README

Search for Filagora that does not tie you to one engine. It defines a SearchEngine contract and a portable filter tree, ships a Meilisearch engine and a database fallback, and keeps the index fresh and healthy with commands. Requires filagora/filagora-core.

The package is at 0.1: the API may change between 0.x minor versions. There is no frontend: you build the search pages.

How it fits together

A package (or your app) registers a SearchIndex: one kind of thing, how a model becomes a Document, and a fallback() that answers the same query from the database.
You search through Search: app(Search::class)->search('threads', new SearchQuery('paypal', $filter)). It returns ids; you load the rows and drop anything the index describes wrongly (an index is only as fresh as its last sync).
The engine is filagora-search.engine: database (the default; no server), array (tests), or one you add with SearchEngines::extend(). If the engine throws, searches are answered from the database for circuit.cooldown_seconds and the failure is logged.
Filters are engine-neutral: new All(new Equals('forum_id', 3), new Range('reply_count', gte: 2)).

Keeping the index fresh

Saving, deleting or restoring a model that belongs to a registered index updates its document (after the transaction commits). Set FILAGORA_SEARCH_QUEUE=true to do that on the queue. Failures are logged and never break the save.

php artisan filagora:search:sync-settings            # create indexes and apply their settings
php artisan filagora:search:reindex --flush          # send every searchable row
php artisan filagora:search:reindex --index=threads --from-id=5000 --batch-size=200 --sleep-ms=50
php artisan filagora:search:backfill threads filters # one field group, as partial updates
php artisan filagora:search:check-health             # counts and failed tasks; non-zero on a problem

filagora:doctor also reports the engine's health. With the database engine all of these say there is nothing to do.

Using Meilisearch

FILAGORA_SEARCH_ENGINE=meilisearch
MEILISEARCH_HOST=http://localhost:7700
MEILISEARCH_KEY=your-key

The engine speaks Meilisearch's HTTP API directly. Changes are applied by Meilisearch after it has answered, so a failure shows up as a failed task; syncSettings() waits for its own tasks and MeilisearchEngine::waitForTasks() waits for the rest. Every word of a query must match, and a total that reaches meilisearch.max_total_hits (1000, Meilisearch's counting limit) is reported with totalIsCapped.

Writing an engine

Implement Filagora\Search\Contracts\SearchEngine, register it, and run the conformance checks in your tests. If your engine applies writes later, pass a closure that waits until it has caught up:

foreach (EngineConformance::checks($engine, fn () => $engine->waitForTasks()) as $name => $check) {
    it("conforms: {$name}", $check);
}

Install

composer require filagora/filagora-search

Nothing to migrate: the package owns no tables. It uses the database engine until you set FILAGORA_SEARCH_ENGINE. Packages that have searchable things (forums, cms) register their own indexes when this package is installed.

Testing

composer test

Tests run against a real MariaDB database named testing.

Licence

MIT. See LICENSE.

filagora / filagora-search

Package info

Security

Statistics