Splitting light

Season 2 Episode 24

Hackathon

If you are no longer interested in the newsletter, please unsubscribe

Several of the team members had gone to School 42. A tuition free university created by the owner of Scaleway, Xavier Niels. Several would be an understatement. Out of the 14 people, more than half had gone there. School 42 frequently organized hackathons. We decided it was the perfect opportunity for us to organize one.

Théo doing the hackathon presentation (1)

Our goal was to have students find use cases that used both Carbon14, the cold storage platform and the Object Storage platform. We had people sign up and their credentials were prepared for the hackathon.

On the start date, a Friday evening, we presented the products. Their objective for them was to find use cases to use both products as a single product.

Florent speaking with Loic on the side and Théo on the floor behind (1)

On the team storage side there was a code and deployment freeze for the duration of the hackathon. Even if we found a bug, we would not fix it. Deploying or fixing the bug could create disruptions and unfairly disadvantage some teams over others.

Over the weekend, we established a rotation of people to be on site, at the school, to help and mentor students. We projected the cluster dashboard over the wall in one of the rooms.

I remember being there during Saturday night. It brought back a lot of nostalgia of my own time at Epitech, a sister university with a similar organization.

One of the students came to me. He was having issues with his project. I had him show what he had done and tell me what he was trying to do. This student set up a hadoop cluster on S3 and connected it to Carbon14 with code. As I looked through what he showed me, we found something odd. The endpoint was correctly set. The region entered correctly. Credentials were good. But the request failed because it could not resolve the domain.

We cracked open the hadoop source code and followed the logic. There it was. Hadoop composed the final URL relying exactly on what AWS did. On AWS, the URL was s3.REGION.domain, but, we naively had done REGION.s3.domain.

Hadoop could not connect to the object storage because we, the storage team, had deviated from the standard.

I looked at alternatives to unlock his issue. Modifying the local resolver didn’t work. No admin rights. Patching hadoop could work but it would require understanding the logic fully and rebuilding it completely. Neither of us had done it before. It would take too much time.

Quentin & Ramzy helping students, inside School 42 (2)

On the server side, even if we didn’t have a deployment freeze, changing this on the cluster level would require too many changes. There was nothing we could do. We talked about the issue for some time. I explained to the student what was wrong and went into some of the details.

I could see he was disappointed. He had this big idea and an issue that was out of his hands made his idea impossible to work.

The next day, students did presentations. They came up in front of everyone and presented their ideas. Immediately we understood something was wrong. Some students had built working Proof of Concepts (PoC), but some had only done a pitch deck. It was our fault. We had not given precise enough instructions. For us it was clear we wanted a PoC but we had not explicitly told them. A few students were mad because the PoC was a criteria. There was nothing we could do.

Loic sitting on top right, Ramzy on the right and Florian in the middle in the back, inside School 42 (2)

We learned two things from this hackathon. To be clear and precise in our ask and our presentation. Most importantly to not deviate from the standard. The rack was dismantled and shipped to Amsterdam where we had room in a datacenter. We would be able to launch before DC5 was ready. While the rack was on the road, we tweaked the configuration to be exactly like AWS.