memfaultd Built-in Metrics

Built-in System Metric Collection

With built-in system Metric collection enabled, memfaultd captures readings on the state and health of your Device which are aggregated as part of Metric Reports uploaded to Memfault.

When built-in system Metric collection is enabled, readings from collectd that share a top-level namespace with one of the below Metric groups will be dropped. This is to avoid double-counting or aggregating readings that may be calculated differently.

Aggregation Types

Like StatsD Metric readings, memfaultd uses different aggregation types internally when aggregating Metrics it collects itself. The different aggregation types are listed below:

• Histogram: An average of all readings captured within a Metric Report, weighted evenly. Some Histogram Metrics also report minimum and maximum values -- see Min/Max Metrics below.
• Gauge: Stores the most recent value for a Metric. When a new reading is received, the old value is discarded and replaced with the new reading's value.
• Counter: A simple monotonically increasing sum. New readings within the duration of a Metric Report are added to the sum of all previous readings captured within that Metric Report.
• String: Stores a string value as a report tag. Used to capture identifying information such as Device names or OUI values.

Min/Max Metrics

For Histogram Metrics, memfaultd normally reports only the average of all readings captured during a Metric Report interval. When min/max reporting is enabled for a Metric, memfaultd will emit three values per interval instead:

• <metric_name> -- the average of all readings
• <metric_name>_min -- the minimum reading observed
• <metric_name>_max -- the maximum reading observed

This is useful for detecting spikes or drops that would otherwise be hidden by averaging -- for example, a brief CPU spike to 100% that averages out to 40% over the heartbeat interval.

Default min/max Metrics

The following built-in Histogram Metrics report min/max values automatically:

Metric pattern	Description
cpu_usage_pct	Overall CPU usage
memory_pct	Overall memory usage
cpu_usage_<process>_pct	Per-process CPU usage
memory_<process>_pct	Per-process memory usage
interface/<interface>/bytes_per_second/rx	Network interface RX throughput
interface/<interface>/bytes_per_second/tx	Network interface TX throughput
thermal/*	All thermal zone metrics

Adding custom min/max Metrics

You can configure additional Histogram Metrics to report min/max values using the metrics.min_max_metrics configuration option. For example, to track min/max for custom application Metrics:

{
  "metrics": {
    "min_max_metrics": ["my_custom_metric", "my_service_latency"]
  }
}

This will cause my_custom_metric and my_service_latency to each emit _min and _max values alongside their averages in every Metric Report. These custom entries are merged with the default min/max metrics listed above.

note

Min/max reporting only applies to Histogram Metrics. Gauge and Counter Metrics are not affected by this setting.

CPU

Metric	Description	Aggregation Type
cpu/cpu/percent/idle	The percent of time this core spent in the idle state	Histogram
cpu/cpu/percent/system	The percent of time this core spent in the system state	Histogram
cpu/cpu/percent/user	The percent of time this core spent in the user state	Histogram
cpu/cpu/percent/iowait	The percent of time this core spent in the iowait state	Histogram
cpu/cpu/percent/irq	The percent of time this core spent in the irq state	Histogram
cpu/cpu/percent/softirq	The percent of time this core spent in the softirq state	Histogram
cpu/cpu/percent/nice	The percent of time this core spent in the nice state	Histogram

Memory

Metric	Description	Aggregation Type
memory/memory/free	The amount of free memory on the system in bytes	Histogram
memory/memory/used	The amount of used memory on the system in bytes	Histogram
memory/memory/cached	The amount of cached memory on the system in bytes	Histogram
memory/memory/buffered	The amount of buffered memory on the system in bytes	Histogram
memory/memory/slab_recl	The amount of reclaimable slab memory on the system in bytes	Histogram
memory/memory/slab_unrecl	The amount of unreclaimable slab memory on the system in bytes	Histogram

Virtual Memory

Metric	Description	Aggregation Type
memory/vm/swaps_in_per_second	Pages swapped in per second from /proc/vmstat	Histogram
memory/vm/swaps_out_per_second	Pages swapped out per second from /proc/vmstat	Histogram
memory/vm/pages_in_per_second	Pages paged in per second from /proc/vmstat	Histogram
memory/vm/pages_out_per_second	Pages paged out per second from /proc/vmstat	Histogram

Temperature

Metric	Description	Aggregation Type
thermal/<thermal zone type>/temp	Each thermal zone temperature reading in degrees Celsius	Histogram

Network Interfaces

By default, memfaultd will detect interfaces whose name does not start with lo, tun, dummy, veth, or usb (to avoid automatically tracking virtual interfaces) and monitor them via the following metrics.

A specific set of interfaces can be specified via the metrics.system_metric_collection.network_interfaces configuration.

Metric	Description	Aggregation Type
interface/<interface>/bytes_per_second/rx	Bytes per second received on this interface	Histogram
interface/<interface>/bytes_per_second/tx	Bytes per second sent on this interface	Histogram
interface/<interface>/errors_per_second/rx	Errors per second for RX traffic on this interface	Histogram
interface/<interface>/errors_per_second/tx	Errors per second for TX traffic on this interface	Histogram
interface/<interface>/dropped_per_second/rx	Packets dropped per second for RX traffic on this interface	Histogram
interface/<interface>/dropped_per_second/tx	Packets dropped per second for TX traffic on this interface	Histogram
interface/<interface>/packets_per_second/rx	Packets received per second on this interface	Histogram
interface/<interface>/packets_per_second/tx	Packets sent per second on this interface	Histogram
interface/<interface>/total_bytes/rx	Total bytes received on this interface since the last reading	Counter
interface/<interface>/total_bytes/tx	Total bytes sent on this interface since the last reading	Counter
interfaces/<interface>/rssi	RSSI of the connection on this interface to a connected AP/router (wireless only)	Histogram

Wireless Networks

Metric	Description	Aggregation Type
wireless/oui/local_<device>	Local OUI (Organizationally Unique Identifier) for the specified Device	String
wireless/oui/ap_<device>	Access Point OUI (Organizationally Unique Identifier) associated with the specified local Device	String

Process-level Metrics

memfaultd will capture the following metrics for processes specified in the metrics.system_metric_collection.processes configuration.

note

<name> is the filename of the process's executable. If it is longer than 16 characters, it will be truncated to the first 16 characters in the filename.

Metric	Description	Aggregation Type
processes/<name>/rss_bytes	Resident set size of the process in bytes	Histogram
processes/<name>/vm_bytes	Virtual memory size of the process in bytes	Histogram
processes/<name>/num_threads	Number of threads in the process	Histogram
processes/<name>/cpu/percent/user	Percent of CPU time the process spent in user mode	Histogram
processes/<name>/cpu/percent/system	Percent of CPU time the process spent in kernel mode	Histogram
processes/<name>/pagefaults/minor	Minor page faults (not requiring disk I/O) since the last reading	Histogram
processes/<name>/pagefaults/major	Major page faults (requiring disk I/O) since the last reading	Histogram

Disk Space Metrics

By default, memfaultd will detect disks whose ID starts with /dev (excluding loop and ram devices) and track their disk space with these metrics.

Other disks or partitions can be specified via the metrics.system_metric_collection.disk_space configuration.

Metric	Description	Aggregation Type
disk_space/<disk>/free_bytes	Bytes on the corresponding disk or partition that are unused	Histogram
disk_space/<disk>/used_bytes	Bytes on the corresponding disk or partition that are in use	Histogram

Diskstats Metrics

By default, memfaultd will collect disk I/O statistics from /proc/diskstats for all Devices (excluding loop and ram Devices). A specific set of Devices can be specified via the metrics.system_metric_collection.diskstats configuration.

Metric	Description	Aggregation Type
diskstats/<device>/reads_per_second	Read operations per second on this Device	Histogram
diskstats/<device>/writes_per_second	Write operations per second on this Device	Histogram

MMC Disk Metrics

For eMMC/MMC block Devices (Devices with names matching mmcblk*), memfaultd collects additional disk health and identification Metrics.

Metric	Description	Aggregation Type
diskstats/<disk>/lifetime_remaining_pct	Estimated lifetime remaining percentage (type A)	Gauge
diskstats/<disk>/lifetime_b_remaining_pct	Estimated lifetime remaining percentage (type B)	Gauge
diskstats/<disk>/bytes_written	Bytes written to this Device since the last reading	Counter
diskstats/<disk>/total_size_bytes	Total size of the Device in bytes	Gauge
diskstats/<disk>/name	Product name of the Device	String
diskstats/<disk>/manufacturer_id	Manufacturer ID of the Device	String
diskstats/<disk>/manufacture_date	Manufacture date of the Device	String
diskstats/<disk>/revision	Firmware Revision of the Device	String
diskstats/<disk>/serial	Serial number of the Device	String

Log Counter Metrics

memfaultd includes a number of built-in Log Filtering rules that increment counter metrics when a matching log line is received.

Metric	Description	Aggregation Type
oomkill_<process_name>	Count of times process_name has been killed by the OOM killer	Counter
systemd_restarts_<service_name>	Count of times the service_name service has been restarted	Counter

note

This metrics require the logging feature to be enabled and for logs to be sent to memfaultd.