Jobs randomly result in composer load class fail without even trying to process it

Question

Context

• Laravel Version: 9.5.1
• Octane version: 1.2.4
• PHP Version: 8.1.3
• Database Driver & Version: PGSQL (12.9)
• Queue driver: Redis (with 8 workers on queue "low")
• Queue failed driver: Disabled (NULL)

Description:

Random instances of the same Job lead to composer autoload error. (Failed to open stream: No such file or directory). I use pm2 to create the workers, the job doesn't even appear at the screen, just print directly in the storage/logs/laravel.log

Job code: https://prnt.sc/pCtZMRf8Er9f
(The backoff seems to work, despite of the job not being processed at the pm2 logs)

EmptyJob: https://prnt.sc/7z9myHScPrTh (For later testing purposes)

• First error: https://prnt.sc/om8PCthikwEs
• Second error (after some tries to fix it): https://prnt.sc/o7EIa2qW6wYO

Storage/logs: https://prnt.sc/P5CL-MkAl5jq

Steps To Reproduce:

Unfortunately I can't reference a repository. The error seems too much random to replicate, sometimes I can run 20 jobs in a row flawless, I tried to test it on my local machine, but the WSL always give a curl timedout. (In my local machine, the Empty Job failed once, despite not having any code at all)

(The error also happens when I dispatch multiple empty jobs)
https://prnt.sc/7z9myHScPrTh
After running 3000 EmptyJobs on 8 workers, I got 8140 lines of logs of:

include(/home/ubuntu/backend/vendor/composer/../facade/ignition/src/QueryRecorder/Query.php): Failed to open stream: No such file or directory 

Answer 1

Turns out that this was a bad cache from opcache, rebooting the machine solved the issue, if you are facing the same issue, clearing the opcache may work.