# Backing up my WordPress blogs

The holidays is the time I have to fix up all the niggling system things that have been failing since last Christmas. This year involves a sad but inevitable migration from FreeBSD on Arp Networks to ArchLinux on Linode because I’d grown less fond of /usr/ports over the past decade. And, while I’m changing, why not do something drastically different?

More on that in a moment though. First, I used to have a process to backup my WordPress blogs automatically. I even had those backups in git and hosted in private repositories on three different services. That all broke in the migration and I never really bothered to fix it. Secretly, I kinda hoped everything would meltdown and I’d permanently lose all the data. That would fix the need to back up anything.

Previously, I used the root user with mysqldump even though I know that was a bad idea. It’s an easy and convenient idea because the root user is already there. It’s bad because it’s the root user, which can do anything, but it also needs a password. In a non-interactive session, such as a cron job, you end up exposing that (or putting it in plaintext in /etc/my.conf):

But I can create another user which doesn’t have a password so I can run it like this:

I can create a user with no password. That’s insecure in general, but just about everything this user can get at is already public. I’m a big fan of multiple database users with precise permissions. I grant the minimal permissions I think mysqldump needs in my case:

No big whoop.

The next part is harder. I need to get this to run unattended, but I’m on ArchLinux now. There’s no cron. The crontab was easy and life was good. Edit a file to add single lines, save it, then stuff happens on the schedule you specified. I didn’t need to know anything about how it worked to use it. I can install various packages to do it.

But, my personal contract with ArchLinux is that I’d do things the way it wants me to do it. Instead of cron, there’s systemd. It’s arrangement is not so simple, but it’s not really designed for users tasks either. It’s designed to get a system up and running.

I could easily have set up a system-level systemd unit to run my WordPress backup command, but I’m a bit allergic to root-level automated tasks that aren’t related to system operation. Instead, I need to setup user-level units for my personal tasks.

First, user-level units do their thing when the user is logged in, and not when they aren’t. That’s no good. I want it to run on its schedule regardless of my login status (although I could complicate it with a cronjob from elsewhere to log in, which I’ve done for some systems). But, I can setup a “linger” for my user:

At least, I should be able to do that. Running it give an incorrect error message. The issue in GitHub lays it out, but on my up-to-date system I still have the problem:

Peeking under the hood, it doesn’t find the directory it wants and doesn’t create it if it’s missing, nor does it warn if the permissions are incorrect. Instead, I fall back on my Unix roots to make the file myself:

I can check that its enabled:

Now my units will continue to do their job even if I am not logged in.

But, I have to create the unit. I’ve made simple, system-level units and remember that pain. I know I have to make some files, but where do I put them? Not in the group-writable /etc/systemd/system because I’m a normal user who’s not in the wheel group.

The ArchLinux SystemD/User has good instructions.

However, almost everything I’ve found deals with setting system-level units. That makes me nervous in the same way that putting my user tasks in a root crontab would. The ArchLinux SystemD/User is pretty good.

I need to put something in ~/.config/systemd/user, although that’s a bit deep. I’ll give myself a shortcut:

I started with a test that wouldn’t do anything else complicated. A sort-of “Hello World”. Here’s ~/.config/systemd/user/test.service that executes the date command, and appends the output to a file in my home directory, and to do this every 60 seconds:

The file is there, but who cares? Not systemd, until you tell it to care:

My new unit shows up in the list of services:

But, its status is still inactive. It’s merely enabled, not running:

Just like with the system units, I need to start it. Now its doing its thing:

After this, I know that something is working. I stop and disable it:

Checking the status again I see that it’s not active, along with a short history:

Here’s the actual unit, with the date command replaced by a program that does all of the hard work:

The file is there, so I enable and start it:

But, of course, I made a mistake and I changed the file. I need to pick up the changes while the service was already running. Since my file has changed, I can’t stop the service (because units can have instructions on what to do then):

So I reload everything to get my changes:

And finally, here’s the program I run. I use mysqldump to get out the SQL version, but I also use the wp tool to export to an XML format. It also reads the wp-config.php file to get the database name.