Optimized unbiased statistical analysis of partially sampled traces without completeness information

A computer-implemented method for reporting transaction trace data for a computer transaction executing in a distributed computing environment, comprising:

• receiving, by an agent, span data from a sensor instrumented in a given method executed by a monitored computer transaction, where the span data describes a portion of execution of the monitored computer transaction performed by the given method and includes a unique identifier for the monitored computer transaction;
• retrieving, by the agent, a shared sampling number for the monitored computer transaction from a data store;
• randomly selecting, by the agent, a value for the shared sampling number and storing the value for the shared sampling number in response to the shared sampling number not being present in the data store, where value for the shared sampling number is randomly selected from a limited set of values;
• detecting, by the agent, an event of the monitored computer transaction that crosses an execution boundary of a thread, a process or a host computing system and making the unique identifier and the shared sampling number for the monitored computer transaction accessible to other agents in response to detecting said event;
• determining, by the agent, a sampling probability for the span data, where the sampling probability defines a percentage of span data reported by the agent and a value for the sampling probability is selected from the limited set of values;
• comparing, by the agent, the shared sampling number to the sampling probability;
• appending, by the agent, the sampling probability to the span data;
• and
• sending, by the agent, the span data as a sampled span data record via a network to a monitoring server, where the sampled span data record is sent to the monitoring server in response to the shared sampling number being less than the sampling probability.