SET BLOB in dynamic SQL: unclear buffer reuse after Firebird 4.0.2 patch

Buffer Allocation Behavior

In Firebird 4.0.2, the server allocates a fresh temporary buffer for each SET BLOB execution within dynamic SQL. These buffers are not reused across different statement executions, even within the same transaction. Following the 4.0.2 memory-leak patches, the engine enforces a stricter allocation-deallocation cycle to ensure internal structures are properly freed.

Memory Release Lifecycle

Temporary buffers used for BLOB staging are released immediately after the statement completes, rather than waiting for the transaction commit. However, because Firebird utilizes internal memory pools, the memory may not be returned to the operating system immediately. Instead, it is marked as available for the next single-statement request within that attachment.

Impact of Event Monitoring

Enabling Event Monitoring for BLOB write events introduces additional overhead. While it does not change the fundamental allocation logic of the SET BLOB statement, it requires the server to maintain additional tracking metadata for each write operation. In high-throughput scenarios involving BLOBs larger than 1 GB, this can lead to a higher memory high-water mark, as the server must track the event state alongside the data buffer.

Verification Steps

To verify if your specific environment is reusing buffers or leaking memory, use the following approach within a single long-running transaction:

1. Execute a loop of SET BLOB statements using a consistent BLOB size.
2. Monitor the MON$MEMORY_USAGE table during the loop.
3. Analyze the trend:
   • Plateau: If memory usage peaks and then stabilizes, the buffers are being correctly freed and reused by the pool.
   • Linear Growth: If memory usage climbs steadily with every execution, a leak is present or buffers are not being released.

SELECT MON$MEMORY_USAGE FROM MON$ATTACHMENTS WHERE MON$ATTACHMENT_ID = CURRENT_CONNECTION;

Assumption and Diagnostic Requirement

This analysis assumes the BLOBs are being handled via standard BLOB_SUB_TYPE storage. If you are using a custom BLOB ID management strategy or external file-linking, the allocation path changes. To refine this recommendation, please specify if the 1 GB+ BLOBs are being streamed in chunks or passed as a single contiguous parameter.