On 3/12/24 05:44, Szűcs Roland wrote:
I tried cbor-based and json-based approaches but none of them worked. The number of documents I tried to index in one request was 15000 but the document size is small: 0.5kb.
Make batch size a parameter and run timings with different sizes. In my setup I found that 5000 docs is the sweet spot for single node and 3000: for a 2-node mirrored (num shards = 1, num replicas = 2) "cloud".
I strongly suspect at 15000 you're DoS'ing your server(s). Dima
