🚶🚶🚶 rust-ingress

This is a POC for ingress solution using Apache Kafka with two services for ingressing and storing messages into Postgres. The code is a happy-path spaghetti to test whether this is even going to work.

Running

Prerequisites: Docker

docker-compose up

Testing

For a simple test:

curl --location --request POST 'http://0.0.0.0:8000/api/v1/message' \
    --header 'Content-Type: application/json' \
    --data-raw '{
        "ts": "1530228282",
        "sender": "testy-test-service",
        "message": {
            "foo": "bar",
            "baz": "bang"
        },
        "sent-from-ip": "1.2.3.4",
        "priority": 2
    }'

Prerequisites: Node.js

# install load testing tooling
npm install -s artillery@latest

# start blasting
artillery run load-test.yaml

Here's resourse usage for a run sending 20k requests over 60 seconds on WSL2(probably the most inefficient):

CONTAINER ID   NAME                          CPU %     MEM USAGE / LIMIT     MEM %     NET I/O           BLOCK I/O   PIDS
08e0e50ea4b9   rust-axum-kafka-ingress-1     0.12%     101.8MiB / 12.44GiB   0.80%     21MB / 16.6MB     0B / 0B     17
d3af0d3e5b91   rust-axum-kafka-worker-1      1.68%     15.89MiB / 12.44GiB   0.12%     12.8MB / 11.1MB   0B / 0B     17
1448a81750a8   rust-axum-kafka-kafka-1       1.61%     424.3MiB / 12.44GiB   3.33%     10.4MB / 8.59MB   0B / 0B     86
3c6f10adab04   rust-axum-kafka-db-1          0.88%     24.11MiB / 12.44GiB   0.19%     7.44MB / 6.32MB   0B / 0B     8
46f32ebaa1be   rust-axum-kafka-zookeeper-1   0.07%     94.45MiB / 12.44GiB   0.74%     81.9kB / 65.2kB   0B / 0B     70

Why Apache Kafka?

High performance(proven 100k RPS) and fault tolerance. Note: Fault tolerance is not really explored in this project as it is not even using a Kafka cluster.

Why Rust?

I stumbled upon fede1024/rust-rdkafka which promised a pretty good performance.

The ingres service is using a relatively new tokio-rs/axum framework, which is speculated to have better performance than Go and offers slightly better ergonomics over the previous HTTP Server frameworks.

Why Dockerfile is so complicated?

It contains a build optimization from Packaging a Rust web service using Docker.

TODO:

• the worker needs to batch inserts as it is getting pretty slow
• run Kafka as a cluster
• the ingress is okay CPU-wise, but suspiciously high memory usage(needs investigation)
• add processing result output in the worker
• run load tests on a better hardware on Linux
• organize code better, add tests