I am using the Couchbase Java SDK and need to update a field in a large number of documents (100k–200k).

The update depends on three fields, and I must update a fourth field inside a transactional context.

I am considering several approaches but I am unsure which is the most efficient and scalable.

Use Case

• Couchbase , Java

• Update condition: based on fieldA, fieldB, fieldC

• Update target: fieldD

• Estimated affected documents: 100k–200k

• Must run updates in a transaction

Approaches I am considering

1. Keyset pagination + batch update

• Fetch document IDs in chunks using keyset pagination.

• In each chunk, either:

  • Run a N1QL UPDATE ... USE KEYS [...] query

  • Perform KV mutateIn operations

I tried this but didn't perform not even better than N1Ql must have done something wrong

2. Query to fetch keys + multi-threaded batch updates

• Run SELECT META().id ... WHERE to get all matching document IDs.

• Execute N1QL update batches across multiple threads.

3. Single N1QL update statement

UPDATE bucket
SET fieldD = 
WHERE fieldA = ... AND fieldB = ... AND fieldC = ...;

• Simplest approach, but unsure about performance and transactional behavior for large datasets.

Questions

1. Which approach is recommended for updating 100k–200k Couchbase documents in a transaction?

2. Is keyset pagination + USE KEYS faster or safer than a single large N1QL update?

3. Does multi-threading improve performance inside a transaction?

4. Are there best practices for bulk transactional updates using the Java SDK?