Overview
The goal of dyn.log
is to be a comprehensive, dynamic, configuration-based logging package for R. While there are several excellent logging solutions already in the R ecosystem, I found none of them were both robust functionality-wise, and completely generic i.e., they are designed exclusively for use within a specific environment (shiny/plumber/etc).
This is the raison d’être for dyn.log
: a flexible logging solution that:
- Integrates seamlessly
- set an R option or call
init_logging
in your .Rprofile and your logger is attached (and will re-attach when yourm(list = ls())
).
- set an R option or call
- Agnostics of project type, so it can be deployed across the entire application stack
- Reusable
- log layouts, levels, associations, etc. can be highly-customized for both shiny apps and plumber services, then reused across packages running in different environments (docker containers).
Demos
In order to keep the package as light as possible dependency-wise, there are no demos that come integrated with the package. However, separate repositories are coming soon to demonstrate the above scenarios and will be linked here.
Installation
GitHub
You can install the development version of dyn.log from GitHub with:
remotes::install_github("bmoretz/dyn.log")
Basic Usage
For basic/most common usage simply install the package from one of the above sources, load the package, initialize the logger, and a logging instance will show up in your global environment (by default, named ‘Logger’):
library(dyn.log)
init_logger()
var1 <- "abc"; var2 <- 123; var3 <- runif(1)
Logger$debug("my log message - var1: {var1}, var2: {var2}, var3: {var3}")
You can also skip the call to init_logger by setting a global option that specifies the configuration you wish to use, i.e., placing:
options("dyn.log.config" = "default")
In your .Rprofile will automatically configure the default logger and the global logging instance will be attached when you call:
The “dyn.log.config” variable can be either a predefined configuration (name) in the package, or a path to a local file that you have pre-customized. This is useful for sharing a single bespoke log configuration across multiple packages or projects.
Logging
There are three main components of a log message, each of them are covered in detail in their respective vignettes. For more detail about how logging works and how you can customize it, please see the package site:
- Levels
- The levels you want to have accessible in your logger.
- Formats
- The types that define contextual information to be logged in a message.
- Layouts
- Containers for format objects that define the rendering specifications for a log message.
The logging functionality is exposed by a R6 class, LogDispatch, is accessible as a global variable called, by default, Logger. The Logger will have methods that correspond to the log levels that are defined in its yaml configuration, which makes logging intuitive. Log messages are automatically assumed to be in standard glue format so local environment variables are captured in messages.
Simple Example
The “out of the box” (OTB) configuration specifies a default vanilla log format that displays the level that was logged, the current time-stamp (with the default TS format), and the log message:
var1 <- "abc"; var2 <- 123; var3 <- runif(1)
Logger$debug("my log message - var1: {var1}, var2: {var2}, var3: {var3}")
Configuration
Everything about dyn.log is configuration driven, the package comes with a basic configuration default.yaml, show below it its entirety and broken down in the sections that follow:
For a detailed look at customizing these settings please see Configurations vignette online.
variable: Logger
settings:
threshold: TRACE
callstack:
max: 5
start: -1
stop: -1
levels:
- name: TRACE
description: Designates finer-grained informational events than the DEBUG.
severity: 600
log_style: !expr crayon::make_style("antiquewhite3")$bold
msg_style: !expr crayon::make_style('gray60')
- name: DEBUG
description: Designates fine-grained events that are most useful to debug an application.
severity: 500
log_style: !expr crayon::make_style('deepskyblue2')$bold
msg_style: !expr crayon::make_style('gray75')
- name: INFO
description: Designates messages that highlight progress at a coarse-grained level.
severity: 400
log_style: !expr crayon::make_style('dodgerblue4')$bold
msg_style: !expr crayon::make_style('gray100')
- name: SUCCESS
description: Designates that the operation was unencumbered and completed successfully.
severity: 300
log_style: !expr crayon::make_style('chartreuse')$bold
msg_style: !expr crayon::bgGreen$bold$black
- name: WARN
description: Designates potentially harmful situations that should be investigated.
severity: 350
log_style: !expr crayon::make_style('darkorange')$bold
msg_style: !expr crayon::bgYellow$bold$black
- name: ERROR
description: Designates error events that might still allow the application to continue running.
severity: 200
log_style: !expr crayon::make_style('firebrick1')$bold
msg_style: !expr crayon::make_style('gray100')
- name: CRITICAL
severity: !expr 100L
description: Designates severe error events that could lead the application to abort.
log_style: !expr crayon::make_style('violetred4')$bold
msg_style: !expr crayon::make_style('violetred4', bg = T)$bold
- name: FATAL
description: Designates very severe error events that will presumably lead the application to abort.
severity: 100
log_style: !expr crayon::make_style('firebrick')$bold
msg_style: !expr crayon::bgRed$bold$white
layouts:
- association: default
seperator: ' '
new_line: \n
formats: new_fmt_log_level(),
new_fmt_timestamp(crayon::silver$italic),
new_fmt_log_msg()
- association: level_msg
seperator: ' '
new_line: \n
formats: new_fmt_log_level(),
new_fmt_log_msg()
Logger Variable
The first setting, variable, defines the name of the global variable you want to access the logger with. The default is Logger, but you can easily change it to: log, my_log, msg or any other value (as long as it’s a valid R variable name). The LogDispatch object is also a singleton, so you always access the logger directly:
nums <- paste0(round(rnorm(25, 0, 5), digits = 2), collapse = ", ")
LogDispatch$new()$warn("These numbers '{nums}' are out of the expected range.")
Settings
The settings node contains the core settings of the log dispatcher, by attribute. These are covered in detail in the Configuration section of the manual.
Levels
The levels node contains the log levels you want available in your environment. When a log level is defined in the configuration, it automatically becomes accessible via a first-class function on the dispatcher, e.g.:
Logger$info("This will be logged with 'INFO' severity level")
You can view all configured log levels, and get a quick summary about them by calling display_log_levels():
The default logging configuration closely resembles the fairly ubiquitous log4j scheme. For a detailed look at log levels refer to the Levels vignette online.
Layouts
Every log message needs to have a format so the dispatcher knows what to render on a log call. Formats are defined in the yaml config and comes with some basic ones pre-configured.
The default log layout is a standard format: {LEVEL} - {TIMESTAMP} - {MSG}, with space as a separator between format objects.
Customizing a Log Message
Log message layouts are exposed as an S3 type in the package called log_layout. Layouts are composed from a series of objects that inherit from fmt_layout.
new_log_layout(
format = list(
new_fmt_metric(crayon::green$bold, "sysname"),
new_fmt_metric(crayon::red$bold, "release"),
new_fmt_line_break(),
new_fmt_log_level(),
new_fmt_timestamp(crayon::silver$italic),
new_fmt_log_msg()
),
seperator = '-',
association = "custom"
)
Logger$info("my log message - var1: {var1}, var2: {var2}, var3: {var3}", layout = "custom")
For a detailed look at layouts refer to the Layouts vignette online.
Logging Associations
One thing you may have noticed about the previous log layout definition was the association parameter. Associations are a useful way to build a customized log layout for your custom R6 types. This can be especially useful in larger applications, such as plumber services or shiny dashboards.
A TestObject is defined as below, who’s primary responsibly is to assign a randomly generated identifier to the instance via the constructor. There is also a method on the object that will call the logger with some local scope variables that will be logged as well.
TestObject <- R6::R6Class(
classname = "TestObject",
public = list(
id = NULL,
initialize = function() {
self$id <- private$generate_id()
},
test_method = function() {
a <- "test"; b <- 123; c <- runif(1)
Logger$info("these are some variables: {a} - {b} - {c}")
}
),
private = list(
generate_id = function(n = 15) {
paste0(sample(LETTERS, n, TRUE), collapse = '')
}
)
)
obj <- TestObject$new()
With the above class defined, we can create a custom log layout that associated with this R6 type with a new log layout:
new_log_layout(
format = list(
new_fmt_literal(crayon::cyan$bold, "Object Id:"),
new_fmt_cls_field(crayon::bgCyan$silver$bold, "id"),
new_fmt_line_break(),
new_fmt_log_level(),
new_fmt_timestamp(crayon::silver$italic),
new_fmt_log_msg(),
new_fmt_line_break(),
new_fmt_metric(crayon::green$bold, "sysname"),
new_fmt_metric(crayon::red$bold, "nodename"),
new_fmt_literal(crayon::blue$bold, "R Version:"),
new_fmt_metric(crayon::blue$italic$bold, "r_ver"),
new_fmt_line_break()
),
association = "TestObject"
)
# notice above, "Logger$info" is called inside the context of the Test Object,
# and the variables are scoped to inside the function.
obj$test_method()
Logger$debug("this is a normal log msg")
As you can see, only when the logger is invoked from inside the class that has a custom layout associated with it does the custom layout get used. The follow-up log call (outside the class scope) reverts back to the standard layout settings.
For a detailed look at customizing a layout for a specific type, please see Configurations vignette online for an example.
Acknowledgments
- R Core for developing and maintaining such an amazing language.
- R Studio for building an incredible open-source ecosystem.
- Hadley Wickham for being super-human.
- Jim Hester for all the fantastic r-lib/actions (covr, lintr & build, pkgdown, etc).
- Brody Gaslam for developing the fansi package which is responsible for all the the pretty logger output in the documentation.
- Everyone in the #rstats community for being inclusive, welcoming and incredibly knowledgeable.