Server-side vs client-side timeout enforcement for long-running UNWIND queries in Neo4j 5.x

Decision Point

Neo4j 5.x provides two distinct mechanisms for handling long-running queries: server-side timeouts via tx.timeout() and client-side cancellation through Bolt channel close(). The choice between these approaches creates different resource implications for queries processing large UNWIND datasets that can trigger memory pressure.

Server-side timeout enforcement using tx.timeout() requires Neo4j 4.2+ and terminates queries at the database level, immediately freeing server resources. Client-side cancellation via Bolt channel close() may leave server resources occupied until the query completes or times out naturally, potentially causing connection pool exhaustion.

The constraint is that UNWIND operations on large datasets can consume significant memory, and standard timeout mechanisms may not catch memory pressure promptly. Neo4j 5.x introduced enhanced transaction metadata tracking that improves visibility into timeout-related transaction states, but the behavior differs between client and server enforcement approaches.

Key questions:

• Which timeout enforcement approach provides more predictable resource cleanup for UNWIND operations processing millions of records?
• How does the error messaging and transaction state visibility differ between server-side tx.timeout() and client-initiated cancellation in Neo4j 5.x?
• Does client-side cancellation properly propagate through connection pools to prevent zombie transactions when handling large UNWIND datasets?