Concept Guide

Document 365/385
Metric reports are generated and values are added to the report at the rate they are produced by the backend
services. Reports that contain metrics with different reporting characteristics will have different numbers of
metric values in the resulting reports that match the rate at which the backend daemons report data for these
metrics.
There may a variance in Metric Value count variance because the rate at which metrics are ingested is
clocked by backend services reporting the metrics. There is some expected variance in the number of metric
values that appear in a report above what you might expect purely by doing the math of “RecurrenceInterval” /
SensorInterval”. Thus, a one minute report that has a metric with a five second sensor interval will not
necessarily have exactly 12 entries.
The Metric Reports will not return any metric that has a NULL or invalid value regardless of the suppression
or heartbeat implementation. Periodic reports configured with no suppress repeated metrics option shall
stream last read data if the server is powered off.
Custom reports are limited to a total of 48 total report definitions, including 24 pre-canned report definitions
and potentially 24 custom report definitions.
Reports with a specific configuration:
NVMeSMARTData - NVMeSMARTData is only supported for SSD (PCIeSSD/NVMe Express) drives
with PCIe bus protocol (not behind SWRAID).
StorageDiskSMARTData report is only supported for SSD drives with SAS/SATA bus protocol and
behind the BOSS controller.
StorageSensor report is only supported for the drives in non-raid mode and not behind the BOSS
controller.
GPGPUStatistics report is only available in specific GPGPU models that support ECC memory
capability (GP102GL [Tesla P40]).
FanSensor report gets generated only for Monolithic servers. For modular servers, the report is empty
(with "MetricValues@odata.count": 0).
When server is powered off only these sensor readings are available: PSU Temperature, System
Board Inlet, and Exhaust Temperatures on monolithic servers; only System Board Inlet and Exhaust
temperatures on modular servers.
When a report is enabled but the device hardware is not present, no report is generated. For
instance, if a GPU card is not present in the system and the GPUMetrics report is pulled, the result
would be an empty report with "MetricValues@odata.count": 0.
When report RecurrenceInterval set to 0s, the report can only be pulled and it cannot be streamed.
The report will be single instance and non-repeating data, if available at that time of pull.
If some metrics common across multiple report definitions and SuppressRepeatedMetricValue set to
true and MetricReportHeartbeatInterval set to either less than or greater than the ReportTimeSpan on
any report then the metric suppression behavior changes and the common metrics are not
suppressed.
If a custom report created with metrics with different sensing intervals, the report would contain only
the metrics with lower sensing interval, for example setting the RecurrenceInterval less than the lower
sensing interval of the metrics.