How do you handle upserts and deletes at scale in Pinecone?

Pinecone upserts insert or overwrite vectors by ID in batches, typically 100 to 1000 vectors per request for best throughput. Deletes can target specific IDs, all vectors in a namespace, or filtered subsets by metadata. Writes are eventually consistent, so a just-written vector may not appear in queries for a short window, typically under a second in serverless indexes. For high-volume ingest, use concurrent batches and the async client, and monitor index stats to confirm vector counts reach expected values.