You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently it's not possible to specify the number of shards/replicas when working with OpenSerach vector store.
This is problematic when using Langchain with OpenSearch in a multi-tenant setting, where we want to have data segregation but still use the same OpenSearch cluster => one tenant per index.
This setting hits rapidly the max_shards_per_node limit of OpenSearch cluster.
PS: it is possible to modify the max_shards_per_node for an OpenSearch cluster, but having the flexibility when using Langchain is also necessary (+ in our case number of documents per tenant is quite low, so one shard is more than enough).