Hacker Cup Round 1 Recap

TLDR: Hacker Cup Round 1 is a wrap! We’ll be holding Round 2 on November 15th (instead of November 1st) to allocate sufficient time for stress testing.

Thanks to everyone who participated, and the judges hope you enjoyed the problems!

First off, congratulations to Geothermal for winning, and our top 3:

1. Geothermal, solving the set in 35 minutes and 35 seconds
2. ksun48, who solved C and D first and came back for B2
3. tourist, for placing third despite solving the set 12 seconds before Geothermal

Additionally, I’d like to give a huge thanks to the people who helped make the round possible, including the authors:

• A1/A2:SecondThread, with diagrams from Alex Li
• B1/B2: anhkha1004
• C: Intellegent, with additional writing from SecondThread and Alex Li
• D: Intellegent, with additional writing from SecondThread and Alex Li

And to the judges at Meta, who spent their entire weekend answering 1,600 clarification requests during and after the round to help save the contest, including especially Joyce Lynn and Wesley May (wjomlex).

What was up with the technical difficulties?

This year, we received an unprecedented number of submissions ( +88% YoY). For comparison:

Unfortunately, our underlying database simply could not keep up with this amount of traffic. This affected everything that relied on that database, from registering for the contest, to viewing the scoreboard, to making and judging submissions. Here’s our requested traffic (orange) and delivered traffic (purple):

Under normal circumstances, these lines should match perfectly.

In truth, we suspect the orange line is also likely an underestimate of the true demand for two reasons:

1. This only reflects people who could register, load pages, and make the requests. Others likely wanted to register for the contest after it started, but couldn’t. (They would need to load the contest and write to the DB to register for it)
2. The orange line shows unexpectedly consistent peaks. Usually we see more variation than this, especially during the beginning and end of the contest. This suggests those regions may be getting clipped because individuals face issues before getting to the place where they even need to make a submission.

What are we doing about it?

1. We’ll be running stress tests of our backend throughout this week to measure and ideally reproduce these failures.
2. After reproducing, we'll be improving our database structure to make sure we can handle the levels of traffic we need.

We want to make sure we execute on this well, and that will take some time and attention to get right. We’ll plan to host Round 2 on November 15th (instead of November 1st). This should give us the time we need to run these tests, and fix the underlying issues we find, without decreasing quality of the problems.

Scaling pains like this are unfortunately common, and they are especially unforgiving in cases like coding contests where all your traffic comes in one huge burst. All things considered, I'm thankful that we were able to save the round, and for the opportunity to build an even more reliable system in the future.

We hope you enjoyed the problems, our transparency, and the chance to compete in Hacker Cup, and that you are looking forward to Round 2 and a more bullet-proof platform as much as we are. We'll see you on the scoreboard on November 15th!