Using Snowflake Connector for Kafka with Snowpipe Streaming¶

Required properties¶

Add or edit your connection settings in the Kafka connector properties file. For more information, see Configuring the Kafka connector.

snowflake.ingestion.method

Required only if using the Kafka connector as the streaming ingest client. Specifies whether to use Snowpipe Streaming or standard Snowpipe to load your Kafka topic data. The supported values are as follows:

• SNOWPIPE_STREAMING

• SNOWPIPE (default)

No additional settings are required to choose the backend service to queue and load topic data. Configure additional properties in your Kafka connector properties file as usual.

snowflake.role.name

Access control role to use when inserting the rows into the table.

Client optimization properties¶

enable.streaming.client.optimization

Specifies whether to enable one-client optimization. This property is supported by Kafka connector release version 2.1.2 and later. It is enabled by default.

With one-client optimization, only one client is created for multiple topic partitions per Kafka connector. This feature can reduce client runtime and lower migration cost by creating larger files.

Values:

• true

• false

Default:

true

Note that in a high throughput scenario (for example, 50 MB/s per connector), enabling this property can result in a higher latency or cost. We recommend that you disable this property for high-throughput scenarios.

Buffer and polling properties¶

buffer.flush.time

Number of seconds between buffer flushes; each flush results in insert operations for the buffered records. The Kafka connector calls the Snowpipe Streaming API once after each flush.

The minimum value supported for the buffer.flush.time property is 1 (in seconds). For higher average data flow rates, we suggest that you decrease the default value for improved latency. If cost is a greater concern than latency, you could increase the buffer flush time. Be careful to flush the Kafka memory buffer before it becomes full to avoid out-of-memory exceptions.

Values:

• Minimum: 1

• Maximum: No upper limit

Default:

10

Note that Snowpipe Streaming automatically flushes data every one second, which is different from the buffer flush time for the Kafka connector. After the Kafka buffer flush time is reached, data will be sent with one second latency to Snowflake through Snowpipe Streaming. For more information, see Snowpipe Streaming latency.

buffer.count.records

Number of records buffered in memory per Kafka partition before ingesting to Snowflake.

Values:

• Minimum: 1

• Maximum: No upper limit

Default:

10000

buffer.size.bytes

Cumulative size in bytes of records buffered in memory per the Kafka partition before they are ingested in Snowflake as data files.

The records are compressed when they are written to data files. As a result, the size of the records in the buffer may be larger than the size of the data files created from the records.

Values:

• Minimum: 1

• Maximum: No upper limit

Default:

20000000 (20 MB)

snowflake.streaming.max.client.lag

Specifies how often Snowflake Ingest Java (https://github.com/snowflakedb/snowflake-ingest-java) flushes the data to Snowflake, in seconds.

A low value keeps the latency low, but it might result in a worse query performance especially when snowflake.streaming.enable.single.buffer is enabled. For more information, see the recommended latency configurations for Snowpipe Streaming.

Values:

• Minimum: 1 second

• Maximum: 600 seconds

Default:

120 seconds for version 3.0.0 or later, 1 second otherwise

snowflake.streaming.enable.single.buffer

Specifies whether to enable single buffer for Snowpipe Streaming and to skip buffering data in the connector’s internal buffer.

This property is supported by the Kafka connector version 2.3.1 and later.

Streaming connector uses internal buffer alongside with the one provided by Snowflake Ingest Java (https://github.com/snowflakedb/snowflake-ingest-java). Setting this property to true makes Kafka connector skip the internal buffer in order to achieve lower latency.

Note that setting this property to true makes buffer.flush.time and buffer.count.records irrelevant.

Values:

• true

• false

Default:

true for version 3.0.0 and later, false otherwise

In addition to the Kafka connector properties, note the Kafka consumer max.poll.records property, which controls the maximum number of records returned by Kafka to Kafka Connect in a single poll. The default value of 500 can be increased, but be mindful of memory constraints. For more information about this property, see the documentation for your Kafka package:

• Apache Kafka (https://kafka.apache.org/documentation/#consumerconfigs_max.poll.records)

• Confluent (https://docs.confluent.io/platform/current/installation/configuration/consumer-configs.html#consumerconfigs_max.poll.records)

Error handling and DLQ properties¶

errors.tolerance

Specifies how to handle errors encountered by the Kafka connector:

This property supports the following values:

Values:

• NONE: Stop loading data when the first error is encountered.

• ALL: Ignore all errors and continue to load data.

Default:

NONE

errors.log.enable

Specifies whether to write error messages to the Kafka Connect log file.

This property supports the following values:

Values:

• TRUE: Write error messages.

• FALSE: Do not write error messages.

Default:

FALSE

errors.deadletterqueue.topic.name

Specifies the name of the DLQ (dead-letter queue) topic in Kafka for delivering messages to Kafka that could not be ingested into Snowflake tables. For more information, see Dead-letter Queues (in this topic).

Values:

Custom text string

Default:

None