Tripitaka

Node.js CI NPM version NPM downloads Maintainability Test Coverage Code Style Discover zUnit

Tripitaka is a low dependency, no frills logger, designed to play nicely with tools like fluentd and Elasticsearch. It is named after the buddhist monk from the TV series, Monkey due to shared values of simplicity and mindfulness, and also because Tripitaka is a term given to ancient collections of Buddhist scriptures, which loosely connects with logging. I wrote Tripitaka because, sadly my previous logger of choice, winston has fallen into disrepair.

TL;DR

const { Logger } = require("tripitaka");
const logger = new Logger();
const book = {
  title: "Monkey",
  author: "Wu Ch'eng-en",
  ISBN10: "9780140441116",
};
logger.info("Retrieved book", { book });

NODE_ENV=production node index.js
{"level":"INFO","message":"Retrieved book","book":{"title":"Monkey","author":"Wu Ch'eng-en","ISBN10":"9780140441116"},"timestamp":"2022-05-27T18:21:17.371Z"}

Design Principles

Tripitaka intentionally ships with only two transports. A streams-based transport which will write to stdout and stderr (or other streams which you supply), and an event emitter based transport which will emit events using the global process object (or another emitter which you supply). This library holds the opinion that external files, database and message brokers are all far better handled with a data collector such as fluentd, but you can write your own transports if you so wish. Tripitaka also eschews child loggers. These are useful for stashing context, but more elegantly implemented via AsyncLocalStorage or continuation-local-storage. See the express example for how.

Usage

Tripitaka supports the same logging levels as console, i.e.

The function arguments are always the same, a message and a context, e.g.

logger.info("How blissful it is, for one who has nothing", {
  env: process.env.NODE_ENV,
});

Assuming the default configuration, this will write the following to stdout when run in a production environment

{
  "env": "production",
  "message": "How blissful it is, for one who has nothing",
  "level": "INFO"
}

If you use the context processor (enabled by default), the context may be an Object, Array or Error. Both errors and array are automatically nested under configurable attributes, which default to “error” and “items” respectively, e.g.

logger.info("How blissful it is, for one who has nothing", [1, 2, 3]);
logger.error("I forbid it!", new Error("Oooh, Demons!"));

{"items":[1,2,3],"message":"How blissful it is, for one who has nothing","level":"INFO"}
{"error":{"message":"Oooh, Demons!","stack":"..."},"message":"Oooh, Demons!","level":"ERROR"}

If you use the empty processor (enabled by default), and you neglect to log a message, Tripitaka will report this

logger.info({ env: process.env.NODE_ENV });

{
  "message": "Empty message logged at Test._fn (/opt/acuminous/tripitaka/index.js:9:5)",
  "env": "production"
}

The exception to this is when you are just logging an Error, in which case the log record message will default to the error message e.g.

logger.error(new Error("Oooh, Demons!"));

{
  "error": { "message": "Oooh, Demons!", "stack": "..." },
  "message": "Oooh, Demons!",
  "level": "ERROR"
}

Customisation

You can customise this output through the use of processors and transports. By default Tripitaka ships with the following configuration.

const { Logger, Level, processors, transports } = require("tripitaka");
const { context, timestamp, json, human } = processors;
const { stream } = transports;

const logger = new Logger({
  level: Level.INFO,
  processors: [context(), timestamp(), process.env.NODE_ENV === "production" ? json() : human()],
  transports: [stream()],
});

Suppressing logs

You can suppress logs by setting the logging level as when you create a Logger instance as above, or by calling logger.disable(). You can re-enable the logger by calling logger.enable().

Processors

A processor is a function you can use to mutate the Tripitaka log record before it is delivered to the transports. Since processors are chained together in an array, the record can be mutated over a series of steps.

The processor is called with a single object containing the following properties:

name type notes
level Level  
message string  
ctx object  
record any Initialised to a shallow clone of the context. Be careful not to mutate nested attributes

example

const logger = new Logger({
  processors: [
    context(),
    ({ record }) => {
      return { ...record, timestamp: new Date() };
    },
    json(),
  ],
});

The out-of-the-box processors are as follows…

augment

Augments the record with the supplied source. If attributes are common to both the record and the source, the source wins. Use with AsyncLocalStorage as a substitute for child loggers. See the express example for how.

name type required default notes
source object or function yes    

Object example

Use an object when the source data is static

const source = { env: process.env.NODE_ENV };
const logger = new Logger({
  processors: [context(), augment({ source }), json()],
});
logger.info("How blissful it is, for one who has nothing");

{
  "env": "production",
  "message": "How blissful it is, for one who has nothing",
  "level": "INFO"
}

Function example

Use a function when the source data is dynamic

const source = () => ({ timestamp: new Date() });
const logger = new Logger({
  processors: [context(), augment({ source }), json()],
});
logger.info("How blissful it is, for one who has nothing");

{
  "timestamp": "2021-03-28T17:43:12.012Z",
  "message": "How blissful it is, for one who has nothing",
  "level": "INFO"
}

buffer

The buffer processor outputs the record as a buffer, optionally encoding it before doing so. For this processor to work, the record must previously have been converted to a string.

name type required default notes
inputEncoding string no    
outputEncoding string no    

example

const logger = new Logger({
  processors: [context(), json(), buffer({ outputEncoding: "hex" })],
});
logger.info("How blissful it is, for one who has nothing");

7b226c6576656c223a22494e464f222c226d657373616765223a22486f7720626c69737366756c2069742069732c20666f72206f6e652077686f20686173206e6f7468696e67227d

context

Performs a shallow copy of the context into the record. It also understands how to handle errors - without it they will not serialize correctly. It is best to put this processor first in the list of processors, as if another processor fires first, it may incorrectly handle the error object.

The processor operates with the following logic:

It has the following options:

name type required default notes
arrayField string no ‘items’ If the context is an instance of Array, it will be nested under an attribute with this name
errorField string no ‘error’ If the context is an instance of Error, it will be nested under an attribute with this name
stack boolean no true Controls whether the stack trace will be logged

example

const logger = new Logger({
  processors: [context({ errorField: "err", stack: false }), json()],
});
logger.error("I forbid it!", new Error("Oooh, Demons!"));

{
  "error": { "message": "Oooh, Demons!" },
  "message": "I forbid it!",
  "level": "ERROR"
}

empty

Logs a message when attempts are made to log an undefined, null or empty message

example

const logger = new Logger({
  processors: [context({ errorField: "err", stack: false }), empty(), json()],
});
logger.error(undefined);

{
  "message": "Empty message logged at Test._fn (/opt/acuminous/tripitaka/index.js:9:5)",
  "level": "ERROR"
}

human

Converts the record into a human readable form. Only intended for local use.

It has the following options:

name type required default notes
serializer function no null  
indent number no undefined  
decycler function no () => {} Determines how circular references are handled. The default behaviour is to silently drop the attribute
colours integer no undefined Defaults to auto detecting colour support. Specify 0 to disable, 1 for basic colours, 2 for 256 colour support or 3 for Truecolour

example

const logger = new Logger({
  processors: [context(), human()],
});
logger.info("How blissful it is, for one who has nothing", {
  pid: process.pid,
  ...process.memoryUsage(),
});

2021-03-28 18:15:23 INFO  How blissful it is, for one who has nothing
{
  "pid": 53072,
  "rss": 31997952,
  "heapTotal": 6938624,
  "heapUsed": 5361608,
  "external": 356316,
  "arrayBuffers": 25566
}

include

Copies the specified paths into a new log record. This is useful to avoid logging every property from a noisy object, including potentially senstive ones (I’m looking at you AxiosError!). Please note, this processor uses rfdc with ‘circles’ mode enabled so may be a little slow.

It has the following options:

name type required default notes
basePath string no   Specifies the base path to work from. Other properties will be copied verbatim
paths array no [] Specifies the paths (relative to any base path) of the fields to include
precondition function no () => true A function which must return true for the processor to run

example

const logger = new Logger({
  processors: [context(), include({ basePath: "error", paths: ["request.method", "request.url", "response.status"]), json()],
});
logger.error(httpError);

{
  "message": "NOT FOUND",
  "level": "ERROR",
  "error": {
    "message": "NOT FOUND",
    "request": {
      "method": "GET",
      "url": "http://httpbin.org/status/404"
    },
    "response": {
      "status": 404
    }
  }
}

Paths can reference arrays (e.g. items[0]), however the resulting document will still be yielded as an object, i.e.

const logger = new Logger({
  processors: [context(), include({ paths: ["items[1]", "items[2]"]), json()],
});
logger.info("How blissful it is, for one who has nothing", { items: ["a", "b", "c", "d"]});

{
  "items": {
    "1": "b",
    "2": "c"
  },
  "message": "How blissful it is, for one who has nothing",
  "level": "INFO"
}

index

Creates a sub document of simple values from the specified paths. This is useful to avoid mapping explosion when writing logs to Elasticsearch. The idea is to disable dynamic mapping by default in your Elasticsearch configuration, and specifically enable it only for the named sub document. Since the processor only copies fields with simple values into the index, you should remain in control of the Elasticsearch index, but still be able to search by key terms and inspect the full log context.

It has the following options:

name type required default notes
field string no ‘fields’ Specifies the name of the sub document
paths array no [] Specifies the paths of the fields to map
reportComplexTypes boolean no false Causes the processor to throw an error if value type is an object, function or symbol

NaN and Infinite values are always silently dropped as they could cause the field to by dynamically mapped as a string instead of a number.

example

const reportComplexTypes = process.env.NODE_ENV !== "production";
const logger = new Logger({
  processors: [context(), index({ field: "@fields", paths: ["character.name"], reportComplexTypes }), json()],
});
logger.info("How blissful it is, for one who has nothing", {
  character: { name: "Monkey", nature: "Irrepressible" },
});

{
  "message": "How blissful it is, for one who has nothing",
  "level": "INFO",
  "character": { "name": "Monkey", "nature": "Irrepresible" },
  "@fields": { "name": "Monkey" }
}

json

Uses json-stringify-safe to safely convert the Tripitaka record to a json string.

It has the following options:

name type required default notes
serializer function no null  
indent number no undefined  
decycler function no () => {} Determines how circular references are handled. The default behaviour is to silently drop the attribute

example

const logger = new Logger({
  processors: [context(), json()],
});
logger.info("How blissful it is, for one who has nothing", {
  env: process.env.NODE_ENV,
});

{
  "env": "production",
  "message": "How blissful it is, for one who has nothing",
  "level": "INFO"
}

timestamp

Adds a timestamp. It has the following options:

name type required default notes
field string no ‘timestamp’ Specifies the name of the timestamp attribute
getTimestamp function no () => new Date(); Overrides how the timestamp is aquired (useful for fixing the timestamp when testing)

example

const logger = new Logger({
  processors: [context(), timestamp({ field: "ts" }), json()],
});
logger.info("How blissful it is, for one who has nothing", {
  env: process.env.NODE_ENV,
});

{
  "ts": "2021-03-28T18:31:21.035Z",
  "env": "production",
  "message": "How blissful it is, for one who has nothing",
  "level": "INFO"
}

Transports

Transports are functions which write the Tripitaka record somewhere. The only parameter is an object, which should container the following properties.

name type notes
level Level  
record any Likely to be an object, string or a Buffer. It all depends on the processors you have selected

The available transports are

stream

The stream transport writes a string to an output stream based on the level. It has the following options:

name type required default notes
level Level no Level.TRACE The minimum log level for this transport
streams object no See notes By default TRACE, DEBUG and INFO messages will be output to stdout, while WARN and ERROR messages routed to stderr

example

const logger = new Logger({
  transports: [
    stream({
      streams: {
        [Level.TRACE.name]: process.stdout,
        [Level.DEBUG.name]: process.stdout,
        [Level.INFO.name]: process.stdout,
        [Level.WARN.name]: process.stdout,
        [Level.ERROR.name]: process.stderr,
      },
    }),
  ],
});
logger.info("How blissful it is, for one who has nothing", {
  env: process.env.NODE_ENV,
});

emitter

The emitter transport emits a Tripitaka record as an event, which can be useful when testing. It has the following options:

name type required default notes
level Level no Level.TRACE The minimum log level for this transport
emitter EventEmitter no process Specify your own event emitter rather than the global process object
events object no See notes By default all log levels will be emitted with the ‘log’ event. Think twice about changing this to ‘error’, since unhandled error events will kill your node process.

example

const logger = new Logger({
  transports: [
    emitter({
      events: {
        [Level.TRACE.name]: "log_trace",
        [Level.DEBUG.name]: "log_debug",
        [Level.INFO.name]: "log_info",
        [Level.WARN.name]: "log_warn",
        [Level.ERROR.name]: "log_error",
      },
    }),
  ],
});
logger.info("How blissful it is, for one who has nothing", {
  env: process.env.NODE_ENV,
});

Asynchronous Transports

If one or more of the transports is asynchronous and you want to ensure all messages have been written before terminating your application, you must wait for the logger.waitForTransports method to yield. This method takes an optional timeout specified in milliseconds. e.g.

process.once("SIGTERM", () => {
  logger
    .waitForTransports(1000)
    .then(() => {
      process.exit();
    })
    .catch((err) => {
      console.error(err);
      process.exit(1);
    });
});

Once you have called logger.waitForTransports any subsequent messages will be still be accepted and will prevent the promise from resolving until they have been processed. If your application logs intensively logger.waitForTransports could therefore block indefinitely unless you specify a timeout.