Utility macro to build static metrics for the rust-prometheus library.
MetricVec
(i.e. CounterVec
, GaugeVec
or HistogramVec
) is slow. However if every possible values for labels are
known, each metric in the MetricVec
can be cached to avoid the runtime cost.
For example, the following code can be slow when it is invoked multiple times:
some_counter_vec.with_label_values(&["label_1_foo", "label_2_bar"]).inc();
It is because we are retriving a specific Counter
according to values each time and to ensure thread-safety there is
a lock inside which makes things worse.
We can optimize it by caching the counter by label values:
// init before hand
let foo_bar_counter = some_counter.with_label_values(&["label_1_foo", "label_2_bar"]);
foo_bar_counter.inc();
So far everything seems good. We achieve the same performance as Counter
for CounterVec
. But what if there are many
labels and each of them has many values? We need to hand-craft a lot of code in this way.
That's what this crate solves. This crate provides a macro that helps you do the optimization above without really introducing a lot of templating code.
-
Add to
Cargo.toml
:[dependencies] prometheus-static-metric = "0.4"
-
Add to
lib.rs
:use prometheus_static_metric;
Use the make_static_metric!
to define all possible values for each label. Your definition will be expanded to a real
struct
for easy access while keeping high-performance.
use prometheus_static_metric::make_static_metric;
make_static_metric! {
pub struct MyStaticCounterVec: Counter {
"method" => {
post,
get,
put,
delete,
},
"product" => {
foo,
bar,
},
}
}
fn main() {
let vec = CounterVec::new(Opts::new("foo", "bar"), &["method", "product"]).unwrap();
let static_counter_vec = MyStaticCounterVec::from(&vec);
static_counter_vec.post.foo.inc();
static_counter_vec.delete.bar.inc_by(4.0);
assert_eq!(static_counter_vec.post.bar.get(), 0.0);
assert_eq!(vec.with_label_values(&["post", "foo"]).get(), 1.0);
assert_eq!(vec.with_label_values(&["delete", "bar"]).get(), 4.0);
}
For heavier scenario that a global shared static-metric might not be effecient enough, you can use make_auto_flush_static_metric!
macro, which will store data in local thread storage, with a custom rate to flush to global MetricVec
.
use prometheus::*;
use lazy_static:lazy_static;
use prometheus_static_metric::{auto_flush_from, make_auto_flush_static_metric};
make_auto_flush_static_metric! {
pub label_enum FooBar {
foo,
bar,
}
pub label_enum Methods {
post,
get,
put,
delete,
}
pub struct Lhrs: LocalIntCounter {
"product" => FooBar,
"method" => Methods,
"version" => {
http1: "HTTP/1",
http2: "HTTP/2",
},
}
}
lazy_static! {
pub static ref HTTP_COUNTER_VEC: IntCounterVec =
register_int_counter_vec ! (
"http_requests_total",
"Number of HTTP requests.",
& ["product", "method", "version"] // it doesn't matter for the label order
).unwrap();
}
lazy_static! {
// You can also use default flush duration which is 1 second.
// pub static ref TLS_HTTP_COUNTER: Lhrs = auto_flush_from!(HTTP_COUNTER_VEC, Lhrs);
pub static ref TLS_HTTP_COUNTER: Lhrs = auto_flush_from!(HTTP_COUNTER_VEC, Lhrs, std::time::Duration::from_secs(1));
}
fn main() {
TLS_HTTP_COUNTER.foo.post.http1.inc();
TLS_HTTP_COUNTER.foo.post.http1.inc();
assert_eq!(
HTTP_COUNTER_VEC
.with_label_values(&["foo", "post", "HTTP/1"])
.get(),
0
);
::std::thread::sleep(::std::time::Duration::from_secs(2));
TLS_HTTP_COUNTER.foo.post.http1.inc();
assert_eq!(
HTTP_COUNTER_VEC
.with_label_values(&["foo", "post", "HTTP/1"])
.get(),
3
);
}
Please take a look at examples directory for more.