https://github.com/rhpvorderman/cromwell-mysql-starter

Scripts and information on starting a per user cromwell mysql database
https://github.com/rhpvorderman/cromwell-mysql-starter

Last synced: 3 months ago
JSON representation

Scripts and information on starting a per user cromwell mysql database

• Host: GitHub
• URL: https://github.com/rhpvorderman/cromwell-mysql-starter
• Owner: rhpvorderman
• License: mit
• Created: 2022-02-22T15:44:24.000Z (over 3 years ago)
• Default Branch: main
• Last Pushed: 2022-03-02T10:41:26.000Z (about 3 years ago)
• Last Synced: 2025-01-17T08:43:58.996Z (4 months ago)
• Size: 3.91 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.rst
  • License: LICENSE

Awesome Lists containing this project

README

        
Creating a per-user MySQL database for Cromwell on HPC

=======================================================

Use case

--------

On a compute cluster that uses a resource scheduler such as SLURM, multiple

users can login to a login node and submit jobs to the cluster. The SLURM

resource scheduler does some job accouting based on the user's ID and the

user's department

A Cromwell server does not work well in this use case, because it runs under

a single user. This means the job accounting does not work properly.

Cromwell's `run` command is used in these cases, but without a backing database

there is no possibility for call-caching, which is vital for production

workflows.

There are several options to alleviate this problem:

1. Use a MySQL database and share a common database between all users.

    pros:

        + Highly efficient database calls

    cons:

        + Shared database can get locked by any of the users.

        + Security risk: confidential information under the GDPR such as

          patient IDs can end up in the database, where everyone can read it.

2. Use Cromwell's in-memory database with a persistence file. Documented `here `_:

    pros:

        + Allows per-project databases to be set up easily.

        + Everyone can share the same configuration as the database can be

          set up in the cromwell executions directory.

        + Permissions handled by filesystem permissions. No security risks.

    cons:

        + More memory usage required by cromwell.

        + The database consumes enormous amounts of disk space. (Multiple

          gigabytes).

        + Cromwell's interaction with the database is slow and on very large

          projects timeouts may occur when reading in the database persistence

          file.

3. Create a separate Cromwell server for every user that asks for it.

    pros:

        + Solves the problem with user permissions, filesystem permissions and

          per-user job accounting.

    cons:

        + Enormous maintenance burden.

Currently we use option number 2 on our cluster. Unfortunately it does not

scale towards thousands of jobs.

Option 4: Allow creation of a per user MySQL database using command line scripts

--------------------------------------------------------------------------------

MySQL has a server-based architecture. Setting up a server is usuallly not

possible without root permissions, however there are some features that enable

setting up MySQL for non-root users.

+ Mysql is available trough the conda-forge channel and can be installed with

  miniconda.

+ Disabling of networking altogether

+ Communicating via file socket

This way a per user database can be set up by the users themselves without

intervention.

step 1: Installing mysql

++++++++++++++++++++++++++++++++++++++++++++++++++++

Set up miniconda with the conda-forge channel. Then:

    conda create -y -n mysql mysql

    conda activate mysql

step 2: Setting up a default configuration

++++++++++++++++++++++++++++++++++++++++++++++++++++

`mysqld` config values can be overwritten on the command line or can be set

in a personal config file: `~/.my.cnf`.

Set the following values::

    [mysqld]

    datadir=/path/to/mysqldb

    skip-networking=TRUE

    socket=/path/to/socket

step 3: Set up the cromwell configuration

++++++++++++++++++++++++++++++++++++++++++

.. code-block::

    database {

      profile = "slick.jdbc.MySQLProfile$"

      db {

        driver = "com.mysql.cj.jdbc.Driver"

        url = "jdbc:mysql:file:///path/to/socket/cromwell?rewriteBatchedStatements=true"

        user = "user"

        password = "pass"

        connectionTimeout = 5000

      }

    }

step 4: Start

+++++++++++++++++++++++++++++++++++++

+ On first start: ``mysqld --initialize``

+ Start the database ``mysqld -D``