Critical bug: current_group and current_key were inside the while loop, causing them to be reset on each batch iteration. When an incomplete group spanned a batch boundary, it would be: 1. Buffered at end of batch N (in local current_group) 2. LOST when loop continued (new local variables created) 3. Re-fetched and yielded again in batch N+1 This caused the same consolidated record to be inserted many times. Solution: Move current_group and current_key OUTSIDE while loop to persist across batch iterations. Incomplete groups now properly merge across batch boundaries without duplication. Algorithm: - Only yield groups when we're 100% certain they're complete - A group is complete when the next key differs from current key - At batch boundaries, incomplete groups stay buffered for next batch - Resume always uses last_completed_key to avoid re-processing This fixes the user's observation of 27 identical rows for the same consolidated record. 🤖 Generated with Claude Code Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
2.0 KiB
2.0 KiB