What order should the following drush commands be ran?

  • config-import
  • updatedb
  • entity-updates

Also, I see entity-updates fail a lot due to field_delete_data* tables existing. How can I delete them as part of my automated deployment?

  • 1
    You shouldn't delete those tables yourself unless you're absolutely sure they are not being referenced. Cron should clean them up. See my question I asked about this recently drupal.stackexchange.com/questions/253055/… – Kevin Jan 24 '18 at 16:49
  • The issue is that I cannot wait for cron to do the tidy up, as we do deployments as and when – Paul Canning Jan 24 '18 at 17:15
  • 1
    You should not delete those tables at will - review that question as to the whys. – Kevin Jan 24 '18 at 17:16
  • I've not yet encountered an issue in doing so :/ – Paul Canning Jan 24 '18 at 17:17

Drush 10.3.0 now provides a deploy command to standardize how Drupal deployment works. This command performs the following:

drush updatedb --no-cache-clear
drush cache:rebuild
drush config:import
drush cache:rebuild
drush deploy:hook

Database updates always come before configuration import! Support for automatic entity updates has been removed from Drupal core (change record) and entup has been removed from Drush core. Whenever an entity type or field storage definition needs to be created, changed or deleted, it has to be done via hook_update_N().

Here is the updated deployment routine I'm happiest with.

drush state:set system.maintenance_mode 1
drush cache:rebuild
git pull
composer install --no-dev
drush cache:rebuild
drush deploy
drush state:set system.maintenance_mode 0
drush cache:rebuild

This also means you have to run two releases if you want to uninstall and remove a contrib module. First release to deploy the updated config that disables the module. Second release to deploy the updated composer.json and lock file after you removed it with Composer.

To make real consecutive releases possible you might want to pass the current release's commit SHA1 to the deployment script and then replace git pull with a more exact routine (where $1 is the SHA1):

# If not empty 1st argument passed to the script, do:
if [ -n "$1" ]; then
  git reset --hard "$1"
  git pull

Otherwise the consecutiveness can not be guaranteed when you push two new releases at once or within a short period. As then the first release's triggered git pull will simply pull the latest changes (from the second release), where instead it should pull only the changes included in the first release. See the full sample repo leymannx/drupal-circleci-behat.

Credit for this git snippet goes to CircleCI. This is how they are doing it in their containers.

| improve this answer | |
  • 5
    I'm strongly against using entup in such a workflow. Modules must provide update functions for their entity type changes, entup is not reliable (there are many changes that it can't do and will fail with an exception) and should only be used when developing a not-yet-released entity type. It's something that non-cli users can not execute. – Berdir Jan 25 '18 at 19:25
  • 2
    Since the answer mentions composer: It is recommended to run composer install with autoloader optimization options such as --optimize-autoloader. – marcvangend Feb 11 '18 at 19:42
  • So am I reading this right that you run cache:rebuild 5 times? 2 inside the new drush deploy and 3 outside of that? – Elijah Lynn Aug 26 at 21:04
  • @ElijahLynn – Absolutely. First cache rebuild after maint mode enabled to 100% ensure the whole site picks it up, next rebuild after composer install to 100% ensure Drupal recognises new or updated dependencies, then drush deploy, and final rebuild after maint mode being disabled and deliver a whole fresh site to your users (you can consider implementing some cache warming strategy after that, using Warmer for example). You can maybe skip the first and the last rebuild but you should definitely do the one after composer install. – leymannx Aug 27 at 6:47

The sequence of commands should be:

updatedb (which runs update hooks)

You do not want to run entity-updates because it is deprecated, see https://www.drupal.org/node/3034742. Instead, rely on update hooks (hook_update_N) to properly modify any database schema or necessary configurations.

It is imperative that updatedb is the first command run that boots Drupal after code has been changed. You can see https://www.drupal.org/project/commerce/issues/3100553 for some commentary on what issues might arise when that is not done.

Example Deployment Script

Here is an example deployment script that has had a lot of review and discussion behind it, but consider it a starting point. It's likely there will be adjustments you would want to make. It does assume you're deploying an artifact (notice that composer install is not run on deployment).

drush sset system.maintenance_mode TRUE
# Create a restore point by taking backups of anything that is not in the code repository: database, media, cache
# Checkout the code you are deploying
drush updb
drush cim sync -y || drush cim sync -y
drush cim sync -y
drush sset system.maintenance_mode FALSE
drush cr

You can see https://www.bounteous.com/insights/2020/03/11/automate-drupal-deployments/ for a deeper explanation behind this. See also https://github.com/drush-ops/drush/pull/4359/ which is a PR to include a deploy command in Drush.

I mentioned the script above is a starting point, I've documented some variations that you might want to apply to that script (or any deploy script you use) that might be helpful: https://www.bounteous.com/insights/2020/03/12/automated-drupal-deployment-and-rollback-recipes/

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.