/tmp filling up - is this due to Unit?

[jameshulse]

Hi all, we've been running a PHP application on Nginx Unit on an Ubuntu server for a few months. Recently the hard drive filled up and it looks like there are 100's of files belonging to the unit user in /tmp named like /tmp/php0aq0sp.

Is this from Nginx Unit? If so, is it possible to clear these automatically via Unit, or do we need a cron job to do so ourselves.

Thanks!

[ac000]

Hi,

Unit does create temporary files on occasion but does clean them up.

In this case, looking at the naming, they look like they have been created via the PHP tmpfile() function, but for some reason have not been removed, which can happen when

The file is automatically removed when closed (for example, by calling fclose(), or when there are no remaining references to the file handle returned by tmpfile()), or when the script ends.

Unit hasn't been crashing has it?

[jameshulse]

Thanks for your reply. Looking at the log there are lots of similar lines about processes starting and ending:

Every few seconds a new process ends and the application restarts. Maybe this is related to the PHP worker pool settings.

[ac000]

Yes, it looks like the application is flapping (but not necessarily crashing, I'd expect to see something about signal 11 or 6 in there if it was).

[jameshulse]

The application config in Nginx Unit looks like this. Could the processes section possibly be mis-configured and causing the flapping?

{
    ...
    "applications": {
        "laravel": {
            "type": "php 8.2",
            "processes": {
                "max": 20,
                "spare": 5,
                "idle_timeout": 20
            },
            "root": "/var/www/app/public/",
            "script": "index.php"
        }
    }

}

[ac000]

Possibly. I'd say an idle timeout of 20 is pretty low, without knowing your workload it's hard to say, but I'd probably go for a couple of minutes or so, allows time for things to settle... is your traffic spiky?

[jameshulse]

Thanks. I'll up this to 5 minutes or so and see if that changes things. I would assume it would quickly kill the workers down to the 5 specified and those would sit idle.

Our traffic on this particular environment is very minimal as it is a staging server. In production there is some traffic, but it is an internal tool used a handful of times a day.