Correlating failures with performance in application telemetry data

Computer readable storage hardware storing instructions configured to cause a computing device to perform a process, the process comprising:

• receiving a telemetry stream comprised of telemetry records representing respective performances of an operation by an application, each telemetry record comprising a respective measure of performance of an invocation of the operation, each telemetry record comprising a respective failure indication indicating whether a respective invocation of the operation succeeded or failed, and each telemetry record comprising a time corresponding to a respective invocation of the operation;
• parsing the telemetry stream to compute an aggregate performance measure for a segment of the telemetry stream based on the measures of performance in the segment;
• based on the aggregate performance measure, determining that the segment corresponds to a period of performance degradation of the application;
• computing an aggregate failure rate for the segment of the telemetry stream from the failure indications thereof;
• computing an aggregate failure performance from the measures of telemetry records in the segment whose failure indications indicate failures, and computing an aggregate success performance from the measures of telemetry records in the segment whose failure indications indicate success;
• and
• based on the aggregate failure rate, the aggregate failure performance, and the aggregate success performance, determining whether the period of performance degradation is positively or negatively correlated with failures of the operation during the period of performance degradation.

A computing device comprising:

• processing hardware;
• storage hardware storing instructions configured to cause the processing hardware to perform a process comprising:
• receiving a telemetry stream comprising entries that respectively represent invocations of an operation by an application;
• identifying performance-degraded portions of a telemetry stream, wherein a performance-degraded portion is identified by determining that the entries in the performance-degraded portion collectively indicate degraded performance of the application;
• computing collective success-performance measures for the performance-degraded portions, respectively, and computing collective failure-performance measures for the performance-degraded portions, respectively, wherein a success-performance measure for a performance-degraded portion is computed from entries thereof that indicate success of the operation, and wherein a failure-performance measure for a performance-degraded portion is computed from entries thereof that indicate failure of the operation;
• computing failure rates for the performance-degraded portions, wherein a failure rate for a performance-degraded portion is computed by determining which entries therein indicate failure of the operation;
• and
• according to the collective success-performance measures, the collective failure-performance measures, and the failure rates, determining correlations between performance degradations of the application for performance-degraded portions, respectively, and failures of the operation.