streaming-export-safety

Use SXSSFWorkbook with a window size (e.g., 100 rows) instead of XSSFWorkbook, which keeps the entire sheet in memory. Enable temp-file compression, write to an OutputStream, and always call dispose() in a finally block to remove temporary files.

XSSFWorkbook loads the whole workbook into memory, while SXSSFWorkbook streams rows beyond a sliding window to temp files. SXSSF cannot randomly access flushed rows, does not support formula recalculation or cloneSheet, and requires dispose() to clean up.

Use encoding/csv with csv.NewWriter and write rows one at a time. You must call w.Flush() (commonly via defer) because the writer buffers data in memory; forgetting Flush leaves all rows unwritten to disk.

Yes, create the workbook with Workbook(write_only=True) and append rows via ws.append(). In write-only mode you cannot use ws.cell() for random access or modify previously written rows.

Synchronous exports taking over 30 seconds get cut off by Nginx or gateways and block HTTP threads. Convert the endpoint to submit an async task that returns a task ID immediately, generates the file in the background, uploads it to object storage, and exposes a status polling endpoint.

Use JSONL when exporting very large datasets, since each line is an independent JSON object that downstream consumers can parse incrementally. Standard JSON arrays require reading the entire file before parsing, which reintroduces memory pressure.