Answer
Metrics (cheap numbers over time): is something wrong, and how much? They drive dashboards and alerts. Traces (one request across services): where is it slow or failing? Logs (detailed events): why did this request fail? A typical investigation goes alert → trace → logs.
Why
They cost differently: a metric costs the same at 10 or 10,000 requests a second, while logs and traces grow with traffic, so they are often sampled. Sharing a trace id across all three lets you jump between them.