Logging Performance

Updated as of 2021-11-03

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

# Match the behaviour of other logging packages and write to the console.
logger::log_appender(logger::appender_stdout)

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.")
}

The loggit Package

The following is what I believe to be a typical loggit setup. Since we only want to log to the console, set the output file to /dev/null. In addition, loggit does not have a notion of thresholds, so there is no “do nothing” operation to test.

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

if (.Platform$OS.type == "unix") {
  loggit::set_logfile("/dev/null")
} else {
  loggit::set_logfile("nul")
}

loggit_info <- function() {
  loggit::loggit("INFO", "Info message.")
}

The rlog Package

The rlog package currently has no configuration options other than the threshold, which is controlled via an environment variable and defaults to hiding debug-level messages:

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

rlog_info <- function() {
  rlog::log_info("Info message.")
}

rlog_debug <- function() {
  rlog::log_debug("Debug message.")
}

Test All Loggers

Debug messages should print nothing.

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

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

log4r_info()
#> INFO  [2021-11-03 16:36:56] Info message.
cat_info()
#> INFO  [2021-11-03 16:36:56] Info message.
logging_info()
#> 2021-11-03 16:36:56 INFO::Info message.
fl_info()
#> INFO [2021-11-03 16:36:56] Info message.
logger_info()
#> INFO [2021-11-03 16:36:56] Info message.
lgr_info()
#> INFO  [16:36:56.760] Info message.
loggit_info()
#> {"timestamp": "2021-11-03T16:36:56+0000", "log_lvl": "INFO", "log_msg": "Info message."}
rlog_info()
#> 2021-11-03 16:36:56 [INFO] 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(),
  loggit = loggit_info(),
  rlog = rlog_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(),
  rlog = rlog_debug(),
  times = 500,
  control = list(warmups = 50)
)

How long does it take to print messages?

print(info_bench, order = "median")
#> Unit: microseconds
#>           expr   min     lq     mean median     uq     max neval
#>            cat  27.3  38.40  54.8350  42.20  53.40  2811.2   500
#>          log4r  27.5  51.55  60.5710  59.30  66.95  1063.6   500
#>           rlog  63.2  96.60 109.0474 107.00 118.25  1057.1   500
#>         logger 224.9 284.25 316.7576 298.50 311.85  3344.8   500
#>  futile.logger 329.6 371.10 417.8544 384.55 400.60  6622.7   500
#>        logging 334.5 383.55 430.3466 398.00 416.00  7507.2   500
#>            lgr 581.8 623.50 723.2016 636.95 657.80 32785.8   500
#>         loggit 763.6 798.40 842.3320 818.75 841.80  3912.4   500

How long does it take to decide to do nothing?

print(debug_bench, order = "median")
#> Unit: microseconds
#>           expr   min     lq     mean median     uq    max neval
#>            cat   1.7   2.70   5.7314   4.20   5.60  788.6   500
#>          log4r   8.8  10.80  15.9736  13.35  16.50 1022.9   500
#>           rlog  11.6  15.10  21.2040  20.70  23.45  872.8   500
#>            lgr  14.2  18.40  38.2624  25.00  29.80 3101.5   500
#>         logger  22.8  28.40  43.6884  32.50  43.70 3053.3   500
#>        logging  21.8  28.80  41.3526  43.20  48.95 1148.5   500
#>  futile.logger 228.1 250.65 267.0700 262.35 276.85 1212.5   500