filog

FiLog: a Meteor 1.9 logging package
===================================


FiLog is a highly configurable isomorphic logger for Meteor applications,
supporting configurable logging strategies and custom processing and log sending
destinations.

Out of the box, it can log to the console (browser, server), the Meteor database,
any Syslog server, and includes a "tee" logger to send to multiple destinations.
On the browser, it can log to its own Meteor server over DDP, or to a
centralized FiLog log aggregation server over HTTP.

FiLog is based upon low-level NPM packages::

* Stack capture and standardization: https://www.npmjs.com/package/tracekit
* Meteor user capture in call stack: https://www.npmjs.com/package/callsite

Read the Documentation.

Configuration and usage
-----------------------

#### Main ideas:

* A logged event is a (level, message, context) triplet, in which
* level is an RFC5424 number (0..7)
* message is a plain string or can be converted to one automatically
* context is a plain Object
* the package has no configuration of its own, it is entirely
configured by its users, the client- and server-side Meteor parts of the
application.
* Both client and server applications may capture exception stack traces and log
them for plain errors (window.onerror handling) and exception reporting.
This is triggered by the logger.arm() method, and can be disabled by the
logger.disarm() method.

#### Typical use case:

- client/main.js:
- instantiates any number of "processors": instances of classes derived from
ProcessorBase. These are able to add information to a message on its way
to the logging destination. Current concrete processors are:
- BrowserProcessor: adds browser-related information, such as user
agent, platform, operating system, client memory
- MeteorUserProcessor: adds user-related information if any is
available. Assumes the application is configured with some Meteor
accounts package, like fgm:accounts-drupal.
- RoutingProcessor: adds request-related information, such as the
path being served.
- instantiates any number of "senders": instances of classes derived from
SenderBase. These are able to receive a (level, message, context)
triplet ("event") and forward it somwhere else. Current concrete senders
are:
- ConsoleSender: output the message using console.log
- MeteorClientHttpSender: HTTP POST the message to the Meteor server
present at the URL defined by the Meteor-standard ROOT_URL
environment variable.
- MongodbSender: store the event in a collection in the Meteor MongoDB
database instance (or the minimongo on the client).
- NullSender: ignore the message.
- SyslogSender: send the event to syslog, on server only: including it on
client will fail.
- TeeSender: send the event to all sender instances passed to its
constructor as an array. Useful to send logs to multiple destinations.
- instantiate a sending "strategy": instance of a class derived from
StrategyBase. There are able to decide, based on an event, where it
should be sent by available senders. They may also modify the logger
instance at the last step of its construction, Two concrete strategies are
currently available:
- LeveledStrategy: based on the level of the message, it defines three
severity levels: low, medium, and high, as well as the breakpoints
between them in terms of RFC5424 levels, and associates a sender
instance with each of these levels
- TrivialStrategy uses a single sender for all messages. Especially useful
for early work client-side, where you want everything to be stored to
collect as much information as possible from a limited number of clients.
May also be useful for tests, to simplify test setup.
- constructs a ClientLogger instance, passing it the strategy instance, like:

let logger = new ClientLogger(new LeveledStrategy(
new NullSender(),
new ConsoleSender(),
new MeteorClientHttpSender(Meteor.absoluteUrl('logger'))
));

- adds processor instances to the logger instance, like:

logger.processors.push(
new BrowserProcessor(),
new RoutingProcessor(),
new MeteorUserProcessor(Meteor)
);

- is now able to log events, like:

logger.warn("Some warning condition", { foo: "bar" });

- with this configuration:
- logger applies processors to add browser, user, and routing information
to the message context
- since warnings are considered worthy of storage (default configuration
of the LeveledStrategy constructor), the logger passes the now-rich
message to the Meteor server thanks to the MeteorClientHttpSender
sender
- message arrives server-side.
- server/main.js
- is configured in much the same way as the client, so it has its own logger,
typically configured with just a MongodbSender instance for all levels.

let sender = new MongodbSender(Mongo, 'logger');
let logger = new ServerLogger(
new LeveledStrategy(sender, sender, sender),
WebApp);
logger.processors.push(new MeteorUserProcessor());

- in the example above, it receives the POSTed event, and proceed to log it
- because it has already been processed client-side, it is logged "raw",
without applying additional processors to it
- the MongoDb sender instance stores the event.
- the server may also initiate a logging action, for example in a publish
function or Meteor method. It can use the same API:

logger.error("Some server condition", { baz: "quux" }

- since the message is server-originated, the server processors will be
applied, to add the Meteor user information to the event
- the MongoDb sender instance stores the event.

Format note
-----------

- Whatever the enabled processors, the Logger.log() method adds a
millisecond-level timestamp to each context, under the timestamp.log key.
- Senders designed to exit the system, like the MongoDB sender, or a syslog or
logstash Beats forwarder, are expected to add another millisecond-level
timestamp, under the timestamp.store key.

These two timestamps are here to alleviate any issue resulting from a clock
difference between clients and servers.

Any sender can add extra keys to the context, under the timestamp main key,
to enable timing diagnostics.

Running tests
-------------

The module contains tests. Some of them are unit tests and need nothing special
to run, while others are currently implemented as integration tests and assume
you have a working harness project using Filog, exposed on
http://localhost:3000.

Start by compiling the package:

* meteor yarn run ts-compile

Then you can run :

* just unit tests with meteor yarn run test-unit
* just integration tests with meteor yarn run test-integration
* both tests with meteor yarn run test
* both tests including coverage generation with meteor yarn run cover

To run integration tests, you need to run the provided test_harness project in
one terminal, and the tests in another one. Alternatively, you could also roll
your own bespoke test harness, which will need to have Filog configured.

#### Example server-side code in the test_harness/ directory

This file is needed to allow Filog to operate on the /logger URL: otherwise,
Meteor will handle it natively and return a 200 with the default application
page, failing the integration tests.

#### Terminal 1

``bash
cd (filog_dir)/test_harness
meteor yarn
meteor run --port 3100
`

This example uses port 3100 to avoid conflicting with existing applications on
the default Meteor port (3000). To use another port, be sure to change it in
__tests__/integration/harness.ts too.


#### Terminal 2

`bash
cd (filog_dir)
meteor yarn
meteor yarn run compile
meteor yarn run test
meteor yarn run cover
`

TIP Reading the .travis.yml` file can be useful too.

License