To get a SymmetricDS node running, it needs to be given an identity and it needs to know how to connect to the database it will be synchronizing. A typical way to specify this is to place properties in the symmetric.properties file. When started up, SymmetricDS reads the configuration and state from the database. If the configuration tables are missing, they are created automatically (auto creation can be disabled). Basic configuration is described by inserting into the following tables (the complete data model is defined in Appendix A, Data Model).

During start up, triggers are verified against the database, and database triggers are installed on tables that require data changes to be captured. The Route, Pull and Push Jobs begin running to synchronize changes with other nodes.

Each node requires properties that allow it to connect to a database and register with a parent node. To give a node its identity, the following properties are used:

When a new node is first started, it is has no information about synchronizing. It contacts the registration server in order to join the network and receive its configuration. The configuration for all nodes is stored on the registration server, and the URL must be specified in the following property:

When deploying to an application server, it is common for database connection pools to be found in the Java naming directory (JNDI). In this case, set the following property:

For a deployment where the database connection pool should be created using a JDBC driver, set the following properties:

A node, a single instance of SymmetricDS, is defined in the NODE table. Two other tables play a direct role in defining a node, as well The first is NODE_IDENTITY. The only row in this table is inserted in the database when the node first registers with a parent node. In the case of a root node, the row is entered by the user. The row is used by a node instance to determine its node identity.

The following SQL statements set up a top-level registration server as a node identified as "00000" in the "corp" node group.

insert into SYM_NODE 
  (node_id, node_group_id, external_id, sync_enabled)
values
  ('00000', 'corp', '00000', 1);

insert into SYM_NODE_IDENTITY values ('00000');

The second table, NODE_SECURITY has rows created for each child node that registers with the node, assuming auto-registration is enabled. If auto registration is not enabled, you must create a row in NODE and NODE_SECURITY for the node to be able to register. You can also, with this table, manually cause a node to re-register or do a re-initial load by setting the corresponding columns in the table itself. Registration is discussed in more detail in Section 4.7, “Opening Registration”.

Node Groups are straightforward to configure and are defined in the NODE_GROUP table. The following SQL statements would create node groups for "corp" and "store" based on our retail store example.

insert into SYM_NODE_GROUP 
  (node_group_id, description)
values
  ('store', 'A retail store node');

insert into SYM_NODE_GROUP 
  (node_group_id, description)
values
  ('corp', 'A corporate node');

Similarly, Node Group links are established using a data event action of 'P' for Push and 'W' for Pull ("wait"). The following SQL statements links the "corp" and "store" node groups for synchronization. It configures the "store" nodes to push their data changes to the "corp" nodes, and the "corp" nodes to send changes to "store" nodes by waiting for a pull.

insert into SYM_NODE_GROUP_LINK
  (source_node_group, target_node_group, data_event_action)
values
  ('store', 'corp', 'P');

insert into SYM_NODE_GROUP_LINK
  (source_node_group, target_node_group, data_event_action)
values
  ('corp', 'store', 'W');

By categorizing data into channels and assigning them to TRIGGERs, the user gains more control and visibility into the flow of data. In addition, SymmetricDS allows for synchronization to be enabled, suspended, or scheduled by channels as well. The frequency of synchronization and order that data gets synchronized is also controlled at the channel level.

The following SQL statements setup channels for a retail store. An "item" channel includes data for items and their prices, while a "sale_transaction" channel includes data for ringing sales at a register.

insert into SYM_CHANNEL 
  (channel_id, processing_order, max_batch_size, max_batch_to_send, 
   extract_period_millis, batch_algorithm, enabled, description)
values
  ('item', 10, 1000, 10,  0, 'default', 1, 'Item and pricing data');

insert into SYM_CHANNEL 
  (channel_id, processing_order, max_batch_size, max_batch_to_send, 
   extract_period_millis, batch_algorithm, enabled, description)
values
  ('sale_transaction', 1, 1000, 10,  60000, 'transactional', 1, 
   'retail sale transactions from register');

Batching is the grouping of data, by channel, to be transferred and committed at the client together. There are three different out-of-the-box batching algorithms which may be configured in the batch_algorithm column on channel.

Node registration is the act of setting up a new NODE and NODE_SECURITY so that when the new node is brought online it is allowed to join the system. Nodes are only allowed to register if rows exist for the node and the registration_enabled flag is set to 1. If the auto.registration SymmetricDS property is set to true, then when a node attempts to register, if registration has not already occurred, the node will automatically be registered.

SymmetricDS allows you to have multiple nodes with the same external_id. Out of the box, openRegistration will open a new registration if a registration already exists for a node with the same external_id. A new registration means a new node with a new node_id and the same external_id will be created. If you want to re-register the same node you can use the reOpenRegistration() JMX method which takes a node_id as an argument.

An initial load is the process of seeding tables at a target node with data from its parent node. An initial load can not occur until after a node is registered. An initial load is requested by setting the initial_load_enabled column on NODE_SECURITY to 1 on the row for the target node in the parent node's database. The next time the target node synchronizes, reload batches will be inserted. At the same time reload batches are inserted, all previously pending batches for the node are marked as successfully sent.

SymmetricDS recognizes that an initial load has completed when the initial_load_time column on the target node is set to a non null value.

Reload batches are inserted in order according to the initial_load_order column on TRIGGER_ROUTER. Initial load data is always queried from the source database table. All data is passed through the configured router to filter out data that might not be targeted at a node.

A more efficient way to subset the data for a load is to provide an initial_load_select clause on TRIGGER_ROUTER. If an initial_load_select clause is provided, data will not be passed through the configured router during initial load. In cases where custom routing is done using a feature like Section 4.6.2.4, “Relational Router”, an initial_load_select clause will always need to be provided because the router would not function properly with initial load data.

insert into sym_trigger (TRIGGER_ID,SOURCE_CATALOG_NAME,
  SOURCE_SCHEMA_NAME,SOURCE_TABLE_NAME,CHANNEL_ID,
  SYNC_ON_UPDATE,SYNC_ON_INSERT,SYNC_ON_DELETE,
  SYNC_ON_INCOMING_BATCH,NAME_FOR_UPDATE_TRIGGER,
  NAME_FOR_INSERT_TRIGGER,NAME_FOR_DELETE_TRIGGER,
  SYNC_ON_UPDATE_CONDITION,SYNC_ON_INSERT_CONDITION,
  SYNC_ON_DELETE_CONDITION,EXTERNAL_SELECT,
  TX_ID_EXPRESSION,EXCLUDED_COLUMN_NAMES,
  CREATE_TIME,LAST_UPDATE_BY,LAST_UPDATE_TIME) 
  values ('SALE_TRANSACTION_DEAD',null,null,
  'SALE_TRANSACTION','transaction',
  0,0,0,0,null,null,null,null,null,null,null,null,null,
  current_timestamp,'demo',current_timestamp);

insert into sym_router (ROUTER_ID,TARGET_CATALOG_NAME,TARGET_SCHEMA_NAME,
  TARGET_TABLE_NAME,SOURCE_NODE_GROUP_ID,TARGET_NODE_GROUP_ID,ROUTER_TYPE,
  ROUTER_EXPRESSION,SYNC_ON_UPDATE,SYNC_ON_INSERT,SYNC_ON_DELETE,
  CREATE_TIME,LAST_UPDATE_BY,LAST_UPDATE_TIME) 
  values ('CORP_2_STORE',null,null,null,
  'corp','store',null,null,1,1,1,
  current_timestamp,'demo',current_timestamp);
   
insert into sym_trigger_router (TRIGGER_ID,ROUTER_ID,INITIAL_LOAD_ORDER,
  INITIAL_LOAD_SELECT,CREATE_TIME,LAST_UPDATE_BY,LAST_UPDATE_TIME) 
  values ('SALE_TRANSACTION_DEAD','CORP_2_REGION',100,null,
   current_timestamp,'demo',current_timestamp);
   

SymmetricDS allows tables to be synchronized bi-directionally. Note that an outgoing synchronization does not process changes during an incoming synchronization on the same node unless the trigger was created with the sync_on_incoming_batch flag set. If the sync_on_incoming_batch flag is set, then update loops are prevented by a feature that is available in most database dialects. More specifically, during an incoming synchronization the source node_id is put into a database session variable that is available to the database trigger. Data events are not generated if the target node_id on an outgoing synchronization is equal to the source node_id.

By default, only the columns that changed will be updated in the target system.

More complex conflict resolution strategies can be accomplished by using the IDataLoaderFilter extension point which has access to both old and new data.

As shown in Section 3.2, “Organizing Nodes”, there may be scenarios where data needs to flow through multiple tiers of nodes that are organized in a tree-like network with each tier requiring a different subset of data. For example, you may have a system where the lowest tier may by a computer or device located in a store. Those devices may connect to a server located physically at that store. Then the store server may communicate with a corporate server for example. In this case, the three tiers would be device, store, and corporate. Each tier is typically represented by a node group. Each node in the tier would belong to the node group representing that tier.

A node will always push and pull data to other node groups according to the node group link configuration. A node can only pull and push data to other nodes that are represented node table in its database and having sync_enabled = 1. Because of this, a tree-like hierarchy of nodes can be created by having only a subset of nodes belonging to the same node group represented at the different branches of the tree.

If auto registration is turned off, then this setup must occur manually by opening registration for the desired nodes at the desired parent node and by configuring each node's registration.url to be the parent node's URL. The parent node is always tracked by the setting of the parent's node_id in the created_at_node_id column of the new node. When a node registers and downloads its configuration it is always provided the configuration for nodes that might register with the node itself based on the Node Group Links defined in the parent node.

When deploying a multi-tiered system it may be advantageous to have only one registration server, even though the parent node of a registering node could be any of a number of nodes in the system. In SymmetricDS the parent node is always the node that a child registers with. The REGISTRATION_REDIRECT table allows a single node, usually the root server in the network, to redirect registering nodes to their true parents. It does so based on a mapping found in the table of the external id (registrant_external_id) to the parent's node id (registration_node_id).

For example, if it is desired to have a series of regional servers that workstations at retail stores get assigned to based on their external_id, the store number, then you might insert into REGISTRATION_REDIRECT the store number as the registrant_external_id and the node_id of the assigned region as the registration_node_id. When a workstation at the store registers, the root server send an HTTP redirect to the sync_url of the node that matches the registration_node_id.

The SymmetricDS software allows for outgoing and incoming changes to be synchronized to/from other databases. The node that initiates a synchronization connection is the client, and the node receiving a connection is the host. Because synchronization is configurable to push or pull in either direction, the same node can act as either a client or a host in different circumstances.

The SymmetricDS software consists of a series of background jobs, managers, Servlets, and services wired together via dependency injection using the Spring Framework.

As a client, the node runs the router job, push job and pull job on a timer thread. The router job uses services to create batches that are targeted at certain nodes. The push job uses services to extract and stream data to another node (that is, it pushes data). The response from a push is a list of batch acknowlegements to indicate that data was loaded. The pull job uses services to load data that is streamed from another node (i.e., it pulls data). After loading data, a second connection is made to send a list of batch acknowlegements.

As a host, the node waits for incoming connections that pull, push, or acknowledge data changes. The push Servlet uses services to load data that is pushed from a client node. After loading data, it responds with a list of batch acknowledgements. The pull Servlet uses services to extract, and stream data back to the client node. The ack Servlet uses services to update the status of data that was loaded at a client node. The router job batches and routes data.

By default, data is extracted from the source database into memory until a threshold size is reached. If the threshold size is reached, data is streamed to a temporary file in the JVM's default temporary directory. Next, the data is streamed to the target node across the transport layer. The receiving node will cache the data in memory until the threshold size is reached, writing to a temporary file if necessary. At last, the data is loaded into the target database by the data loader. This step by step approach allows for extract time, transport time, and load time to all be measured independently. It also allows database resources to be used most optimally.

The transport manager handles the incoming and outgoing streams of data between nodes. The default transport is based on a simple implementation over HTTP. An internal transport is also provided. It is possible to add other implementations, such as a socket-based transport manager.

Node communication over HTTP is represented in the following figure.

Figure 4.1. Node Communication

The StandaloneSymmetricEngine is wrapper API that can be used to directly start the client services only. The SymmetricWebServer is a wrapper API that can be used to directly start both the client and host services inside a Jetty web container. The SymmetricLauncher provides command line tools to work with and start SymmetricDS.

The frequency of data synchronization is controlled by the coordination of a series of asynchronous events.

After data is captured, the first event that occurs is the routing of the captured DATA rows. Data is routed by the Route Job. It is a single background task that inserts into DATA_EVENT and OUTGOING_BATCH. The Route Job determines which nodes data will be sent to, as well as how much data will be batched together for transport. When the start.route.job SymmetricDS property is set to true, the frequency that routing occurs is controlled by the job.routing.period.time.ms. Each time data is routed, the DATA_REF table is updated with the id of the last contiguous data row to have been processed. This is done so the query to find unrouted data is optimal.

After data is routed, it awaits transport to the target nodes. Transport can occur when a client node is configured to pull data or when the host node is configured to push data. These events are controlled by the Push and the Pull Jobs. When the start.pull.job SymmetricDS property is set to true, the frequency that data is pulled is controlled by the job.pull.period.time.ms. When the start.push.job SymmetricDS property is set to true, the frequency that data is pushed is controlled by the job.push.period.time.ms. Data is extracted by channel from the source database's DATA table at an interval controlled by the extract_period_millis column on the CHANNEL table. The last_extract_time is always recorded, by channel, on the NODE_CHANNEL_CTL table for the host node's id. When the Pull and Push Job run, if the extract period has not passed according to the last extract time, then the channel will be skipped for this run. If the extract_period_millis is set to zero, data extraction will happen every time the jobs run.

SymmetricDS also provides the ability to configure windows of time when synchronization is allowed. This is done using the NODE_GROUP_CHANNEL_WINDOW table. A list of allowed time windows can be specified for a node group and a channel. If one or more windows exist, then data will only be extracted and transported if the time of day falls within the window of time specified. The configured times are always for the target node's local time. If the start_time is greater than the end_time, then the window crosses over to the next day.

All data loading may be disabled by setting the dataloader.enable property to false. This has the effect of not allowing incoming synchronizations, while allowing outgoing synchronizations. All data extractions may be disabled by setting the dataextractor.enable property to false. These properties can be controlled by inserting into the root server's PARAMETER table. These properties affect every channel with the exception of the 'config' channel.

SymmetricDS examines the current configuration, corresponding database triggers, and the underlying tables to determine if database triggers need created or updated. The change activity is recorded on the TRIGGER_HIST table with a reason for the change. The following reasons for a change are possible:

A configuration entry in Trigger without any history in Trigger Hist results in a new trigger being created (N). The Trigger Hist stores a hash of the underlying table, so any alteration to the table causes the trigger to be rebuilt (S). When the last_update_time is changed on the Trigger entry, the configuration change causes the trigger to be rebuilt (C). If an entry in Trigger Hist is missing the corresponding database trigger, the trigger is created (T).

The process of examining triggers and rebuilding them is automatically run during startup and each night by the SyncTriggersJob. The user can also manually run the process at any time by invoking the syncTriggers() method over JMX. The SyncTriggersJob is enabled by default to run at 15 minutes past midnight. If SymmetricDS is being run from a collection of servers (multiple instances of the same Node running against the same database), then locking should be enable to prevent database contention. The following runtime properties control the behavior of the process.

With the proper configuration SymmetricDS can publish XML messages of captured data changes to JMS during routing or transactionally while data loading synchronized data into a target database. The following explains how to publish to JMS during synchronization to the target database.

The XmlPublisherDataLoaderFilter is a IDataLoaderFilter that may be configured to publish specific tables as an XML message to a JMS provider. See Chapter 6, Extending SymmetricDS for information on how to configure an extension point. If the publish to JMS fails, the batch will be marked in error, the loaded data for the batch will be rolled back and the batch will be retried during the next synchronization run.

The following is an example extension point configuration that will publish four tables in XML with a root tag of 'sale'. Each XML message will be grouped by the batch and the column names identified by the groupByColumnNames property which have the same values.

<?xml version="1.0" encoding="UTF-8"?>
<beans xmlns="http://www.springframework.org/schema/beans"
    xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
    xmlns:context="http://www.springframework.org/schema/context"
    xsi:schemaLocation="http://www.springframework.org/schema/beans 
           http://www.springframework.org/schema/beans/spring-beans-3.0.xsd
           http://www.springframework.org/schema/context
           http://www.springframework.org/schema/context/spring-context-3.0.xsd">
 
    <bean id="configuration-publishingFilter" 
      class="org.jumpmind.symmetric.integrate.XmlPublisherDataLoaderFilter">
        <property name="xmlTagNameToUseForGroup" value="sale"/>
        <property name="tableNamesToPublishAsGroup">
            <list>
               <value>SALE_TX</value>
               <value>SALE_LINE_ITEM</value>
               <value>SALE_TAX</value>
               <value>SALE_TOTAL</value>
            </list>            
        </property>
        <property name="groupByColumnNames">
            <list>
               <value>STORE_ID</value>
               <value>BUSINESS_DAY</value>
               <value>WORKSTATION_ID</value>
               <value>TRANSACTION_ID</value>
            </list>                      
        </property>
        <property name="publisher">
           <bean class="org.jumpmind.symmetric.integrate.SimpleJmsPublisher">
               <property name="jmsTemplate" ref="definedSpringJmsTemplate"/>
           </bean>
        </property>        
    </bean>     
</beans>

The publisher property on the XmlPublisherDataLoaderFilter takes an interface of type IPublisher. The implementation demonstrated here is an implementation that publishes to JMS using Spring's JMS template. Other implementations of IPublisher could easily publish the XML to other targets like an HTTP server, the file system or secure copy it to another server.

The above configuration will publish XML similiar to the following:

<?xml version="1.0" encoding="UTF-8"?>
<sale xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" 
  id="0012010-01-220031234" nodeid="00001" time="1264187704155">
  <row entity="SALE_TX" dml="I">
    <data key="STORE_ID">001</data>
    <data key="BUSINESS_DAY">2010-01-22</data>
    <data key="WORKSTATION_ID">003</data>
    <data key="TRANSACTION_ID">1234</data>
    <data key="CASHIER_ID">010110</data>
  </row>        
  <row entity="SALE_LINE_ITEM" dml="I">
    <data key="STORE_ID">001</data>
    <data key="BUSINESS_DAY">2010-01-22</data>
    <data key="WORKSTATION_ID">003</data>
    <data key="TRANSACTION_ID">1234</data>
    <data key="SKU">9999999</data>  
    <data key="PRICE">10.00</data>
    <data key="DESC" xsi:nil="true"/>
  </row>        
  <row entity="SALE_LINE_ITEM" dml="I">
    <data key="STORE_ID">001</data>
    <data key="BUSINESS_DAY">2010-01-22</data>
    <data key="WORKSTATION_ID">003</data>
    <data key="TRANSACTION_ID">1234</data>
    <data key="SKU">9999999</data>  
    <data key="PRICE">10.00</data>
    <data key="DESC" xsi:nil="true"/>
  </row>
  <row entity="SALE_TAX" dml="I">
    <data key="STORE_ID">001</data>
    <data key="BUSINESS_DAY">2010-01-22</data>
    <data key="WORKSTATION_ID">003</data>
    <data key="TRANSACTION_ID">1234</data>
    <data key="AMOUNT">1.33</data>  
  </row>
  <row entity="SALE_TOTAL" dml="I">
    <data key="STORE_ID">001</data>
    <data key="BUSINESS_DAY">2010-01-22</data>
    <data key="WORKSTATION_ID">003</data>
    <data key="TRANSACTION_ID">1234</data>
    <data key="AMOUNT">21.33</data>  
  </row>
</sale>
            

To publish JMS messages during routing the same pattern is valid, with the exception that the extension point would be the XmlPublisherDataRouter and the router would be configured by setting the router_type of a ROUTER to the Spring bean name of the registered extension point. Of course, the router would need to be linked through TRIGGER_ROUTERs to each TRIGGER table that needs published.