Get up and running quickly with Open edX services.
If you are seeking info on the Vagrant-based devstack, please see https://openedx.atlassian.net/wiki/display/OpenOPS/Running+Devstack. This project is meant to replace the traditional Vagrant-based devstack with a multi-container approach driven by Docker Compose. It is still in the beta testing phase.
These docs might be out of date. Please see the updated docs at https://edx.readthedocs.io/projects/edx-installing-configuring-and-running/en/latest/installation/index.html.
Tickets or issues should be filed in Jira under the platform project: https://openedx.atlassian.net/projects/PLAT/issues
You should run all make
commands described below on your local machine, not from within a VM (virtualenvs are ok, and in fact recommended) as these commands are for standing up a new docker based VM.
This project requires Docker 17.06+ CE. We recommend Docker Stable, but Docker Edge should work as well.
NOTE: Switching between Docker Stable and Docker Edge will remove all images and settings. Don't forget to restore your memory setting and be prepared to provision.
For macOS users, please use Docker for Mac. Previous Mac-based tools (e.g. boot2docker) are not supported.
Docker for Windows may work but has not been tested and is not supported.
Linux users should not be using the overlay
storage driver. overlay2
is tested and supported, but requires kernel version 4.0+. Check which storage driver your docker-daemon is configured to use:
You will also need the following installed:
- make
- python pip (optional for MacOS)
New images for our services are published frequently. Assuming that you've followed the steps in Getting Started below, run the following sequence of commands if you want to use the most up-to-date versions of the devstack images.
This will stop any running devstack containers, pull the latest images, and then start all of the devstack containers.
All of the services can be run by following the steps below. For analyticstack, follow Getting Started on Analytics.
NOTE: Since a Docker-based devstack runs many containers, you should configure Docker with a sufficient amount of resources. We find that configuring Docker for Mac with a minimum of 2 CPUs and 6GB of memory works well.
Install the requirements inside of a Python virtualenv.
The Docker Compose file mounts a host volume for each service's executing code. The host directory defaults to be a sibling of this directory. For example, if this repo is cloned to
~/workspace/devstack
, host volumes will be expected in~/workspace/course-discovery
,~/workspace/ecommerce
, etc. These repos can be cloned with the command below.You may customize where the local repositories are found by setting the DEVSTACK_WORKSPACE environment variable.
Be sure to share the cloned directories in the Docker -> Preferences... -> File Sharing box.
Pull any changes made to the various images on which the devstack depends.
Run the provision command, if you haven't already, to configure the various services with superusers (for development without the auth service) and tenants (for multi-tenancy).
NOTE: When running the provision command, databases for ecommerce and edxapp will be dropped and recreated.
The username and password for the superusers are both
edx
. You can access the services directly via Django admin at the/admin/
path, or login via single sign-on at/login/
.Default:
Provision using docker-sync:
Start the services. This command will mount the repositories under the DEVSTACK_WORKSPACE directory.
NOTE: it may take up to 60 seconds for the LMS to start, even after the
make dev.up
command outputsdone
.Default:
Start using docker-sync:
After the services have started, if you need shell access to one of the services, run make <service>-shell
. For example to access the Catalog/Course Discovery Service, you can run:
To see logs from containers running in detached mode, you can either use "Kitematic" (available from the "Docker for Mac" menu), or by running the following:
To view the logs of a specific service container run make <service>-logs
. For example, to access the logs for Ecommerce, you can run:
To reset your environment and start provisioning from scratch, you can run:
For information on all the available make
commands, you can run:
The provisioning script creates a Django superuser for every service.
Email: edx@example.com
Username: edx
Password: edx
The LMS also includes demo accounts. The passwords for each of these accounts is edx
.
Username | |
---|---|
audit | audit@example.com |
honor | honor@example.com |
staff | staff@example.com |
verified | verified@example.com |
Analyticstack can be run by following the steps below.
NOTE: Since a Docker-based devstack runs many containers, you should configure Docker with a sufficient amount of resources. We find that configuring Docker for Mac with a minimum of 2 CPUs and 6GB of memory works well for analyticstack. If you intend on running other docker services besides analyticstack ( e.g. lms, studio etc ) consider setting higher memory.
- Follow steps 1 and 2 from Getting Started section.
Before running the provision command, make sure to pull the relevant docker images from dockerhub by running the following commands:
Run the provision command to configure the analyticstack.
Start the analytics service. This command will mount the repositories under the DEVSTACK_WORKSPACE directory.
NOTE: it may take up to 60 seconds for Hadoop services to start.
To access the analytics pipeline shell, run the following command. All analytics pipeline job/workflows should be executed after accessing the shell.
To see logs from containers running in detached mode, you can either use "Kitematic" (available from the "Docker for Mac" menu), or by running the following command:
To view the logs of a specific service container run
make <service>-logs
. For example, to access the logs for Hadoop's namenode, you can run:To reset your environment and start provisioning from scratch, you can run:
NOTE: Be warned! This will remove all the containers and volumes initiated by this repository and all the data ( in these docker containers ) will be lost.
For information on all the available
make
commands, you can run:
- For running acceptance tests on docker analyticstack, follow the instructions in the Running analytics acceptance tests in docker guide.
- For troubleshooting docker analyticstack, follow the instructions in the Troubleshooting docker analyticstack guide.
Each service is accessible at localhost
on a specific port. The table below provides links to the homepage of each service. Since some services are not meant to be user-facing, the "homepage" may be the API root.
Service | URL |
---|---|
Credentials | http://localhost:18150/api/v2/ |
Catalog/Discovery | http://localhost:18381/api-docs/ |
E-Commerce/Otto | http://localhost:18130/dashboard/ |
LMS | http://localhost:18000/ |
Notes/edx-notes-api | http://localhost:18120/api/v1/ |
Studio/CMS | http://localhost:18010/ |
Sometimes you may need to restart a particular application server. To do so, simply use the docker-compose restart
command:
<service>
should be replaced with one of the following:
- credentials
- discovery
- ecommerce
- lms
- edx_notes_api
- studio
If you'd like to add some convenience make targets, you can add them to a local.mk
file, ignored by git.
The ecommerce image comes pre-configured for payments via CyberSource and PayPal. Additionally, the provisioning scripts add the demo course (course-v1:edX+DemoX+Demo_Course
) to the ecommerce catalog. You can initiate a checkout by visiting http://localhost:18130/basket/add/?sku=8CF08E5 or clicking one of the various upgrade links in the LMS. The following details can be used for checkout. While the name and address fields are required for credit card payments, their values are not checked in development, so put whatever you want in those fields.
- Card Type: Visa
- Card Number: 4111111111111111
- CVN: 123 (or any three digits)
- Expiry Date: 06/2025 (or any date in the future)
PayPal (same for username and password): devstack@edx.org
Docker Compose files useful for integrating with the edx.org marketing site are available. This will NOT be useful to those outside of edX. For details on getting things up and running, see https://openedx.atlassian.net/wiki/display/OpenDev/Marketing+Site.
If you want to modify an installed package – for instance edx-enterprise
or completion
– clone the repository in ~/workspace/src/your-package
. Next, ssh into the appropriate docker container (make lms-shell
), run pip install -e /edx/src/your-package
, and restart the service.
There are Docker CI Jenkins jobs on tools-edx-jenkins that build and push new Docker images to DockerHub on code changes to either the configuration repository or the IDA's codebase. These images are tagged according to the branch from which they were built (see NOTES below). If you want to build the images on your own, the Dockerfiles are available in the edx/configuration
repo.
NOTES:
- edxapp and IDAs use the
latest
tag for configuration changes which have been merged to master branch of their repository andedx/configuration
. - Images for a named Open edX release are built from the corresponding branch of each repository and tagged appropriately, for example
hawthorn.master
orhawthorn.rc1
. - The elasticsearch used in devstack is built using elasticsearch-devstack/Dockerfile and the
devstack
tag.
BUILD COMMANDS:
The build commands above will use your local configuration, but will pull application code from the master branch of the application's repository. If you would like to use code from another branch/tag/hash, modify the *_VERSION
variable that lives in the ansible_overrides.yml
file beside the Dockerfile
. Note that edx-platform is an exception; the variable to modify is edx_platform_version
and not EDXAPP_VERSION
.
For example, if you wanted to build tag release-2017-03-03
for the E-Commerce Service, you would modify ECOMMERCE_VERSION
in docker/build/ecommerce/ansible_overrides.yml
.
- Set the
OPENEDX_RELEASE
environment variable to the appropriate image tag; "hawthorn.master", "zebrawood.rc1", etc. Note that unlike a server install,OPENEDX_RELEASE
should not have the "open-release/" prefix. - Use
make dev.checkout
to check out the correct branch in the local checkout of each service repository once you've set theOPENEDX_RELEASE
environment variable above. make pull
to get the correct images.
All make
target and docker-compose
calls should now use the correct images until you change or unset OPENEDX_RELEASE
again. To work on the master branches and latest
images, unset OPENEDX_RELEASE
or set it to an empty string.
We use database dumps to speed up provisioning and generally spend less time running migrations. These dumps should be updated occasionally - when database migrations take a prolonged amount of time or we want to incorporate changes that require manual intervention.
To update the database dumps:
- Destroy and/or backup the data for your existing devstack so that you start with a clean slate.
- Disable the loading of the existing database dumps during provisioning by commenting out any calls to
load-db.sh
in the provisioning scripts. This disabling ensures a start with a completely fresh database and incorporates any changes that may have required some form of manual intervention for existing installations (e.g. drop/move tables). - Provision devstack with
make provision
. - Dump the databases and open a pull request with your updates:
You can run Django migrations as normal to apply any changes recently made to the database schema for a particular service. For example, to run migrations for LMS, enter a shell via make lms-shell
and then run:
Alternatively, you can discard and rebuild the entire database for all devstack services by re-running make dev.provision
or make dev.sync.provision
as appropriate for your configuration. Note that if your branch has fallen significantly behind master, it may not include all of the migrations included in the database dump used by provisioning. In these cases, it's usually best to first rebase the branch onto master to get the missing migrations.
To access a MySQL or Mongo shell, run the following commands, respectively:
Log into the LMS shell, source the edxapp
virtualenv, and run the makemigrations
command with the devstack_docker
settings:
Also, make sure you are aware of the Django Migration Don'ts as the edx-platform is deployed using the red-black method.
JavaScript packages for Node.js are installed into the node_modules
directory of the local git repository checkout which is synced into the corresponding Docker container. Hence these can be upgraded via any of the usual methods for that service (npm install
, paver install_node_prereqs
, etc.), and the changes will persist between container restarts.
Unlike the node_modules
directory, the virtualenv
used to run Python code in a Docker container only exists inside that container. Changes made to a container's filesystem are not saved when the container exits, so if you manually install or upgrade Python packages in a container (via pip install
, paver install_python_prereqs
, etc.), they will no longer be present if you restart the container. (Devstack Docker containers lose changes made to the filesystem when you reboot your computer, run make down
, restart or upgrade Docker itself, etc.) If you want to ensure that your new or upgraded packages are present in the container every time it starts, you have a few options:
- Merge your updated requirements files and wait for a new edxops Docker image for that service to be built and uploaded to Docker Hub. You can then download and use the updated image (for example, via
make pull
). The discovery and edxapp images are buit automatically via a Jenkins job. All other images are currently built as needed by edX employees, but will soon be built automatically on a regular basis. See How do I build images? for more information. - You can update your requirements files as appropriate and then build your own updated image for the service as described above, tagging it such that
docker-compose
will use it instead of the last image you downloaded. (Alternatively, you can temporarily editdocker-compose.yml
to replace theimage
entry for that service with the ID of your new image.) You should be sure to modify the variable override for the version of the application code used for building the image. See How do I build images?. for more information. - You can temporarily modify the main service command in
docker-compose.yml
to first install your new package(s) each time the container is started. For example, the part of the studio command which reads...&& while true; do...
could be changed to...&& pip install my-new-package && while true; do...
. - In order to work on locally pip-installed repos like edx-ora2, first clone them into
../src
(relative to this directory). Then, inside your lms shell, you canpip install -e /edx/src/edx-ora2
. If you want to keep this code installed across stop/starts, modifydocker-compose.yml
as mentioned above.
Optimized static assets are built for all the Open edX services during provisioning, but you may want to rebuild them for a particular service after changing some files without re-provisioning the entire devstack. To do this, run the make target for the appropriate service. For example:
To rebuild static assets for all service containers:
You can usually switch branches on a service's repository without adverse effects on a running container for it. The service in each container is using runserver and should automatically reload when any changes are made to the code on disk. However, note the points made above regarding database migrations and package updates.
When switching to a branch which differs greatly from the one you've been working on (especially if the new branch is more recent), you may wish to halt the existing containers via make down
, pull the latest Docker images via make pull
, and then re-run make dev.provision
or make dev.sync.provision
in order to recreate up-to-date databases, static assets, etc.
If making a patch to a named release, you should pull and use Docker images which were tagged for that release.
The LMS and CMS read many configuration settings from the container filesystem in the following locations:
/edx/app/edxapp/lms.env.json
/edx/app/edxapp/lms.auth.json
/edx/app/edxapp/cms.env.json
/edx/app/edxapp/cms.auth.json
Changes to these files will not persist over a container restart, as they are part of the layered container filesystem and not a mounted volume. However, you may need to change these settings and then have the LMS or CMS pick up the changes.
To restart the LMS/CMS process without restarting the container, kill the LMS or CMS process and the watcher process will restart the process within the container. You can kill the needed processes from a shell within the LMS/CMS container with a single line of bash script:
LMS:
CMS:
From your host machine, you can also run make lms-restart
or make studio-restart
which run those commands in the containers for you.
See the Pycharm Integration documentation.
LMS and Studio use a devpi container to cache PyPI dependencies, which speeds up several Devstack operations. See the devpi documentation.
It's possible to debug any of the containers' Python services using PDB. To do so, start up the containers as usual with:
This command starts each relevant container with the equivalent of the '--it' option, allowing a developer to attach to the process once the process is up and running.
To attach to the LMS/Studio containers and their process, use either:
Set a PDB breakpoint anywhere in the code using:
and your attached session will offer an interactive PDB prompt when the breakpoint is hit.
To detach from the container, you'll need to stop the container with:
or a manual Docker command to bring down the container:
After entering a shell for the appropriate service via make lms-shell
or make studio-shell
, you can run any of the usual paver commands from the edx-platform testing documentation. Examples:
Tests can also be run individually. Example:
If you want to see the browser being automated for JavaScript or bok-choy tests, you can connect to the container running it via VNC.
Browser | VNC connection |
---|---|
Firefox (Default) | vnc://0.0.0.0:25900 |
Chrome (via Selenium) | vnc://0.0.0.0:15900 |
On macOS, enter the VNC connection string in the address bar in Safari to connect via VNC. The VNC passwords for both browsers are randomly generated and logged at container startup, and can be found by running make vnc-passwords
.
Most tests are run in Firefox by default. To use Chrome for tests that normally use Firefox instead, prefix the test command with SELENIUM_BROWSER=chrome SELENIUM_HOST=edx.devstack.chrome
.
To run the base set of end-to-end tests for edx-platform, run the following make target:
If you want to use some of the other testing options described in the edx-e2e-tests README, you can instead start a shell for the e2e container and run the tests manually via paver:
The browser running the tests can be seen and interacted with via VNC as described above (Firefox is used by default).
If you are having trouble with your containers, this sections contains some troubleshooting tips.
If a container stops unexpectedly, you can look at its logs for clues:
docker-compose logs lms
Make sure you have the latest code and Docker images.
Pull the latest Docker images by running the following command from the devstack directory:
Pull the latest Docker Compose configuration and provisioning scripts by running the following command from the devstack directory:
Lastly, the images are built from the master branches of the application repositories (e.g. edx-platform, ecommerce, etc.). Make sure you are using the latest code from the master branches, or have rebased your branches on master.
Sometimes containers end up in strange states and need to be rebuilt. Run make down
to remove all containers and networks. This will NOT remove your data volumes.
Sometimes you just aren't sure what's wrong, if you would like to hit the reset button run make dev.reset
.
Running this command will perform the following steps:
- Bring down all containers
- Reset all git repositories to the HEAD of master
- Pull new images for all services
- Compile static assets for all services
- Run migrations for all services
It's good to run this before asking for help.
If you want to completely start over, run make destroy
. This will remove all containers, networks, AND data volumes.
In case you botched a migration or just want to start with a clean database.
Open up the mysql shell and drop the database for the desired service:
make mysql-shell mysql DROP DATABASE (insert database here)
From your devstack directory, run the provision script for the service. The provision script should handle populating data such as Oauth clients and Open edX users and running migrations:
./provision-(service_name)
If you notice that the ownership of some (maybe all) files have changed and you need to enter your root password when editing a file, you might have pulled changes to the remote repository from within a container. While running git pull
, git changes the owner of the files that you pull to the user that runs that command. Within a container, that is the root user - so git operations should be ran outside of the container.
To fix this situation, change the owner back to yourself outside of the container by running:
Most of the paver
commands require a settings flag. If omitted, the flag defaults to devstack
, which is the settings flag for vagrant-based devstack instances. So if you run into issues running paver
commands in a docker container, you should append the devstack_docker
flag. For example:
While running make static
within the ecommerce container you could get an error saying:
To fix this, remove the directory manually outside of the container and run the command again.
If you see the error no space left on device
on a Mac, Docker has run out of space in its Docker.qcow2 file.
Here is an example error while running make pull
:
...
32d52c166025: Extracting [==================================================>] 1.598 GB/1.598 GB
ERROR: failed to register layer: Error processing tar file(exit status 1): write /edx/app/edxapp/edx-platform/.git/objects/pack/pack-4ff9873be2ca8ab77d4b0b302249676a37b3cd4b.pack: no space left on device
make: *** [pull] Error 1
Try this first to clean up dangling images:
If you are still seeing issues, you can try cleaning up dangling volumes.
Warning: In most cases this will only remove volumes you no longer need, but this is not a guarantee.
While provisioning, some have seen the following error:
This issue can be worked around, but there's no guaranteed method to do so. Rebooting and restarting Docker does not seem to correct the issue. It may be an issue that is exacerbated by our use of sync (which typically speeds up the provisioning process on Mac), so you can try the following:
Once you get past the issue, you should be able to continue to use sync versions of the make targets.
While provisioning, some have seen the following error:
This error is an indication that your docker process died during execution. Most likely, this error is due to running out of memory. Try increasing the memory allocated to Docker.
On the Mac, this often manifests as the hyperkit
process using a high percentage of available CPU resources. To identify the container(s) responsible for the CPU usage:
Once you've identified a container using too much CPU time, check its logs; for example:
The most common culprit is an infinite restart loop where an error during service startup causes the process to exit, but we've configured docker-compose
to immediately try starting it again (so the container will stay running long enough for you to use a shell to investigate and fix the problem). Make sure the set of packages installed in the container matches what your current code branch expects; you may need to rerun pip
on a requirements file or pull new container images that already have the required package versions installed.
Docker for Mac has known filesystem issues that significantly decrease performance for certain use cases, for example running tests in edx-platform. To improve performance, Docker Sync can be used to synchronize file data from the host machine to the containers.
Many developers have opted not to use Docker Sync because it adds complexity and can sometimes lead to issues with the filesystem getting out of sync.
You can swap between using Docker Sync and native volumes at any time, by using the make targets with or without 'sync'. However, this is harder to do quickly if you want to switch inside the PyCharm IDE due to its need to rebuild its cache of the containers' virtual environments.
If you are using macOS, please follow the Docker Sync installation instructions before provisioning.
Check your version and make sure you are running 0.4.6 or above:
If not, upgrade to the latest version:
If you are having issues with docker sync, try the following:
The performance improvements provided by cached consistency mode for volume mounts introduced in Docker CE Edge 17.04 are still not good enough. It's possible that the "delegated" consistency mode will be enough to no longer need docker-sync, but this feature hasn't been fully implemented yet (as of Docker 17.12.0-ce, "delegated" behaves the same as "cached"). There is a GitHub issue which explains the current status of implementing delegated consistency mode.