Framework Dependencies

Galaxy is a large Python application with a long list of Python module dependencies. As a result, the Galaxy developers have made significant effort to provide these dependencies in as simple a method as possible while remaining compatible with the Python packaging best practices. Thus, Galaxy’s runtime setup procedure makes use of virtualenv for package environment isolation, pip for installation, and wheel to provide pre-built versions of dependencies.

In addition to framework dependencies, as of Galaxy 18.01, the client (UI) is no longer provided in its built format in the development (dev) branch of the source repository. The built client is still provided in release_YY.MM branches and the master branch.

How it works

Upon startup (with run.sh), the startup scripts will:

  1. Create a Python virtualenv in the directory .venv.
  2. Unset the $PYTHONPATH environment variable (if set) as this can interfere with installing dependencies.
  3. Download and install packages from the Galaxy project wheel server, wheels.galaxyproject.org, as well as the Python Package Index (aka PyPI) , using pip.
  4. Start Galaxy using .venv/bin/python.

Options

A variety of options to run.sh are available to control the above behavior:

  • --skip-venv: Do not create or use a virtualenv.
  • --skip-wheels: Do not install wheels.
  • --no-create-venv: Do not create a virtualenv, but use one if it exists at .venv or if $VIRTUAL_ENV is set (this variable is set by virtualenv’s activate).
  • --replace-pip/--no-replace-pip: Do/do not upgrade pip if necessary.

Managing dependencies manually

Create a virtualenv

Using a virtualenv in .venv under the Galaxy source tree is not required. More complicated Galaxy setups may choose to use a virtualenv external to the Galaxy source tree, which can be done either by not using run.sh directly (an example of this can be found under the Scaling and Load Balancing documentation) or using the --no-create-venv option, explained in the Options section. It is also possible to force Galaxy to start without a virtualenv at all, but you should not do this unless you know what you’re doing.

To manually create a virtualenv, you will first need to obtain virtualenv. There are a variety of ways to do this:

  • pip install virtualenv
  • brew install virtualenv
  • Install your Linux distribution’s virtualenv package from the system package manager (e.g. apt-get install python-virtualenv).
  • Download the virtualenv source from PyPI, untar, and run the virtualenv.py script contained within as python virtualenv.py /path/to/galaxy/virtualenv

Once this is done, create a virtualenv. In our example, the virtualenv will live in /srv/galaxy/venv and the Galaxy source code has been cloned to /srv/galaxy/server.

$ virtualenv /srv/galaxy/venv
New python executable in /srv/galaxy/venv/bin/python
Installing setuptools, pip, wheel...done.
$ . /srv/galaxy/venv/bin/activate
(venv)$

Next, in galaxy.yml, set the virtualenv option in the uwsgi section to point to your new virtualenv:

---
uwsgi:
    #...
    virtualenv: /srv/galaxy/venv

Install dependencies

Normally, run.sh calls common_startup.sh, which creates the virtualenv and installs dependencies. You can call this script yourself to set up Galaxy pip and the dependencies without creating a virtualenv using the --no-create-venv option:

(venv)$ PYTHONPATH= sh /srv/galaxy/server/scripts/common_startup.sh --no-create-venv
Requirement already satisfied: pip>=8.1 in /home/nate/.virtualenvs/test/lib/python2.7/site-packages
Collecting numpy==1.9.2 (from -r requirements.txt (line 4))
  Downloading https://wheels.galaxyproject.org/packages/numpy-1.9.2-cp27-cp27mu-manylinux1_x86_64.whl (10.2MB)
    100% |████████████████████████████████| 10.2MB 21.7MB/s
Collecting bx-python==0.7.3 (from -r requirements.txt (line 5))
  Downloading https://wheels.galaxyproject.org/packages/bx_python-0.7.3-cp27-cp27mu-manylinux1_x86_64.whl (2.1MB)
    100% |████████████████████████████████| 2.2MB 97.2MB/s

    ...

Installing collected packages: numpy, bx-python, ...
Successfully installed numpy-1.9.2 bx-python-0.7.3 ...

Warning

If your $PYTHONPATH is set, it may interfere with the dependency installation process. Without --no-create-venv the $PYTHONPATH variable will be automatically unset, but we assume you know what you’re doing and may want it left intact if you are using --no-create-venv. If you encounter problems, try unsetting $PYTHONPATH as shown in the example above.

Dependency management complications

Certain deployment scenarios or other software may complicate Galaxy dependency management. If you use any of these, relevant information can be found in the corresponding subsection below.

Galaxy job handlers

All Galaxy jobs run a metadata detection step on the job outputs upon completion of the tool. The metadata detection step requires many of Galaxy’s dependencies. Because of this, it’s necessary to make sure the metadata detection step runs in Galaxy’s virtualenv. If you run a relatively simple Galaxy deployment (e.g. run.sh) then this is assured for you automatically. In more complicated setups (running under supervisor and/or the virtualenv used to start Galaxy is not on a shared filesystem) it may be necessary to make sure the handlers know where the virtualenv (or a virtualenv containing Galaxy’s dependencies) can be found.

If the virtualenv cannot be located, you will see job failures due to Python ImportError exceptions, like so:

Traceback (most recent call last):
  File "/srv/galaxy/tmp/job_working_directory/001/set_metadata_RK41sy.py", line 1, in <module>
        from galaxy_ext.metadata.set_metadata import set_metadata; set_metadata()
  File "/srv/galaxy/server/lib/galaxy_ext/metadata/set_metadata.py", line 23, in <module>
        from sqlalchemy.orm import clear_mappers
ImportError: No module named sqlalchemy.orm

If this is the case, you can instruct jobs to activate the virtualenv with an env tag in job_conf.xml:

<destination id="cluster" runner="drmaa">
    <!-- ... other destination params -->
    <env file="/cluster/galaxy/venv/bin/activate" />
</destination>

If your Galaxy server’s virtualenv isn’t available on the cluster you can create one manually using the instructions under Managing dependencies manually.

Pulsar

If using Pulsar’s option to set metadata on the remote server, the same conditions as with Galaxy job handlers apply. You should create a virtualenv on the remote resource, install Galaxy’s dependencies in to it, and set an <env> tag pointing to the virtualenv’s activate as in the Galaxy job handlers section. Instructions on how to create a virtualenv can be found under the Managing dependencies manually section.

Conda

Conda and virtualenv are incompatible. However, Conda provides its own environment separation functionality in the form of Conda environments. Starting Galaxy with Conda Python will cause Galaxy to create a _galaxy_YY.MM environment into which Galaxy framework dependencies will be installed.

To instruct Galaxy to use a different environment name, set the $GALAXY_CONDA_ENV environment variable.

If you use the recommended setup of sourcing $CONDA_ROOT/etc/profile.d/conda.sh as of Conda 4.4, Conda Python is not automatically added to your $PATH, so Galaxy will not use Conda for its framework dependencies unless you start it with an environment activated that has Conda Python installed. This is because Galaxy defaults to creating a virtualenv unless the python on $PATH is Conda Python.

Tip

If you would like to force Galaxy to use Conda with Conda 4.4 or later, the simplest method is:

  1. Ensure .venv does not exist.
  2. Activate the base environment with conda activate base

Galaxy will not install in to the base environment.

Caution

Versions of Galaxy prior to 18.05 could install in to the base/root Conda environment. Consult the correct version of the documentation for your version of Galaxy.

Installing dependencies with conda (instead of pip)

Because Conda package names typically match PyPI package names, you can install Conda versions of what dependencies are available from conda-forge and Bioconda using a script provided with Galaxy, but this is not necessary. If you do not run the script, the dependencies will simply be installed via pip.

$ conda config --add channels conda-forge
$ conda config --add channels bioconda
$ conda create --name galaxy --file <(lib/galaxy/dependencies/conda-file.sh)
Filtering out requirements not available in conda... done
Solving environment: done

## Package Plan ##

  environment location: /srv/galaxy/conda/envs/galaxy

  added / updated specs:
    - anyjson==0.3.3
    #...
    - whoosh==2.7.4


The following NEW packages will be INSTALLED:

    anyjson:            0.3.3-py27_1            conda-forge
    #...
    zlib:               1.2.8-3                 conda-forge

Proceed ([y]/n)?

Preparing transaction: done
Verifying transaction: done
Executing transaction: done
#
# To activate this environment, use
#
#     $ conda activate galaxy
#
# To deactivate an active environment, use
#
#     $ conda deactivate

Next, activate the environment and run pip to fetch the remaining dependencies:

$ conda activate galaxy
$ pip install --index-url https://wheels.galaxyproject.org/simple/ --extra-index-url https://pypi.python.org/simple/ -r requirements.txt
Requirement already satisfied: pip>=8.1 in /srv/galaxy/conda/envs/galaxy/lib/python2.7/site-packages
#...
Collecting SQLAlchemy==1.0.15 (from -r requirements.txt (line 8))
  Downloading https://wheels.galaxyproject.org/packages/SQLAlchemy-1.0.15-cp27-cp27mu-manylinux1_x86_64.whl (1.0MB)
    100% |████████████████████████████████| 1.0MB 48.6MB/s
#...
Installing collected packages: SQLAlchemy, ...
Successfully installed SQLAlchemy-1.0.15 ...
$ uwsgi --yaml config/galaxy.yml
[uWSGI] getting YAML configuration from config/galaxy.yml
[uwsgi-static] added mapping for /static/style => static/style/blue
[uwsgi-static] added mapping for /static => static
*** Starting uWSGI 2.0.15 (64bit) on [Thu Jan 25 11:57:17 2018] ***

You may encounter the following traceback when starting Galaxy:

Traceback (most recent call last):
  File "lib/galaxy/main.py", line 278, in <module>
    main()
  File "lib/galaxy/main.py", line 274, in main
    app_loop(args, log)
  File "lib/galaxy/main.py", line 124, in app_loop
    log=log,
  File "lib/galaxy/main.py", line 91, in load_galaxy_app
    from galaxy.app import UniverseApplication
  File "/srv/galaxy/server/lib/galaxy/app.py", line 30, in <module>
    from galaxy.visualization.data_providers.registry import DataProviderRegistry
  File "/srv/galaxy/server/lib/galaxy/visualization/data_providers/registry.py", line 15, in <module>
    from galaxy.visualization.data_providers import genome
  File "/srv/galaxy/server/lib/galaxy/visualization/data_providers/genome.py", line 17, in <module>
    from bx.bbi.bigbed_file import BigBedFile
ImportError: cannot import name BigBedFile

If this is the case, uninstall bx-python from Conda and reinstall it from the Galaxy wheel:

$ conda remove bx-python
Solving environment: done

## Package Plan ##

  environment location: /srv/galaxy/conda/envs/galaxy

  removed specs:
    - bx-python


The following packages will be REMOVED:

    bx-python: 0.7.3-py27_0 bioconda

Proceed ([y]/n)?

Preparing transaction: done
Verifying transaction: done
Executing transaction: done
$ pip install --index-url https://wheels.galaxyproject.org/simple bx-python
Collecting bx-python
  Downloading https://wheels.galaxyproject.org/packages/bx_python-0.7.3-cp27-cp27mu-manylinux1_x86_64.whl (2.1MB)
    100% |████████████████████████████████| 2.2MB 66.1MB/s
Installing collected packages: bx-python
Successfully installed bx-python-0.7.3

uWSGI

run.sh should automatically set --virtualenv on uWSGI’s command line. However, you can override this using the virtualenv option in the uwsgi section of galaxy.yml as described in the Managing dependencies manually section.

Unpinned dependencies

Danger

Unpinned dependencies may be useful for development but should not be used in production. Please do not install unpinned dependencies unless you know what you’re doing. While the Galaxy Committers will do their best to keep dependencies updated, they cannot provide support for problems arising from unpinned dependencies.

Galaxy’s dependencies can be installed either “pinned” (they will be installed at exact versions specified for your Galaxy release) or “unpinned” (the latest versions of all dependencies will be installed unless there are known incompatibilities with new versions). By default, the release branches of Galaxy use pinned versions for three reasons:

  1. Using pinned versions insures that the prebuilt wheels on wheels.galaxyproject.org will be installed, and no compilation will be necesseary.
  2. Galaxy releases are tested with the pinned versions and this allows us to give as much assurance as possible that the pinned versions will work with the given Galaxy release (especially as time progresses and newer dependency versions are released while the Galaxy release receives fewer updates.
  3. Pinning furthers Galaxy’s goal of reproducibility as differing dependency versions could result in non-reproducible behavior.

If you would like to install unpinned versions of Galaxy’s dependencies, install dependencies using the unpinned requirements file, and then instruct Galaxy to start without attempting to fetch wheels:

(venv)$ pip install -r lib/galaxy/dependencies/requirements.txt
(venv)$ deactivate
$ sh run.sh --no-create-venv --skip-wheels

You may be able to save yourself some compiling by adding the argument --index-url https://wheels.galaxyproject.org/simple/ to pip install, but it is possible to install all of Galaxy’s dependencies directly from PyPI.

Adding additional Galaxy dependencies

New packages can be added to Galaxy, or the versions of existing packages can be updated, using pipenv and Galaxy Starforge, Galaxy’s Docker-based build system.

Note

Dependency pinning management is being migrated to pipenv. As of this release, pinning for packages used for Galaxy development are managed by pipenv, but pinning for regular runtime packages are still managed with manual changes to pinned-requirements.txt. See Pull Request #4891 for details.

The process is still under development and will be streamlined and automated over time. For the time being, please use the following process to add new packages and have their wheels built:

  1. Install Starforge (e.g. with pip install starforge or python setup.py install from the source). You will also need to have Docker installed on your system.
  2. Obtain wheels.yml (this file will most likely be moved in to Galaxy in the future) and add/modify the wheel definition.
  3. Use starforge wheel --wheels-config=wheels.yml <wheel-name> to build the wheel. If the wheel includes C extensions, you will probably want to also use the --no-qemu flag to prevent Starforge from attempting to build on Mac OS X using QEMU/KVM.
  4. If the wheel build is successful, submit a pull request to Starforge with your changes to wheels.yml.
  5. A Galaxy Committers group member will need to trigger an automated build of the wheel changes in your pull request. Galaxy’s Jenkins service will build these changes using Starforge.
  6. If the pull request is merged, submit a pull request to Galaxy modifying the files in lib/galaxy/dependencies as appropriate.

You may attempt to skip directly to step 4 and let the Starforge wheel PR builder build your wheels for you. This is especially useful if you are simply updating an existing wheel’s version. However, if you are adding a new C extension wheel that is not simple to build, you may need to go through many iterations of updating the PR and having a `Galaxy Committers group`_ member triggering builds before wheels are successfully built. You can avoid this cycle by performing steps 1-3 locally.