If you are not using key-value operations on Couchbase data nodes, do you not need to allocate a lot of memory?
We plan to use only the Index, Query, and Search (vector) services.
Dociments are stored in kv (data) nodes. Unless all your queries are “covered” you need the same amount. For non-covered queries, n1ql gets the document with a kv get. Less memory will mean that fewer documents are memory resident and accessing the document will be slower.
Is it correct to interpret it as:
“If all queries are index-covered, then the data node doesn’t need a lot of memory”?
The reason I keep asking is that the data volume is so large that I feel the need to increase the hardware memory significantly, which is a financial burden.
I’m only using Query and Vector Search, not KV.
If the data node also requires memory, would it be acceptable if only 50% of the data is resident in memory?
Covered queries only need data from indexes. So if you are never getting documents from the data nodes, then it doesn’t matter if getting documents from data nodes is slow.
If you start inserting many documents at such a rate that in-memory documents need to be persisted to make room for them, then inserting is going to slow down.
This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.