Appearance
Source:
docs/research/scheduler-view-packages.md. Edit it there; this page is generated at build time.
Is there already a package for the in-app scheduler view?
Research for issue 25, which gates itself on exactly this question — "There may also be an existing community package worth using instead of building. Find out before building." — and which sits alongside the release map, because the answer turns out to add a step to the deploy.
Every version claim below was read from the package's own composer.json at a named tag, or from the Laravel framework source already vendored in this repo, and every compatibility claim was tested by resolving it against a copy of this repo's real composer.lock. No badge, no Packagist summary, and no blog post was treated as evidence. Where something could only be established from a README, it says so.
Recommendation
Install spatie/laravel-schedule-monitor for the recording half, and build the Filament page ourselves. No package ships the page issue 25 picked — a verdict banner over a dense table, with a different row shape for the command that only queues — and the three that ship a page are either uninstallable here, editable-by-design (the thing we ruled out), or a fortnight old with no users. But the part of the ticket that was genuinely undecided — "where last-run records land" — has a good, boring, already-written answer, and writing our own onSuccess() / onFailure() plumbing would be reinventing two tables and an event subscriber that Spatie has maintained since 2020.
The second recommendation is a negative one and it matters as much. Do not install a queue monitor to answer the RefreshMerchantPrices half. The repo already has a better, domain-level answer: affiliate_links.price_checked_at, written by the job itself. A scheduled run that dispatched twenty jobs an hour ago and left twenty links with a price_checked_at older than that is a dead worker, stated in the project's own vocabulary, with no new dependency and no second queue_monitors table. croustibat/filament-jobs-monitor would work — it is genuinely Filament 4 compatible and it genuinely records job execution rather than dispatch — but it answers a question we can already answer.
One trap, and it is the reason this document is longer than "yes, use Spatie". The package's own "did this run on time" check, Task::lastRunFinishedTooLate(), returns false for any task that has not been synced (see The trap). Install the package, forget schedule-monitor:sync in the deploy, and the page is serenely green on a box where cron has never run — which is precisely the failure issue 25 exists to make loud, rebuilt out of the tool bought to prevent it.
The four tests
Issue 25 and its chosen design give any candidate four things to pass. They are listed here because most candidates fail on the first one and never reach the interesting ones.
- It must install. Filament 4, Laravel 13, and
php >=8.4.1in practice — 18 locked runtime packages require it, so 8.3 is not an option whatevercomposer.jsonsays. - It must be loud when cron is not running. A page that looks identical installed or not is worthless. This is the job.
- It must distinguish dispatched from ran.
PublishScheduledContentworks inline;RefreshMerchantPricesonly queues. "The scheduler invoked it successfully" is true of the second command even when no worker has drained anything for a week. - It must be read-only. Schedules stay in
routes/console.php. A package whose selling point is editing them from a database is not a weaker option, it is the wrong shape.
Testing compatibility without installing anything
composer why-not cannot answer this question. It only inspects packages already in the project:
$ composer prohibits spatie/laravel-schedule-monitor 4.3.1
Could not find package "spatie/laravel-schedule-monitor" in your projectSo each candidate was resolved against a copy of this repo's composer.json and composer.lock in a scratch directory, with composer require --dry-run. That exercises the real solver against the real locked dependency graph — the only test that distinguishes "the constraint strings look fine" from "it installs" — while the repo's own tree stays untouched. The resolver output quoted below is from those runs.
The candidates
spatie/laravel-schedule-monitor — install it, for the store, not the page
composer.json at tag 4.3.1 (released 2026-06-25):
json
"require": {
"php": "^8.2",
"illuminate/bus": "^11.0|^12.0|^13.0",
"lorisleiva/cron-translator": "^0.3.0|^0.4.0|^0.5",
"nesbot/carbon": "^2.63|^3.0",
"nunomaduro/termwind": "^2.0",
"spatie/laravel-package-tools": "^1.9"
}^11.0|^12.0|^13.0 is the line that matters. Resolved against this repo:
Lock file operations: 2 installs, 0 updates, 0 removals
- Locking lorisleiva/cron-translator (v0.5.2)
- Locking spatie/laravel-schedule-monitor (4.3.1)Two packages, no upgrades, no downgrades, nothing displaced. php: ^8.2 has no upper bound, so 8.4.15 is fine.
It has no Filament page and does not claim one. It is a CLI package: schedule-monitor:sync, schedule-monitor:list, schedule-monitor:verify. What it gives us is everything under those commands.
The service provider subscribes to the framework's own scheduler events — Event::subscribe(ScheduledTaskEventSubscriber::class) in ScheduleMonitorServiceProvider::registerEventHandlers() — which are ScheduledTaskStarting, ScheduledTaskFinished, ScheduledTaskFailed and ScheduledTaskSkipped, all four dispatched by ScheduleRunCommand in the framework vendored here (vendor/laravel/framework/src/Illuminate/Console/Scheduling/ScheduleRunCommand.php). That is the same mechanism as ->onSuccess() / ->onFailure() the ticket names as the obvious route, wired once rather than per task.
Where it lands is two tables, from create_schedule_monitor_tables.php.stub: monitored_scheduled_tasks carries name, type, cron_expression, timezone, last_started_at, last_finished_at, last_failed_at, last_skipped_at and grace_time_in_minutes; monitored_scheduled_task_log_items carries one row per start/finish/ failure/skip with a meta JSON column holding runtime, exit_code, memory and — with ->storeOutputInDb() — the command's output. That is the ticket's column list, already designed.
The genuinely valuable part is not the tables, though, it is Support/ScheduledTasks/ScheduledTasks.php and the Task objects it builds. ScheduledTasks::createForSchedule() walks the live Schedule — the same object schedule:list reads — and hands back Tasks exposing, as public methods, nearly the whole of the picked design:
| Wanted by issue 25 | Task method |
|---|---|
| Human-readable cadence | humanReadableCron() (via lorisleiva/cron-translator) |
| Next due | nextRunAt() |
| Last run | lastRunStartedAt(), lastRunFinishedAt() |
| Last outcome | lastRunFailed(), lastRunSkippedAt() |
| The verdict banner | lastRunFinishedTooLate() |
| Which tasks are not being recorded at all | readyForMonitoringTasks(), unnamedTasks(), duplicateTasks() |
A Filament page can call that collection directly and never touch the Eloquent models. The verdict-banner logic in particular is already written, in Tasks/Task.php:
php
public function lastRunFinishedTooLate(): bool
{
if (! $this->isBeingMonitored()) {
return false;
}
$lastFinishedAt = $this->lastRunFinishedAt()
? $this->lastRunFinishedAt()
: $this->monitoredScheduledTask->created_at->subSecond();
$expectedNextRunStart = $this->nextRunAt($lastFinishedAt);
$shouldHaveFinishedAt = $expectedNextRunStart->addMinutes($this->graceTimeInMinutes());
return $shouldHaveFinishedAt->isPast();
}Note the fallback on the second branch: a task that has never finished is measured from the moment it was synced, so a box where cron has never run goes late on schedule rather than reading "never" forever. That is exactly A's weakness — the quiet "never" row — solved, and it is the weakness the issue comment says has to be designed out rather than inherited.
Two caveats on this package, both real.
"Notify me when a task doesn't run" is an Oh Dear feature, not an in-app one. The README's headline promise — "Oh Dear will send you a notification whenever a scheduled task doesn't run on time or fails" — is about a paid SaaS, and grace_time_in_minutes lives under the oh_dear key in the published config. We are not buying Oh Dear and do not need to: lastRunFinishedTooLate() reads that config value locally and computes the verdict in-process, with no API token configured. But anyone skimming the README will conclude the alerting needs a subscription, and that is not what we are taking.
schedule-monitor:sync has to run on every deploy. From the README: "Every time you deploy your application, you should execute the schedule-monitor:sync command… We highly recommend adding this command to the script that deploys your production environment." It reads the live schedule and writes a row per task. Two consequences for the map: it becomes a task in the Deployer recipe (issue 18), and — the README again — "Running the sync command will remove any other cron monitors that you've defined other than the application schedule." The sync is destructive to monitored_scheduled_tasks, and log items cascade on delete.
Naming: the package derives a name per task automatically (for commands, the command name), so neither of our two needs ->monitorName(). Both are Schedule::command(...) with a class argument, which CommandTask handles.
mvenghaus/filament-plugin-schedule-monitor — dead, and not close
This is the obvious candidate — a Filament page over exactly the package recommended above — and it cannot be installed. composer.json at tag 3.0.1, the latest release, published 2024-02-04:
json
"require": {
"php": "^8.2",
"spatie/laravel-schedule-monitor": "^3.5.0"
}The ^3.5.0 is fatal. Spatie's 3.x line never gained Laravel 13: 3.10.3, the last 3.x tag, requires "illuminate/bus": "^9.0|^10.0|^11.0|^12.0". The resolver against this repo:
Problem 1
- Root composer.json requires mvenghaus/filament-plugin-schedule-monitor ^3.0 -> satisfiable by
mvenghaus/filament-plugin-schedule-monitor[3.0.0, 3.0.1].
- mvenghaus/filament-plugin-schedule-monitor[3.0.0, ..., 3.0.1] require
spatie/laravel-schedule-monitor ^3.5.0 -> satisfiable by
spatie/laravel-schedule-monitor[3.5.0, ..., 3.10.3].
- spatie/laravel-schedule-monitor[3.10.1, ..., 3.10.3] require
illuminate/bus ^9.0|^10.0|^11.0|^12.0 -> found illuminate/bus[...] but these were not loaded,
likely because it conflicts with another require.Worth noticing how it fails, because it is a trap for the next person: the plugin declares no filament/filament constraint at all. Nothing in its composer.json would have stopped it installing next to Filament 4 and breaking at runtime. It is only blocked because its transitive Spatie dependency is pinned to a line that stops at Laravel 12. A Composer error here is luck, not diligence.
And even if it installed, it would not be the design. MonitoredScheduledTaskResource is a bare table over the Eloquent model — eight TextColumns, four of them raw timestamps (last_started_at, last_finished_at, last_failed_at, last_skipped_at), no computed verdict, no banner, no lastRunFinishedTooLate(). On a dead box every one of those cells reads em-dash. That is variant A with its weakness fully intact, which is the option the ticket already rejected on its own.
Last release February 2024, two releases total. Treat it as unmaintained.
croustibat/filament-jobs-monitor — installs, works, solves a problem we don't have
composer.json at tag v4.6.0 (2026-08-24):
json
"require": {
"php": "^8.2|^8.3|^8.4|^8.5",
"filament/filament": "^4.0|^5.0",
"illuminate/contracts": "^10.0|^11.0|^12.0|^13.0",
"spatie/laravel-package-tools": "^1.13.5"
}It resolves clean — Lock file operations: 1 install — and it is a genuine Filament 4 plugin, actively released, 283 stars. (Same code at ultraviolettes/filament-jobs-monitor; the org moved, the Composer name did not.)
It also does the right thing on test 3. QueueMonitorProvider hooks Queue::before / after / failing / exceptionOccurred and writes to a queue_monitors table with started_at, finished_at, failed and exception_message. That is execution, not dispatch — the distinction the issue comment calls non-optional. It needs no trait on the job and works on the database queue driver this repo uses.
It is still not what we should install, for one reason: RefreshAffiliateLinkPrices already records its own completion in the domain. affiliate_links.price_checked_at — added deliberately in 2026_09_06_150002_add_price_checked_at_to_affiliate_links_table.php as "when it last got an answer", distinct from price_updated_at — is written by the job. So "the scheduler dispatched at 10:00 and the oldest due link was last checked at 04:00" is a dead-worker verdict in the project's own vocabulary, computed from a table we already maintain, with nothing installed. A second queue_monitors table logging every job in the app to answer one question about one job is a worse trade, and the pruning (retention_days: 7) and a Shield resource come with it.
Keep it in mind if queue health ever becomes a question in its own right rather than a footnote to this page. It is the right package for that question.
codenzia/filament-system-tools — installs, closest on paper, too new to trust
The only candidate that claims the actual thing — Filament 4 and scheduler and cron-liveness detection. composer.json at tag v0.3.2:
json
"require": {
"php": "^8.3",
"filament/filament": "^4.0 || ^5.0",
"filament/tables": "^4.0 || ^5.0",
"spatie/laravel-package-tools": "^1.15.0"
}It resolves clean against this repo. And its BackgroundWorkerInspector is a direct, honest attack on test 2 — a heartbeat file touched whenever schedule:run fires, declared stale after 120 seconds, so a box with no cron reads dead rather than empty:
php
public const HEARTBEAT_FRESHNESS_SECONDS = 120;That is the right instinct and it is the only package here that has it. Three reasons not to take it anyway.
The scale is wrong. It is a kitchen sink: a log viewer, a database backup/restore tool, a schema-aware data-migration wizard, a table browser, and — on the admin panel — an arbitrary SQL query runner (src/Livewire/SqlQueryRunner.php). We want one read-only page. Adding a remote-SQL-execution surface to the admin panel to get it is not a trade worth making, and Shield would need a permission decision per page, not one.
The evidence is thin in the strict sense. Zero stars, no GitHub releases at all (two tags, v0.1.0 and v0.3.2), last pushed the day before this was written. There is no track record to read. Its README's feature list is the only description of what the scheduler page shows, and a README is not a primary source for behaviour — the code confirms the heartbeat and confirms it lists scheduled tasks with next-run time, but there is no per-task run history anywhere in it, so "last outcome" and "duration" are simply absent.
The licence needs a human. LICENSE.md is dual MIT / proprietary, MIT applying "for open source projects released under an OSI-approved open source license". This repo is private. That is a question for the owner, not an agent, and it is reason enough on its own to stop here.
husam-tariq/filament-database-schedule and zpm-packages/cron-manager-filament — wrong by design, and also uninstallable
Both came up in the search; both are the option the ticket explicitly closed. husam-tariq's README describes moving schedules into the database so they "can be changed, activated, inactivated or deleted via your panel without the need for a new application deployment" — which is the source of truth leaving the repo, stated as the feature. zpm-packages goes further and manages the host's actual crontab. Read-only is settled; neither is an option regardless of whether it installs.
They also do not install, which makes the point cheaply. husam-tariq/filament-database-schedule at v5.0 requires "filament/filament": "^5.0":
- husam-tariq/filament-database-schedule v5.0 requires filament/filament ^5.0 -> found
filament/filament[v5.0.0, ..., v5.8.1] but it conflicts with your root composer.json
require (^4.0).zpm-packages/cron-manager-filament on main requires "filament/filament": "~5.0" and "livewire/livewire": "^4.0". (main, not a tag — the repo had no release to read, which is itself worth knowing.)
Laravel Pulse — installs; has no scheduler recorder
laravel/pulse at v1.8.1 takes illuminate/*: ^13.0 and resolves clean here (pulling doctrine/sql-formatter and laravel/sentinel). It is simply not a scheduler tool. The full recorder list at that tag is CacheInteractions, Exceptions, Queues, Servers, SlowJobs, SlowOutgoingRequests, SlowQueries, SlowRequests, UserJobs, UserRequests — and the Pulse documentation's card list matches it exactly. Nothing listens to ScheduledTaskFinished. Pulse would tell us a queue is backing up; it would say nothing about whether schedule:run fired.
Two further frictions if it were ever wanted for its own sake: "Pulse's first-party storage implementation currently requires a MySQL, MariaDB, or PostgreSQL database" — this repo runs SQLite in development and in the e2e suite — and its dashboard is its own Livewire route with its own viewPulse gate, outside Filament and outside Shield.
Laravel Horizon — out on infrastructure, and only ever half the problem
Horizon is queue-only by construction; it has no view of the scheduler at all. Even for the queue half it does not fit. From the Horizon documentation:
Laravel Horizon requires that you use Redis to power your queue. Therefore, you should ensure that your queue connection is set to
redisin your application'sconfig/queue.phpconfiguration file.
This repo's config/queue.php defaults to env('QUEUE_CONNECTION', 'database') and .env.example sets QUEUE_CONNECTION=database. Adopting Horizon means adopting Redis as queue infrastructure on the box the map is about. Its composer.json at v5.49.0 also hard-requires ext-pcntl and ext-posix, and running it means a supervised long-lived php artisan horizon process. That is a queue-infrastructure decision, entirely separable from issue 25, and nothing here argues for making it now.
The declarative half needs no package at all
Worth stating plainly because it bounds how much any package is actually buying. schedule:list ships a --json flag — vendor/laravel/framework/src/Illuminate/Console/Scheduling/ScheduleListCommand.php, whose displayJson() emits expression, command, description, next_due_date, next_due_date_human, timezone, has_mutex, repeat_seconds and environments per event. Run against this repo just now:
json
[{"expression":"* * * * *","command":"php artisan app:publish-scheduled-content",
"description":"Publish Review\/Comparison\/Listicle\/Article rows whose scheduled time has passed.",
"next_due_date":"2026-09-14 09:23:00 +00:00","next_due_date_human":"15 seconds from now",
"timezone":"UTC","has_mutex":false,"repeat_seconds":null,"environments":[]},
{"expression":"0 * * * *","command":"php artisan app:refresh-merchant-prices",
"description":"Queue price refreshes for affiliate links whose prices are due, respecting each merchant's TTL.",
"next_due_date":"2026-09-14 10:00:00 +00:00","next_due_date_human":"37 minutes from now",
"timezone":"UTC","has_mutex":false,"repeat_seconds":null,"environments":[]}]A page would not shell out for this — it would resolve Illuminate\Console\Scheduling\Schedule from the container and read ->events(), which is what both ScheduleListCommand and Spatie's ScheduledTasks do. The prototype already established this half is free. has_mutex is a small bonus the ticket did not ask for: it is true while withoutOverlapping() is holding the lock, so a RefreshMerchantPrices run that hung shows as "still running" rather than "hasn't run".
The one thing the framework does not give is a human-readable cadence — ScheduleListCommand prints the raw * * * * *. lorisleiva/cron-translator, which comes in with Spatie, is where "Every minute" comes from.
The trap: an unsynced task reads as healthy
This is the finding that most changes how the build should go.
lastRunFinishedTooLate() opens with if (! $this->isBeingMonitored()) { return false; }, and isBeingMonitored() is ! is_null($this->monitoredScheduledTask) — a row in monitored_scheduled_tasks, created only by schedule-monitor:sync. So a task that was never synced is not "late". It is not anything. Every task-level verdict on the page comes back green.
That is the exact shape of failure issue 25 was opened to prevent, reintroduced by the fix. A box where the deploy forgot schedule-monitor:sync, or where the migration ran but the sync did not, shows a healthy page with a healthy banner.
The page must therefore treat "this task is not being recorded" as its own loud state, not as absence of a problem. Spatie hands us the predicate for free: ScheduledTasks::createForSchedule()->readyForMonitoringTasks() is precisely the set of named, monitorable tasks with no row — non-empty means the sync has not run, and the banner should say so in those words rather than showing green. unnamedTasks() and duplicateTasks() are the two other silent-miss cases and are equally cheap to surface.
What the page still has to do itself
None of it is large, and all of it is the part that was picked rather than the part that was available:
- The verdict banner, aggregating
lastRunFinishedTooLate()/lastRunFailed()across tasks, plus the unsynced state above. No package renders this. - Two row shapes. An inline command's row can say "ran, succeeded, 1.2s". A queuing command's row must not, because that sentence is true of a dead worker. The row for
app:refresh-merchant-pricesshould report the dispatch and the domain outcome — the count of due links whoseprice_checked_atpredates the last scheduler run — which is a query againstaffiliate_links, not against anything Spatie stores. This is C's distinction, which the issue comment made non-negotiable, and it is the one thing no candidate ships in any form. - Shield. Any new page needs
shield:generate --all --panel=adminand the hardcoded list indatabase/seeders/ShieldSeeder.phpupdating, or it is invisible to everyone includingsuper_admin.config/filament-shield.phphaspages.prefix => 'view'anddiscovery.discover_all_pages => false, so a page inapp/Filament/Pages/in the admin panel generates oneView:<PageClass>permission. (This is the one place a package would have been no easier: mvenghaus and croustibat each register Filament resources, which would need the full twelve-permission resource set generated and seeded, not one.) - A pruning schedule.
monitored_scheduled_task_log_itemsgrows by four rows an hour plus 1,440 a day from the every-minute command. Spatie's answer is Laravel's model pruning —Schedule::command('model:prune', ['--model' => MonitoredScheduledTaskLogItem::class])->daily()against the config'sdelete_log_items_older_than_days(default 30). That is a third entry inroutes/console.php, which the page will then display, which is fine. - The deploy step.
schedule-monitor:syncaftermigrateand before the symlink flip, in whatever issue 18 produces.
On the repo's own guardrails, the published migration is pure Schema::create in up(), so scripts/check-migration-safety.php — which reads only up() — passes it without an exemption comment. Nothing about this needs the destructive-migration escape hatch.
Where the evidence is thin
Stated plainly, so nobody later mistakes confidence for verification:
codenzia/filament-system-tools' scheduler page was read, not run. The heartbeat mechanism and the absence of per-task run history are from the source atv0.3.2; what the page actually looks like is from its README, which is a secondary source about its own code. Given zero stars and no releases, nobody else has verified it either.- The Filament plugin directory could not be searched exhaustively.
filamentphp.com/pluginsrenders its search client-side, so a fetch of?search=schedulereturns the unfiltered default listing. Candidates were found through web search against that domain and through GitHub instead. A plugin listed there and not indexed would have been missed. - No candidate was installed or exercised in a browser. Everything here is constraint resolution and source reading. Per
CLAUDE.md's standing warning about Filament assumptions, whichever way this goes should still be looked at in a real browser before it is called done. lastRunFinishedTooLate()'s behaviour on a genuinely dead cron is reasoned from the code, not observed. The fallback tocreated_at->subSecond()says a never-run task goes late one grace period after the first missed window, and that reading is straightforward, but it was not run against a box with cron stopped.- Two spellings of the same code.
croustibat/filament-jobs-monitorandultraviolettes/filament-jobs-monitorshare tags and release dates and are evidently the same project after an org move. Which name Packagist will carry long-term is not something the repositories state. - Spatie 4.x's
UPGRADING.mdwas not read. Nothing here depends on 3.x → 4.x migration notes since we would install 4.x fresh, but if anything in this document is ever cross-checked against writing about the 3.x line, that line stops at Laravel 12 and describes a different tool.