The day you launch, your instinct is to keep adding. The day after, if you're honest, you realise the job changed while you weren't looking. It's not "what can I build next" anymore. It's "what happens when something breaks and I'm asleep."
Yesterday was load-testing day. Ten thousand simulated bookings hammered the system via JMeter, and I watched it bend but not break. That's not the same as trusting it to survive the real internet unsupervised. Simulated load is polite. Real traffic is drunk, impatient, and arrives at 3 AM.
A booking platform moves real money. Razorpay tells us a payment succeeded by sending a webhook -- a little message that says "this one's paid." Almost always it arrives in a second. But "almost always" is a terrifying phrase when it's attached to someone's rupees.
The failure I can't watch
What if the webhook never arrives? A guest pays, the confirmation should fire, and the message that tells us so gets lost to a flaky network. Now there's a booking stuck in limbo -- paid by the guest, unconfirmed by us. Nobody's at fault, and everyone's unhappy.
So today I built a reconciler: a quiet background job that hunts for bookings stuck unpaid on a missed webhook and settles them against what actually happened at Razorpay. If a webhook keeps failing, it doesn't fail silently -- it messages me. I'd rather be woken up than find out from a customer.
The unglamorous layer
Alongside it: a proper staging-to-production pipeline so I never deploy untested code to real users again. Daily encrypted database backups -- because data without backups is a promise you can't keep. Sentry error tracking so I see crashes the moment they happen, not when someone complains. And an ops runbook: the exact steps to deploy, rollback, and recover if everything goes sideways at once.
None of this shows up on a screen a host will ever see. A host will never know the reconciler exists unless it saves their booking. They'll never read the runbook unless I use it to rescue their data at 2 AM. That's the point. Infrastructure is the work that earns its keep by being invisible.
Reliability isn't a feature you add later. It's the floor you build on.
— Note to self, day two of being live
A stranger trusted us with their money yesterday during load testing. Tomorrow, real strangers will do the same for real. Today's work was making sure that trust survives the things I can't stay awake to watch.