Logging Performance

The following vignette presents benchmarks for log4r against all general-purpose logging packages available on CRAN:

Each logging package features slightly different capabilities, but these benchmarks are focused on the two situations common to using all of them:

  1. Logging simple messages to the console; and
  2. Deciding not to log a message because it is below the threshold.

The first of these is likely the most common kind of logging done by end users, although some may chose to log to files, over HTTP, or to the system log (among others). Yet a benchmark of these other scenarios would largely show the relative expense of these operations, instead of the overhead of the logic performed by the logging packages themselves.

The second measures the performance impact of leaving logging messages in running code, even if they are below the current threshold of visibility. This is another measure of overhead for each logging package.

Using cat()

As a reference point, we can measure how long it takes R itself to write a simple message to the console:

cat_debug <- function() {
  cat() # Print nothing.
}

cat_info <- function() cat(
  "INFO  [", format(Sys.time(), "%Y-%m-%d %H:%M:%S", usetz = FALSE),
  "] Info message.", sep = ""
)

The log4r Package

The following is a typical log4r setup:

log4r_logger <- log4r::logger(threshold = "INFO")

log4r_info <- function() {
  log4r::info(log4r_logger, "Info message.")
}

log4r_debug <- function() {
  log4r::debug(log4r_logger, "Debug message.")
}

The futile.logger Package

The following is a typical futile.logger setup:

requireNamespace("futile.logger")
#> Loading required namespace: futile.logger

futile.logger::flog.logger()

fl_info <- function() {
  futile.logger::flog.info("Info message.")
}

fl_debug <- function() {
  futile.logger::flog.debug("Debug message.")
}

The logging Package

The following is what I believe to be a typical logging setup:

requireNamespace("logging")
#> Loading required namespace: logging

logging::basicConfig()

logging_info <- function() {
  logging::loginfo("Info message.")
}

logging_debug <- function() {
  logging::logdebug("Debug message.")
}

The logger Package

The following is what I believe to be a typical logger setup:

requireNamespace("logger")
#> Loading required namespace: logger

logger_info <- function() {
  logger::log_info("Info message.")
}

logger_debug <- function() {
  logger::log_debug("Debug message.")
}

The lgr Package

The following is what I believe to be a typical lgr setup:

requireNamespace("lgr")
#> Loading required namespace: lgr

lgr_logger <- lgr::get_logger("perf-test")
lgr_logger$set_appenders(list(cons = lgr::AppenderConsole$new()))
lgr_logger$set_propagate(FALSE)

lgr_info <- function() {
  lgr_logger$info("Info message.")
}

lgr_debug <- function() {
  lgr_logger$debug("Debug message.")
}

Test All Loggers

Debug messages should print nothing.

log4r_debug()
cat_debug()
logging_debug()
fl_debug()
logger_debug()
lgr_debug()

Info messages should print to the console. Small differences in output format are to be expected.

log4r_info()
#> INFO  [2019-09-04 17:46:26] Info message.
cat_info()
#> INFO  [2019-09-04 17:46:26] Info message.
logging_info()
#> 2019-09-04 17:46:26 INFO::Info message.
fl_info()
#> INFO [2019-09-04 17:46:26] Info message.
logger_info()
#> INFO [2019-09-04 17:46:26] Info message.
lgr_info()
#> INFO  [17:46:26.100] Info message.

Benchmarks

The following benchmarks all loggers defined above:

info_bench <- microbenchmark::microbenchmark(
  cat = cat_info(),
  log4r = log4r_info(),
  futile.logger = fl_info(),
  logging = logging_info(),
  logger = logger_info(),
  lgr = lgr_info(),
  times = 500,
  control = list(warmups = 50)
)

debug_bench <- microbenchmark::microbenchmark(
  cat = cat_debug(),
  log4r = log4r_debug(),
  futile.logger = fl_debug(),
  logging = logging_debug(),
  logger = logger_debug(),
  lgr = lgr_debug(),
  times = 500,
  control = list(warmups = 50)
)

How long does it take to print messages?

How long does it take to decide to do nothing?