Skip to content

Files

Latest commit

 

History

History
164 lines (99 loc) · 9.56 KB

CHANGES.md

File metadata and controls

164 lines (99 loc) · 9.56 KB

Breaking and notable changes

14 October 2021

  • Add the ability to skip the rm -rf step in bower-install.sh and npm-install.sh, when bower.json or package.json has changed, by exporting BOWER_INSTALL_SH_SKIP_CLEAN=1 or NPM_INSTALL_SH_SKIP_CLEAN=1.

6 April 2021

Breaking

  • django-compressor and ixc-whitenoise will now include a .$GIT_COMMIT suffix in their manifest filename, so we can run new and old versions of a codebase side by side with a shared volume.

    When upgrading:

    • To disable this behaviour for local development, where you do not want to re-compress or collect static files for every commit, add the following to your .env.develop.secret file:

      # Configure static file manifests. Comment out to save versioned files with a
# `$GIT_COMMIT` suffix.
export COMPRESS_OFFLINE_MANIFEST='manifest.json'
export STATICFILES_MANIFEST='staticfiles.json'

13 November 2020

Breaking

  • Use tuple instead of list for template dirs, context processors, loaders. Check if you are using += [..., ] and change to (..., )

17 September 2020

Breaking

  • setup.sh and setup-tests.sh no longer automatically run any npm script, even if SETUP_NPM_RUN or SETUP_TESTS_NPM_RUN is defined.

    Instead, export SETUP_COMMAND and SETUP_TESTS_COMMAND, which can be any shell command instead of the name of an npm script.

    This allows us to execute any combination of npm scripts in sequence (via run-s) or parallel (via run-p), without cluttering up package.json with near identical wrapper scripts.

    We can also execute any shell script or bash command.

    When upgrading:

    • Add export SETUP_COMMAND='npm run -l build --if-present' and export SETUP_TESTS_COMMAND='npm run -l build --if-present' to your .env file to restore the old behaviour.

    • Or add export SETUP_COMMAND='...' to your .env file, as required.

  • The COMPRESS_OFFLINE setting is now True. Offline compression is required when using WhiteNoise without autorefresh, which is not used in production because it is only intended for development and has had a serious security issue in the past.

    When upgrading:

    • Add "compress": "manage.py compress --verbosity=0" to the package.json file.
    • Add export SETUP_COMMAND='run-s -l compress ...' to your .env file.
    • Add run-s -l collectstatic compress to Dockerfile.
    • Confirm that your {% compress %} blocks do not have any dynamic content (changes with context), or set IXC_COMPRESSOR_REQUEST, IXC_COMPRESSOR_GLOBAL_CONTEXT and IXC_COMPRESSOR_OPTIONAL_CONTEXT accordingly.

Notable

  • Supervisor no longer wraps its programs with entrypoint.sh. This was a convenience during troubleshooting so we could edit a dotenv file and restart the program, without having to restart supervisor or the container itself.

    Now that entrypoint.sh is optional, it is no longer appropriate to always wrap programs with it.

    If you are still using entrypoint.sh, in a pinch you can live edit the supervisor template to reinstate the wrapper.

8 September 2020

Breaking

  • setup.sh now runs the setup npm script instead of build, to better reflect its purpose. To upgrade, rename your build script to setup in package.json or export SETUP_NPM_RUN=build in your env file(s).
  • The hard coded collectstatic and compress commands have been removed from runtests.sh. Instead, the setup npm script is run by setup-tests.sh. If you need to execute different commands for setup and tests, export SETUP_TESTS_NPM_RUN=foo in your env file(s).

27 March 2019

Breaking

  • Rename SUPERVISORD_* environment variables to SUPERVISOR_* for consistency with Supervisor's own environment variables.
  • Remove logentries program from supervisor.tmpl.conf. Add it back via SUPERVISOR_TMPL_CONF, if needed.
  • Run a single gunicorn process by default. Explicitly set GUNICORN_WORKERS=auto for the previous behaviour (2x CPU cores +1 for sync, 1x CPU cores for async).
  • Run a single nginx worker process by default. Explicitly set NGINX_WORKER_PROCESSES=auto for the previous behaviour (1 per CPU core).
  • Remove www redirect via nginx when NGINX_REDIRECT_SERVER_NAME is defined. Do the redirect in your app or with https://github.com/ixc/nginx-proxy-redirect

Notable

  • Add prefix-logs.sh and use it in supervisor.include.tmpl.conf. Send program logs directly to stdout and stderr with a prefix, so we don't need dockerize -stdout ... -stderr ... anymore.
  • Send nginx access log to stdout instead of gunicorn, now that we can send all supervisor program logs directly to stdout and stderr with a prefix.
  • Add SUPERVISOR_NUMPROCS environment variable, for programs that don't internally run multiple worker processes.
  • Each process will have SUPERVISOR_PROCESS_NUM environment variable set, which can be used to increment the base NGINX_PROXY_PORT (default: 8080) port number.
  • Configure nginx to load balance all processes via the default server, AND allow direct access to each process via s{{ SUPERVISOR_PROCESS_NUM }}.* subdomains.
  • Define default environment variables once in shell scripts instead of config templates, where they are often needed multiple times.
  • Disable nginx CPU affinity. Allow workers to execute on any available process.
  • Move nginx config from lower to higher contexts (e.g. from server to http) where possible.

10 July 2018

Breaking

  • djcelery.schedulers.DatabaseScheduler is no longer hard coded in celerybeat.sh. If you want to continue using the database scheduler, add CELERYBEAT_SCHEDULER = 'djcelery.schedulers.DatabaseScheduler' to your project settings.

  • Offline compression is no longer enabled in compressor.py settings, because we now have persistent shared volumes on Docker for AWS. If you want to continue using offline compression, add COMPRESS_OFFLINE = True to your project settings.

  • git-secret is no longer supported. Switch to transcrypt, before upgrading.

  • SASS/SCSS precompiler has been removed from compressor.py settings, because it forks a compiler on every single request when offline compression is disabled. Instead:

    • Add RUN entrypoint.sh compile-scss.sh to Dockerfile before collecting static files.
    • Add static/COMPILED/ to .dockerignore and /static/COMPILED/ to .gitignore.
    • Reference the compiled CSS in templates, instead of the source SCSS. For example: {% static "COMPILED/scss/foo.css" %} instead of {% static "scss/foo.scss" }.
    • Run compile-sass.sh --watch during development.

  • Setuptools has vendored its dependencies again. Update Dockerfile and go.sh, changing pip install ... -r <(grep -v setuptools requirements.txt) to pip install ... -r requirements.txt

  • setup-django.sh has been removed in favour of setup.sh and setup-wait.sh. Add the former to a new setup service with no health check. Replace setup-django.sh with the latter in existing services, and remove start_period from health checks.

  • Dockerize is now a required.

Notable

  • If your project uses Python 3, you must export PYTHON_VERSION=python3 in Dockerfile and specify --python=python3 in go.sh when creating the virtualenv.

  • go.sh no longer uses the active virtualenv instead of var/go.sh-venv when VIRTUAL_ENV is set. If you want to continue using a virtualenv in a non-standard location, add PROJECT_VENV_DIR='path/to/venv/dir' to your .env.base file.

  • Editable packages are now always installed in $PROJECT_DIR/src, when run with Docker or go.sh. If you want to install to a different location, add PIP_SRC='path/to/src/dir' to your .env.local file.

  • Stub setup.py files are no longer needed to add the project directory to the Python path.

  • Gunicorn now uses gevent async worker by default. If you want to continue using the default sync worker, add GUNICORN_WORKER_CLASS='sync' to your .env.base file.

  • runserver.sh now runs Gunicorn instead of the Django dev server, with a single async worker, autoreloading enabled, and simplified logging. If you want to run the Django dev server, run manage.py runserver 0.0.0.0:8000.

  • bash.sh sleeps forever when there is no tty. Use this as the command in a shell service with no replicas, which you can scale up on demand then exec into to troubleshoot.

  • You no longer need to manually terminate existing connections before executing SETUP_POSTGRES_FORCE=1 setup-postgres.sh to drop and recreate a database.

  • SETUP_POSTGRES_FORCE=1 setup-postgres.sh will now prompt for confirmation when there is a tty, for safety.

  • You can now exclude unwanted table data when restoring from a source database via SRC_PGDUMP_EXTRA. For example, SRC_PGDUMP_EXTRA='--exclude-table-data django_session --exclude-table-data FOO ...'

  • Management commands executed in Dockerfile should be wrapped with entrypoint.sh to ensure the runtime environment is properly configured.

16 January 2018

  • Config templates are now rendered to the $PROJECT_DIR/var/etc directory for easier discovery and to avoid writing to potentially read-only file systems.

    The default templates can be overridden by setting the LOGENTRIES_TMPL_CONF, NEWRELIC_TMPL_CONF, NGINX_TMPL_CONF, SUPERVISOR_TMPL_CONF, or SUPERVISOR_INCLUDE_TMPL_CONF environment variables in your dotenv file.

  • supervisor.sh no longer attempts to render $PROJECT_DIR/etc/supervisor.tmpl.conf as an alternative to the default include config (nginx proxy).

    Instead, you should explicitly set SUPERVISOR_INCLUDE_TMPL_CONF=$PROJECT_DIR/etc/supervisord.include.tmpl.conf in your dotenv file.