Failed jobs you can replay

01 Replay one, or all of them

failed() runs once, when the job has given up. Keep it small: report the exception, and leave the order row alone if handle() might have written it already.

class Example
{
    public function failed(Throwable $exception): void
    {
        report($exception);
    }
}

php artisan queue:failed
php artisan queue:retry 8f3c
php artisan queue:retry all

Read the exception before you retry. Until handle() is safe to run twice — lesson 8 — don't replay a job that charges a card. Retry is safe when handle() can run twice and still leave the order paid once.

02 Keep the order id next to the exception

report() stores the exception. It does not, by itself, say which order it belonged to. failed() still has the constructor arguments, so the log line can name them. Leave the order row alone when handle() might have written it already. A second write here can mark a paid order as failed.

class Example
{
    public function failed(Throwable $exception): void
    {
        logger()->error('MarkOrderPaid gave up', [
            'order_id' => $this->orderId,
            'attempt' => $this->attempt,
            'message' => $exception->getMessage(),
        ]);

        report($exception);
    }
}

queue:retry copies that same payload back onto the queue. It does not build a new attempt key. If the logged attempt is already on the order, the idempotent handle() returns and the retry is a no-op. That is the outcome you want after a worker died at the end of a successful charge.