Index aliases

Aliases allow you to use an index under a different name.This very useful for zero downtime deployments.

There are three aliases created: a read alias, a write alias and a history alias. The read alias is used for reading, the write index is used for writing. The history aggregates all old indices, which can be pruned. When updating an index it is recreated to a new index with a unique name. The “write” alias is pointed to the new index and all Scout updates will be forwarded to the “write” index. After all entities are imported the “read” alias will also be pointed to the new index.

If you wish to keep the old indices set prune_old_aliases to false in config/explorer.php

Using aliases

A model is only using index aliases if it implements the Aliased interface or enabled in the configuration (see mapping).

After that, any time you use the elastic:update command a new index will be created and when the insertion of models is done the alias will be pointed to the new index.

<?php

namespace App\Models;

use Illuminate\Database\Eloquent\Factories\HasFactory;
use Illuminate\Database\Eloquent\Model;
use JeroenG\Explorer\Application\Explored;
use JeroenG\Explorer\Application\Aliased;
use Laravel\Scout\Searchable;

class Post extends Model implements Explored, Aliased
{
    use HasFactory;
    use Searchable;

    //...
}

return [
    'indexes' => [
        'posts' => [
            'aliased' => true,
            'properties' => [
                'id' => 'keyword',
                'title' => 'text',
                'created_at' => 'date',
                'published' => 'boolean',
                'author' => 'nested',
            ],
        ],
    ],
];

Be aware that if you currently already have indices and would like to move to using aliases you will need to delete those indices before configuring the aliases. In Elasticsearch a given name can only be either an index or alias, not both and this cannot be changed on-the-fly.

Note on updating aliases

When you update a model, Laravel Scouts will update the index. When you use index aliases, a new index is created and the alias is being pointed to the nex one. What you don’t want is for the alias to be pointing to the new index before Elasticsearch is done with indexing all documents. To prevent this, the alias update is done in a job that is dispatched to the queue. If there is no queue it will still be done in the background, but it will be done synchronously. This could still be enough of a “delay” for Elasticsearch to finish indexing, so there is no immediate need to set up a queue.