Upgrading the application

…or, upgrading the software of Scheduler itself.

The home of the Scheduler software is GitHub

If you already have a working Scheduler installation then this information should be embedded within it but for reference, the URL for fetching the software is https://github.com/XronosSchedulingLtd/scheduler.git

At any given moment, the “master” branch on GitHub will point to the latest released version, whilst in addition there are tags for each release, of the form “v1.2.0”.

You might choose to track the master branch (this is the default option) or you can lock your installation to a chosen version. The latter is useful if you need to do a progressive upgrade from a very old version - see below.

Cron jobs

Before you start to do an upgrade, think about when your various cron jobs are configured to run. They happen in the background and some of them can take quite a few hours to run - those which check all the events for student clashes and flag them.

If you start doing an upgrade whilst a cron job is running, odd things could happen.

Release notes

Before any upgrade, read the release notes. These are in a file called RELEASE_NOTES in the root directory of the application, and you can read it directly on GitHub.

There may be specific instructions there relating to a particular release. Release v0.28.3 is of particular note because it involves switching to a later version of Ruby, and thus needs extra steps.

Backup

Before you start, always take a fresh backup of your database. You know it makes sense.

Check current version

Before you start you need to know two things:

  • What version of software do you currently have?

  • What version is your system set up to request?

The first of these is easy - unless you have a very old version of the software, the version number is displayed in the bottom right-hand corner of the main screen of the application.

Note

If you have admin privileges within the application you can also get additional information by hovering your mouse pointer over this version number. The hover text will tell you the both the version of Ruby and the version of Rails in use.

If your version is too old to display the version number like this then you will find it in a file called “VERSION” in the root directory of the application.

Then you want to know what version of software your system is currently configured to expect. This will affect how it behaves when you connect to GitHub for an update.

The ‘git status’ command will give you this information. As the scheduler user, change into the SCHEDULER_DIR and type ‘git status’ and you should see something like this:

$ . ~/etc/whichsystem
$ cd $SCHEDULER_DIR
$ git status
On branch master
Your branch is up-to-date with 'origin/master'
nothing to commit, working directory clean
$

If you have changed any local configuration files that last line won’t appear and instead you will see a list of the changed files.

Note

The “Your branch is up-to-date…” line is potentially confusing. It doesn’t mean that your branch is necessarily up-to-date with the ‘master’ branch held on GitHub. ‘master’ and ‘origin/master’ are both branches on your local machine. What it means is, the last time your system checked with GitHub it was up to date.

The interesting line of output however is the first one - “On branch master”. That tells us that our system is set up to follow the latest release from GitHub, even if we haven’t got it yet. If instead you get something like this:

$ git status
HEAD detached at v1.2.3
...

Then your system is currently locked to version 1.2.3 of the software. We’ll see why you might want to have it like this in upcoming sections of this page.

Basic steps

Assuming your system is on branch ‘master’ and you simply need to upgrade by one version, the steps are as follows:

  • Stop the application

  • Install the new version of software from GitHub

  • Install any required support packages

  • Apply any required database structure changes

  • Pre-compile the CSS and JavaScript assets

  • Re-start the application

Note that some care is needed if you have let your installation get out of date. It is not always possible to go straight from a very old version to the current version. See below for instructions on how to update a very old version.

For this simple case, your upgrade session should look like this:

$ . ~/etc/whichsystem
$ cd $SCHEDULER_DIR
$ bin/delayed_job stop
$ sudo systemctl stop puma
[sudo] password for scheduler:
$ git pull
$ bundle install
$ rake db:migrate
$ rake assets:precompile
$ sudo systemctl start puma
$ bin/delayed_job start

Note that, unless you’ve taken a very long time over this, you won’t be prompted for your password a second time.

Note

The “git pull” command both fetches the new version from GitHub and places it into your application’s root directory. That’s fine for this simple upgrade case.

Upgrading in steps

If your system is more than one release behind GitHub then the general rule is to upgrade one release at a time. If you have only two or three to do then this isn’t too hard. If you have more, then see the next section.

Note

Upgrading by several versions all in one go may be fine, but there are circumstances in which it will not work.

Imagine you’re on version 1 of the software.

The migration to version 2 causes an update to all the user records.

The migration to version 3 introduces a new field to the user record and adds an application software constraint that this field must always be populated. It then immediately populates the field.

If you try to go straight from version 1 to version 3, then the new (version 3) software will be installed before any of the database updates are done. Then the database migrations are executed sequentially and the one for version 2 will fail because the constraint requiring the new field to be populated is already in place but the field does not exist.

If however you upgrade from version 1 to version 2, then from version 2 to version 3, all will be well.

Let’s assume you are currently running v1.2.8 of the software and the current version is v1.2.11. We’ll assume further that your installation is on the ‘master’ branch - you just haven’t got around to updating for a while.

Warning

At this point, your system’s idea of what the ‘master’ branch is differs from GitHub’s. We need to bring them gently back into line.

Start by shutting down nginx (as above) and then lock your system to v1.2.8 of the software. Once you’ve done that you can bring down all the new versions from GitHub:

$ . ~/etc/whichsystem
$ cd $SCHEDULER_DIR
$ bin/delayed_job stop
$ sudo systemctl stop puma
[sudo] password for scheduler:
$ git checkout v1.2.8
$ git fetch

Note the use of “git fetch” instead of “git pull”. “git pull” says, “Grab the latest of everything and put my chosen version in my application directory.” If you do that whilst your system is on the ‘master’ branch then you’ll jump straight to the latest version - not what is wanted in this case.

The “git fetch” command on the other hand says, “Fetch down anything new which you can find on GitHub, but don’t do anything with it for now - just fetch it”.

You’re now in a position to apply the updates one by one. Still working on our imagined task of going from v1.2.8 to v1.2.11 you would do the following:

$ git checkout v1.2.9
$ bundle install
$ rake db:migrate
$ git checkout v1.2.10
$ bundle install
$ rake db:migrate
$ git checkout v1.2.11
$ bundle install
$ rake db:migrate
$ rake assets:precompile

And then assuming that v1.2.11 is indeed the latest version we can put the system back into its original state with:

$ git checkout master
$ sudo systemctl start puma
$ bin/delayed_job start

Note

When you check out an explicit version like v1.2.9 you get what looks like a dire warning from Git about your system being in a “Detached HEAD” state. This is not a problem, because you’re not doing development work on the system - just installing specific versions of the code.

If you have set yourself up with a test system then you can short-circuit this process by trying the update on your test system first. If all the database migrations run without error then it will be fine to go straight from your start version to your target version (but see note about v0.28.3 below).