Laravel Horizon/Queues - jobs stuck if frankenphp is used

[lukasamd]

What happened?

Hey, we use FrankenPHP on all envs with Laravel. Some time ago we've switched from using php to frankenphp php-cli for stuff like Scheduler and Horizon. We also added this to AppServiceProvider boot method:

        SupervisorCommandString::$command = 'exec /usr/local/bin/frankenphp php-cli artisan horizon:supervisor';
        WorkerCommandString::$command = 'exec /usr/local/bin/frankenphp php-cli artisan horizon:work';

^^ It's to ensure sub-processes will also use frankenphp php-cli.

At the beginning everything was fine, but last time our developers started to report issues with queues / strange behavior.
For example chained jobs, really small and simple: main job pushed 20 chained jobs and only first 3-4 were processed, then next one stuck in pending state and the whole chain failed after exceeding retry_after from queues config

There are a lot of factors like horizon/queues config, machines config etc but everything works fine after switching back to php artisan horizon and removing mentioned part from AppServiceProvider

Right now we are not sure - is it fine to use frankenphp php-cli for long-running jobs like scheduler/horizon or other? It has some nice advantages like reacting properly on system signals (used especially on Kubernetes and lara scheduler). If possible, we do not want to use supervisor

We've checked other parts like databases/possible deadlock issues etc. and can exclude them

Build Type

Docker (Debian Trixie)

Worker Mode

Yes

Operating System

GNU/Linux

CPU Architecture

x86_64

PHP configuration

Frankenphp 1.9.1 based on Debian Trixie with PHP 8.4
Installed extensions:

RUN install-php-extensions \
    bcmath \
    exif \
    gd \
    gettext \
    igbinary \
    imagick \
    intl \
    mongodb \
    opcache \
    pcntl \
    pgsql \
    pdo_pgsql \
    redis \
    xsl \
    yaml \
    zip \
    zstd \
    @composer

On Kubernetes we use non-root, read-only filesystem (securityContext) without any privileges but it does not matter - problem exists even on local

---

php.ini config for instances with running Horizon / Schduler

max_execution_time = 0
max_input_time = 0
max_input_vars = 25000
memory_limit = 1024M
post_max_size = 2048M
upload_tmp_dir = /tmp
upload_max_filesize = 2048M

error_reporting = E_ALL & ~E_DEPRECATED & ~E_USER_DEPRECATED

opcache.enable = 1
opcache.enable_cli = 1

opcache.memory_consumption = 256
opcache.interned_strings_buffer = 32
opcache.max_accelerated_files = 50000
opcache.validate_timestamps = 0
opcache.file_cache = /tmp/php-opcache

opcache.jit = tracing
opcache.jit_buffer_size = 256M

Relevant log output

No logs, we just see jobs randomly stuck - because of that queues are slower

[henderkes]

You should prefer using regular php over frankenphp php-cli for the time being. There are plans to make the latter a full replacement, but for the time being, it isn't.

[sneycampos]

i'm using php instead frankenphp php-cli and works great in production for 1y+

[lukasamd]

Ok, thanks for the info guys! I think it can be added to documentation to make 100% clear we should not use frankenphp php-cli to start some long-time processes

[francoism90]

Is this still true in 2026?

[henderkes]

It should generally be fine, but it's still not a full cli sapi replacement.

[jalexmelendez]

Is this still true in 2026?

Yes. I tried to use it for long running workers deployed on prem and it keeps crashing, I'll compile my workers using static php and let you know how it goes

[francoism90]

Yep, also switched back.

[henderkes]

Upstream PR in PHP is still open, it should hopefully make it into 8.6. It will definitely not land in 8.5 or earlier, though.

[henderkes]

Reopening as a reminder, perhaps a llm can figure out how to reproduce this in an open source project.

[jalexmelendez]

Hello everyone, I came up with a simple temporary solution. I ran the frankenphp php-cli command wrapped inside a bash script that will re-run the command if the system kills the queue worker, here it is. (To kill the worker close the terminal or add a escape sequence to till the loop)

#!/usr/bin/env bash
# Perpetual Laravel queue worker (FrankenPHP static php-cli).
# Usage: ./worker.sh [queue:work options...]
# Example: ./worker.sh --queue=default --sleep=3 --tries=3 --timeout=0

set -euo pipefail

cd "$(dirname -- "$0")"

FRANKENPHP="${FRANKENPHP:-./frankenphp}"
RESTART_SLEEP="${RESTART_SLEEP:-1}"

if [[ ! -x "$FRANKENPHP" ]]; then
    echo "worker.sh: not executable: $FRANKENPHP" >&2
    exit 1
fi

while true; do
    if ! "$FRANKENPHP" php-cli artisan queue:work "$@"; then
        echo "worker.sh: exited $?. Restarting in ${RESTART_SLEEP}s..." >&2
    else
        echo "worker.sh: exited. Restarting in ${RESTART_SLEEP}s..." >&2
    fi
    sleep "$RESTART_SLEEP"
done

Run this on the same dir as your executable