Skip to content

Latest commit

 

History

History
305 lines (244 loc) · 9.8 KB

help.md

File metadata and controls

305 lines (244 loc) · 9.8 KB
Raw

Help

Log rotation

Use a separate tool for log rotation: We recommend logrotate. Consider we output our logs to /var/log/myapp.log like so:

$ node server.js > /var/log/myapp.log

We would rotate our log files with logrotate, by adding the following to /etc/logrotate.d/myapp:

/var/log/myapp.log {
       su root
       daily
       rotate 7
       delaycompress
       compress
       notifempty
       missingok
       copytruncate
}

The copytruncate configuration has a very slight possibility of lost log lines due to a gap between copying and truncating - the truncate may occur after additional lines have been written. To perform log rotation without copytruncate, see the Reopening log files help.

Reopening log files

In cases where a log rotation tool doesn't offer a copy-truncate capabilities, or where using them is deemed inappropriate, pino.destination is able to reopen file paths after a file has been moved away.

One way to use this is to set up a SIGUSR2 or SIGHUP signal handler that reopens the log file destination, making sure to write the process PID out somewhere so the log rotation tool knows where to send the signal.

// write the process pid to a well known location for later
const fs = require('fs')
fs.writeFileSync('/var/run/myapp.pid', process.pid)

const dest = pino.destination('/log/file')
const logger = require('pino')(dest)
process.on('SIGHUP', () => dest.reopen())

The log rotation tool can then be configured to send this signal to the process after a log rotation event has occurred.

Given a similar scenario as in the Log rotation section a basic logrotate config that aligns with this strategy would look similar to the following:

/var/log/myapp.log {
       su root
       daily
       rotate 7
       delaycompress
       compress
       notifempty
       missingok
       postrotate
           kill -HUP `cat /var/run/myapp.pid`
       endscript
}

Saving to multiple files

See pino.multistream.

Log Filtering

The Pino philosophy advocates common, pre-existing, system utilities.

Some recommendations in line with this philosophy are:

  1. Use grep: 
    $ # View all "INFO" level logs
$ node app.js | grep '"level":30'
  2. Use jq: 
    $ # View all "ERROR" level logs
$ node app.js | jq 'select(.level == 50)'

Transports and systemd

systemd makes it complicated to use pipes in services. One method for overcoming this challenge is to use a subshell:

ExecStart=/bin/sh -c '/path/to/node app.js | pino-transport'

Log to different streams

Pino's default log destination is the singular destination of stdout. While not recommended for performance reasons, multiple destinations can be targeted by using pino.multistream.

In this example we use stderr for error level logs and stdout as default for all other levels (e.g. debug, info, and warn).

const pino = require('pino')
var streams = [
  {level: 'debug', stream: process.stdout},
  {level: 'error', stream: process.stderr},
  {level: 'fatal', stream: process.stderr}
]

const logger = pino({
  name: 'my-app',
  level: 'debug', // must be the lowest level of all streams
}, pino.multistream(streams))

How Pino handles duplicate keys

Duplicate keys are possibly when a child logger logs an object with a key that collides with a key in the child loggers bindings.

See the child logger duplicate keys caveat for information on this is handled.

Log levels as labels instead of numbers

Pino log lines are meant to be parseable. Thus, Pino's default mode of operation is to print the level value instead of the string name. However, while it is possible to set the useLevelLabels option, we recommend using one of these options instead if you are able:

  1. If the only change desired is the name then a transport can be used. One such transport is pino-text-level-transport.
  2. Use a prettifier like pino-pretty to make the logs human friendly.

Pino with debug

The popular debug is used in many modules across the ecosystem.

The pino-debug module can capture calls to debug loggers and run them through pino instead. This results in a 10x (20x in asynchronous mode) performance improvement - even though pino-debug is logging additional data and wrapping it in JSON.

To quickly enable this install pino-debug and preload it with the -r flag, enabling any debug logs with the DEBUG environment variable:

$ npm i pino-debug
$ DEBUG=* node -r pino-debug app.js

pino-debug also offers fine grain control to map specific debug namespaces to pino log levels. See pino-debug for more.

Unicode and Windows terminal

Pino uses sonic-boom to speed up logging. Internally, it uses fs.write to write log lines directly to a file descriptor. On Windows, unicode output is not handled properly in the terminal (both cmd.exe and powershell), and as such the output could be visualized incorrectly if the log lines include utf8 characters. It is possible to configure the terminal to visualize those characters correctly with the use of chcp by executing in the terminal chcp 65001. This is a known limitation of Node.js.

Mapping Pino Log Levels to Google Cloud Logging (Stackdriver) Severity Levels

Google Cloud Logging uses severity levels instead log levels. As a result, all logs may show as INFO level logs while completely ignoring the level set in the pino log. Google Cloud Logging also prefers that log data is present inside a message key instead of the default msg key that Pino uses. Use a technique similar to the one below to retain log levels in Google Cloud Logging

const pino = require('pino')

// https://cloud.google.com/logging/docs/reference/v2/rest/v2/LogEntry#logseverity
const PinoLevelToSeverityLookup = {
  trace: 'DEBUG',
  debug: 'DEBUG',
  info: 'INFO',
  warn: 'WARNING',
  error: 'ERROR',
  fatal: 'CRITICAL',
};

const defaultPinoConf = {
  messageKey: 'message',
  formatters: {
    level(label, number) {
      return {
        severity: PinoLevelToSeverityLookup[label] || PinoLevelToSeverityLookup['info'],
        level: number,
      }
    },
    log(message) {
      return { message }
    }
  },
}

module.exports = function createLogger(options) {
  return pino(Object.assign({}, options, defaultPinoConf))
}

Avoid Message Conflict

As described in the message documentation, when a log is written like log.info({ msg: 'a message' }, 'another message') then the final output JSON will have "msg":"another message" and the 'a message' string will be lost. To overcome this, the logMethod hook can be used:

'use strict'

const log = require('pino')({
  level: 'debug',
  hooks: {
    logMethod (inputArgs, method) {
      if (inputArgs.length === 2 && inputArgs[0].msg) {
       inputArgs[0].originalMsg = inputArgs[0].msg
      }
      return method.apply(this, inputArgs)
    }
  }
})

log.info('no original message')
log.info({ msg: 'mapped to originalMsg' }, 'a message')

// {"level":30,"time":1596313323106,"pid":63739,"hostname":"foo","msg":"no original message"}
// {"level":30,"time":1596313323107,"pid":63739,"hostname":"foo","msg":"a message","originalMsg":"mapped to originalMsg"}

Exit logging (deprecated for Node v14+)

In pino v7, The following piece of documentation is not needed in Node v14+ and it will emit a deprecation notice.

When a Node process crashes from uncaught exception, exits due to a signal, or exits of it's own accord we may want to write some final logs – particularly in cases of error.

Writing to a Node.js stream on exit is not necessarily guaranteed, and naively writing to an asynchronous logger on exit will definitely lead to lost logs.

To write logs in an exit handler, create the handler with pino.final:

process.on('uncaughtException', pino.final(logger, (err, finalLogger) => {
  finalLogger.error(err, 'uncaughtException')
  process.exit(1)
}))

process.on('unhandledRejection', pino.final(logger, (err, finalLogger) => {
  finalLogger.error(err, 'unhandledRejection')
  process.exit(1)
}))

The finalLogger is a special logger instance that will synchronously and reliably flush every log line. This is important in exit handlers, since no more asynchronous activity may be scheduled.