As I am performing some typesense tuning/performan...
# community-help
t
As I am performing some typesense tuning/performance experiments together with Claude, there is two problems we run into where the configuration does not apply as expected. Created two issues with the findings. The first is about configuration not being applied as it should when using arguments when starting typesense. But noticed that when using ENV variables it works as a workaround, so I know the tuning is very effecient when its applied. But probably more people that might experience that the settings doesnt work as expected and dont think its a good way to tune the system, when in fact it seems to be a bug. https://github.com/typesense/typesense/issues/2679 And the second "problem" found. I tried to set different --num-documents-parallel-load values to see how it affects the startup speed. But as I did not notice any significant difference over certain values so I asked claude to source dive typesense, and it seems to be capped by a hardcoded limit of 250 MB that overrules the --num-documents-parallel-load setting. https://github.com/typesense/typesense/issues/2683 Would love to hear teams feedback if there is some institutional knowledge that makes the 250 MB limit make sense, or if it should be configurable. Or if its just a AI hallucination 🙂 Have a nice day! And as always - thanks for a awesome product!
k
👋 The rocksdb configuration issue is certain a problem. We will fix it. As for batch size, the 250 MB limit existed in the pre-embedding generation era where you just didn't have such large document payloads. We will have to revisit it now.
t
Nice! Yes embeddings is a blessing and a curse 🙂 Thanks for response!