Adaptive backoff provides a mechanism to intelligently back off use of rate
limited or failure operations through sources simples structs, Backoffs
. Backoffs
take input on success and failure, then return a duration to wait.
For an adaptive backoff, as the failure and success calls increase the returned duration eventually converges on a value to avoid rate limiting of requests.
Include within your Cargo.toml
:
adaptive_backoff = "0.2"
And follow an example below:
Below is an example of an adapative ExponentialBackoff
which works from a queue
and converges on a minimal delay duration between calls. It grows by a factor of
2.0
to a maximum of 300
seconds.
use std::time::Duration;
use adaptive_backoff::prelude::*;
let mut backoff = ExponentialBackoffBuilder::default()
.factor(2.0)
.max(Duration::from_secs(300))
.adaptive()
.build()
.unwrap();
while let Some(item) = queue.pop() {
loop {
match worker_iter(&conn, &item).await {
Ok(_) => {
delay_for(backoff.success()).await;
break;
}
Err(_) => delay_for(backoff.fail()).await,
}
}
}
If adaptive is omitted from the example above, a simple backoff is returned.
Its API lacks success()
and fail()
, instead it can only return increasing
delays with wait()
until reset()
is called to return it.
use std::time::Duration;
use adaptive_backoff::prelude::*;
let mut backoff = ExponentialBackoffBuilder::default()
.factor(2.0)
.max(Duration::from_secs(30))
.build()
.unwrap();
while let Some(item) = queue.pop() {
loop {
match worker_iter(&conn, &item).await {
Ok(_) => {
delay_for(backoff.wait()).await;
break;
}
Err(_) => delay_for(backoff.wait()).await,
}
}
backoff.reset();
}
There are tests for backoff implementations which contain example use of the external API with expected output. See both the simple exponential example and the adaptive exponential backoff example.