Skip to content

README.md

Latest commit

 

History

History
96 lines (65 loc) · 2.86 KB

README.md

File metadata and controls

96 lines (65 loc) · 2.86 KB
Raw

Feldera Nexmark benchmark in SQL

Feldera includes two versions of the Nexmark benchmark: one implemented in Rust and the other in SQL. This is the SQL version of the benchmark, which is less mature and less complete than the Rust version. To run the Rust version of the benchmark, use cargo bench --bench nexmark.

There are three big steps:

  1. Start a Kafka (or Redpanda) broker and bring up a Feldera instance.

  2. Generate Nexmark input events into the Kafka broker. This is how you control the number of input events in the benchmark.

  3. Run the benchmark itself using the generated events.

The actual benchmark in step 3 can run multiple times using the same data generated in step 2.

The following sections explain each step.

Step 1: Start Kafka and Feldera

This step has the following sub-steps:

  1. Start or obtain access to a Kafka (or Redpanda) broker. One way to do that is to start a Redpanda container on the local system:

     docker run --name redpanda -p 9092:9092 --rm -itd docker.redpanda.com/vectorized/redpanda:v23.2.3 redpanda start --smp 2

  2. Bring up a Feldera instance, e.g. by running (in a separate terminal):

    cargo run -p pipeline-manager --features pg-embed

Step 2: Generate data into Kafka

This step has the following sub-steps:

  1. Save host:port for accessing the broker into $KAFKA_BROKER, i.e. if you started Redpanda with the command suggested in the previous step:

    KAFKA_BROKER=localhost:9092

  2. Delete any existing events already generated into the Kafka broker, using the rpk utility from Redpanda. If this is the first time you're generating events into this broker, then you can skip this step (but it doesn't hurt to run it):

    rpk topic -X brokers=$KAFKA_BROKER delete bid auction person

  3. Generate events into the Kafka broker using a command like the following:

    cargo run  -p dbsp_nexmark --example generate --features with-kafka -- --max-events 10000000 -O bootstrap.servers=$KAFKA_BROKER

    The command above generates 10 million events. This is a moderate number that you should feel free to increase or decrease.

  4. Optionally, if you want to verify that the requested number of events was generated, you can use the following command using the rpk utility that comes with Redpanda. It should print exactly the number pased as --max-events earlier:

    for topic in auction bid person; do rpk -X brokers=$KAFKA_BROKER topic consume -f '%v' -o :end $topic; done | wc -l

Step 3: Run the benchmark

This step has the following sub-steps:

  1. Save the API URL for the instance into $FELDERA_API, e.g.:

    FELDERA_API=http://localhost:8080

  2. Run the benchmark itself:

    python3 benchmark/feldera-sql/run.py --api-url $FELDERA_API --kafka-broker $KAFKA_BROKER