package manager.
Documentation
Documentation for MariaDB products is available in the Library. For the documentation on specific products, see the links below:
© 2021 MariaDB
Licensed under the Creative Commons Attribution 3.0 Unported License and the GNU Free Documentation License.
https://mariadb.com/kb/en/sample-platform-x3-implementation-for-transactional-and-analytical-workloads/