Altinity Datasets for ClickHouse

Welcome! altinity-datasets loads test datasets for ClickHouse. It is inspired by Python libraries that automatically load standard datasets for quick testing.

Getting Started

Altinity-datasets requires Python 3.5 or greater. The clickhouse-client executable must be in the path to load data.

Before starting you must install the altinity-datasets package using pip3. Following example shows install into a Python virtual environment. First command is only required if you don't have clickhouse-client already installed on the host.

sudo apt install clickhouse-client
sudo pip3 install altinity-datasets

Many users will prefer to install within a Python3 virtual environment, for example:

python3 -m venv my-env
. my-env/bin/activate
pip3 install altinity-datasets

You can also install a current version directly from Github:

pip3 install git+https://github.com/altinity/altinity-datasets.git

To remove altinity-datasets run the following command:

pip3 uninstall altinity-datasets

Using datasets

The ad-cli command manages datasets. You can see available commands by typing ad-cli -h/--help. All subcommands also accept -h/--help options.

Listing repos

Let's start by listing repos, which are locations that contain datasets.

ad-cli repo list

This will return a list of repos that have datasets. For the time being there is just a built-in repo that is part of the altinity-datasets package.

Finding datasets

Next let's see the available datasets.

ad-cli dataset search

This gives you a list of datasets with detailed descriptions. You can restrict the search to a single dataset by typing the name, for example ad-cli search wine. You can also search other repos using the repo file system location, e.g., ad-cli search wine --repo-path=$HOME/myrepo.

Loading datasets

Now, let's load a dataset. Here's a command to load the iris dataset to a ClickHouse server running on localhost.

ad-cli dataset load iris

Here is a more complex example. It loads the iris dataset to the iris_new database on a remote server. Also, we parallize the upload with 10 threads.

ad-cli load iris --database=iris_new --host=my.remote.host.com --parallel=10

The command shown above is typical of the invocation when loading on a server that has a large number of cores and fast storage.

Note that it's common to reload datasets expecially during development. You can do this using ad-cli load --clean. IMPORTANT: This drops the database to get rid of dataset tables. If you have other tables in the same database they will be dropped as well.

Dumping datasets

You can make a dataset from any existing table or tables in ClickHouse that reside in a single database. Here's a simple example that shows how to dump the weather dataset to create a new dataset. (The weather dataset is a built-in that loads by default to the weather database.)

ad-cli dataset dump weather

There are additional options to control dataset dumps. For example, we can rename the dateset, restrict the dump to tables that start with 'central', compress data, and overwrite any existing data in the output directory.

ad-cli dataset dump new_weather -d weather --tables='^central' --compress \
  --overwrite

Extra Connection Options

The dataset load and dump commands by default connect to ClickHouse running on localhost with default user and empty password. The following example options connect using encrypted communications to a specific server with explicit user name and password. The last option suppresses certificate verification.

ad-cli dataset load iris -H 127.0.0.1 -P 9440 \
-u special -p secret --secure --no-verify 

Note: To use --no-verify you must also ensure that clickhouse-client is configured to accept invalid certificates. Validate by logging in using clickhouse-client with the --secure option. Check and correct settings in /etc/clickhouse-client/config.xml if you have problems.

Repo and Dataset Format

Repos are directories on the file system. The exact location of the repo is known as the repo path. Data sets under the repo are child directories that in turn have subdirectories for DDL commands and data. The following listing shows part of the organization of the built-ins repo.

built-ins/
  iris/
    data/
      iris/
        iris.csv
    ddl/
      iris.sql
    manifest.yaml
  wine/
    data/
      wine/
        wine.csv
    ddl/
      wine.sql
    manifest.yaml

To create your own dataset you can dump existing tables using ad-cli dataset dump or copy the examples in built-ins. The format is is simple.

• The manifest.yaml file describes the dataset. If you put in extra fields they will be ignored.
• The DDL directory contains SQL scripts to run. By convention these should be named for the objects (i.e., tables) that they create.
• The data directory contains CSV data. There is a separate subdirectory for each table to be loaded. Its name must match the table name exactly.
• CSV files can be uncompressed .csv or gzipped .csv.gz. No other formats are supported and the file types must be correctly specified.

You can place new repos in any location you please. To load from your own repo run a load command and use the --repo-path option to point to the repo location. Here's an example:

ad-cli dataset load mydataset --repo-path=$HOME/my-repo

Development

To work on altinity-datasets clone from Github and install.

git clone https://github.com/altinity/altinity-datasets.git
cd altinity-datasets
python3 setup.py develop 

After making changes you should run tests.

cd tests
python3 -m unittest --verbose

The following commands build an installable and push to pypi.org. PyPI account credentials must be set in TWINE_USERNAME and TWINE_PASSWORD.

python3 setup.py sdist
twine upload --repository-url https://upload.pypi.org/legacy/ dist/*

Code conventions are enforced using yapf and flake8. Run the dev-format-code.sh script to check formatting.

Run tests as follows with virtual environment set. You will need a ClickHouse server with a null password on the default user.

cd tests
python3 -m unittest -v

Errors

Out-of-date pip3 causes installation failure

If pip3 installs with the message error: invalid command 'bdist_wheel' you may need to upgrade pip. Run pip3 install --upgrade pip to correct the problem.

Materialized views cannot be dumped

ad-cli will fail with an error if you try to dump a database that has materialized views. The workaround is to omit them from the dump operation using a table regex as shown in the following example:

ad-cli dataset dump nyc_taxi_rides --repo-path=.  --compress --parallel=6 \
--tables='^(tripdata|taxi_zones|central_park_weather_observations)$'

--no-verify option fails on self-signed certs

When using ad-cli --secure together with --no-verify options you need to also configure clickhouse-client to skip certificate verification. This only applies when the certificate is self-signed. You must change /etc/clickhouse-client/config.xml as follows to skip certificate validation:

<config>
    <openSSL>
        <client> <!-- Used for connection to server's secure tcp port -->
            ...
            <invalidCertificateHandler>
                <name>AcceptCertificateHandler</name>
            </invalidCertificateHandler>
        </client>
    </openSSL>
    ...
</config>

Limitations

The most important are:

• Error handling is spotty. If clickhouse-client is not in the path things may fail mysteriously.
• Datasets have to be on the local file system. In the future we will use cloud object storage such as S3.

Please file issues at https://github.com/Altinity/altinity-datasets/issues. Pull requests to fix problems are welcome.