We tend to see a pretty strong relationship betwee...
# community-help
t
We tend to see a pretty strong relationship between pending write batches and CPU throttling. Would anyone have recommendations on like write rates to aim for on specific clusters for this stuff, or how they reduce the overhead of writes? Would removing faceting from fields get us significant reductions here, that sorta thing?
j
Batching multiple writes into a fewer API calls using the import endpoint would give you the most write performance gains
I would recommend at least 1K writes per API call (and then increasing the client-side timeout)
Besides that, indexing time is indeed proportional to the number of fields in the schema, whether faceting is enabled for a field, whether sorting is enabled for a string field, if characters specified in token_separators end up leading to a lot of words, etc
t
@Jason Bosco - Thank you for this advice! Wow, we are typically doing one write per API call usually 😅
😄 1
@Jason Bosco - We have been doing some troubleshooting and picking up some low hanging fruit, and our write performance is already looking amazingly better. Thank you for all of your help and advice, and of course an amazing product!!!
🎉 1
🙌 1
j
That's great to hear! Happy to help