Closed RPKosec closed 1 month ago
how do we send them in batches?
how do we send them in batches?
I added draft PR, how I envisioned the solution. This way, no socket warnings are triggered and throttling error is omitted, but it still has way out in case of to many exceptions in a row or if any other exception is thrown to avoid infinite loop.
When using AWS Bedrock Embeddings on larger number of documents (larger Confluence space load for instance), AWS will return throttling error, which in turn will fail the upsert to vector store.
Steps to reproduce the behavior:
Expected behavior When upserting larger number of documents using AWS Bedrock Embeddings, no error should be raised. Throttling error should be mitigated using batches (with batch size settable by user). If throttling happens due to to many requests, retry with timeout between retries (increasing with iterations, which should be limited and limit set by user).
Flow Screenshots Basic example of flow, producing an error:
Setup
Additional context Error trace of the exception: