Terminated process/502 on syscall timeouts

[oprypkhantc]

• Octane Version: 1.4.2
• Laravel Version: 9.52.4
• PHP Version: 8.1.9
• Server & Version: Swoole 5.0.1

Description:

When a request is timed out due to a syscall (curl, for example), response is not returned, causing a 502. Worst of all, master process gets restarted for whatever reason, causing a downtime before Octane restarts.

The exit code is 139 (segfault), meaning it is a bug in Swoole allowing Octane to do something wrong. Yet, it's very likely Octane is doing something wrong here, given it's relation to max_execution_time.

Steps To Reproduce:

Set max_execution_time in config/octane.php to 5; add this route:

Route::get('test', function () {
	Http::timeout(40)->connectTimeout(40)->get('https://hub.dummyapis.com/delay?seconds=10');
});

start Octane and navigate to your_domain/test. You will get a 502 if you're using a reverse proxy, or just a cut-off response if you're hitting Octane directly. You'll also see that the Octane process has terminated with an exit code of 139.

[oprypkhantc]

It's possible to debug this down to C++ Swoole code; the segfault is happening somewhere in there based on the coredump I got from production. I don't have the specific line number, only the function.

I can try debugging it more to get a line number if that's needed, but I'm not very familiar with neither C++ nor debugging it :) My only guess is that that function uses reference to the worker to send the response, and the worker gets killed by Octane prior to sending the response.

[smortexa]

@sy-records It seems #510 doesn't work anymore.

[sy-records]

I will test it, I also found before that it causes the main process to exit, probably because of the Swoole5 modification.

[sy-records]

@oprypkhantc You are right, a coredump has occurred and I will give feedback to Swoole Team.

[sy-records]

Swoole is fixed, please test it with master branch. @oprypkhantc

[driesvints]

Thanks @sy-records

[joe820912boy]

I got the timeout error in the following

"2023/04/08 03:12:14 [error] 801885#801885: *46084951 upstream timed out (110: Connection timed out) while reading response header from upstream, client: x.x.x.x, server: server_nam, request: "POST /path/to/url", upstream: "http://127.0.0.1:8000/path/to/url", host: "host.name""

Have someone that faced this before?

I search that it seems occurs between proxy and server.

Is it solvable that we extend "proxy_read_timeout"?

Or there are any solutions?

[kolaente]

I'm still hitting this with swoole 5.0.3 and octane 2.0.0.

Looks like this only happens when I access very specific routes in my app and only while being logged in with certain users.

Should I open a new issue?

[NargesHajiani]

The issue still exists in Swoole v5.0.3 and Octane v2.0.1.

Reproduction:

Route::get('/health', function (Request $request) {
    sleep(10); // max_execution_time = 5
});

[NathanFreeman]

I'll take a look later.