Is co-partitioning required for KStream-to-GlobalKTable joins?

Is co-partitioning required for KStream-to-GlobalKTable joins?

• A. Yes
• B. No
• C. Only for certain topics
• D. It depends on partitions

Answer: (B)

GlobalKTable joins don’t require co-partitioning because the table data is replicated to every stream task. When you join a KStream with a GlobalKTable, each task has a local copy of the GlobalKTable, so the join can look up the table by key on the same task regardless of how the KStream is partitioned. This eliminates the need to align partitions between the stream and the table. Keep in mind, though, that replicating the GlobalKTable means more resource usage, since the table’s data is duplicated across all tasks. This trade-off is often acceptable for small or frequently looked-up lookup tables. For standard KStream-to-KTable joins, co-partitioning is typically required, which is why this distinction matters.

GlobalKTable joins don’t require co-partitioning because the table data is replicated to every stream task. When you join a KStream with a GlobalKTable, each task has a local copy of the GlobalKTable, so the join can look up the table by key on the same task regardless of how the KStream is partitioned. This eliminates the need to align partitions between the stream and the table.

Keep in mind, though, that replicating the GlobalKTable means more resource usage, since the table’s data is duplicated across all tasks. This trade-off is often acceptable for small or frequently looked-up lookup tables. For standard KStream-to-KTable joins, co-partitioning is typically required, which is why this distinction matters.