A tiny key-value store with write-ahead logging for my databases course

This was the final project for my undergrad databases course. The assignment was to build a persistent key-value store in Java, and the grading rubric cared about crash recovery. I did the write-ahead log part myself; two teammates handled the B-tree index and the CLI. My piece was: append each put/delete to a log file before touching the in-memory map, then replay the log on startup. I got fsync working so a kill -9 mid-write wouldn't lose committed data. Where it fell apart: the log grew forever because I never wrote compaction or checkpointing, so restart time got slow with a big log. We had maybe two weeks, so I cut checkpointing and just documented it as a known limitation in the README. It was never used by anyone real, it was graded and that's it. In the interview I led with that framing, said 'this was a class project, here's the part I owned,' then walked the durability path. The interviewer kept pushing on what happens if the machine dies during fsync itself, and I had to reason about torn writes on the spot rather than pretend I'd handled it.