This is more for the typesense team, but have y'all explored optimizations, or have any recommendations for large-ish datasets? My use case is 500M-1B records of ~200bytes. Search time's are okay, but cold starts at this size are the bottleneck right now
j
Jason Bosco
03/16/2026, 12:06 PM
Building the in-memory index does take time proportionally to the amount of docs you have in the collection.
You could speed it up using the following parameters (assuming you have enough cores to handle the additional parallellization):
Copy code
num-collections-parallel-load
Copy code
num-documents-parallel-load
Jason Bosco
03/16/2026, 12:07 PM
Another way to side-step the load time is by running a 3-node cluster, so that if you need to upgrade the cluster, you can do it one node at a time, so the other two nodes in the cluster will continue serving traffic, while the 3rd node is re-loading data