The Scheduler Nightmare
When we added the "Schedule" button to Vocal Spark, we thought it would be a weekend project.
"Just send a POST request at a specific time," we said. "How hard can it be?"
Famous last words.
The Timezone Trap
Time is an illusion, especially in software development.
The User: "I want this posted at 9 AM." The Server: "Which 9 AM? UTC? EST? The 9 AM where you are now, or where you will be next week?"
We initially stored everything in local time. Big mistake. When a user traveled from New York to London, their entire content calendar shifted by 5 hours, posting "Good Morning" tweets at 2 PM.
The Fix: We rewrote the entire database to store timestamps in UTC and calculate the offset dynamically based on the user's browser locale at the moment of viewing, while locking the execution time to their profile settings.
API Hell
Connecting to social platforms is like negotiating with five different bureaucracies that all speak different languages.
- Twitter/X: Strict rate limits and constantly changing endpoints.
- LinkedIn: Complex token refresh rotations that expire if you look at them wrong.
- Instagram: Requires media sizing so precise that being 1 pixel off causes a failure.
We built a Retry & Queue System. If LinkedIn's API hiccups (which it does), our system doesn't just fail. It waits 30 seconds, tries again. Waits 2 minutes, tries again. Only after 5 exponential backoff attempts do we alert the user.
The "Ghost" Post Issue
During beta, we had a terrifying issue: Ghost Posts. The UI said "Posted", but the social media feed was empty.
It turned out to be a race condition. The UI updated optimistically before the server confirmed the transaction.
We switched to a Websocket-based status system. Now, you see the "Publishing..." spinner until the exact millisecond we get the 200 OK from the platform. It feels slower by 300ms, but it is 100% accurate. Trust is faster than speed.
Building a scheduler taught us one thing: The simple features are often the most complex to engineer robustly.
