This article explains why storing log messages alone is insufficient for robust operation of a software service. Metrics also need to be gathered and stored. https://medium.com/@copyconst…/logs-and-metrics-6d34d3026e38
tl;dr - Log volume can spike dramatically when user activity increases, especially when things go wrong. This makes it possible for an alerting system based on logs to be swamped. For a metrics system, volume increases with the number of metrics collected. This is stable and much less likely to fail or slow down during a crisis.