About topic statistics¶
Table of Contents
ROS 2 provides the integrated measurement of statistics for messages received by any subscription. Allowing a user to collect subscription statistics enables them to characterize the performance of their system or aid in diagnosis of any present issues.
The measurements provided are the received message age and received message period. For each measurement the statistics provided are the average, maximum, minimum, standard deviation, and sample count. These statistics are calculated in a moving window.
Each statistic set is calculated in constant time and constant memory by using the utilities implemented in the libstatistics_collector package. When a new message is received by a subscription, this is a new sample for calculation in the current measurement window. The average calculated is simply a moving average. The maximum, minimum,and sample count are updated upon receipt of each new sample, whereas the standard deviation is calculated using Welford’s online algorithm.
Received message period
Uses the system clock to measure the period between received messages
Received message age
Requires a message to have a timestamp populated in the header field in order to calculate the age of the message as sent from a publisher.
By default, Topic Statistics measurements are not enabled. After enabling this feature for a specific node via the subscription configuration options, both received message age and received message period measurements are enabled for that specific subscription.
The data is published as a statistics_msg/msg/MetricsMessage
at a configurable period (default 1 second) to a configurable topic (default
Note that the publishing period also serves as the sample collection window period.
Since received message period requires a message timestamp in a header field, empty data is published. That is, all statistics values are NaN if no timestamp is found. Publishing NaN values instead of not publishing at all avoids the absence of a signal problem and is meant to explicitly show that a measurement could not be made.
The first sample of each window for the received message period statistic does not yield a measurement. This is because calculating this statistic requires knowing the time the previous message arrived, so subsequent samples in the window yield measurements.
Similar to ROS 1 Topic Statistics, both message age and message period are calculated, albeit from the subscription side. Other ROS 1 metrics, e.g., the number of dropped messages or traffic volume, are currently not provided.