Skip to content

Latest commit

History

History
97 lines (86 loc) 路 4.61 KB

SOURCE.md

File metadata and controls

97 lines (86 loc) 路 4.61 KB

Kafka Connect Redis - Source

Subscribes to Redis channels/patterns (including keyspace notifications) and writes the received messages to Kafka.

WARNING Delivery of keyspace notifications is not reliable for Redis clusters. Keyspace notifications are node-local and adding new upstream nodes to your Redis cluster may involve a short period where events on the new node are not picked up until the connector discovers the node and issues a SUBSCRIBE command to it. This is a limitation of keyspace notifications that the Redis organization would like to overcome in the future.

Record Schema

Key

Avro

{
    "namespace": "io.github.jaredpetersen.kafkaconnectredis",
    "name": "RedisSubscriptionEventKey",
    "type": "record",
    "fields": [
        {
            "name": "channel",
            "type": "string"
        },
        {
            "name": "pattern",
            "type": [null, "string"]
        }
    ]
}

Connect JSON

{
    "name": "io.github.jaredpetersen.kafkaconnectredis.RedisSubscriptionEventKey",
    "type": "struct",
    "fields": [
        {
            "field": "channel",
            "type": "string",
            "optional": false
        },
        {
            "field": "pattern",
            "type": "string",
            "optional": true
        }
    ]
}

Value

Avro

{
    "namespace": "io.github.jaredpetersen.kafkaconnectredis",
    "name": "RedisSubscriptionEventValue",
    "type": "record",
    "fields": [
        {
            "name": "message",
            "type": "string"
        }
    ]
}

Connect JSON

{
    "name": "io.github.jaredpetersen.kafkaconnectredis.RedisSubscriptionEventValue",
    "type": "struct",
    "fields": [
        {
            "field": "message",
            "type": "string",
            "optional": false
        }
    ]
}

Partitions

Records are partitioned using the DefaultPartitioner class. This means that the record key is used to determine which partition the record is assigned to.

In the case of subscribing to Redis keyspace notifications, it may be useful to avoid partitioning the data so that multiple event types can arrive in order as a single event stream. This can be accomplished by configuring the connector to publish to a Kafka topic that only contains a single partition, forcing the DefaultPartitioner to only utilize the single partition.

The plugin can be configured to use an alternative partitioning strategy if desired. Set the configuration property connector.client.config.override.policy to value All on the Kafka Connect worker (the overall Kafka Connect application that runs plugins). This will allow the override of the internal Kafka producer and consumer configurations. To override the partitioner for an individual connector plugin, add the configuration property producer.override.partitioner.class to the connector plugin with a value that points to a class implementing the Partitioner interface, e.g. org.apache.kafka.clients.producer.internals.DefaultPartitioner.

Parallelization

Splitting the workload between multiple tasks is possible via the configuration property tasks.max. The connector splits the work based on the number of configured channels/patterns. If the max tasks configuration exceeds the number of channels/patterns, the number of channels/patterns will be used instead as the maximum.

Configuration

Connector Properties

Name Type Default Importance Description
topic string High Topic to write to.
redis.uri string High Redis connection information provided via a URI string.
redis.cluster.enabled boolean false High Target Redis is running as a cluster.
redis.channels string High Redis channels to subscribe to separated by commas.
redis.channels.patterns.enabled boolean High Redis channels use patterns (PSUBSCRIBE).