Multivariate auto-differential correlation engine in a computing environment

1. A computer-implemented method implemented in a database system within an on-demand computing services environment, the method comprising:

retrieving from the database system a metric data stream for a plurality of metrics, each metric measuring a respective computing characteristic, the metric data stream including a plurality of values for each of a sequence of time intervals, each value corresponding with a respective one of the metrics;

determining, using one or more processors, a plurality of metric correlation matrices for the metrics, each metric correlation matrix associated with a respective time period in the metric data stream, each time period including a respective plurality of the time intervals;

selecting, using one or more processors, a subset of comparison metric correlation matrices from the plurality of metric correlation metric matrices;

determining via a processor a plurality of anomaly scores each corresponding with a respective time period, the plurality of anomaly scores including a designated anomaly score for a designated time period determined by comparing a designated metric correlation matrix associated with the designated time period with the selected subset of comparison metric correlation metric matrices;

determining a plurality of moving average anomaly scores, each moving average anomaly score representing a respective average of the anomaly scores associated with a respective time window; and

when it is determined that the designated anomaly score exceeds a designated threshold, storing the designated anomaly score in the database system.

2. The computer-implemented method recited in claim 1 , the method further comprising:

performing seasonality detection on the metric stream to identify a plurality of season periods, one or more of the metric values varying systematically with the season periods.

3. The computer-implemented method recited in claim 2 , the method further comprising:

determining a respective annotation value for each of the identified season periods via dynamic historic window annotation, the annotation value being selected from a set of season period classification values.

4. The computer-implemented method recited in claim 3 , wherein each of the subset of comparison metric correlation matrices corresponds to a respective one of the identified season periods.

5. The computer-implemented method recited in claim 4 , wherein the subset of metric correlation matrices are selected based on the annotation values.

6. The computer-implemented method recited in claim 2 , wherein performing seasonality detection comprises applying seasonal and trend decomposition using Loess.

7. The computer-implemented method recited in claim 1 , wherein each of the metrics corresponds to a respective value measured at a computing system.

8. The computer-implemented method recited in claim 7 , wherein the computing system represents a portion of the on-demand computing services environment, the on-demand computing services environment configured to provide computing services to a plurality of service recipients via the internet.

9. The computer-implemented method recited in claim 1 , wherein the database system includes a multitenant database storing information associated with a plurality of tenants.

10. The computer-implemented method recited in claim 1 , wherein the database system includes a dynamic-schema database in which database column attributes are dynamically determined based on object relational mapping.

11. A computing system within an on-demand computing services environment, the computing system configured to perform a method comprising:

retrieving from the database system a metric data stream for a plurality of metrics, each metric measuring a respective computing characteristic, the metric data stream including a plurality of values for each of a sequence of time intervals, each value corresponding with a respective one of the metrics;

determining, using one or more processors, a plurality of metric correlation matrices for the metrics, each metric correlation matrix associated with a respective time period in the metric data stream, each time period including a respective plurality of the time intervals;

selecting, using one or more processors, a subset of comparison metric correlation matrices from the plurality of metric correlation metric matrices;

determining via a processor a plurality of anomaly scores each corresponding with a respective time period, the plurality of anomaly scores including a designated anomaly score for a designated time period determined by comparing a designated metric correlation matrix associated with the designated time period with the selected subset of comparison metric correlation metric matrices;

determining a plurality of moving average anomaly scores, each moving average anomaly score representing a respective average of the anomaly scores associated with a respective time window; and

when it is determined that the designated anomaly score exceeds a designated threshold, storing the designated anomaly score in the database system.

12. The computing system recited in claim 11 , the method further comprising:

performing seasonality detection on the metric stream to identify a plurality of season periods, one or more of the metric values varying systematically with the season periods.

13. The computing system recited in claim 12 , the method further comprising:

determining a respective annotation value for each of the identified season periods via dynamic historic window annotation, the annotation value being selected from a set of season period classification values.

14. The computing system recited in claim 13 , wherein each of the subset of comparison metric correlation matrices corresponds to a respective one of the identified season periods.

15. The computing system recited in claim 14 , wherein the subset of metric correlation matrices are selected based on the annotation values.

16. The computing system recited in claim 12 , wherein performing seasonality detection comprises applying seasonal and trend decomposition using Loess.

17. The computing system recited in claim 11 , wherein each of the metrics corresponds to a respective value measured at a computing system.

18. The computing system recited in claim 17 , wherein the computing system represents a portion of an on-demand computing services environment configured to provide computing services to a plurality of service recipients via the internet.

19. One or more non-transitory computer readable media having instructions stored thereon for performing a computer-implemented method implemented in a database system within an on-demand computing services environment, the method comprising:

retrieving from the database system a metric data stream for a plurality of metrics, each metric measuring a respective computing characteristic, the metric data stream including a plurality of values for each of a sequence of time intervals, each value corresponding with a respective one of the metrics;

determining, using one or more processors, a plurality of metric correlation matrices for the metrics, each metric correlation matrix associated with a respective time period in the metric data stream, each time period including a respective plurality of the time intervals;

selecting, using one or more processors, a subset of comparison metric correlation matrices from the plurality of metric correlation metric matrices;

determining via a processor a plurality of anomaly scores each corresponding with a respective time period, the plurality of anomaly scores including a designated anomaly score for a designated time period determined by comparing a designated metric correlation matrix associated with the designated time period with the selected subset of comparison metric correlation metric matrices;

determining a plurality of moving average anomaly scores, each moving average anomaly score representing a respective average of the anomaly scores associated with a respective time window; and

when it is determined that the designated anomaly score exceeds a designated threshold, storing the designated anomaly score in the database system.