Why government portals fail on the last day

Placeholder article. Replace with your own editorial before publishing.

Every recruitment cycle produces the same graph. Applications trickle in for three weeks, and then two-thirds of the total arrives in the final seventy-two hours — with a spike in the last four.

Capacity planned against the average is off by an order of magnitude at exactly the moment failure is most visible.

What actually breaks

It is rarely raw compute. In our experience the failures cluster:

• Database connection exhaustion. Application servers scale out, each opens a pool, and the database hits its connection ceiling.
• File upload storage. Every applicant uploads a photograph, a signature and supporting documents. Storage throughput, not application logic, saturates.
• Payment gateway timeouts. The gateway is also serving everyone else's deadline. Retry logic that assumes success becomes a duplicate-charge generator.
• Synchronous email and SMS. A confirmation message sent inside the request path ties your availability to a third party's.

What we do about it

Queue anything that does not have to be synchronous. Put uploads on object storage with pre-signed URLs so files never traverse the application tier. Make payment initiation idempotent. And load-test against the real closing-day curve, replayed from the previous cycle's logs — not against a flat average.

None of this is exotic. It is just planned for before the deadline rather than during it.