Laravel 13.36 landed today with a focused set of improvements around queue workers, validation, and database connections. None of these are headline features, but they solve real problems that show up in production: workers pinning write connections, missing queue configuration, and verbose validation rules. Here is what changed and why it matters.
Background queue connection in the default config
Since Laravel 11, the skeleton application has shipped with a background queue connection intended for low-priority work. The framework's own config/queue.php never received it, so any project that hadn't published the config would hit an InvalidArgumentException when dispatching to background.
// config/queue.php (framework default, now in 13.36)
'connections' => [
'sync' => [...],
'database' => [...],
'redis' => [...],
'deferred' => [...],
'background' => [ // ← new in 13.36
'driver' => 'database',
'table' => 'jobs',
'queue' => 'background',
'retry_after' => 90,
],
'failover' => [...],
// ...
],If you already published the config and defined your own background connection, nothing changes — configs are merged. If you haven't published it, the connection now works out of the box. This is a small fix that removes a confusing error for new projects and anyone using the skeleton defaults.
Flushing sticky connections between jobs
When a database connection is marked sticky (the default for read/write splitting), a write operation pins subsequent reads to the primary connection. In a long-running queue worker, a single job that writes can cause the next 10,000 jobs to read from the primary — defeating the purpose of replicas.
Laravel 13.36 adds an opt-in flag to reset the sticky state between jobs:
// In a service provider, e.g. AppServiceProvider
use Illuminate\Queue\Worker;
Worker::$flushStickyConnections = true;The framework calls $connection->flushStickyConnections() after each job completes. The flag is opt-in for 13.x to avoid behavioural changes in existing applications; the plan is to make it the default in Laravel 14.
If you run queue workers against a read-replica setup, enable this. It restores the intended read distribution without requiring a deploy or worker restart cycle to unpick the pinned connection.
Validation rule builders completed
Laravel has been migrating validation rules to a fluent builder API. With 13.36, the remaining rules now have builders, so you can write:
use Illuminate\Validation\Rules;
$request->validate([
'email' => Rules\Email::rule()->domain('example.com'),
'password' => Rules\Password::rule()->min(12)->mixedCase()->numbers()->symbols(),
'avatar' => Rules\File::rule()->image()->max(2048)->dimensions(Rule\Dimensions::rule()->minWidth(200)),
'tags' => Rules\Array::rule()->min(1)->max(10),
'metadata' => Rules\Json::rule(),
]);instead of the string-based syntax. The builder approach gives you IDE autocomplete, static analysis support, and a single place to discover available options. It also makes complex rules composable:
$passwordRule = Rules\Password::rule()->min(10)->mixedCase();
$request->validate([
'password' => $passwordRule,
'password_confirmation' => ['required', $passwordRule],
]);All core rules now have builders. If you've been holding off because your favourite rule was missing, the gap is closed.
DeleteWhenMissingModels and FailOnTimeout on queued mailables
Queued jobs have long supported the DeleteWhenMissingModels and FailOnTimeout attributes. As of 13.36, mailables that implement ShouldQueue can use them too:
use Illuminate\Queue\Attributes\DeleteWhenMissingModels;
use Illuminate\Queue\Attributes\FailOnTimeout;
use Illuminate\Contracts\Queue\ShouldQueue;
use Illuminate\Mail\Mailable;
#[DeleteWhenMissingModels]
#[FailOnTimeout(120)]
class OrderShipped extends Mailable implements ShouldQueue
{
public function __construct(public Order $order) {}
}DeleteWhenMissingModels deletes the queued mail if the related Eloquent model (the Order in this example) no longer exists when the job runs — useful when an order is cancelled before the email sends. FailOnTimeout marks the job as failed if it exceeds the given seconds, rather than letting it retry indefinitely.
Both attributes work on queued notifications as well, since notifications use the same queuing mechanism under the hood.
Batch Redis queue notification pushes
When you queue many notifications at once — for example, notifying thousands of users of a new announcement — Laravel now batches the Redis RPUSH calls into a single pipeline round-trip. The change is internal to RedisQueue; no application code changes are required. In benchmarks with 10,000 notifications, the enqueue time dropped from ~2.1 s to ~380 ms.
If you use Redis queues and send bulk notifications, you get the speedup automatically on 13.36.
Schema::getColumn() is now case-insensitive
Schema::hasColumn() has always compared column names case-insensitively (via strtolower). Schema::getColumn() and Schema::getColumnType() did not, which led to inconsistent behaviour:
// Before 13.36: false on PostgreSQL if the column is "email" but you ask for "Email"
Schema::hasColumn('users', 'Email'); // true
Schema::getColumn('users', 'Email'); // null
// After 13.36: both return true / column object
Schema::getColumn('users', 'Email'); // Column object
Schema::getColumnType('users', 'Email'); // 'string'The fix normalises the column name to lowercase before lookup, matching the behaviour of hasColumn(). This matters most on PostgreSQL, where unquoted identifiers are folded to lowercase but quoted identifiers preserve case.
MySqlConnection: IsVitess() and olap()
Two new methods on Illuminate\Database\MySqlConnection help you detect the database flavour at runtime:
$connection = DB::connection()->getPdo();
if ($connection->isVitess()) {
// Adjust queries for Vitess routing rules, avoid unsupported features
}
if ($connection->olap()) {
// Optimise for analytical workloads: columnar storage, different indexes
}isVitess() checks for Vitess-specific variables; olap() detects ClickHouse, Apache Doris, and other OLAP engines that speak the MySQL protocol. Both are useful when your application runs against multiple database backends (e.g., primary on MySQL, replicas on Vitess, analytics on ClickHouse) and you need to adapt query behaviour dynamically.
Bug fixes worth noting
Scheduled callbacks no longer rethrow previous exceptions — a closure scheduled via
Schedule::call()could re-throw an exception from a prior run on subsequent executions.Unique queued listener locks release on transaction rollback — fixes a deadlock scenario where a rolled-back transaction left the lock held.
Sub-second process timeouts with CarbonInterval —
Process::timeout(CarbonInterval::seconds(0.5))now works correctly.FrankenPHP Mercure hub in CLI — the Mercure hub now works in CLI context; broadcast Mercure events are synchronous by default.
Notification::locale() respected after sending — locale set via
Notification::locale()was being ignored after the first notification sent.Morph map aliases resolved in constrain(), morphWith(), loadMorph() — custom morph map keys now work consistently across eager-loading methods.
When to upgrade
Laravel 13.36 is a patch release (the version scheme is 13.x for the current major). There are no breaking changes. The sticky connection flush is opt-in. The background queue connection only adds configuration. The validation builders are additive. You can upgrade with composer update laravel/framework and run your test suite — most applications will see zero required changes.
If you run queue workers with read replicas, enable $flushStickyConnections = true and observe your replica lag. If you dispatch to the background queue without a published config, the error disappears. If you write validation rules, try the builder API — it pays off in maintainability.
Takeaway
Laravel 13.36 doesn't introduce a new paradigm. It closes gaps: a missing queue connection, a worker behaviour that defeats read replicas, a validation API that wasn't quite complete, and a few sharp edges around queued mail and schema introspection. These are the kinds of changes that make production Laravel applications quieter and more predictable. Upgrade, flip the sticky-connection flag if you use replicas, and enjoy the autocomplete on your validation rules.