The test is done on Tinybird Free trial.

Benchmarking a database often requires deep expertise and fine-tuning. Here, our goal is merely to test the default experience of a new user, i.e. someone who does not invest the time to optimize performance.

Testing is semi-automated.

The system has a timeout of 10s; after that, it recommends to optimize (rewrite) the query.

Load time and data size in the results are set to 0, as Tinybird did not indicate these resources.

Head to https://www.tinybird.co and create an account.

Tinybird supports data inserts from various sources. We are going to use S3 to load a Parquet file into Tinybird. Since Tinybird limits the file size to 1 GB, and the test data set is larger than that, we split it into smaller chunks using ClickHouse:

INSERT INTO FUNCTION s3('https://hitsparquet.s3.eu-west-3.amazonaws.com/data/hits_{_partition_id}.parquet', '', '', 'Parquet')
PARTITION BY rand() % 50
SELECT * FROM hits

Importing files with sizes a little bit less than 1 GB did not always work. We instead used 50 files of around 280 MB each. You will need to use the auto mode to make sure all the files are read.

Once the data is inserted you can create the endpoints needed to run the benchmark using pipes. run.sh will iterate through each endpoint.