Multiple instances of Horizon all connecting to single "use" Horizon configuration

[sts-ryan-holton]

Hi 👋

Looking to gain some insights here. I'm currently running a large scale web application using Laravel Horizon. We're processing in the region of 30 million combined queued jobs per hour and have already gone to efforts to split out Horizon queues into different Redis instances, which has helped.

The issue though, is that Laravel Horizon utilises whatever the "default" connection is to store meta information to function, and we're seeing bottlenecks in the number of connections:

/*
|--------------------------------------------------------------------------
| Horizon Redis Connection
|--------------------------------------------------------------------------
|
| This is the name of the Redis connection where Horizon will store the
| meta information required for it to function. It includes the list
| of supervisors, failed jobs, job metrics, and other information.
|
*/

'use' => 'default',

Was wondering how we can get around this? Is there anyway to distribute the meta information across Horizon instances whilst maintaining data integrity. Our current Horizon config looks like:

<?php

use Illuminate\Support\Str;

return [

    /*
    |--------------------------------------------------------------------------
    | Horizon Domain
    |--------------------------------------------------------------------------
    |
    | This is the subdomain where Horizon will be accessible from. If this
    | setting is null, Horizon will reside under the same domain as the
    | application. Otherwise, this value will serve as the subdomain.
    |
    */

    'domain' => env('HORIZON_DOMAIN'),

    /*-----------------------------------------------------------
    | Horizon Path
    |--------------------------------------------------------------------------
    |
    | This is the URI path where Horizon will be accessible from. Feel free
    | to change this path to anything you like. Note that the URI will not
    | affect the paths of its internal API that aren't exposed to users.
    |
    */

    'path' => env('HORIZON_PATH', 'horizon'),

    /*
    |--------------------------------------------------------------------------
    | Horizon Redis Connection
    |--------------------------------------------------------------------------
    |
    | This is the name of the Redis connection where Horizon will store the
    | meta information required for it to function. It includes the list
    | of supervisors, failed jobs, job metrics, and other information.
    |
    */

    'use' => 'default',

    /*
    |--------------------------------------------------------------------------
    | Horizon Redis Prefix
    |--------------------------------------------------------------------------
    |
    | This prefix will be used when storing all Horizon data in Redis. You
    | may modify the prefix when you are running multiple installations
    | of Horizon on the same server so that they don't have problems.
    |
    */

    'prefix' => env(
        'HORIZON_PREFIX',
        Str::slug(env('APP_URL', 'url'), '_').Str::slug(env('APP_NAME', 'laravel'), '_').'_horizon:'

    ),

    /*
    |--------------------------------------------------------------------------
    | Horizon Route Middleware
    |--------------------------------------------------------------------------
    |
    | These middleware will get attached onto each Horizon route, giving you
    | the chance to add your own middleware to this list or change any of
    | the existing middleware. Or, you can simply stick with this list.
    |
    */

    'middleware' => ['web'],

    /*
    |--------------------------------------------------------------------------
    | Queue Wait Time Thresholds
    |--------------------------------------------------------------------------
    |
    | This option allows you to configure when the LongWaitDetected event
    | will be fired. Every connection / queue combination may have its
    | own, unique threshold (in seconds) before this event is fired.
    |
    */

    'waits' => [
        'redis:default' => 60,
    ],

    /*
    |--------------------------------------------------------------------------
    | Job Trimming Times
    |--------------------------------------------------------------------------
    |
    | Here you can configure for how long (in minutes) you desire Horizon to
    | persist the recent and failed jobs. Typically, recent jobs are kept
    | for one hour while all failed jobs are stored for an entire week.
    |
    */

    'trim' => [
        'recent' => 1,
        'pending' => 1,
        'completed' => 1,
        'recent_failed' => 5,
        'failed' => 5,
        'monitored' => 1,
    ],

    /*
    |--------------------------------------------------------------------------
    | Silenced Jobs
    |--------------------------------------------------------------------------
    |
    | Silencing a job will instruct Horizon to not place the job in the list
    | of completed jobs within the Horizon dashboard. This setting may be
    | used to fully remove any noisy jobs from the completed jobs list.
    |
    */

    'silenced' => [
        App\Jobs\ProcessAnalytic::class,
        App\Jobs\StoreApiRequestLog::class,
        App\Jobs\ProcessModelObserver::class,
        App\Jobs\ArchiveApplication::class,
        App\Jobs\Archive\StoreArchiveRecordInArchive::class,
        Laravel\Telescope\Jobs\ProcessPendingUpdates::class,
    ],

    /*
    |--------------------------------------------------------------------------
    | Metrics
    |--------------------------------------------------------------------------
    |
    | Here you can configure how many snapshots should be kept to display in
    | the metrics graph. This will get used in combination with Horizon's
    | `horizon:snapshot` schedule to define how long to retain metrics.
    |
    */

    'metrics' => [
        'trim_snapshots' => [
            'job' => 48,
            'queue' => 48,
        ],
    ],

    /*
    |--------------------------------------------------------------------------
    | Fast Termination
    |--------------------------------------------------------------------------
    |
    | When this option is enabled, Horizon's "terminate" command will not
    | wait on all of the workers to terminate unless the --wait option
    | is provided. Fast termination can shorten deployment delay by
    | allowing a new instance of Horizon to start while the last
    | instance will continue to terminate each of its workers.
    |
    */

    'fast_termination' => true,

    /*
    |--------------------------------------------------------------------------
    | Memory Limit (MB)
    |--------------------------------------------------------------------------
    |
    | This value describes the maximum amount of memory the Horizon master
    | supervisor may consume before it is terminated and restarted. For
    | configuring these limits on your workers, see the next section.
    |
    */

    'memory_limit' => 4096,

    /*
    |--------------------------------------------------------------------------
    | Queue Worker Configuration
    |--------------------------------------------------------------------------
    |
    | Here you may define the queue worker settings used by your application
    | in all environments. These supervisors and settings handle all your
    | queued jobs and will be provisioned by Horizon during deployment.
    |
    */

    'defaults' => [
        'supervisor-archive' => [
            'connection' => 'archive',
            'queue' => ['archive-prepare', 'archive-store'],
            'balance' => 'auto',
            'autoScalingStrategy' => 'time',
            'maxProcesses' => 1,
            'maxTime' => 60,
            'maxJobs' => 1000,
            'memory' => 512,
            'tries' => 1,
            'timeout' => 180,
            'nice' => 0,
            'sleep' => 3,
        ],
        'supervisor-csv-exports' => [
            'connection' => 'csv-exports',
            'queue' => ['csv-exports'],
            'balance' => false,
            'autoScalingStrategy' => 'time',
            'maxProcesses' => 1,
            'maxTime' => 60,
            'maxJobs' => 1000,
            'memory' => 2048,
            'tries' => 1,
            'timeout' => 700,
            'nice' => 0,
            'sleep' => 3,
        ],
        'supervisor-statistics' => [
            'connection' => 'horizon_statistics_queue',
            'queue' => ['statistics'],
            'balance' => 'auto',
            'autoScalingStrategy' => 'time',
            'maxProcesses' => 1,
            'maxTime' => 60,
            'maxJobs' => 1000,
            'memory' => 512,
            'tries' => 1,
            'timeout' => 90,
            'nice' => 0,
            'sleep' => 1,
        ],
        'supervisor-aggregates' => [
            'connection' => 'horizon_aggregates_queue',
            'queue' => ['aggregates'],
            'balance' => 'auto',
            'autoScalingStrategy' => 'time',
            'maxProcesses' => 1,
            'maxTime' => 60,
            'maxJobs' => 1000,
            'memory' => 512,
            'tries' => 1,
            'timeout' => 90,
            'nice' => 0,
            'sleep' => 1,
        ],
        'supervisor-charts' => [
            'connection' => 'horizon_charts_queue',
            'queue' => ['charts'],
            'balance' => 'auto',
            'autoScalingStrategy' => 'time',
            'maxProcesses' => 1,
            'maxTime' => 60,
            'maxJobs' => 1000,
            'memory' => 512,
            'tries' => 1,
            'timeout' => 90,
            'nice' => 0,
            'sleep' => 1,
        ],
        'supervisor-model-rules' => [

            // TODO: RH 10/04/2025 utilising charts queue behind the scenes until we measure how
            // much resource we might need for this.
            'connection' => 'horizon_model_rules_queue',

            'queue' => ['model-rules'],
            'balance' => 'auto',
            'autoScalingStrategy' => 'time',
            'maxProcesses' => 1,
            'maxTime' => 60,
            'maxJobs' => 1000,
            'memory' => 512,
            'tries' => 1,
            'timeout' => 90,
            'nice' => 0,
            'sleep' => 3,
        ],
        'supervisor-automations' => [
            'connection' => 'horizon_automations_queue',
            'queue' => ['automations'],
            'balance' => 'auto',
            'autoScalingStrategy' => 'time',
            'maxProcesses' => 1,
            'maxTime' => 60,
            'maxJobs' => 1000,
            'memory' => 512,
            'tries' => 1,
            'timeout' => 700,
            'nice' => 0,
            'sleep' => 0,
        ],
        'supervisor-fast-jobs' => [
            'connection' => 'redis-short-running',
            'queue' => ['on-demand-runs-now', 'redirects', 'listeners', 'observers', 'fraud', 'notifications', 'redis-short-running', 'default'],
            'balance' => 'auto',
            'autoScalingStrategy' => 'time',
            'maxProcesses' => 1,
            'maxTime' => 60,
            'maxJobs' => 1000,
            'memory' => 512,
            'tries' => 1,
            'timeout' => 40,
            'nice' => 0,
            'sleep' => 0,
        ],
        'supervisor-slow-jobs' => [
            'connection' => 'redis-long-running',
            'queue' => ['on-demand-runs-now', 'applications', 'pingtrees', 'icicle', 'redis-long-running',  'default'],
            'balance' => 'auto',
            'autoScalingStrategy' => 'time',
            'maxProcesses' => 1,
            'maxTime' => 60,
            'maxJobs' => 1000,
            'memory' => 512,
            'tries' => 1,
            'timeout' => 700,
            'nice' => 0,
            'sleep' => 0,
        ],
    ],

    'environments' => [
        'production' => [
            'supervisor-archive' => [
                'minProcesses' => 1,
                'maxProcesses' => 25,
                'balanceMaxShift' => 1,
                'balanceCooldown' => 1,
            ],
            'supervisor-csv-exports' => [
                'minProcesses' => 1,
                'maxProcesses' => 1,
                'balanceMaxShift' => 1,
                'balanceCooldown' => 1,
            ],
            'supervisor-statistics' => [
                'minProcesses' => 1,
                'maxProcesses' => 50,
                'balanceMaxShift' => 3,
                'balanceCooldown' => 2,
            ],
            'supervisor-aggregates' => [
                'minProcesses' => 1,
                'maxProcesses' => 50,
                'balanceMaxShift' => 3,
                'balanceCooldown' => 2,
            ],
            'supervisor-charts' => [
                'minProcesses' => 1,
                'maxProcesses' => 50,
                'balanceMaxShift' => 3,
                'balanceCooldown' => 2,
            ],
            'supervisor-model-rules' => [
                'minProcesses' => 1,
                'maxProcesses' => 50,
                'balanceMaxShift' => 3,
                'balanceCooldown' => 2,
            ],
            'supervisor-automations' => [
                'minProcesses' => 3,
                'maxProcesses' => 75,
                'balanceMaxShift' => 3,
                'balanceCooldown' => 2,
            ],
            'supervisor-fast-jobs' => [
                'minProcesses' => 1,
                'maxProcesses' => 75,
                'balanceMaxShift' => 3,
                'balanceCooldown' => 2,
            ],
            'supervisor-slow-jobs' => [
                'minProcesses' => 3,
                'maxProcesses' => 75,
                'balanceMaxShift' => 3,
                'balanceCooldown' => 2,
            ],
        ],

        'local' => [
            'supervisor-archive' => [
                'minProcesses' => 1,
                'maxProcesses' => 5,
                'balanceMaxShift' => 1,
                'balanceCooldown' => 1,
            ],
            'supervisor-csv-exports' => [
                'minProcesses' => 1,
                'maxProcesses' => 1,
                'balanceMaxShift' => 1,
                'balanceCooldown' => 1,
            ],
            'supervisor-statistics' => [
                'minProcesses' => 1,
                'maxProcesses' => 8,
                'balanceMaxShift' => 5,
                'balanceCooldown' => 1,
            ],
            'supervisor-aggregates' => [
                'minProcesses' => 1,
                'maxProcesses' => 8,
                'balanceMaxShift' => 5,
                'balanceCooldown' => 1,
            ],
            'supervisor-charts' => [
                'minProcesses' => 1,
                'maxProcesses' => 8,
                'balanceMaxShift' => 5,
                'balanceCooldown' => 1,
            ],
            'supervisor-model-rules' => [
                'minProcesses' => 1,
                'maxProcesses' => 8,
                'balanceMaxShift' => 5,
                'balanceCooldown' => 1,
            ],
            'supervisor-automations' => [
                'minProcesses' => 3,
                'maxProcesses' => 40,
                'balanceMaxShift' => 5,
                'balanceCooldown' => 1,
            ],
            'supervisor-fast-jobs' => [
                'minProcesses' => 3,
                'maxProcesses' => 40,
                'balanceMaxShift' => 5,
                'balanceCooldown' => 1,
            ],
            'supervisor-slow-jobs' => [
                'minProcesses' => 3,
                'maxProcesses' => 40,
                'balanceMaxShift' => 5,
                'balanceCooldown' => 1,
            ],
        ],
    ],
];

[henzeb]

We have setup multiple separate horizon instances with each their own cache. You just need different endpoints to access the dashboard that belongs to that instance. All instances including web knows about each other's caches and can dispatch jobs to the correct connection, back and forth. You would have for example a slow-job worker (horizon) and a fast job worker both as fully separate cache-horizon combination.