🌲 Timber - Master your Elixir apps with structured logging

1. Data quality. The usefulness of your logs starts here. This is why we ship libraries that structure logs from within your application; a fundamental difference from parsing. Not only is it much more stable, but we can include data you couldn't obtain otherwise.
2. Human readability. Structuring your logs doesn't mean they have to be unreadable. Timber augments your logs with structured data. Meaning we do not alter the original log message, we simply attach metadata to it. And our console is specifically designed to give you access to this data, without compromising readability. 😮
3. Reliable downstream consumption. All log events adhere to a normalized, shared, schema that follows semantic versioning and goes through a standard release process. This means you can rely on the structure of your logs and interact consistently with them across apps of any language: queries, graphs, alerts, and other downstream consumers.
4. Zero risk of code debt or lock-in. Logging is a standard that has been around since the dawn of computers. It's built into every language, framework, and library. Timber adheres strictly to the default Logger interface. There are no special APIs, and no need to pepper your app with Timber specific code. It's just better logging. If you choose to stop using Timber, you can do so without consequence.
5. Long term retention. Timber is designed on modern big-data principles. As a result, we can offer 6+ months of retention at prices cheaper than alternatives offering <1 month. This allows you to unlock your logs for purposes beyond debugging.

1. Captures and structures your framework and 3rd party logs. (see next question)
2. Adds useful context to every log line. (see next question)
3. Allows you to easily add tags and timings to log. (see Usage)
4. Provides a framework for logging custom structured events. (see Usage)
5. Offers transport strategies to send your logs to the Timber service.

Out of the box you get everything in the Timber.Events namespace:

1. Controller Call Event
2. Exception Event
3. HTTP Client Request Event (outgoing)
4. HTTP Client Response Event
5. HTTP Server Request Event (incoming)
6. HTTP Server Response Event
7. SQL Query Event
8. Template Render Event
9. ...more coming soon, file an issue to request.

We also add context to every log, everything in the Timber.Contexts namespace. Context is structured data representing the current environment when the log line was written. It is included in every log line. Think of it like join data for your logs:

1. HTTP Context
2. Organization Context
3. Server Context
4. System Context
5. Runtime Context
6. User Context
7. ...more coming soon, file an issue to request.

They'll continue to work as expected. Timber adheres strictly to the default Logger interface and will never deviate in any way.

In fact, traditional log statements for non-meaningful events, debug statements, etc, are encouraged. In cases where the data is meaningful, consider logging a custom event.

No client, no special API, no magic, just use Logger as normal:

Logger.info("My log message")

# My log message @metadata {"level": "info", "context": {...}}

Tags provide a quick way to filter and identify logs:

Logger.info("My log message", tags: ["tag"])

# My log message @metadata {"level": "info", "tags": ["tag"], "context": {...}}

• In the Timber console use the query: tags:tag.

Timings allow you to easily capture one-off timings in your code; a simple way to benchmark code execution:

timer = Timber.start_timer()
# ... code to time ...
time_ms = Timber.duration_ms(timer)
Logger.info("Task complete", tags: ["my_task"] time_ms: time_ms)

# Task complete @metadata {"level": "info", "tags": ["my_task"], "time_ms": 56.4324, "context": {...}}

• In the Timber console use the query: tags:my_task time_ms>500
• The Timber console will also display this value inline with your logs. No need to include it in the log message, but you certainly can if you'd prefer.

Custom events can be used to structure information about events that are central to your line of business like receiving credit card payments, saving a draft of a post, or changing a user's password. You have 2 options to do this:

1. Log a map (simplest)

The simplest way to send an event and kick the tires:

event_data = %{customer_id: "xiaus1934", amount: 1900, currency: "USD"}
Logger.info("Payment rejected", event: %{payment_rejected: event_data})

# Payment rejected @metadata {"level": "warn", "event": {"payment_rejected": {"customer_id": "xiaus1934", "amount": 100, "reason": "Card expired"}}, "context": {...}}

1. Log a struct (recommended)

Defining structs for your important events just feels oh so good :) It creates a strong contract with down stream consumers and gives you compile time guarantees.

def PaymentRejectedEvent do
  use Timber.Events.CustomEvent, type: :payment_rejected

  @enforce_keys [:customer_id, :amount, :currency]
  defstruct [:customer_id, :amount, :currency]

  def message(%__MODULE__{customer_id: customer_id}) do
    "Payment rejected for #{customer_id}"
  end
end

event = %PaymentRejectedEvent{customer_id: "xiaus1934", amount: 1900, currency: "USD"}
message = PaymentRejectedEvent.message(event)
Logger.info(message, event: event)

# Payment rejected @metadata {"level": "warn", "event": {"payment_rejected": {"customer_id": "xiaus1934", "amount": 100, "reason": "Card expired"}}, "context": {...}}

• In the Timber console use queries like: payment_rejected.customer_id:xiaus1934 or payment_rejected.amount>100
• Also, notice there is no mention of Timber in the above code. Just plain old logging.

What about regular Hashes, JSON, or logfmt?

Go for it! Timber will parse the data server side, but we highly recommend the above examples. Providing a :type allows timber to classify the event, create a namespace for the data you send, and make it easier to search, graph, alert, etc.

Logger.info(%{key: "value"})
# {"key": "value"} @metadata {"level": "info", "context": {...}}

Logger.info('{"key": "value"}')
# {"key": "value"} @metadata {"level": "info", "context": {...}}

Logger.info("key=value")
# key=value @metadata {"level": "info", "context": {...}}

Context is additional data shared across log lines. Think of it like join data. For example, the http.request_id is included in the context, allowing you to view all log lines related to that request ID. Not just the lines that contain the value.

1. Add a map (simplest)

The simplest way to add context is:

Timber.add_context(%{build: %{version: "1.0.0"}})
Logger.info("My log message")

# My log message @metadata {"level": "info", "context": {"build": {"version": "1.0.0"}}}

This adds context data keyspaces by build.

1. Add a struct (recommended)

Just like events, we recommend defining your custom contexts. It makes a stronger contract with downstream consumers.

def BuildContext do
  use Timber.Contexts.CustomContext, type: :build
  @enforce_keys [:version]
  defstruct [:version]
end

Timber.add_context(%BuildContext{version: "1.0.0"})
Loger.info("My log message")

# My log message @metadata {"level": "info", "context": {"build": {"version": "1.0.0"}}}

• In the Timber console use a query like: context.build.version:1.0.0

Replace any existing config :logger calls with:

# config/config.exs

config :logger,
  backends: [Timber.LoggerBackend],
  handle_otp_reports: false # Timber handles errors, structures them, and adds additional metadata

config :timber, :capture_errors, true

👉 Skip if you are not using Plug.

# lib/my_app/endpoint.ex

plug Plug.Logger # <--- REMOVE THIS LINE

...

# ADD THESE LINES
# Insert at the bottom, immediately before `plug MyApp.Router`
plug Timber.Integrations.ContextPlug
plug Timber.Integrations.EventPlug

plug MyApp.Router

• Be sure to insert these plugs at the bottom of your endpoint.ex file, immediately before plug MyApp.Router. This ensures Timber captures the request ID and other useful context.

👉 Skip if you are not using Phoenix.

# config/config.exs

config :my_app, MyApp.Endpoint,
  instrumenters: [Timber.Integrations.PhoenixInstrumenter]

Now that Timber is handling logging, disable Phoenix logging with:

# my_app/web.ex

def controller do
  quote do
    use Phoenix.Controller, log: false # <--- Add log: false
  end
end

👉 Skip if you are not using Ecto.

# config/config.exs

config :my_app, MyApp.Repo,
  loggers: [{Timber.Integrations.EctoLogger, :log, [:info]}]

Now that Timber is all set up, we want to make sure it's development friendly:

# config/dev.exs

config :timber, :io_device,
  colorize: true,
  format: :logfmt,
  print_timestamps: true,
  print_log_level: true,
  print_metadata: false

The recommended strategy for Heroku is to setup a log drain. To get your Timber log drain URL:

👉 Add your app to Timber

Timber does not force an HTTP client on you. The following instruction utilize the Timber default Timber.Transports.HTTP.HackneyClient. This is a highly efficient client that utilizes hackney, batching, stay alive connections, connection pools, and msgpack to deliver logs with high throughput and little overhead. If you'd like to use another client see Timber.Transports.HTTP.Client.

1. Add HTTP dependencies to mix.exs:

# Elixir >= 1.4? Adding the applications list is optional.
def application do
  [applications: [:hackney, :timber]] # <-- Be sure to add hackney!
end

def deps do
  [
    {:timber, "~> 1.0"},
    {:hackney, "~> 1.6"} # <-- ADD ME
  ]
end

1. Configure Timber to use the Network transport in config/config.exs:

# config/config.exs

config :timber,
  transport: Timber.Transports.Network,
  api_key: System.get_env("TIMBER_LOGS_KEY")

config :timber, :http_transport,
  http_client: Timber.Transports.HTTP.HackneyClient,
  api_key: System.get_env("TIMBER_LOGS_KEY")

1. Obtain your Timber API 🔑 by adding your app in Timber.

2. Assign your API key to the TIMBER_LOGS_KEY environment variable.

• Note: we use the Network transport so that we can upgrade protocols in the future if we deem it more efficient. For example, TCP. If you want to use strictly HTTP, simply replace Timber.Transports.Network with Timber.Transports.HTTP.

Checkout our docs for a comprehensive list of install instructions.