Feature Request: Add Timeout Parameter to Embedder Configuration

[miiton]

ref: meilisearch/meilisearch#5337

Description

I would like to request adding a timeout parameter to the embedder configuration settings. This would allow users to configure custom timeout values when working with embedder models, especially for local embedders that take longer to process.

Current Behavior

Currently, when using embedders like Ollama running locally, the default timeout seems to be around 30 seconds. This is too short for some embedder models, resulting in numerous timeout errors during document processing.

Requested Behavior

Add a timeout parameter to the embedder configuration that would allow setting a custom timeout duration in seconds. For example:

// MEILI_HOST/indexes/docs/settings/embedders
{
  "default": {
    "source": "ollama",
    "url": "http://ollama:11434/api/embed",
    "model": "nomic-embed-text",
    "timeout_seconds": 300,   // <------- THIS!
    "documentTemplate": "pageTitle: '{{doc.title}}' | documentType: '{{doc.document_type}}' | productName: '{{doc.product_name}}' | documentLanguage: '{{doc.document_language}}' | programmingLanguage: '{{doc.programming_language}}' | content: '{{doc.chunk_text}}'"
  }
}

Use Case

Although the practical performance implications might make this setup less ideal for production environments, having an adjustable timeout is valuable for:

1. Testing and experimenting with different embedding models
2. Development environments where performance is not the primary concern
3. Specialized use cases where longer processing times are acceptable

[macraig]

Hi @miiton , thanks for the feedback. Your feature request is reasonable, especially for local or slower embedders.

That said, we're being cautious about expanding the embedder configuration API surface. Introducing a timeout parameter likely means adding several related settings (e.g. number of tries, max duration of a single try before going to the next try, max delay between retries, batching logic, etc.). Since our current API is already quite dense, we'll need to figure out which parameters are important to add or not, and how to add them in a coherent way.

We’re discussing how best to group these kinds of settings. Your use case helps us shape that thinking, so we really appreciate you flagging it.