Node metrics

A Corda node exports a number of metrics for the purpose of monitoring the health of the node via JMX. These metrics are described below.

For more information on how to monitor a node, see Node administration, Node monitoring and logging, and Node monitoring scenarios.

Metric accuracy and performance

Histogram and timer-based metrics use internal data reservoirs to compute percentile-based metric data. By default, the metrics produced by Corda nodes rely on the reservoir implementation provided by the DropWizard metrics library, which is an exponentially decaying reservoir (EDR). These reservoirs sample the data and do not keep a full set of data over time. This can cause outlying data points to be missed if they are not included in the sample set.

Alternatively, a node can be configured to use a time window reservoir by adding the following configuration block to the node.conf file:

enterpriseConfiguration {
    metricsConfiguration {
        reservoirType = TIME_WINDOW // Can also be EDR, but this is the default if this item is absent
        timeWindow = <Duration> // Optional - will default to 5m if not specified
    }

The reservoirType configuration option can be set as either TIME_WINDOW or EDR, and will default to EDR if the configuration block is missing.

Time window data reservoirs record all data points across the specified time window. This ensures that outliers are accurately recorded for metrics calculation, and contain a bias towards more recent results.

Using time window reservoirs increases heap memory usage proportionally to the amount of load on the system.

Attachments

Caches

A Corda node maintains a number of caches. For each of the metrics below, the name of the cache must be supplied in the component field to show metrics for that cache.

There are two types of caches: size-based and weight-based. Size-based caches are measured in the number of entries in the cache, while weight-based caches are measured in the bytes of memory occupied by the entries.

Flows

Actions

Actions are reified IO actions to execute as part of state machine transitions. These metrics are only exposed when the relevant action gets executed for the first time.

Metering

P2P

Other metrics