deepset-ai · bogdankostic · Jun 18, 2026 · Jun 17, 2026
@@ -111,6 +111,10 @@ def run(
                 logger.warning("Could not read {source}. Skipping it. Error: {error}", source=source, error=e)
                 continue
 
+            if not bytestream.data:
+                logger.warning("Skipping {source} because it is empty.", source=source)
+                continue
+
             try:
                 text = extract(bytestream.data.decode("utf-8"), **merged_extraction_kwargs)
             except Exception as conversion_e:

@@ -0,0 +1,6 @@
+---
+fixes:
+  - |
+    ``HTMLToDocument`` now skips ``ByteStream`` objects with empty content instead of
+    passing them to trafilatura. This prevents noisy lxml parse error logs (such as
+    "Document is empty") when a fetcher emits an empty stream.
@@ -156,6 +156,23 @@ def test_run_error_handling(self, caplog):
             assert "Could not read non_existing_file.html" in caplog.text
             assert results["documents"] == []
 
+    def test_run_empty_bytestream(self, caplog):
+        """
+        Test that an empty ByteStream is skipped without invoking extraction,
+        so no noisy lxml parse errors are emitted.
+        """
+        empty_stream = ByteStream(data=b"")
+        empty_stream.mime_type = "text/html"
+        converter = HTMLToDocument()
+
+        with patch("haystack.components.converters.html.extract") as mock_extract:
+            with caplog.at_level(logging.WARNING):
+                results = converter.run(sources=[empty_stream])
+
+        assert results["documents"] == []
+        mock_extract.assert_not_called()
+        assert "because it is empty" in caplog.text
+
     def test_mixed_sources_run(self, test_files_path):
         """
         Test if the component runs correctly if the input is a mix of paths and ByteStreams.