Performance Metering

This guide explains how to send performance metrics about the inmanta server to influxdb.

The inmanta server has a built-in pyformance instrumentation for all API endpoints and supports sending the results to influxdb.

Configuration summary

To enable performance reporting, set the options as found under influxdb in the server configuration file.

For example:

[influxdb]
# The hostname of the influxdb server
host = localhost
# The port of the influxdb server
port = 8086
# The name of the database on the influxdb server
name = inmanta
tags= environment=prod,az=a

Setup guide

1. To install influxdb, follow the instructions found at docs.influxdata.com.

2. Create a database to send the data to:

   influx
   CREATE DATABASE inmanta
   

3. Update the inmanta config file, add the following block

   [influxdb]
   # The hostname of the influxdb server
   host = localhost
   # The port of the influxdb server
   port = 8086
   # The name of the database on the influxdb server
   name = inmanta
   

4. Restart the inmanta server.

5. [optional] install grafana, follow the instructions found at https://grafana.com/grafana/download

6. [optional] load the inmanta dashboard found at https://grafana.com/grafana/dashboards/10089-inmanta-api-performance/

Reported Metrics

This section assumes familiarity with influxdb. See here.

All metrics are reported under the measurement metrics. Different measurements are distinguished by a tag called key.

Two main types of metrics are reported: 1. Metrics related to API performance 2. Others

API performance metrics

Each API method is reported with a key=rpc.{endpoint_name}. The endpoint_name is the server’s internal name for the endpoint.

To know which url corresponds to which method, please consult either

• the operationId field of the OpenAPI spec or

• the method names in inmanta.protocol.methods and inmanta.protocol.methods_v2

The fields available for each API endpoint are (cfr metrics timer):

field	type	description
15m_rate	float	fifteen-minute exponentially-weighted moving average of the request rate
5m_rate	float	five-minute exponentially-weighted moving average of the request rate
1m_rate	float	one-minute exponentially-weighted moving average of the request rate
mean_rate	float	mean of the request rate
min	float	minimal observed request latency
50_percentile	float	median (50 percentile) observed request latency
75_percentile	float	75 percentile observed request latency
95_percentile	float	95 percentile observed request latency
99_percentile	float	99 percentile observed request latency
999_percentile	float	999 percentile observed request latency
max	float	maximal observed request latency
avg	float	average observed latency
std_dev	float	standard deviation of the observed latency
count	float	number of calls seen since server start
sum	float	total wall-time spent executing this call since server start

Other Metrics

Key	Type	Unit	Description
self.spec.cpu	int	ns	The result of a small CPU benchmark, executed every second. Provides a baseline for machine performance.