The straw that broke the pipelines back

Necessity is the mother of invention. Well shit...Awaaaaayyy we go!

I had a great weekend, coding my little heart out. I am on the final mile of the fileorbit.cloud beta. I added Stripe Checkout, and I am testing it. Looks great. One small snag. My emails won't send.

I added email templates to the Tranquil PDS that look sexy. I was wicked proud of it. I wasn't thinking much of it and added my subscription emails to be sent to the PDS since I am lazy and thought it sends emails, NBD. Welp that turned out to be dumb.

I learned this by adding tracing since my Stripe Checkout webhook was working wonderfully. My quota was being increased and decreased based on if I canceled or subscribed, yet I kept seeing lifecycle email failed (ledger change kept). A log I had designed. Lifecycle meaning the circle of life for a subscription: they add extra to move data quickly or have more space for a trip, then cancel a month later after they needed to tighten their budget.

As I said, necessity is the mother of invention! So I looked in the PDS logs but its busy in there. Firehose this, error that, but nothing that signaled email failed. So I turned to a trusty tool I've used at many jobs: tracing. After some hardcore coding this weekend I got my tracing working. Then I saw it:

PDS sendEmail failed: (404 Not Found): Account not found

...OMG WHY AM I SO DUMB. Maybe because I am on my last braincell and its 12am. The account I am using to test is this one, and its hosted with BlueSky not on FileOrbit.

I sighed and rolled up my sleeves. Time to add a new module. I made an email service, but I am cheap and don't have Netflix money so I made a module that is its own service. Instead of it running on its own box I would treat it like my billing service: a standalone web service that deploys with its own port and routing via Caddy, yet with a twist, a systemd service!!! I know right, crazy stuff. Back to the basics from when I was selfhosting a homelab. This gave me best of both worlds. Loosely coupled APIs that one day can be easily peeled out of the one server instance and horizontally scaled on its own. For a garage software project this is good enough.

Here comes the straw... Tangled.sh has a free pipeline they refer to as Spindles. It is simple but it does the job. There is one catch: it has a timeout of 15m unless you configure your own hosted runner. I love the cloud for this (someone else's server). I don't have to manage it. When it fails I post to Discord and poor Lewis sighs and fixes it while I am at my day job. So what do you know, I hit the limit with no turning back.

So now I am knee deep into a real process for releases where I have parallel Spindles building the gateway, file explorer, billing, and email dependencies and uploading them into a private artifact store via DO Spaces that I previously used for sccache to speed up the pipelines to avoid the timeout.

Well this sucks, but the silver lining is now I have a stateless pipeline. I am forced to create a release trigger via tagging which gets created on a new desktop app publish, and boom, now I have a performant and well documented process for updating the web application and desktop apps.

While I have this cooking I started a project I call Sleeping Beauty. It is a self hosted Spindle helper for installing bare metal Spindles so I can now, instead of making the desktop apps be the release tag trigger, work on the app and test it out, set a patch, minor, or major tag via the Nx release process, push it, and both my MacBook and Windows desktop will build and publish my desktop apps. Its still early but I am really excited to be forced to mature my process. This will allow me to release with consistency and with ease.

If you made it this far in my ramble I appreciate you!

Until next time, peace nerds and normies!

~ Triforce