Skip to content

candleindark/datalad-registry

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

codecov

Datalad-Registry is now live at registry.datalad.org.


Preface

We fully test Datalad-Registry with Podman. Nevertheless, you should be able to launch a Datalad-Registry instance using Docker with little to no deviation from this guide.

Testing and Development Setup

Prerequisites

  • The following dependencies are needed in a system in order to test and develop Datalad-Registry:

  • Datalad-Registry's package version is determined by versioningit using Git tags in the repo the package is built from. Please ensure your clone of the original Datalad-Registry repo includes all the tags for a successful build. (Direct clones of the original likely have all those tags. For clones of a fork of the original, you may need to add the original as a remote and fetch all the tags from it.)

To run tests

  1. Setup

    1. On Debian systems, install the necessary dependencies for Python PostgreSQL libs:

      sudo apt-get install libpq-dev python3-dev

    2. Install Datalad-Registry for testing in the Python virtual environment for this project (as mentioned in the Prerequisites section):

      pip install -e .[test]

    3. Launch the needed components of DataLad-Registry for testing from a subshell with needed environment variables loaded from env.test. (Note: using a subshell avoids polluting the current shell with the environment variables from env.test):

      (set -a && . ./env.test && set +a && podman-compose -f docker-compose.test.yml up -d)

  2. Test execution

    1. Launch the tests from a subshell with the needed environment variables loaded from env.test:

      (set -a && . ./env.test && set +a && python -m pytest -s -v)

  3. Teardown

    When the testing is done, you can bring down the components of Datalad-Registry launched.

    1. Bring down the components of Datalad-Registry launched from a subshell with the needed environment variables loaded from env.test:

      (set -a && . ./env.test && set +a && podman-compose -f docker-compose.test.yml down)

To develop

  1. Setup

    1. On Debian systems, install the necessary dependencies for Python PostgreSQL libs:

      sudo apt-get install libpq-dev python3-dev

    2. Install Datalad-Registry for development in the Python virtual environment for this project (as mentioned in the Prerequisites section):

      pip install -e .[dev]

    3. Set values for needed environment variables by creating a .env.dev file

      template.env is a template for creating the .env.dev file. It lists all the needed environment variables with defaults. We will use it to create the .env.dev file.

      1. Create the .env.dev file by copying the template.env file to .env.dev:

        cp template.env .env.dev

      2. Modify the .env.dev file according to your needs by adjusting the values for usernames, passwords, etc.

      note: we git ignore all .env files.

    4. Launch the needed components of DataLad-Registry for development from a subshell with needed environment variables loaded from .env.dev. (Note: using a subshell avoids polluting the current shell with the environment variables from .env.dev):

      (set -a && . ./.env.dev && set +a && podman-compose -f docker-compose.yml -f docker-compose.dev.override.yml up -d --build)

  2. Development

    At this point, the proper development environment is set up. However, please note the following characteristics of this development environment:

    1. The current directory at the host is bind-mounted to the /app directory within the web service container.
    2. The subdirectory ./instance at the host is bind-mounted to the /app/instance directory within the web service container to serve as the instance folder for the Flask application run by the web service container.
    3. The web service container runs the Flask application in debug mode and reacts to changes in the current directory, the codebase, at the host machine.
    4. All other component services of Datalad-Registry, as defined in docker-compose.yml, do not react to changes in the codebase at the host machine. (Note: This behavior is the result of a design choice. The worker service, a Celery worker, for example, should not react to changes in the codebase at the host machine for the tasks it executes may not always be idempotent.)
      1. To realize the changes in the codebase at the host machine in other component services, you needed to bring down all the components of Datalad-Registry as specified in the following Teardown section and relaunch them according to the above Setup section.
  3. Teardown

    When done with developing, you can bring down the components of Datalad-Registry launched.

    1. Bring down the components of Datalad-Registry launched from a subshell with the needed environment variables loaded from .env.dev:

      (set -a && . ./.env.dev && set +a && podman-compose -f docker-compose.yml -f docker-compose.dev.override.yml down)

Read-Only Mode

Datalad-Registry can operate in a read-only mode. In this mode, Datalad-Registry consists of two services, a web service that accepts only read-only requests and a read-only database service, as define in docker-compose.read-only.yml. The read-only database service is a replica of the database service in an instance of Datalad-Registry that allows both reads and writes, operating in PRODUCTION or DEVELOPMENT mode.

To set up Datalad-Registry to run in read-only mode involves the following steps:

  1. Configure the database service of an existing instance of Datalad-Registry that allows both reads and writes, operating in PRODUCTION or DEVELOPMENT mode, to be the primary database service that the database service in the read-only instance of Datalad-Registry will replicate from.

    1. Create a role in the primary database service for replication by executing the following SQL command via psql or any other PostgreSQL client.
      CREATE ROLE <replica_user> WITH REPLICATION LOGIN ENCRYPTED PASSWORD '<password>';
      where <replica_user> is the name of the role and <password> is the password for the role to access the primary database service.
    2. Modify the postgresql.conf configuration file of the primary database service.
      1. Enable the wal_level configuration parameter and set its value to replica.
      2. Enable the wal_log_hints configuration parameter and set its value to on.
      3. Enable the wal_keep_size configuration parameter and set its value to 1024.
    3. Modify the pg_hba.conf configuration file of the primary database service.
      1. Add the following line to the end of the file.
        host replication   <replica_user>  <replica_source_ip>/32   md5
        
        where <replica_user> is the name of the role created two steps before, and <replica_source_ip> is the IP address of the replica database service in the read-only instance of Datalad-Registry. (Note: <replica_source_ip>/32 as a whole specifies a range of IP addresses that a replica database service can connect from.)
    Note:

    The location of the postgresql.conf and pg_hba.conf configuration files depends on the individual setup of PostgreSQL. All PostgreSQL setups defined in all the Docker Compose files in this project store the postgresql.conf and pg_hba.conf configuration files in /var/lib/postgresql/data. The application of any changes in the postgresql.conf file requires a restart of the PostgreSQL service. The application of any changes in the pg_hba.conf file can be accomplished by either restarting the PostgreSQL service or executing the following SQL command via psql or any other PostgreSQL client.

    SELECT pg_reload_conf();
  2. Set up the database service of the read-only instance of Datalad-Registry to be a read-only replica of the primary database service.

    1. Take a base backup of the primary database service at a node that is to be served as the read-only replica.
      1. Start this node with an empty data directory.
        1. Uncomment the line command: ["tail", "-f", "/dev/null"] in docker-compose.read-only.yml. (This allows starting of the node without starting the PostgreSQL service and populating the data directory.)
        2. Start the node by executing the following command.
          (set -a && . ./.env.read-only && set +a && podman-compose -f docker-compose.read-only.yml up read-only-db -d)
          where .env.read-only is a file containing the needed environment variables. You can use the template.env.read-only file as a template to construct this file.
      2. Run the base backup command inside the node.
        1. Get into the BASH shell of the node by running the following command.
          podman exec -it <name_of_the_node> /bin/bash
          where <name_of_the_node> is the name of the container that is the node.
        2. Run the following command pg_basebackup.
          pg_basebackup -h <primary_ip> -p <port_number> -U <replica_user> -X stream -C -S replica_1 -v -R -W -D /var/lib/postgresql/data
          where <primary_ip> is the IP address of the primary database service, <port_number> is the port number of the primary database service, and <replica_user> is the name of the role created in the primary database in step 1.
        3. Once the backup is complete, exit the bash shell of the node by running the following command.
          exit
      3. Stop the node by running the following command.
        (set -a && . ./.env.read-only && set +a && podman-compose -f docker-compose.read-only.yml down)
      4. Restore the docker-compose.read-only.yml file to its original state by commenting out the line command: ["tail", "-f", "/dev/null"].

After going through the above steps, the initial setup for running Datalad-Registry in read-only mode is complete. To start the read-only instance of Datalad-Registry, just run the following command.

(set -a && . ./.env.read-only && set +a && podman-compose -f docker-compose.read-only.yml up -d --build)

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Python 94.5%
  • HTML 3.6%
  • Other 1.9%