We're seeing a large number of the following error on Datadog:

2024-09-17 21:39:35 UTC | CORE | ERROR | (comp/dogstatsd/server/server.go:634 in errLog) | Dogstatsd: error parsing metric message '"openai.tokens.completion:undefined|d|#service:<service>,env:prod,version:1.29399.0,error:0,org:<org>,endpoint:/v1/embeddings,model:text-embedding-ada-002"': could not parse dogstatsd metric values: strconv.ParseFloat: parsing "undefined": invalid syntax

(the error is specifically the float conversion of undefined and we're only seeing it for embeddings). We have the OpenAI integration enabled which appears to be working for chat completion requests.

I briefly looked through the corresponding code in the tracer:

is the value that is parsed, and is set here:

This appears to be the right part of the code given that the openai response is for an embeddings.create request:

I checked with version 4.62.0 of the openai npm SDK and observed that both the return type and the actual response don't have body.usage.completion_tokens as a valid field of the response:

https://github.com/openai/openai-node/blob/7110f07cb79f27f69341b01b75f8a1f102ceab21/src/resources/embeddings.ts#L41-L56

Is there something I'm missing or does the tracing code need to account for this field not being present?