[6.x] Fix stale asset listings across queued jobs

[ryanmitchell]

I had Claude try and figure out Rob's issue, and it pinpointed the blink cache on the AssetContainer's contents().

This is its logic:

In a long-running queue:work worker, Blink persists for the entire process lifetime. All jobs share the same AssetContainerContents instance. The isWorker() guards on all(), filteredFilesIn(), and filteredDirectoriesIn() are there to prevent stale in-memory state from one job bleeding into the next. But metaFilesIn() has no isWorker() guard:

// AssetContainerContents.php:182
public function metaFilesIn($folder, $recursive)
{
if (isset($this->metaFiles[$key = $folder.($recursive ? '-recursive' : '')])) {
return $this->metaFiles[$key]; // ← No isWorker() check — always stale across jobs
}
// ...
return $this->metaFiles[$key] = $files;
}

More importantly, the filteredFiles/filteredDirectories instances are only cleared via add(). If some other code path reads from the container's contents after job A runs (before the next add() call in job B), it gets the Blink-cached instance with the stale $this->files.

Why Horizon works

With php artisan queue:work, a single process runs all jobs and Blink accumulates indefinitely. With Horizon, each horizon:work subprocess handles a bounded number of jobs before being replaced (via memory limits or maxProcesses). Fresh subprocesses get fresh Blink state. This is the most likely explanation for the Horizon/queue:work asymmetry.

Why private containers are affected more than public

This is the trickier part. The private() check only affects URL generation — it has no effect on listing or caching code paths. The actual difference is almost certainly how the assets are being queried in your front-end code:

• Public assets may be queried/linked by direct URL (e.g. asset.url → /assets/image.jpg), which requires no Stache lookup — it just works if the file is on disk
• Private assets must be served through a controller that calls Asset::find() or goes through queryAssets() — requiring the Stache to know about them

If the listing cache or Stache is stale, private assets simply don't exist from Statamic's perspective, while public assets might still be reachable by URL.

The clean fix

The real solution is to not Blink-cache the AssetContainerContents instance, since the whole point of isWorker() guards inside it is to defeat the in-memory caching anyway

This way, workers always get a fresh AssetContainerContents instance per call (no shared state between jobs), while normal web requests still get the per-request Blink cache benefit. The metaFilesIn() stale-data gap also disappears since there's no cross-job instance sharing.

Closes #14576

[ryanmitchell]

Rob confirmed this fixed his issue.

[robdekort]

How nice. Thanks!