Kafka provides an integration point through Source and Sink Connectors. The GeodeKafkaSource allows Geode to be a data source for Kafka The GeodeKafkaSink allows Geode to consume data off of topics and store data from Kafka.
- Kafka is installed and is up and running. See the Kafka quickstart for more info: Kafka Quickstart
- A Geode Cluster with at least one locator and one server and regions to source from and sink to.
- Topics created in Kafka to source from and sink to.
Installation of the connector is similar in process to other Kafka Connectors. For now, we will follow the guide for Manual Installation.
In summary, we will use the standalone worker for this example.
- Build the jar into a known (and Kafka accessible) location (in the geode-kafka-connector project run './gradlew shadowJar' and the artifacts will be in build/lib/
- Modify the connect-standalone.properties and point to where the connector is installed.
plugin.path=(Path to your clone)/geode-kafka-connector/build/libs/
#depending on object type, you might want to modify the converter (for manually testing we can use the JSON or string converter)
#key.converter=
#value.converter=
- Create and modify connect-geode-sink.properties file, for example
name=geode-kafka-sink
connector.class=GeodeKafkaSink
tasks.max=1
topic-to-regions=[someTopicToSinkFrom:someRegionToConsume]
topics=someTopicToSinkFrom
locators=localHost[10334]
- Create and modify connect-geode-source.properties files
name=geode-kafka-source
connector.class=GeodeKafkaSource
tasks.max=1
region-to-topics=[someRegionToSourceFrom:someTopicToConsume]
locators=localHost[10334]
- Run bin/connect-standalone.sh config/connect-standalone.properties config/connect-geode-source.properties config/connect-geode-sink.properties
| Property | Required | Description | Default |
|---|---|---|---|
| locators | no, but... | A comma separated string of locators that configure which locators to connect to | localhost[10334] |
| topic-to-regions | yes | A comma separated list of "one topic to many regions" bindings. Each binding is surrounded by brackets. For example "[topicName:regionName], [anotherTopic: regionName, anotherRegion]" | [gkctopic:gkcregion] |
| security-client-auth-init | no | Point to class that implements the AuthInitialize Interface | |
| null-values-mean-remove | no | If set to true, when topics send a SinkRecord with a null value, we will convert to an operation similar to region.remove instead of putting a null value into the region | true |
- The topic-to-regions property allows us to create mappings between topics and regions. A single one-to-one mapping would look similar to "[topic:region]" A one-to-many mapping can be made by comma separating the regions, for example "[topic:region1,region2]" This is equivalent to both regions being consumers of the topic.
| Property | Required | Description | Default |
|---|---|---|---|
| locators | no, but... | A comma separated string of locators that configure which locators to connect to | localhost[10334] |
| region-to-topics | yes | A comma separated list of "one region to many topics" mappings. Each mapping is surrounded by brackets. For example "[regionName:topicName], "[anotherRegion: topicName, anotherTopic]" | [gkcregion:gkctopic] |
| security-client-auth-init | no | Point to class that implements the AuthInitialize Interface | |
| security-username | no | Supply a username to be used to authenticate with Geode. Will automatically set the security-client-auth-init to use a SystemPropertyAuthInit if one isn't supplied by the user | null |
| security-password | no | Supply a password to be used to authenticate with Geode | null |
| geode-connector-batch-size | no | Maximum number of records to return on each poll | 100 |
| geode-connector-queue-size | no | Maximum number of entries in the connector queue before backing up all Geode cq listeners sharing the task queue | 10000 |
| load-entire-region | no | Determines if we should queue up all entries that currently exist in a region. This allows us to copy existing region data. Will be replayed whenever a task needs to re-register a cq | true |
| durable-client-id-prefix | no | Prefix string for tasks to append to when registering as a durable client. If empty string, will not register as a durable client | "" |
| durable-client-timeout | no | How long in milliseconds to persist values in Geode's durable queue before the queue is invalidated | 60000 |
| cq-prefix | no | Prefix string to identify Connector cq's on a Geode server | cqForGeodeKafka |
- The region-to-topics property allows us to create mappings between regions and topics. A single one-to-one mapping would look similar to "[region:topic]" A one-to-many mapping can be made by comma separating the topics, for example "[region:topic1,topic2]" This is equivalent to the region be a producer for both topics
- Consider modifying Kafka Properties like tasks.max in the source and sink parameters.
Extra Details
- Each source task has information and will push off the shared queue to Kafka, however only one task will register a cq with Apache Geode
- Each sink task is able to update any of the configured Apache Geode region.
- Formatters - Possibly a JSON to and from PDX formatter
- Security - security settings for Geode
- Dynamic Region creation - Dynamically create regions when topics are created (filter what names to look for and what types of regions to create)
- Allow a single worker to connect to multiple Geode Clusters?