Reduce conversion memory footprint

app/services/conversion.py

@@ -0,0 +1,47 @@
+from dataclasses import dataclass
+from pathlib import Path
+
+from app.services.exporter import ExportResult
+from app.services.parser import create_parse_session
+from app.services.processing import ProcessingOptions, filter_records, sort_records
+from app.services.storage import ResultMetadata, persist_result
+
+
+@dataclass(slots=True)
+class ConversionResult:
+    metadata: ResultMetadata
+    export_result: ExportResult
+    parsed_count: int
+    filtered_count: int
+
+
+def convert_uploaded_log(
+    stream,
+    options: ProcessingOptions,
+    output_dir: Path,
+    output_format: str,
+    preview_record_limit: int,
+) -> ConversionResult:
+    """Convert an uploaded log into a persisted export with a small in-memory preview.
+
+    Parsing, filtering, and export writing are streamed to keep memory usage low.
+    Sorting still materializes the filtered records because global ordering by datetime
+    or severity requires seeing the whole filtered result set first.
+    """
+    parse_session = create_parse_session(stream)
+    sorted_records = sort_records(filter_records(parse_session.iter_records(), options), options)
+    metadata, export_result = persist_result(
+        output_dir=output_dir,
+        records=sorted_records,
+        union_keys=parse_session.union_keys(),
+        mode=options.mode,
+        output_format=output_format,
+        preview_record_limit=preview_record_limit,
+    )
+
+    return ConversionResult(
+        metadata=metadata,
+        export_result=export_result,
+        parsed_count=parse_session.parsed_count,
+        filtered_count=len(sorted_records),
+    )