Description:
------------
Hi,

I have a FPM (unix socket) + nginx setup.

I have a couple of scripts, invoked from cron through nginx, that are designed to run lengthy tasks.

The very first thing these scripts do is to invoke fastcgi_finish_request() (after going through some routing logic).

They don't produce any output, neither before nor after invoking fastcgi_finish_request(). I did try writing to stdout and/or seanding headers before the call, the behavior remains the same.

Seldomly (roughly one request out of 10), FPM doesn't seem to notify nginx, which results in nginx reporting a timeout (which is configured to 30 seconds):

	2020/08/03 18:38:33 [error] 82940#103032: *41412 upstream timed out (60: Operation timed out) while reading response header from upstream, client: XX.XXX.XXX.XXX, server: XXXXXXXXX.XXXXXXXXXX.com, request: "GET /cron/XXXXXXXXXX HTTP/1.1", upstream: "fastcgi://unix:/tmp/XXXXXXXXX.sock", host: "XXXXXXXXX.XXXXXXXXXX.com"

Even when the 30 seconds timeout is not reached, I see instances of scripts running for more that a couple of milliseconds one would except when fastcgi_finish_request() is invoked straight away (first column = response time in seconds):

	0.001 XX.XXX.XXX.XXX - - [04/Aug/2020:14:34:17 +0200] "GET /cron/XXXXXXXXXX HTTP/1.1" 200 5 "-" "cron"
	17.253 XX.XXX.XXX.XXX - - [04/Aug/2020:14:34:17 +0200] "GET /cron/XXXXXXXXXX HTTP/1.1" 200 5 "-" "cron"

I have no idea where to start to try debugging this issue.

Any ideas?