A bug occurred in Swoole-v5.1.2, please report it about swoole-src

[modrink]

Same here, using Laravel, Octane, Swoole

Image built using the following Dockerfile:

FROM php:8.3-cli-alpine

WORKDIR /var/www/html

RUN curl -sSLf \
    -o /usr/local/bin/install-php-extensions \
    https://github.com/mlocati/docker-php-extension-installer/releases/download/2.2.11/install-php-extensions \
    && echo "4d8a1bbd2d53a4ce8d194647da9dc2ff37768a26cdb46f459bfc7642d170eb85147534e850bcb503b7b38f04ddb6228f2ef0e50619349ba02fe79aa138452cbe  /usr/local/bin/install-php-extensions" | sha512sum -c - \
    && chmod +x /usr/local/bin/install-php-extensions \
    && install-php-extensions \
    zip \
    pdo_mysql \
    pgsql \
    pdo_pgsql \
    redis \
    calendar \
    gd \
    exif \
    swoole \
    pcntl \
    && apk add --no-cache --virtual .build-deps npm \
    && npm install -g svgo \
    && apk add --no-cache \
    su-exec \
    supervisor \
    jpegoptim \ 
    optipng \
    pngquant \
    gifsicle \
    ffmpeg \
    && apk del .build-deps

Trying to access the project with Octane & Swoole:

[2024-05-13 14:10:43 #31274.0]  WARNING Server::check_worker_exit_status(): worker(pid=31408, id=16) abnormal exit, status=0, signal=11
A bug occurred in Swoole-v5.1.2, please report it.
The Swoole developers probably don't know about it,
and unless you report it, chances are it won't be fixed.
You can read How to report a bug doc before submitting any bug reports:
>> https://github.com/swoole/swoole-src/blob/master/.github/ISSUE.md 
Please do not send bug reports in the mailing list or personal letters.
The issue page is also suitable to submit feature requests.

OS: Linux 6.1.0-20-amd64 #1 SMP PREEMPT_DYNAMIC Debian 6.1.85-1 (2024-04-11) x86_64
GCC_VERSION: 13.2.1 20231014
OPENSSL_VERSION: OpenSSL 3.1.4 24 Oct 2023
PHP_VERSION : 8.3.7

Additional info:

/var/www/html # php --ri swoole

swoole

Swoole => enabled
Author => Swoole Team <[email protected]>
Version => 5.1.2
Built => May 13 2024 13:08:15
coroutine => enabled with boost asm context
epoll => enabled
eventfd => enabled
signalfd => enabled
spinlock => enabled
rwlock => enabled
openssl => OpenSSL 3.1.4 24 Oct 2023
dtls => enabled
http2 => enabled
json => enabled
c-ares => 1.27.0
zlib => 1.3.1
brotli => E16781312/D16781312
mutex_timedlock => enabled
pthread_barrier => enabled
futex => enabled
mysqlnd => enabled
async_redis => enabled

Directive => Local Value => Master Value
swoole.enable_coroutine => On => On
swoole.enable_library => On => On
swoole.enable_fiber_mock => Off => Off
swoole.enable_preemptive_scheduler => Off => Off
swoole.display_errors => On => On
swoole.use_shortname => On => On
swoole.unixsock_buffer_size => 8388608 => 8388608

/var/www/html # uname -a
Linux 6883120929f9 6.1.0-20-amd64 #1 SMP PREEMPT_DYNAMIC Debian 6.1.85-1 (2024-04-11) x86_64 Linux

/var/www/html # php -v
PHP 8.3.7 (cli) (built: May 10 2024 21:59:17) (NTS)
Copyright (c) The PHP Group
Zend Engine v4.3.7, Copyright (c) Zend Technologies

/var/www/html # gcc -v
/bin/ash: gcc: not found

Can confirm it's working on v5.1.1 (previous build)

[NathanFreeman]

@robinnydahl @modrink Perhaps some commits led to this issue.I'll take some time to test it.

[robinnydahl]

Hi all,

I'm afraid we have not yet had the time to fully make a reproducible case without sharing sensitive details, but I can confirm the issue is present in Swoole 5.1.2 and that reverting back to 5.1.1 solves the issue.

We have tested setting up a clean Laravel project with Octane using Swoole 5.1.2 inside a docker image running from php:8.2-cli-alpine and using mlocati/php-extension-installer to install Swoole. When reaching Swoole on port 8000 the service keeps running, the same goes for when we access it through NGINX inside the image as well.

The NGINX config inside the image looks like this:

map $http_upgrade $connection_upgrade {
    default upgrade;
    ''      close;
}
 
server {
    listen 80;
    listen [::]:80;
    server_name domain.com;
    server_tokens off;
    root /home/forge/domain.com/public;
 
    index index.php;
 
    charset utf-8;
 
    location /index.php {
        try_files /not_exists @octane;
    }
 
    location / {
        try_files $uri $uri/ @octane;
    }
 
    location = /favicon.ico { access_log off; log_not_found off; }
    location = /robots.txt  { access_log off; log_not_found off; }
 
    access_log off;
    error_log  /var/log/nginx/domain.com-error.log error;
 
    error_page 404 /index.php;
 
    location @octane {
        set $suffix "";
 
        if ($uri = /index.php) {
            set $suffix ?$query_string;
        }
 
        proxy_http_version 1.1;
        proxy_set_header Host $http_host;
        proxy_set_header Scheme $scheme;
        proxy_set_header SERVER_PORT $server_port;
        proxy_set_header REMOTE_ADDR $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection $connection_upgrade;
 
        proxy_pass http://127.0.0.1:8000$suffix;
    }
}

The problem actually occurs when we are running the requests through the Kubernetes Ingress Controller which is also using NGINX behind the scenes. After the first request is made through the Kubernetes Ingress Controller the Swoole service shuts down, after this we are also unable to reach the Swoole endpoint directly within the environment as well.

I can concur with @modrink that the issue is that the client closes the connection (HTTP 499), so there seems to be something wrong with the implementation in the HTTP server that was introduced in 5.1.2 and that it's triggered mostly when using HTTP Proxies like NGNIX (with certain settings), Traefik and Cloudflare.

As mentioned we have not yet been able to reproduce this in a way to share yet.

[error1001es]

RUN install-php-extensions pcntl sockets swoole pdo_mysql

this command i installed swoole

[NathanFreeman]

Do you have the xdebug extension? If so, try not using it and see.

[error1001es]

no, i dont use xdebug

[error1001es]

it's error only in 5.1.2, 5.1.1 works fine

[NathanFreeman]

@modrink I can not reproduce it, Is this a consistently reproducible scenario?
I am directly accessing 127.0.0.1:8000 without any business logic.

[modrink]

We have Cloudflare & Traefik V3 in front, but I'll test more when I get the chance. I was not able to reproduce this locally (only on server with the proxy). But Swoole v5.1.1 works and OpenSwoole latest works as well.

With Traefik the occurring errors are HTTP 499 and 502

[linuxd3v]

This happens to me as well. swoole 5.1.2 + xdebug 3.3.2

sandbox-sb.34i5g-swoole8.1 | [2024-05-17 13:07:00 #1.0] WARNING Server::check_worker_exit_status(): worker(pid=154, id=9) abnormal exit, status=0, signal=11
sandbox-sb.34i5g-swoole8.1 | A bug occurred in Swoole-v5.1.2, please report it.
sandbox-sb.34i5g-swoole8.1 | The Swoole developers probably don't know about it,
sandbox-sb.34i5g-swoole8.1 | and unless you report it, chances are it won't be fixed.
sandbox-sb.34i5g-swoole8.1 | You can read How to report a bug doc before submitting any bug reports:
sandbox-sb.34i5g-swoole8.1 | >> https://github.com/swoole/swoole-src/blob/master/.github/ISSUE.md
sandbox-sb.34i5g-swoole8.1 | Please do not send bug reports in the mailing list or personal letters.
sandbox-sb.34i5g-swoole8.1 | The issue page is also suitable to submit feature requests.
sandbox-sb.34i5g-swoole8.1 |
sandbox-sb.34i5g-swoole8.1 | OS: Linux 6.8.8-300.fc40.x86_64 #1 SMP PREEMPT_DYNAMIC Sat Apr 27 17:53:31 UTC 2024 x86_64
sandbox-sb.34i5g-swoole8.1 | GCC_VERSION: 11.4.0
sandbox-sb.34i5g-swoole8.1 | OPENSSL_VERSION: OpenSSL 3.0.2 15 Mar 2022
sandbox-sb.34i5g-swoole8.1 | PHP_VERSION : 8.1.2-1ubuntu2.17
sandbox-sb.34i5g-swoole8.1 |

[NathanFreeman]

@linuxd3v Can you provide a simple script for reproducing the error?

[NathanFreeman]

@linuxd3v I found that the max_nesting_level parameter of xdebug is set to 512. If a certain loop exceeds 512 times, a PHP Fatal error will be thrown, terminating the process.