monolyth / croney
PHP-based CRON scheduler
Requires
- php: >=8.1
- psr/log: ^3.0
- ulrichsg/getopt-php: ^4.0
Requires (Dev)
- gentry/gentry: ^0.16.10
- gentry/toast: ^0.2.0
- toast/unit: ^2.0
README
PHP-based CRON scheduler
Don't you hate having to juggle a gazillion cronjobs for each application? We sure do! We also hate having to adhere to a library-specific code format to circumvent this problem (e.g. Symfony, Laravel... you know who you are).
You know what we'd like to do? We just want to register a bunch of callables and have a central script figure it out. Hello Croney!
Installation
Composer (recommended)
composer require monomelodies/croney
Manual
- Download or clone the repository;
- Add the namespace
Croney
for the pathpath/to/croney/src
to your PSR-4 autoloader.
Setting up the executable
Croney needs to run periodically, so create a simple executable that we will add as a cronjob:
#!/usr/bin/php <?php // Let's assume this is in bin/cron. // It's empty for now.
$ chmod a+x bin/cron $ crontab -e
How often you let Croney run is up to you. The default assumption is every
minute since it is the smallest interval possible on Unix-like systems. We'll
see how to optimise this to e.g. every five minutes later on. For now, register
the cronjob with * * * * *
, i.e. every minute.
The Scheduler
class
At Croney's core is an instance of the Scheduler
class. This is what is used
to run tasks and it takes care (as the name implies) of scheduling them.
In your bin/cron
file:
#!/usr/bin/php
<?php
use Croney\Scheduler;
$schedule = new Scheduler;
Adding tasks
The Scheduler
extends ArrayObject
, so to add a task simply set it. The
simplest way is to add a callable:
#!/usr/bin/php
<?php
// ...
$schedule['some-task'] = function () {
// ...perform the task...
};
This task gets run every minute (or whatever interval you set your cronjob to). A task can be any callable, including class methods (even static ones).
You can also pass a class name, which will then get instantiated and
__invoke
'd. If you need to pass construction arguments, do the instantiating
yourself.
When you've setup all your tasks, call process
on the Scheduler
to actually
run them:
<?php // ... $schedule->process();
Running tasks at specific intervals or times
To have more control over when exactly a task is run, add the Monolyth\Croney\RunAt attribute to your invokable:
<?php
$scheduler['some-task'] =
#[Monolyth\Croney\RunAt("Y-m-d H:m")]
function () {
// ...
};
The parameter to at
is a PHP date string which, when parsed using the run's
current time, should preg_match
it. The above example runs the task every
minute (which is the default assuming your cronjob runs every minute). To run a
task every five minutes instead, you'd write something like this:
<?php $scheduler['some-task'] = #[Monolyth\Croney\RunAt("Y-m-d H:[0-5][05]")] function () { // ... };
Note that the date
function works with placeholders, so if you need to regex
on e.g. a decimal (\d
) you would need to double-escape it. See the PHP manual
page for date
for a list of all valid placeholders.
preg_match
is called without checking string position, i.e. if you would pass only'H'
as the date to match it would run on the hour, but also every minute (since 00-24 are all valid minutes and it would match'i'
as well) and also every day, month and (for all practical purposes since I'm not expecting this library to still be alive by the year 2399 ;)) all years. So be as specific as you need!
Note that the seconds part is irrelevant due to the granularity of cron and should be omitted or your task will likely never run (since the date it is compared to also doesn't include seconds).
Also note that if your task is an __invoke
able object, the RunAt attribute
should be on the __invoke
method, not the object itself.
Running the script less often
We mentioned earlier how you can also choose to run the cronjob less often than every minute, say every five minutes. If you only have tasks that run every five minutes (or multiples of that), that's fine and no further configuration is required. But what if you want to run your cronjob every five minutes, but still be able to schedule tasks based on minutes?
An example of this would be a cronjob that runs every five minutes, defining five tasks, each of which is run one minute after the previous task.
The first parameter when constructing the Scheduler object is actually its duration (in minutes):
<?php $scheduler = new Scheduler(5); // Runs for five minutes
(As you'll have guessed, the default value here is 1
.)
When you call process
, the tasks will actually be run 5 times (once every
minute) and executed when the time is there. E.g.:
<?php $scheduler = new Scheduler(5); // First task, runs only on the first loop $scheduler['first-task'] = #[RunAt("H:[0-5]0")] function () { // ... }; // Second task, runs only on the second loop $scheduler['second-task'] #[RunAt("")] = function () { $this->at('H:[0-5]1'); }; // etc.
Croney calls PHP's sleep
function in between loops.
Croney tries to calculate the actual number of seconds to sleep, so if the tasks from the first loop took, say, 3 seconds in total it sleeps for 57 seconds before the next loop. Note however that this is not exact and does not guarantee that your task will run exactly on the dot. If your task involves time-based operations make sure to "round down" the time to the expected value.
In theory, you could let your script run at midnight on January the first and calculate everything from there. In the real world, this is obviously not practical since any error whatsoever means you have to wait a whole year to see if your fix solved the problem!
Typical values are every 5 or 10 minutes, maybe 30 or 60 on very busy servers.
Long running tasks
Typically a task runs in (micro)seconds, but sometimes one of your tasks will be
"long running". If this is intentional (e.g. a periodic cleanup operation of
user-uploaded files) you would obviously runAt
it at a safe interval, and you
should take care limit stuff in your task itself (e.g. "max 100 files per run").
Still, every so often you'll need to write a task that should run often, but
might in extreme cases take longer than expected to do so.
A fictional example: a task that reads a mailbox (e.g. to push them into a ticketing system). If that mailbox explodes for whatever reason (let's be positive and imagine your application became really popular overnight ;)) this would pose a problem: the previous run might still be reading mails as the next run starts, causing mails to be handled twice. Obviously not desirable.
Croney "locks" each task prior to running, and does not attempt to re-run as long as it is locked. If a run fails due to locking, a warning is logged and the task is retried periodically for as long as the cronjob runs (assuming its RunAt configuration allows that).
The locking is done based on an MD5 hash of the task name. This includes the current working directory, so multiple Croney instances for different projects shouldn't conflict.
Error handling
You can pass an instance of Psr\Log\LoggerInterface
as an argument to the
Scheduler
constructor. This will then be used to log any messages triggered by
tasks, in the way that you specified.
If no logger was defined, all messages go to STDOUT
or STDERR
(see the
included ErrorLogger
).
Should your individual tasks also need logging, you'll need to supply them with their own instance of a logger.
Development
During development, you probably want to run tasks when testing (not just at a specific time), and also probably just a specific task. Croney as of version 0.3 comes with two command line flags for this:
--all|-a
Use this flag to run all tasks, regardless of specified scheduling. Do not do
this in production!
--job=jobname|-jjobname
Run only the specified jobname
. If the job is scheduled for particular times,
you'll likely want to use this in conjunction with the --all
(or -a
) flag.
You might also want to receive some more verbose feedback on what's going on. To
accomplish this, call your executable with the --verbose
(or -v
) flag.
If you need to test tasks in a longer running schedule, it's annoying to have
to wait minutes before your tasks complete. In that case, simply override the
scheduler's internal sleeper
property. An example (that is actually used in
the unit tests for Croney itself):
<?php $scheduler = new class (2) extends Scheduler { public function __construct(int $duration) { parent::__construct($duration); $this->sleeper = new class () extends Sleeper { public function snooze(int $seconds) : void { // parent implementation: simply `sleep($seconds)` // next call is needed to realign the internal timer // so RunAt attributes keep working. $this->advanceInternalClock(); } }; } };
Croney also includes a TestLogger
which simply echoes logged messages, so you
may use output buffering to check them in tests.