Skip to content

Meilisearch

The Wodby Meilisearch catalog stack runs Meilisearch, a search engine for websites and apps, as a single instance. The current services and defaults are documented in wodby/stack-meilisearch.

API keys

Every Meilisearch request except GET /health needs an API key, sent as Authorization: Bearer <key>.

Wodby generates the master key when the app is created. It is the master_key token on the Meilisearch app service's Configuration > Tokens tab, and the MEILI_MASTER_KEY variable inside the container. See Tokens. Use the master key only to manage API keys, and keep it on your servers. Never put it in browser code or in a repository.

Meilisearch creates default keys from the master key. List them with the master key:

curl -H "Authorization: Bearer <master key>" https://<meilisearch domain>/keys
  • Give browsers and mobile apps the Default Search API Key. It can only search.
  • Give backends that add or update documents the Default Admin API Key, or a narrower key you create with POST /keys.

See Meilisearch's security guide.

Changing the master key changes every key derived from it. The default keys and every key you created stop working, and your clients need the new keys.

Connect to Meilisearch

Meilisearch has a public HTTPS domain, so browsers and mobile apps can search it directly. Services in the same app environment can use http://<app service name>:7700 instead.

Wodby does not protect the Meilisearch domain with basic auth, and you should not add one under Endpoints > Auths. Meilisearch reads its API key from the Authorization header, the same header basic auth uses, so search and indexing requests would fail. The API keys protect the domain.

Meilisearch accepts document batches of up to 100 MB. The domain allows 5 minutes per request so that large batches can finish uploading. Do not set the route's request body size: it makes the route hold each request in memory until it is complete. See HTTP settings.

Development and production mode

The mode follows the app environment's type:

  • In dev environments, Meilisearch runs in development mode. Opening the domain in a browser shows Meilisearch's search preview, where you can try searches with an API key.
  • In prod, staging, test, and feature environments, Meilisearch runs in production mode, without the search preview.

In both modes, every request except GET /health needs an API key.

Change configuration

Meilisearch settings are MEILI_* environment variables on the Meilisearch app service. Change them and deploy the service. See Environment variables and Meilisearch's configuration reference.

Index settings, such as searchable, filterable, and sortable attributes, ranking rules, and synonyms, are set per index through the API.

Meilisearch's telemetry is off.

Version updates

A Meilisearch database opens only with the version that wrote it. When a stack upgrade brings a newer Meilisearch, Meilisearch upgrades the database in place when it starts. See Upgrade. Searches keep working during the upgrade, and new tasks wait until it is done. Follow it with GET /tasks?types=upgradeDatabase.

Meilisearch warns that this upgrade is not atomic and can, rarely, corrupt the database. Before you upgrade a production environment, create a dump with POST /dumps, or make sure you can reindex from your application's own data. A failed or stuck upgrade task can be cancelled, which rolls the database back. See Meilisearch's update guide.

New non-production environments upgrade their stack automatically by default. Turn auto-upgrade off for environments whose upgrades you want to control.

Data and backups

The Meilisearch volume (10 GB by default) holds the database. Dumps and snapshots that you create with the API are written to the same volume and use its space. Raise the volume size for large datasets.

The Meilisearch service has no backup or import operation in Wodby. Create dumps or snapshots with the API, or keep the source data in your application so that you can reindex it.

Limits

  • The Meilisearch service is fixed to one replica. It runs one instance on one volume and can't be scaled.
  • The stack requests 0.2 CPU and 512 MiB of memory for Meilisearch and limits it to 2 CPU and 2048 MiB. Indexing uses up to two thirds of the memory limit, so raise the limit for large indexing batches. Change resources under Runtime > Resources.