Is there a way to configure the OpenAI text embedding request timeout?

[jonwwj]

I'm using a custom OpenAI-compatible endpoint for text embeddings via Milvus's built-in embedding function. My endpoint has higher latency than OpenAI's API, and some requests take longer than 30 seconds to complete, which causes them to fail with a timeout error.

Looking at the source code, the timeout appears to be hardcoded to 30 seconds in

milvus/internal/util/function/embedding/openai_embedding_provider.go

Lines 125 to 134 in d7f0751

	provider := OpenAIEmbeddingProvider{
	client: c,
	fieldDim: fieldDim,
	modelName: modelName,
	user: user,
	embedDimParam: dim,
	maxBatch: 128,
	timeoutSec: 30,
	extraInfo: extraInfo,
	}

There is no configuration option in milvus.yaml to override it.

Is there a supported way to adjust this timeout that I'm missing?

Thanks!

[yhmo]

All the callers are using the newOpenAIEmbeddingProvider() to return an OpenAIEmbeddingProvider. But it returns an OpenAIEmbeddingProvider with a hardcoded timeout value 30. So, it is not possible to configure the timeout value right now. We need a code change to expose the "timeoutSec". @junjiejiangjjj

[xiaofan-luan]

working on it