Techniques are provided for application tracing for inline performance analysis. One method comprises obtaining trace events generated by instructions executed in response to trace points in instrumented software; updating, for each trace event, a buffer entry of a sampling buffer that corresponds to a particular processing core and a time window, wherein the buffer entry is identified based on (a) a flow type identifier associated with the instructions, (b) an identifier of a respective trace event, and (c) an identifier of an adjacent trace event to the respective trace event, and wherein the updating comprises updating, for the time window: (i) a first counter indicating a cumulative number of events for the respective and adjacent trace events, and (ii) a second counter indicating a cumulative amount of time between the respective and adjacent trace events; and determining one or more performance metrics associated with the respective and adjacent trace events in the time window using the first and second counters.