Maybe this will also give some more info. The document imports are ran in a queue consumer. That means there could be concurrency of up to 20 import requests per second (all having 20-30 docs). I've been running it turned down for testing so normally there's about 450 total documents that I attempt to import. Maybe that dog piling effect is causing something to choke up? This is the callsite:

const contentResp = await typesense
  .collections(collectionName)
  .documents()
  .import(data, {
    dirty_values: 'reject',
    action: 'emplace',
  });