spm-agent
v2.2.5
Published
Node.js agent framework for SPM by Sematext
Downloads
1,303
Readme
Sematext Agent Framework for Node.js
A framework for monitoring agents written in Node.js for sending metrics to Sematext Cloud or InfluxDB. The Sematext Node.js monitoring agent is built on top of this framework.
Functionality
- Sender interface to Sematext backend receivers
- Buffering metrics in case of network outages
- Reconnect after failures
- Logging functions
- Configuration handling
- Pluggable agents
Example to Implement a Monitoring Agent
Sematext Cloud supports the Influx Line Protocol for metric ingestion.
Agent modules must provide a start
and stop
function.
The function agent.addMetrics(metric)
collects a metric, which is shipped in bulk requests via Influx Line Protocol to the metrics receiver.
All metric objects, must have the mandatory fields measurement
, tags
and fields
:
- measurement - the name of the measurement e.g. 'process.memry'
- tags - key/value pairs, useful for filtering or aggregation of data
- fields - numeric fields with the measurement values
const SpmAgent = require('spm-agent')
const client = new SpmAgent()
const os = require('os')
// configure client, with non-default values
// or use the file ./.spmagentrc in YAML format
// the default configuration contains values for Sematext Cloud US
// SpmAgent.Config.tokens.spm = process.env.MONITORING_TOKEN
SpmAgent.Config.influx = {
dbName: 'metrics',
// change receiver to Sematext Cloud EU
// default is spm-receiver.sematext.com for US region
host: 'spm-receiver.eu.sematext.com',
port: 443,
protocol: 'https'
}
// a monitoring agent needs a start an stop function
class MemoryMonitor {
start (agent) {
this.agent = agent
// initialize your metrics collector ...
// Typically agents collect metrics periodically, every N seconds. The time between // two collection activities is the collectionInterval, specified in milliseconds.
this.tid = setInterval(function () {
const measurement = {
// measurment namespace
measurement: 'process.memory',
timestap: new Date(),
tags: {
role: 'frontend',
'os.host': os.hostname()
// The monitoring token can be set as 'tags.token' value
// Routing a metrics to a different monitoring app
// requires setting the `token` tag
// ,token = process.env.OTHER_APP_TOKEN
},
// metrics names and values
fields: {
rss: process.memoryUsage().rss
}
}
// pass metrics to sender schedule
agent.addMetrics(measurement)
}, SpmAgent.Config.collectionInterval)
}
stop () {
if (this.tid) {
clearInterval(this.tid)
}
}
}
client.createAgent(new SpmAgent.Agent(new MemoryMonitor()))
// log collected metrics to console
// observe all metrics for testing/debuggin
client.on('metrics', console.log)
// observe a specific metric using the measurment name
client.on('process.memory', console.log)
Metric Tags
The agent is able to use environment variables as metrics tags. Define a list of environment variables to be used as metric tags.
Let's assume this is the enviroment of the monitored application:
> env
TERM_PROGRAM=Apple_Terminal
TERM=xterm-256color
SHELL=/bin/bash
TMPDIR=/var/folders/v6/bh2q1xsn27g4d2g54z2ylv100000gn/T/
TERM_PROGRAM_VERSION=404.1
TERM_SESSION_ID=F12E0DBD-BF40-466D-85EC-08E89EDC3440
USER=stefan
PWD=/Users/stefan
HOME=/Users/stefan
LOGNAME=stefan
LC_CTYPE=UTF-8
DISPLAY=/private/tmp/com.apple.launchd.J60ybGpban/org.macosforge.xquartz:0
_=/usr/bin/env
You want to see later the performance metrics aggregated or filtered for specific user.
In this case the agent should collect the user name as tag. You can find the user name in the process environment variable USER
.
Let's also assume the workingdirectory of the application PWD
is another relevant identifier for your monitored application.
To configure the agent collecting USER and PWD as tags you can specify a commaseparated list of environment variables for the the agent configuration:
# add the values of env. variables USER, PWD as tags to metrics
export MONITORING_TAGS_FROM_ENV="USER, PWD"
The generated metrics include then the environment variable name and its value as tags. The result in Influx Line Protocol (ILP) format, produced by the agent for metric called swap
:
swap, USER=stefan,PWD=/Users/stefan out=0i,in=0i 1576765680000000000
Define static tags in key:value
format
export MONITORING_TAGS_FROM_ENV="organisation:sematext, service:api"
The result in in ILP format:
swap, organisation=sematext,service=api out=0i,in=0i 1576765680000000000
Both types of tags can be mixed:
export MONITORING_TAGS_FROM_ENV="organisation:sematext, service:api, USER, PWD"
The result in in ILP format:
swap, organisation=sematext,service=api,USER=stefan,PWD=/Users/stefan,os.host=imac.local out=0i,in=0i 1576765680000000000
The config file entry influx.tagsFromEnv
in .spmagenrc
works as well:
tokens:
spm: 'YOUR_MONITORING_TOKEN'
infra: 'YOUR_INFRA_MONITROING_TOKEN'
influx:
tagsFromEnv: 'organisation:sematext, USER, PWD'
dbName: 'metrics'
host: spm-receiver.sematext.com
protocol: https
port: 443
Contribute
Let us know about monitoring agents you need, maybe you like to contribute with your domain expertise!
Related Modules
- Node.js Monitoring
- spm-agent-nodejs
- Sematext Cloud - full stack observability platform