SharePlex 8.6.6 - Upgrade Guide

Configure Replication to a Kafka target

Overview

The SharePlex Post process can connect and write to a Kafka broker. The data is written as XML records that include the data definitions, the operation type, and the changed column values. This data is written as a sequential series of operations as they occurred on the source, which can then be posted in sequential order to a target database or consumed by an external process or program.

Guidelines for posting to Kafka

• A SharePlex Post process acts as a Kafka producer. A SharePlex Post process writes to a single topic with the default partition of 0 in a single broker. To support multiple topics or multiple brokers, configure multiple Post processes using named Post queues. For more information, see Configure named post queues.
• The SharePlex Post process does not create topics. You must create the topic through Kafka.
• For performance reasons, Post batches messages together before sending them to the Kafka broker. Tuning parameters are available to control the batch size. These can be used to balance performance with latency.

Note: For the platforms, datatypes and operations that are supported when using SharePlex to replicate to Kafka, see the SharePlex Release Notes.

Configure SharePlex on the source

When replicating data to Kafka, configure the source database and SharePlex on the source system as follows.

Enable supplemental logging

On the source system, enable PK/UK supplemental logging in the Oracle source database. SharePlex must have the Oracle key information to build an appropriate key on the target.

SQL> ALTER DATABASE ADD SUPPLEMENTAL LOG DATA (PRIMARY KEY, UNIQUE) COLUMNS;

Set SP_OCT_USE_SUPP_KEYS parameter

On the source system, set the SP_OCT_USE_SUPP_KEYS parameter to a value of 1. This parameter directs SharePlex to use the columns set by Oracle's supplemental logging as the key columns when a row is updated or deleted. When both supplemental logging and this parameter are set, it ensures that SharePlex can always build a key and that the SharePlex key will match the Oracle key.

See the SharePlex Reference Guide for more information about this parameter.

Configure replication

On the source, create a SharePlex configuration file that specifies capture and routing information. The structure that is required in a configuration file varies, depending on your replication strategy, but this shows you the required syntax for routing data to Kafka.

where:

• SID is the Oracle SID of the source Oracle database.
• src_owner.table is the owner and name of the source table.
• !kafka is a required keyword indicating SharePlex is posting to Kafka.
• :tgt_owner.table is optional and specifies the owner and name of the target table. Use if either component is different from that of the source table. Allow no spaces between !kafka and :tgt_owner.table. Type case-sensitive names in the correct case and enclose them within double quotes, as in "MySchema"."MyTable"

• host is the name of the target system.

Note:  For more information, see Create configuration files.

Source configuration example

Configure SharePlex on the target

These instructions configure the SharePlex Post process to connect to Kafka. You must have a running Kafka broker.

To configure post to Kafka

1. Create a Kafka topic.
2. Start sp_cop. (Do not activate the configuration yet.)
3. Run sp_ctrl.

4. Issue the target command to configure posting to a Kafka broker and topic. The following are example commands.

   sp_ctrl> target x.kafkaset kafka broker=localhost:9092

   sp_ctrl> target x.kafka set kafka topic=shareplex

   See View and change Kafka settings for command explanations and options.

View and change Kafka settings

To view current property settings for output to Kafka, use the following command:

target x.kafka show

To change a property setting, use the following command.

target x.kafka [queue queuename] set kafka property=value

where:

• queue queuename is the name of a Post queue. Use this option if there are multiple Post processes.
• property and value are shown in the following table.

Table 12: Kafka target properties

* To avoid latency, if Post detects no more incoming messages, it sends the packet to Kafka immediately without waiting for the threshold to be satisfied.

Set recovery options

If the Kafka process aborts suddenly, or if the machine that it is running on aborts, row changes may be written twice to the Kafka topic. The consumer must manage this by detecting and discarding duplicates.

Every record of every row-change operation in a transaction has the same transaction ID and is also marked with a sequence ID. These attributes are id and msgIdx, respectively, under the txn element in the XML output (see About the XML).

The transaction ID is the SCN at the time the transaction was committed, and the sequence ID is the index of the row change in the transaction. These two values are guaranteed to be the same if they are re-written to the Kafka topic in a recovery situation.

If desired, you can configure Post to include additional metadata with every row-change record by using the following command:

target x.kafka [queue queuename] set metadata property[, property]

Table 13: Optional metadata properties

Example

target x.kafka set metadata time, userid, trans, size

To reset the metadata

target x.kafka [queue queuename] reset metadata

To view the metadata

target x.kafka [queue queuename] show metadata

About the XML

The XML format is separated into operation and schema "types" for easier consumption. They are actually the same when viewed from an XSD perspective and are not distinct types. The template XML represents all possible attributes and elements. The individual XML represents the bare minimum output for each supported operation.

After startup, the first time that Post writes a change record for any given table, it first writes a schema record for that table. Each schema record contains the table name and details of interest for each columns. A schema record is written only once for each table during a Post run, unless there is a change to that schema, and then a new schema record is written. If Post stops and starts, schema records are written again, once for each table as Post receives a change record for it.

Schema record template

<?xml version="1.0" encoding="UTF-8" ?>  
<?opentarget version="1.0" ?>  
<opentarget>  
    <txn  
        id="xs:integer"  
        oracleTxnId="xs:string"   
        commitTime="xs:dateTimeStamp" />  
    <tbl  
        name="xs:string"  
        utcOffset="xs:integer"
        <cmd ops="schema">  
            <schema>  
                <col  
                    name="xs:string"  
                    xmlType="xs:string"  
                    key="xs:boolean"  
                    nullable="xs:boolean"  
                    length="xs:integer"  
                  />  
            </schema>  
        </cmd>  
    </tbl>  
</opentarget>

Table 14: Explanation of schema template (* = optional)

Operation record template

<?xml version="1.0" encoding="UTF-8" ?>
<?opentarget version="1.1" ?>
<opentarget>
    <txn
        id="xs:integer" 
        msgIdx="xs:integer" 
        msgTot="xs:integer" 
        oracleTxnId="xs:string"
        commitTime="xs:dateTimeStamp"
        userId="xs:string" />

    <tbl 
        name="xs:string"
        <cmd ops="xs:string">
            <row id="xs:string">
                <col name="xs:string"></col>
                <lkup>
                    <col name="xs:string"></col>
                </lkup>
            </row>
        </cmd>
    </tbl>
</opentarget>

Table 15: Explanation of operation template (* = optional)

Note: The id and msgIdx attributes together uniquely identify an operation.

Supported datatypes

See the SharePlex Release Notes for a chart that shows how Oracle datatypes are converted to XML.

Sample XML records

Source table

This is the table for which the sample operations are generated.

SQL> desc products

Source DML operations

insert into products values (230117, ‘Hamsberry vintage tee, cherry’, 4099);
commit;
update products set price=3599 where product_id=230117 and price=4099;
commit;
delete products where product_id=230117;
commit;
truncate table products;

Schema record

<?xml version="1.0" encoding="UTF-8"?>
<?opentarget version="1.1"?>
<opentarget>
<txn id="2218316945" commitTime="2014-10-10T13:18:43" userId="85" oracleTxnId="3.10.1339425" />
<tbl name="MFG.PRODUCTS" utcOffset="-5:00">
<cmd ops="schema">
<schema>
<col name="PRODUCT_ID" xmlType="decimal" key="true" nullable="false" length="22" />
<col name="DESCRIPTION" xmlType="string" key="false" nullable="true" length="600" />
<col name="PRICE" xmlType="decimal" key="false" nullable="true" length="22" />
</schema>
</cmd>
</tbl>
</opentarget>

Insert record

<?xml version="1.0" encoding="UTF-8"?>
<?opentarget version="1.1"?>
<opentarget>
<txn id="2218316945" msgIdx="1" msgTot="1" commitTime="2014-10-10T13:18:43" userId="85" oracleTxnId="3.10.1339425" />
<tbl name="MFG.PRODUCTS">
<cmd ops="ins">
<row id="AAAmDbAAEAAApRrAAA">
<col name="PRODUCT_ID">230117</col>
<col name="DESCRIPTION">Hamsberry vintage tee, cherry</col>
<col name="PRICE">4099</col>
</row>
</cmd>
</tbl>
</opentarget>

Update record

<?xml version="1.0" encoding="UTF-8"?>
<?opentarget version="1.1"?>
<opentarget>
<txn id="2218318728" msgIdx="1" msgTot="1" commitTime="2014-10-10T13:19:12" userId="85" oracleTxnId="1.17.970754" />
<tbl name="MFG.PRODUCTS">
<cmd ops="upd">
<row id="AAAmDbAAEAAApRrAAA">
<col name="PRICE">3599</col>
<lkup>
<col name="PRODUCT_ID">230117</col>
<col name="PRICE">4099</col>
</lkup>
</row>
</cmd>
</tbl>
</opentarget>

Delete record

<?xml version="1.0" encoding="UTF-8"?>
<?opentarget version="1.1"?>
<opentarget>
<txn id="2218319446" msgIdx="1" msgTot="1" commitTime="2014-10-10T13:19:25" userId="85" oracleTxnId="5.23.1391276" />
<tbl name="MFG.PRODUCTS">
<cmd ops="del">
<row id="AAAmDbAAEAAApRrAAA">
<lkup>
<col name="PRODUCT_ID">230117</col>
</lkup>
</row>
</cmd>
</tbl>
</opentarget>

Truncate record

<?xml version="1.0" encoding="UTF-8"?>
<?opentarget version="1.1"?>
<opentarget>
<txn id="2218319938" commitTime="1988-01-01T00:00:00" userId="85" oracleTxnId="11.4.939801" />
<tbl name="MFG.PRODUCTS">
<cmd ops="trunc" />
</tbl>
</opentarget>