Gap
models.generateVideos() in @google/genai is not instrumented. The method generates videos from a text description (Veo model) and is exposed as a public generative execution API on the Models class.
Evidence
- SDK source:
src/models.ts in googleapis/js-genai exposes generateVideos() as a public method on the Models class.
- Current instrumentation:
js/src/instrumentation/plugins/google-genai-channels.ts and google-genai-plugin.ts only cover models.generateContent, models.generateContentStream, and models.embedContent.
- Vendor types:
js/src/vendor-sdk-types/google-genai.ts — GoogleGenAIModels interface does not include generateVideos.
- Auto-instrumentation config:
js/src/auto-instrumentations/configs/google-genai.ts has no entry for video generation.
Notes
generateVideos uses a long-running operation (polling) pattern — it returns an operation object that is polled until complete. Instrumentation may need to wrap the full polling lifecycle to capture accurate latency and output, similar to how streaming completions are aggregated. This complexity may warrant a deferred or best-effort approach (e.g., wrapping the initial call and recording when the operation resolves).
Gap
models.generateVideos()in@google/genaiis not instrumented. The method generates videos from a text description (Veo model) and is exposed as a public generative execution API on theModelsclass.Evidence
src/models.tsingoogleapis/js-genaiexposesgenerateVideos()as a public method on theModelsclass.js/src/instrumentation/plugins/google-genai-channels.tsandgoogle-genai-plugin.tsonly covermodels.generateContent,models.generateContentStream, andmodels.embedContent.js/src/vendor-sdk-types/google-genai.ts—GoogleGenAIModelsinterface does not includegenerateVideos.js/src/auto-instrumentations/configs/google-genai.tshas no entry for video generation.Notes
generateVideosuses a long-running operation (polling) pattern — it returns an operation object that is polled until complete. Instrumentation may need to wrap the full polling lifecycle to capture accurate latency and output, similar to how streaming completions are aggregated. This complexity may warrant a deferred or best-effort approach (e.g., wrapping the initial call and recording when the operation resolves).