Datacenter networks require active, automated, high- precision (e.g., per-packet) monitoring infrastructure that operates at fine-grained time-scales (e.g., milliseconds). Building such an infrastructure is a daunting task. For instance, just capturing packet headers for a 10Gbps link at line rate would create 0.9million records per second! In addition, the automated monitoring system must be able to interactively slice-and-dice these logs (especially the most recently captured logs) to detect and diagnose net- work events. Unfortunately, as we show, existing systems support only one of the two desirable properties — high write-rates of logs, or interactive queries on fresh logs. NetPlay is a system that simultaneously supports the two properties using a new data structure — MonoLog (for Monotonic Log). We show that the monotonicity in network logs (that is, the fact that once a packet header is captured, it is neither updated nor deleted) allows MonoLog to significantly simplify existing lock-free concurrency mechanisms. NetPlay, by using MonoLogs for both network logs and indexes to query these logs, thus supports writing millions of records per second and interactively query logs written within last millisecond, while utilizing less than 40% of a single core.