node package manager

Introducing npm Enterprise add-ons. Integrate third-party dev tools into npm…

monitr

Node process monitoring tool

monitr

Nodejs process monitoring module

This package is tested only with Node versions 0.10, 0.12 and 4.0.0.

Note: This module currently works only on Linux operating systems.

This module starts a separate thread within the Nodejs runtime that monitors and collects statistics about the running nodejs process. These statistics are then sent as JSON messages via UDP datagrams over a local domain socket.

Here is the list of data the module reports periodically:

 { status: 
     { pid: <pid of the node process>,
       ts: <current time stamp>,
       cluster: <process group id>,
       reqstotal: <total requests processed by this node process server>,
       utcstart: <when the process was started>,
       events: <number of new reports being processed since last stats reporting>,,
       cpu: <cpu usage>,
       mem: <memory usage>,
       cpuperreq: <cpu usage per request>,
       oreqs: <current open requests count>,
       sys_cpu: <system cpu load>,
       oconns: <current open connections count>,
       user_cpu: <user cpu load>,
       rps: <requests per second>,
       kbs_out: <kbs of data transferred since last stats reporting>,
       elapsed: <time elapsed since last event>,
       kb_trans: <total kbs of data transferred>,
       jiffyperreq: <cpu usage in terms of ticks per request>,
       gc: {
           scavenge: { count: <number>, elapsed_ms: <number>, max_ms: <number> },
           marksweep: { count: <number>, elapsed_ms: <number>, max_ms: <number> }
       }
    }
 }

It provides the running nodejs application with the ability to introspect garbage collection activity by creating read-only properties at process.monitor.gc that reports:

  1. count: number of times GC stop-the-world events occurred
  2. elapsed: cumulative time (in milliseconds) spent in GC

Installation

With npm do:

npm install monitr

Usage

var monitor = require('monitr');
monitor.start();

Spawns a thread and monitors the process. Writes process stats every second to the socket path.

monitor.stop();

Terminates the thread and closes the socket.

monitor.setIpcMonitorPath('/tmp/my-process-stats.mon');

Sets the datagram socket name to write the stats. Defaults to /tmp/nodejs.mon

Health Status

Monitr supports custom health functionality whereby the app can report its own health. The following methods are added to process.monitor to set and get the health information.

setHealthStatus(isDown, statusCode)
isDown()
getStatusCode()
getStatusTimestamp() - Return seconds when setHealthStatus was last called
getStatusDate() - Return Date object

Once setHealthStatus is invoked, the status json, described above, will have following additional fields.

health_status_timestamp: <timestamp when the setHealthStatus was invoked, in sec>,
health_is_down: <app is down or up, boolean>,
health_status_code: <health status code>

Handling HUP events

Monitr installs a custom SIGHUP handler which will optionally print out a NodeJS stack backtrace of the Javascript currently being executed. This can be useful for debugging where a NodeJS process may be stuck.

Due to the method by which the stack backtrace is obtained (installing a v8 Javascript engine debug event listener and simulating a debugger event), there is a performance slowdown for code running while the stack backtrace option is active.

By default the backtrace option is not enabled. You can enable it by setting the showBacktrace property to true, e.g.

monitor.showBacktrace = true

Setting monitor.showBacktrace to false will restore the original performance by removing the debug event listener.

Implementation

It looks up /proc/* files on the system to report CPU Usage. It looks up /proc/pid/* files on the system to report its own stats. process.monitor.* methods are set by lib/monitor.js.

It calls the process.monitor.* methods to report total requests since monitoring started (reqstotal), current requests in flight (oreqs), current open connections (oconns) and total data returned since monitoring started (kb_trans). Note: oreqs may be greater than oconns when keepalive is enabled.

It attaches to the v8 garbage collection hooks to instrument (for each GC type) the following stats for each reporting interval.

  1. count : number of times GC type invoked
  2. elapsed_ms: total elapsed time nodejs thread is blocked
  3. max_ms: maximum time spent blocked by any one GC event

Example

Please refer to examples/README.md for examples showing the use of these functions.

Build Status

Node Badge