kafka-delta-ingest VS rust-rdkafka

Rust can offer incredible cost savings when you can use it in place of spark to interact with your delta lake. One such project was kafka-delta-ingest. The developers were able to reduce the cost of running the pipeline by over 90%. However, most of this stuff is still very experimental and not ready for production but you will definitely be seeing more projects like this just based on how much money can be saved.

This independence from a catalog allows for path based reads and writes. This is handy when writing from Kafka directly to Delta Lake for the first layer of ingestion. You don’t need a catalog (or even Spark). https://github.com/delta-io/kafka-delta-ingest/tree/main/src

As far as I know no. You certainly could use events on a streaming ledger like Kafka or Redpanda and then store to delta with https://github.com/delta-io/kafka-delta-ingest and process them with all the gis goodness of spark. However, this is fairly complicated and much different from a simple postgis drop in replacement. There are specialized meaning faster and more efficient systems out there for specialized tasks such as geo fencing in real-time

kafka-delta-ingest is a good project to get streaming data into a Delta Lake. Here's a great talk on the topic.

What about https://github.com/delta-io/kafka-delta-ingest?

I use this library a lot and it's got some nice touches for how to handle wrapping a C library: https://github.com/fede1024/rust-rdkafka

For this example, we're using rdkafka to build producers and consumers, because it allows us to specify custom headers for each record.

kafka

You might find https://crates.io/crates/rdkafka helpful

I am playing with tokio and rust-rdkafka library, following the examples like this one: https://github.com/fede1024/rust-rdkafka/blob/6fb2c37/examples/asynchronous_processing.rs

The source for the current version of the library can be found on Github. I had to increase the major version because I needed to break the API in order to support all formats supported by the current Schema Registry version. I also added the possibility to set an API key, so it can be used with Confluent Cloud, the cloud offering from Confluent. As part of the latest major refactoring it's also supporting async. This might improve performance of your app, and is also the default for the major Kafka client, more information about why you would want to use async can be found in the async book. The schemas retrieved from the Schema Registry are cached. This way the schema is only retrieved once for each id, and reused for other messages with the same id.

What's wrong with rust-rdkafka?

This is a two-part series to help you get started with Rust and Kafka. We will be using the rust-rdkafka crate which itself is based on librdkafka (C library).