Customizing the Format and the Destination of a Log Record

## Error in get(genname, envir = envir) : object 'testthat_print' not found

In this vignette I suppose that you are already familiar with The Anatomy of a Log Request vignette.

What gets logged?

logger mostly relies on and uses the default log4j log levels and supports suppressing log messages with a lower log level compared to the currently set threshold in the logging namespace:

log_info('Hi, there!')
#> INFO [2021-10-18 23:25:54] Hi, there!
log_debug('How are you doing today?')
log_threshold()
#> [1] 400
#> attr(,"level")
#> [1] "INFO"
#> attr(,"class")
#> [1] "loglevel" "integer"
log_threshold(TRACE)
log_debug('How are you doing today?')
#> DEBUG [2021-10-18 23:25:54] How are you doing today?

So the ?log_threshold function can both get and set the log level threshold for all future log requests.

For the full list of all supported log levels and so thus the possible log level thresholds, see ?log_levels.

If you want to define the log level in a programmatic way, check out ?log_level, eg

log_level(INFO, 'Hi, there!')
#> INFO [2021-10-18 23:25:54] Hi, there!

To temporarily update the log level threshold, you may also find the ?with_log_threshold function useful:

log_threshold(INFO)
log_debug('pst, can you hear me?')
log_info('no')
#> INFO [2021-10-18 23:25:54] no

with_log_threshold(log_debug('pst, can you hear me?'), threshold = TRACE)
#> DEBUG [2021-10-18 23:25:54] pst, can you hear me?
log_info('yes')
#> INFO [2021-10-18 23:25:54] yes

with_log_threshold({
    log_debug('pst, can you hear me?')
    log_info('yes')
}, threshold = TRACE)
#> DEBUG [2021-10-18 23:25:54] pst, can you hear me?
#> INFO [2021-10-18 23:25:54] yes

You can also define your own log level(s) if needed, for example introducing an extra level between DEBUG and INFO:

FYI <- structure(450L, level = 'FYI', class = c('loglevel', 'integer'))
log_threshold(FYI)
log_debug('ping')
log_level(FYI, 'ping')
#> FYI [2021-10-18 23:25:54] ping
log_info('pong')
#> INFO [2021-10-18 23:25:54] pong

Log namespaces

By default, all log messages will be processed by the global logger definition, but you may also use custom namespaces (eg to deliver specific log records to a special destination or to apply a custom log level threshold) and even multiple loggers as well within the very same namespace (eg to deliver all INFO and above log levels in the console and everything below that to a trace log file).

If you specify an unknown namespace in a log request, it will fall back to the global settings:

log_threshold(INFO)
log_trace('Hi, there!', namespace = 'kitchensink')
log_info('Hi, there!', namespace = 'kitchensink')
#> INFO [2021-10-18 23:25:54] Hi, there!

But once you start customizing that namespace, it gets forked from the global settings and live on its own without modifying the original namespace:

log_threshold(TRACE, namespace = 'kitchensink')
log_trace('Hi, there!', namespace = 'kitchensink')
#> TRACE [2021-10-18 23:25:54] Hi, there!
log_info('Hi, there!', namespace = 'kitchensink')
#> INFO [2021-10-18 23:25:54] Hi, there!
log_trace('Hi, there!')

Log message formatter functions

In the above example, we logged strings without any dynamic parameter, so the task of the logger was quite easy. But in most cases you want to log a parameterized string and the formatter function’s task to transform that to a regular character vector.

By default, logger uses glue in the background:

log_formatter(formatter_glue)
log_info('There are {nrow(mtcars)} cars in the mtcars dataset')
#> INFO [2021-10-18 23:25:54] There are 32 cars in the mtcars dataset
log_info('2 + 2 = {2+2}')
#> INFO [2021-10-18 23:25:54] 2 + 2 = 4

If you don’t like this syntax, or want to save a dependency, you can use other formatter functions as well, such as ?formatter_sprintf (being the default in eg the logging and futile.logger packages) or ?formatter_paste, or write your own formatter function converting R objects into string.

Log message layouts

By default, ?log_level and its derivative functions (eg ?log_info) will simply record the log-level, the current timestamp and the message after being processed by glue:

log_info(42)
#> INFO [2021-10-18 23:25:54] 42
log_info('The answer is {42}')
#> INFO [2021-10-18 23:25:54] The answer is 42
log_info('The answers are {1:5}')
#> INFO [2021-10-18 23:25:54] The answers are 1
#> INFO [2021-10-18 23:25:54] The answers are 2
#> INFO [2021-10-18 23:25:54] The answers are 3
#> INFO [2021-10-18 23:25:54] The answers are 4
#> INFO [2021-10-18 23:25:54] The answers are 5

In the above example, first, 42 was converted to a string by the ?formatter_glue message formatter, then the message was passed to the ?layout_simple layout function to generate the actual log record.

An example of another layout function writing the same log messages in JSON:

log_layout(layout_json())
log_info(42)
#> {"time":"2021-10-18 23:25:54","level":"INFO","ns":"global","ans":"global","topenv":"R_GlobalEnv","fn":"eval","node":"nevermind","arch":"x86_64","os_name":"Linux","os_release":"5.14.8-arch1-1","os_version":"#1 SMP PREEMPT Sun, 26 Sep 2021 19:36:15 +0000","pid":1622485,"user":"daroczig","msg":"42"}
log_info('The answer is {42}')
#> {"time":"2021-10-18 23:25:54","level":"INFO","ns":"global","ans":"global","topenv":"R_GlobalEnv","fn":"eval","node":"nevermind","arch":"x86_64","os_name":"Linux","os_release":"5.14.8-arch1-1","os_version":"#1 SMP PREEMPT Sun, 26 Sep 2021 19:36:15 +0000","pid":1622485,"user":"daroczig","msg":"The answer is 42"}
log_info('The answers are {1:5}')
#> {"time":"2021-10-18 23:25:54","level":"INFO","ns":"global","ans":"global","topenv":"R_GlobalEnv","fn":"eval","node":"nevermind","arch":"x86_64","os_name":"Linux","os_release":"5.14.8-arch1-1","os_version":"#1 SMP PREEMPT Sun, 26 Sep 2021 19:36:15 +0000","pid":1622485,"user":"daroczig","msg":"The answers are 1"}
#> {"time":"2021-10-18 23:25:54","level":"INFO","ns":"global","ans":"global","topenv":"R_GlobalEnv","fn":"eval","node":"nevermind","arch":"x86_64","os_name":"Linux","os_release":"5.14.8-arch1-1","os_version":"#1 SMP PREEMPT Sun, 26 Sep 2021 19:36:15 +0000","pid":1622485,"user":"daroczig","msg":"The answers are 2"}
#> {"time":"2021-10-18 23:25:54","level":"INFO","ns":"global","ans":"global","topenv":"R_GlobalEnv","fn":"eval","node":"nevermind","arch":"x86_64","os_name":"Linux","os_release":"5.14.8-arch1-1","os_version":"#1 SMP PREEMPT Sun, 26 Sep 2021 19:36:15 +0000","pid":1622485,"user":"daroczig","msg":"The answers are 3"}
#> {"time":"2021-10-18 23:25:54","level":"INFO","ns":"global","ans":"global","topenv":"R_GlobalEnv","fn":"eval","node":"nevermind","arch":"x86_64","os_name":"Linux","os_release":"5.14.8-arch1-1","os_version":"#1 SMP PREEMPT Sun, 26 Sep 2021 19:36:15 +0000","pid":1622485,"user":"daroczig","msg":"The answers are 4"}
#> {"time":"2021-10-18 23:25:54","level":"INFO","ns":"global","ans":"global","topenv":"R_GlobalEnv","fn":"eval","node":"nevermind","arch":"x86_64","os_name":"Linux","os_release":"5.14.8-arch1-1","os_version":"#1 SMP PREEMPT Sun, 26 Sep 2021 19:36:15 +0000","pid":1622485,"user":"daroczig","msg":"The answers are 5"}

If you need colorized logs highlighting the important log messages, check out ?layout_glue_colors, and for other formatter and layout functions, see the manual of the above mentioned functions that have references to all the other functions and generator functions bundled with the package.

Custom log record layout

To define a custom format on how the log messages should be rendered, you may write your own formatter and layout function(s) or rely on the function generator functions bundled with the logger package, such as ?layout_glue_generator.

This function returns a layout function that you can define by glue-ing together variables describing the log request via ?get_logger_meta_variables, so having easy access to (package) namespace, calling function’s name, hostname, user running the R process etc.

A quick example:

Another example of making use of the generator function is to update the layout to include the Process ID that might be very useful eg when forking, see for example the below code chunk still using the above defined log layout:

f <- function(x) {
    log_info('received {length(x)} values')
    log_success('with the mean of {mean(x)}')
    mean(x)
}
library(parallel)
mclapply(split(runif(100), 1:10), f, mc.cores = 5)
#> nevermind/26448/R_GlobalEnv/FUN 2018-12-02 21:54:11 INFO: received 10 values
#> nevermind/26448/R_GlobalEnv/FUN 2018-12-02 21:54:11 SUCCESS: with the mean of 0.403173440974206
#> nevermind/26449/R_GlobalEnv/FUN 2018-12-02 21:54:11 INFO: received 10 values
#> nevermind/26448/R_GlobalEnv/FUN 2018-12-02 21:54:11 INFO: received 10 values
#> nevermind/26449/R_GlobalEnv/FUN 2018-12-02 21:54:11 SUCCESS: with the mean of 0.538581100990996
#> nevermind/26448/R_GlobalEnv/FUN 2018-12-02 21:54:11 SUCCESS: with the mean of 0.485734378430061
#> nevermind/26450/R_GlobalEnv/FUN 2018-12-02 21:54:11 INFO: received 10 values
#> nevermind/26449/R_GlobalEnv/FUN 2018-12-02 21:54:11 INFO: received 10 values
#> nevermind/26450/R_GlobalEnv/FUN 2018-12-02 21:54:11 SUCCESS: with the mean of 0.580483326432295
#> nevermind/26452/R_GlobalEnv/FUN 2018-12-02 21:54:11 INFO: received 10 values
#> nevermind/26449/R_GlobalEnv/FUN 2018-12-02 21:54:11 SUCCESS: with the mean of 0.461282140854746
#> nevermind/26450/R_GlobalEnv/FUN 2018-12-02 21:54:11 INFO: received 10 values
#> nevermind/26451/R_GlobalEnv/FUN 2018-12-02 21:54:11 INFO: received 10 values
#> nevermind/26450/R_GlobalEnv/FUN 2018-12-02 21:54:11 SUCCESS: with the mean of 0.465152264293283
#> nevermind/26452/R_GlobalEnv/FUN 2018-12-02 21:54:11 SUCCESS: with the mean of 0.618332817289047
#> nevermind/26451/R_GlobalEnv/FUN 2018-12-02 21:54:11 SUCCESS: with the mean of 0.493527933699079
#> nevermind/26452/R_GlobalEnv/FUN 2018-12-02 21:54:11 INFO: received 10 values
#> nevermind/26452/R_GlobalEnv/FUN 2018-12-02 21:54:11 SUCCESS: with the mean of 0.606248055002652
#> nevermind/26451/R_GlobalEnv/FUN 2018-12-02 21:54:11 INFO: received 10 values
#> nevermind/26451/R_GlobalEnv/FUN 2018-12-02 21:54:11 SUCCESS: with the mean of 0.537314630229957

Note that the layout_glue_generator functions also adds a special attribute to the resulting formatting function so that when printing the layout function to the console, the user can easily interpret what’s being used instead of just showing the actual functions’s body:

log_layout()
#> layout_glue_generator(format = "{node}/{pid}/{namespace}/{fn} {time} {level}: {msg}")

For more details on this, see the Writing custom logger extensions vignette.

## reset layout
log_layout(layout_simple)

Delivering log records to their destination

By default, logger will write to the console or stdout via the ?appender_console function:

log_appender()
#> appender_stdout

To write to a logfile instead, use the ?appender_file generator function, that returns a function that can be used in any namespace:

t <- tempfile()
log_appender(appender_file(t))
log_info('where is this message going?')
log_appender()
#> appender_file(file = t)
readLines(t)
#> [1] "INFO [2021-10-18 23:25:54] where is this message going?"
unlink(t)

There’s a similar generator function that returns an appender function delivering log messages to Slack channels:

## load Slack configuration, API token etc from a (hopefully encrypted) yaml file or similar
slack_config <- config::config(...)
## redirect log messages to Slack
log_appender(appender_slack(
    channel   = '#gergely-test',
    username  = 'logger',
    api_token = slack_config$token
), namespace = 'slack')
log_info('Excited about sending logs to Slack!')
#> INFO [2018-11-28 00:21:13] Excited about sending logs to Slack!
log_info('Hi there from logger@R!', namespace = 'slack')

You may find ?appender_tee also useful, that writes the log messages to both stdout and a file.

## reset appender
log_appender(appender_console)

And the are many other appender functions bundled with logger as well, eg some writing to Syslog, Telegram, Pushbullet, a database table or an Amazon Kinesis stream – even doing that asynchronously via appender_async – see Simple Benchmarks on Performance for more details.

Stacking loggers

Note that the ?appender_tee functionality can be implemented by stacking loggers as well, eg setting two loggers for the global namespace: ?appender_console and ?appender_file. The advantage of this approach is that you can set different log level thresholds for each logger, for example:

log_threshold()
#> [1] 400
#> attr(,"level")
#> [1] "INFO"
#> attr(,"class")
#> [1] "loglevel" "integer"

## create a new logger with index 2
log_threshold(TRACE, index = 2)

## note that the original logger still have the same log level threshold
log_threshold()
#> [1] 400
#> attr(,"level")
#> [1] "INFO"
#> attr(,"class")
#> [1] "loglevel" "integer"
log_threshold(index = 2)
#> [1] 600
#> attr(,"level")
#> [1] "TRACE"
#> attr(,"class")
#> [1] "loglevel" "integer"

## update the appender of the new logger
t <- tempfile()
log_appender(appender_file(t), index = 2)

## test both loggers
log_info('info msg')
log_debug('info msg')

readLines(t)
#> [1] "INFO [2021-10-18 23:25:54] info msg" 
#> [2] "DEBUG [2021-10-18 23:25:54] info msg"
unlink(t)