- 4 retry strategies (plus the ability to use your own)
- Optional jitter / randomness to spread out retries and minimize collisions
- Wait time cap
- Callbacks for custom retry logic or error handling
Notes:
- This is a fork. You can find the original project here.
- Now the codebase super strict, and it's covered with tests as much as possible. The original author is great, but the code was smelly :) It's sooo easy, and it took just one my evening... ;)
- I don't like wording "backoff" in the code. Yeah, it's fun but... I believe "retry" is more obvious. Sorry :)
- There is nothing wrong to use import instead of global namespace for function. Don't use old-school practices.
- Static variables with default values are deprecated and disabled. See dump of thoughts below.
- New methods
setJitterPercent|getJitterPercent
,setJitterMinCap|getJitterMinCap
to have fine-tuning. - My project has aliases for backward compatibility with the original. ;)
composer require jbzoo/retry
This library provides sane defaults, so you can hopefully just jump in for most of your use cases.
By default, the retry is quadratic with a 100ms base time (attempt^2 * 100
), a max of 5 retries, and no jitter.
The simplest way to use Retry is with the retry
helper function:
use function JBZoo\Retry\retry;
$result = retry(function() {
return doSomeWorkThatMightFail();
});
If successful $result
will contain the result of the closure. If max attempts are exceeded the inner exception is re-thrown.
You can of course provide other options via the helper method if needed.
Method parameters are $callback
, $maxAttempts
, $strategy
, $waitCap
, $useJitter
.
The Retry class constructor parameters are $maxAttempts
, $strategy
, $waitCap
, $useJitter
.
use JBZoo\Retry\Retry;
$retry = new Retry(10, 'exponential', 10000, true);
$result = $retry->run(function() {
return doSomeWorkThatMightFail();
});
Or if you are injecting the Retry class with a dependency container, you can set it up with setters after the fact. Note that setters are chainable.
use JBZoo\Retry\Retry;
// Assuming a fresh instance of $retry was handed to you
$result = (new Retry())
->setStrategy('constant')
->setMaxAttempts(10)
->enableJitter()
->run(function() {
return doSomeWorkThatMightFail();
});
Important Note: It's a fork. So I left it here just for backward compatibility. Static variables are deprecated and don't work at all!
This is terrible practice! Explicit is better than implicit. ;)
- Example #1. Different parts of your project can have completely different settings.
- Example #2. Imagine what would happen if some third3-party library (in
./vendor
) uses its own default settings. Let's fight! - Example #3. It's just an attempt to store variables in a global namespace. Do you see it?
So the next variables are deprecated, and they don't influence anything.
use JBZoo\Retry\Retry;
Retry::$defaultMaxAttempts;
Retry::$defaultStrategy;
Retry::$defaultJitterEnabled;
Just use dependencies injection or so and don't warm your head.
There are four built-in strategies available: constant, linear, polynomial, and exponential.
The default base time for all strategies is 100 milliseconds.
use JBZoo\Retry\Strategies\ConstantStrategy;
$strategy = new ConstantStrategy(500);
This strategy will sleep for 500 milliseconds on each retry loop.
use JBZoo\Retry\Strategies\LinearStrategy;
$strategy = new LinearStrategy(200);
This strategy will sleep for attempt * baseTime
, providing linear retry starting at 200 milliseconds.
use JBZoo\Retry\Strategies\PolynomialStrategy;
$strategy = new PolynomialStrategy(100, 3);
This strategy will sleep for (attempt^degree) * baseTime
, so in this case (attempt^3) * 100
.
The default degree if none provided is 2, effectively quadratic time.
use JBZoo\Retry\Strategies\ExponentialStrategy;
$strategy = new ExponentialStrategy(100);
This strategy will sleep for (2^attempt) * baseTime
.
In our earlier code examples we specified the strategy as a string:
use JBZoo\Retry\Retry;
use function JBZoo\Retry\retry;
retry(function() {
// ...
}, 10, 'constant');
// OR
$retry = new Retry(10, 'constant');
This would use the ConstantStrategy
with defaults, effectively giving you a 100 millisecond sleep time.
You can create the strategy instance yourself in order to modify these defaults:
use JBZoo\Retry\Retry;
use JBZoo\Retry\Strategies\LinearStrategy;
use function JBZoo\Retry\retry;
retry(function() {
// ...
}, 10, new LinearStrategy(500));
// OR
$retry = new Retry(10, new LinearStrategy(500));
You can also pass in an integer as the strategy, will translate to a ConstantStrategy with the integer as the base time in milliseconds:
use JBZoo\Retry\Retry;
use function JBZoo\Retry\retry;
retry(function() {
// ...
}, 10, 1000);
// OR
$retry = new Retry(10, 1000);
Finally, you can pass in a closure as the strategy if you wish. This closure should receive an integer attempt
and return a sleep time in milliseconds.
use JBZoo\Retry\Retry;
use function JBZoo\Retry\retry;
retry(function() {
// ...
}, 10, function($attempt) {
return (100 * $attempt) + 5000;
});
// OR
$retry = new Retry(10);
$retry->setStrategy(function($attempt) {
return (100 * $attempt) + 5000;
});
You may want to use a fast growing retry time (like exponential) but then also set a max wait time so that it levels out after a while.
This cap can be provided as the fourth argument to the retry
helper function, or using the setWaitCap()
method on the Retry class.
If you have a lot of clients starting a job at the same time and encountering failures, any of the above retry strategies could mean the workers continue to collide at each retry.
The solution for this is to add randomness. See here for a good explanation:
https://aws.amazon.com/ru/blogs/architecture/exponential-backoff-and-jitter
You can enable jitter by passing true
in as the fifth argument to the retry
helper function, or by using the enableJitter()
method on the Retry class.
By default, we use the "FullJitter" approach outlined in the above article, where a random number between 0 and the sleep time provided by your selected strategy is used.
But you can change the maximum time for Jitter with method setJitterPercent(). It's 100 by default. Also you can set min value for jitter with
setJitterMinCap(it's
0` by default).
By default, Retry will retry if an exception is encountered, and if it has not yet hit max retries.
You may provide your own retry decider for more advanced use cases. Perhaps you want to retry based on time rather than number of retries, or perhaps there are scenarios where you would want retry even when an exception was not encountered.
Provide the decider as a callback, or an instance of a class with an __invoke
method. Retry will hand it four parameters: the current attempt, max attempts, the last result received, and the exception if one was encountered. Your decider needs to return true or false.
use JBZoo\Retry\Retry;
$retry = new Retry();
$retry->setDecider(function($attempt, $maxAttempts, $result, $exception = null) {
return someCustomLogic();
});
You can provide a custom error handler to be notified anytime an exception occurs, even if we have yet to reach max attempts. This is a useful place to do logging for example.
use JBZoo\Retry\Retry;
$retry = new Retry();
$retry->setErrorHandler(function($exception, $attempt, $maxAttempts) {
Log::error("On run {$attempt}/{$maxAttempts} we hit a problem: {$exception->getMessage()}");
});
make update
make test-all
MIT
- CI-Report-Converter - Converting different error reports for deep compatibility with popular CI systems.
- Composer-Diff - See what packages have changed after
composer update
. - Composer-Graph - Dependency graph visualization of composer.json based on mermaid-js.
- Mermaid-PHP - Generate diagrams and flowcharts with the help of the mermaid script language.
- Utils - Collection of useful PHP functions, mini-classes, and snippets for every day.
- Image - Package provides object-oriented way to manipulate with images as simple as possible.
- Data - Extended implementation of ArrayObject. Use files as config/array.
- SimpleTypes - Converting any values and measures - money, weight, exchange rates, length, ...