It's thrilling to click The Button and watch Craft update itself. I take advantage of multi-site configurations with every deploy, and my practice thus far is to...

  1. Upgrade Craft in production.
  2. Commit production file system changes via git.
  3. Choose Settings → Backup Database and download the backup file.
  4. Load the remote changes to my local environment via git.
  5. Clear out and locally import the .sql backup I downloaded in step #3.

I specifically update in production because I like thrills, and because I had previously done the opposite (upgrade locally, push live) and ran into problems.

This has been working for me pretty reliably and Craft upgrades always go smoothly, but I doubt that system updates on the production server are the best idea.

Is there a better way?

3 Answers 3


I know your method works flawlessly, but it scares the crap out of me.

Luckily Craft's update setup is magical.

What I do is this.

  1. Make sure my repo is clear of unstaged files, just so the update can be in a commit all by itself.

  2. Run the update on my local dev environment (with a local database) (Test)

  3. Commit the updated files and push

  4. Backup live database just in case

  5. Deploy the files on live. (I use Beanstalk, so it uploads the files for me, but you can also do a git pull if that's your method of deployment.) Craft detects that the database version doesn't match the files and puts up a maintenance message.

  6. Visit /admin and click the update button

  7. Drink.

  • So basically the reverse, with an intriguing last step! I appreciate the answer and am curious about how most people go about this.
    – Matt Stein
    Commented Jun 12, 2014 at 1:12
  • 6
    I like this route in that it gives you a chance to test the update locally first. Also, there could be a step 0 to this in that you could download the DB from production and update your local dev DB first. That way you're actual local testing is as close to production as possible.
    – Natetronn
    Commented Jun 12, 2014 at 2:29
  • 4
    I approve of this answer, especially step #7.
    – Brad Bell
    Commented Jun 12, 2014 at 2:41
  • 3
    For a single version upgrade, this method is smooth, but remember, the more files you deploy, each file deployed gets sent to your server one by one. For larger updates, it's good to consider testing your upgrade like in step #2, and if all goes well, update step #5 to be 'Run the update in your production environment'. If you run the one-click update in production, Craft downloads all the files as a zip and unpacks them on your server which is much quicker and reduces the chance for your site throwing temporary, odd errors while files make it to the server. Commented Jun 15, 2014 at 16:00
  • 2
    Good point. Beanstalk deploys quite quickly, but yes, for a bigger update it could have some errors while the files are uploading. I would suggest temporarily renaming and replacing index.php with a maintenance message if this is the case. Commented Jun 15, 2014 at 17:59

I actually keep all the craft system files gitignored. This way, all the assets are independent of the version of Craft and therefore of the updates. When it's time to run an update, I'll do it on dev first and then I'll run it in production knowing that it'll be fine.

So far I have not had a problem with this approach and it seems to me that it is a more elegant way of doing things.

Now, in my installations I keep separate databases, maintaining the structure the same across all environments. When I need to see the same data from production, in dev, I'll import it manually. I can't say that this side of things is the easiest to work with but since I've had to deal with small databases until now it hasn't been much trouble. I am beginning a new, larger project so... things may change soon. :P

Here's my usual gitignore file for Crafty projects.

# CMS related files

# OS or program files

# Grunt

# Others

The folder public/content is where all the assets generated by craft go. These get ignored since the real data is in storage.

As you can see, storage is ignored here, as I said before, that's because I keep separate databases. That works for me but it's certainly not ideal for larger environments.

  • The problem I have with this is if you have multiple developers on your team, then having one of them download the repo does not give them all the files they need, they still need to obtain Craft files from somewhere else, and it creates extra steps. The team would need a clear process on how to handle this without creating additional issues.
    – Simon East
    Commented Apr 30, 2017 at 21:57

I personally use a slightly different method than what Jeremy recommends.

  1. Clean repo on the git side of things: no unstaged or non committed files
  2. Backup live DB and import in dev
  3. Update dev (test)
  4. Update production
  5. Commit & deploy

Not encountered any problem so far and Craft runs its tests on both environments.

  • This is what I've tested as well, and it works for me.
    – Jon Horton
    Commented Jun 23, 2014 at 12:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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