Just finished the full Datadog SWE loop for a backend role on their APM team. Posting the coding round specifics because I had to piece this together from outdated threads.
The OA / phone screen coding round: One 45-minute live coding interview on CoderPad. One interviewer, camera on. One problem. No multiple choice, no automated test suite, real human watching you the whole time.
The problem I got was graph-based: something like finding the shortest dependency path in a package graph. Medium difficulty on Leetcode's scale, maybe medium-hard. Classic BFS/DFS territory but with a twist in the input format that you need to parse first. They care that you ask clarifying questions before coding.
What tripped people up (per the interviewer's hints): Not verifying edge cases: empty graph, cyclic dependencies, single node. They seem to weight correctness of edge cases heavily.
The onsite coding rounds: Two separate 45-minute rounds. Both live. Problems were again in the medium-to-hard range, one felt like an interval/scheduling problem, the other was more data structure design (like implementing a rate limiter from scratch with a sliding window). The rate limiter one is practically a classic for infra-facing teams.
They explicitly said they don't care which language you use. I used Python, my interviewer was a Go person, didn't matter.
No system design in the coding blocks. Those are separate rounds. The coding rounds are pure algo/DS.
Overall: harder than I expected for a non-FAANG company. The live format means you can't fake it. Budget time for talking through your approach before writing a single line.