14 releases (3 stable)
1.1.1 | Mar 10, 2024 |
---|---|
1.1.0 | Jan 2, 2022 |
1.0.0 | Feb 9, 2020 |
0.9.9-beta.1 | Feb 9, 2020 |
0.2.2 | May 30, 2019 |
#55 in Debugging
13,224 downloads per month
Used in 3 crates
34KB
349 lines
Logging Timers for Rust
This crate provides a couple of simple timers that log messages indicating the elapsed time between their creation and dropping. Messages are output via the log crate.
Timers have names, and the log messages are constructed in such a way that they contain the module, filename and line number of the place where the timer was constructed.
Using the Timer Attributes
The simplest way to get started is to use one of the two attributes time
or stime
to
instrument a function, the name of the function is used as the name of the timer:
use logging_timer::{time, stime};
#[time]
fn find_files(dir: PathBuf) -> Vec<PathBuf> {
let files = vec![];
// expensive operation here
return files;
} // 'TimerFinished' message is logged here
Both attributes accept two optional arguments, to specify the log level and a pattern
into which the name of the function will be substituted. The pattern is helpful to
disambiguate functions when you have many in the same module with the same name: new
might occur many times on different structs, for example.
#[time] // Use default log level of Debug
#[time("info")] // Set custom log level
#[time("info", "FirstStruct::{}")] // Logs "FirstStruct::new()" at Info
#[time("info", "SecondStruct::{}")] // Logs "SecondStruct::new()" at Info
#[time("ThirdStruct::{}")] // Logs "ThirdStruct::new()" at Debug
#[time("never")] // Turn off instrumentation at compile time
Valid values for the log level are error, warn, info, debug and trace. Debug is the default. You can also specify 'never' to completely disable the instrumentation at compile time. The log level should appear first but as shown above can be omitted. The macros distinguish the log level from the pattern by looking for "{}".
Using the Inline Timers
More flexibility, including logging extra information, is provided by the two function-like
macro forms, timer!
and stimer!
. The difference is that timer!
returns a timer that
logs a message only when it is dropped, while stimer!
returns a timer that logs a started
message as soon as it is created, and a finished message when it is dropped. There are also
two corresponding proc-macros called time
and stimer
which can be used to instrument
functions with a timer.
In this example "FIND_FILES" is the name of the timer (using all UPPERCASE for the timer name is optional but helps make the name stand out in the log)
use logging_timer::{timer};
fn find_files(dir: PathBuf) -> Vec<PathBuf> {
let _tmr = timer!("FIND_FILES");
let files = vec![];
// expensive operation here
return files;
} // _tmr is dropped here and a 'TimerFinished' message is logged
You can replace timer!
with stimer!
to get a timer that logs a starting message as
well, giving you a pair of 'bracketing' log messages.
In addition, both timer macros accept format_args! style parameters, allowing you to include extra information in the log messages.
let _tmr = timer!("FIND_FILES", "Directory = {}", dir);
Outputting Intermediate Messages
The executing!
macro allows you to make the timer produce a message before it is dropped.
You can call it as many times as you want. A pseudocode example:
use logging_timer::{timer, executing};
fn find_files(dir: PathBuf) -> Vec<PathBuf> {
let tmr = timer!("FIND_FILES");
let files = vec![];
for dir in sub_dirs(dir) {
// expensive operation
executing!(tmr, "Processed {}", dir);
}
return files;
} // tmr is dropped here and a 'TimerFinished' message is logged
Controlling the Final Message
The finish!
macro also makes the timer log a message, but it also has the side
effect of suppressing the normal drop message. finish!
is useful when you want the final
message to include some information that you did not have access to until the calculation had
finished.
use logging_timer::{timer, executing, finish};
fn find_files(dir: PathBuf) -> Vec<PathBuf> {
let tmr = timer!("FIND_FILES");
let files = vec![];
finish!(tmr, "Found {} files", files.len());
return files;
} // tmr is dropped here but no message is produced.
Setting the log level
By default both timer
and stimer
log at Debug
level. An optional first parameter to
these macros allows you to set the log level. To aid parsing of the macro arguments this
first parameter is terminated by a semi-colon. For example:
let tmr1 = timer!(Level::Warn; "TIMER_AT_WARN");
let tmr2 = stimer!(Level::Info; "TIMER_AT_INFO");
Example of Timer Output
The overall format will depend on how you customize the output format of the log crate, but as an illustrative example:
2019-05-30T21:41:41.847982550Z DEBUG [TimerStarting] [dnscan/src/main.rs/63] DIRECTORY_ANALYSIS
2019-05-30T21:41:41.868690703Z INFO [dnlib::configuration] [dnlib/src/configuration.rs/116] Loaded configuration from "/home/phil/.dnscan/.dnscan.json"
2019-05-30T21:41:41.897609281Z DEBUG [TimerFinished] [dnlib/src/io.rs/67] FIND_FILES, Elapsed=28.835275ms, Dir="/home/phil/mydotnetprojects", NumSolutions=1 NumCsproj=45, NumOtherFiles=12
2019-05-30T21:41:41.955140835Z DEBUG [TimerFinished] [dnlib/src/analysis.rs/93] LOAD_SOLUTIONS, Elapsed=57.451736ms
2019-05-30T21:41:42.136762196Z DEBUG [TimerFinished] [dnlib/src/analysis.rs/108] LOAD_PROJECTS, Elapsed=181.563223ms, Found 43 linked projects and 2 orphaned projects
2019-05-30T21:41:42.136998556Z DEBUG [TimerStarting] [dnscan/src/main.rs/87] CALCULATE_PROJECT_GRAPH
2019-05-30T21:41:42.143072972Z DEBUG [TimerExecuting] [dnscan/src/main.rs/87] CALCULATE_PROJECT_GRAPH, Elapsed=6.075205ms, Individual graphs done
2019-05-30T21:41:42.149218039Z DEBUG [TimerFinished] [dnscan/src/main.rs/87] CALCULATE_PROJECT_GRAPH, Elapsed=12.219438ms, Found 19 redundant project relationships
2019-05-30T21:41:42.165724712Z DEBUG [TimerFinished] [dnscan/src/main.rs/108] WRITE_OUTPUT_FILES, Elapsed=16.459312ms
2019-05-30T21:41:42.166445Z INFO [TimerFinished] [dnscan/src/main.rs/63] DIRECTORY_ANALYSIS, Elapsed=318.48581ms
Here the [Timer*]
blocks are the target
field from log's Record
struct and [dnscan/src/main.rs/63]
is the filename and number from Record
- this captures the place where the timer was
instantiated. The module is also set, but is not shown in these examples.
What is Measured?
The time that is logged is the wall-clock time - the difference between the start of execution of the function and the drop of the timer at the end of the function.
The wall clock time is not necessarily length of time that the function spent executing
on the CPU. This is true for a normal function due to thread pre-emption, but it especially
true for async
functions which may spend a large part of their lives suspended.
For more sophisticated usage you may want to look into the Tokio Tracing crate.
Code Examples
There is also an example program in the examples folder which demonstrates all the different usages. To run, clone the repository and in Linux do
RUST_LOG=debug cargo run --example logging_demo
If you're on Windows, in PowerShell you can do
$env:RUST_LOG="debug"
cargo run --example logging_demo
History
See the CHANGELOG.
Performance
The timer
and stimer
macros return an Option<LoggingTimer>
. The method
log_enabled is
used to check whether logging is enabled at the requested level. If logging is
not enabled then None
is returned. This avoids most calculation in the case
where the timer would be a no-op, such that the following loop will create and
drop 1 million timers in about 4ms on my 2012-era Intel i7.
for _ in 0..1_000_000 {
let _tmr = stimer!("TEMP");
}
In comparison, v0.3 of the library would always return a LoggingTimer
, and
the loop took ten times longer.
An Option<LoggingTimer>
is 104 bytes in size on 64-bit Linux.
Dependencies
~1.5MB
~39K SLoC