Replace stream/filetype detection library with puremagic, 20Mb less RAM usage (#3491)

changedetectionio/async_update_worker.py

@@ -334,6 +334,10 @@ async def async_update_worker(worker_id, q, notification_q, app, datastore):
                             if update_handler.fetcher.content or (not update_handler.fetcher.content and empty_pages_are_a_change):
                                 watch.save_last_fetched_html(contents=update_handler.fetcher.content, timestamp=int(fetch_start_time))
 
+                            # Explicitly delete large content variables to free memory IMMEDIATELY after saving
+                            # These are no longer needed after being saved to history
+                            del contents
+
                             # Send notifications on second+ check
                             if watch.history_n >= 2:
                                 logger.info(f"Change detected in UUID {uuid} - {watch['url']}")
@@ -372,6 +376,12 @@ async def async_update_worker(worker_id, q, notification_q, app, datastore):
                 datastore.update_watch(uuid=uuid, update_obj={'fetch_time': round(time.time() - fetch_start_time, 3),
                                                                'check_count': count})
 
+                # NOW clear fetcher content - after all processing is complete
+                # This is the last point where we need the fetcher data
+                if update_handler and hasattr(update_handler, 'fetcher') and update_handler.fetcher:
+                    update_handler.fetcher.clear_content()
+                    logger.debug(f"Cleared fetcher content for UUID {uuid}")
+
         except Exception as e:
             logger.error(f"Worker {worker_id} unexpected error processing {uuid}: {e}")
             logger.error(f"Worker {worker_id} traceback:", exc_info=True)
@@ -392,7 +402,28 @@ async def async_update_worker(worker_id, q, notification_q, app, datastore):
                         #logger.info(f"Worker {worker_id} sending completion signal for UUID {watch['uuid']}")
                         watch_check_update.send(watch_uuid=watch['uuid'])
 
-                    update_handler = None
+                    # Explicitly clean up update_handler and all its references
+                    if update_handler:
+                        # Clear fetcher content using the proper method
+                        if hasattr(update_handler, 'fetcher') and update_handler.fetcher:
+                            update_handler.fetcher.clear_content()
+
+                        # Clear processor references
+                        if hasattr(update_handler, 'content_processor'):
+                            update_handler.content_processor = None
+
+                        update_handler = None
+
+                    # Clear local contents variable if it still exists
+                    if 'contents' in locals():
+                        del contents
+
+                    # Note: We don't set watch = None here because:
+                    # 1. watch is just a local reference to datastore.data['watching'][uuid]
+                    # 2. Setting it to None doesn't affect the datastore
+                    # 3. GC can't collect the object anyway (still referenced by datastore)
+                    # 4. It would just cause confusion
+
                     logger.debug(f"Worker {worker_id} completed watch {uuid} in {time.time()-fetch_start_time:.2f}s")
                 except Exception as cleanup_error:
                     logger.error(f"Worker {worker_id} error during cleanup: {cleanup_error}")

changedetectionio/content_fetchers/base.py

@@ -64,6 +64,19 @@ class Fetcher():
     # Time ONTOP of the system defined env minimum time
     render_extract_delay = 0
 
+    def clear_content(self):
+        """
+        Explicitly clear all content from memory to free up heap space.
+        Call this after content has been saved to disk.
+        """
+        self.content = None
+        if hasattr(self, 'raw_content'):
+            self.raw_content = None
+        self.screenshot = None
+        self.xpath_data = None
+        # Keep headers and status_code as they're small
+        logger.trace("Fetcher content cleared from memory")
+
     @abstractmethod
     def get_error(self):
         return self.error

changedetectionio/processors/magic.py

@@ -64,24 +64,31 @@ class guess_stream_type():
         # Remove whitespace between < and tag name for robust detection (handles '< html', '<\nhtml', etc.)
         test_content_normalized = re.sub(r'<\s+', '<', test_content)
 
-        # Magic will sometimes call text/plain as text/html!
+        # Use puremagic for lightweight MIME detection (saves ~14MB vs python-magic)
         magic_result = None
         try:
-            import magic
+            import puremagic
 
-            mime = magic.from_buffer(content[:200], mime=True) # Send the original content
-            logger.debug(f"Guessing mime type, original content_type '{http_content_header}', mime type detected '{mime}'")
-            if mime and "/" in mime:
-                magic_result = mime
-                # Ignore generic/fallback mime types from magic
-                if mime in ['application/octet-stream', 'application/x-empty', 'binary']:
-                    logger.debug(f"Ignoring generic mime type '{mime}' from magic library")
-                # Trust magic for non-text types immediately
-                elif mime not in ['text/html', 'text/plain']:
-                    magic_content_header = mime
+            # puremagic needs bytes, so encode if we have a string
+            content_bytes = content[:200].encode('utf-8') if isinstance(content, str) else content[:200]
+
+            # puremagic returns a list of PureMagic objects with confidence scores
+            detections = puremagic.magic_string(content_bytes)
+            if detections:
+                # Get the highest confidence detection
+                mime = detections[0].mime_type
+                logger.debug(f"Guessing mime type, original content_type '{http_content_header}', mime type detected '{mime}'")
+                if mime and "/" in mime:
+                    magic_result = mime
+                    # Ignore generic/fallback mime types
+                    if mime in ['application/octet-stream', 'application/x-empty', 'binary']:
+                        logger.debug(f"Ignoring generic mime type '{mime}' from puremagic library")
+                    # Trust puremagic for non-text types immediately
+                    elif mime not in ['text/html', 'text/plain']:
+                        magic_content_header = mime
 
         except Exception as e:
-            logger.error(f"Error getting a more precise mime type from 'magic' library ({str(e)}), using content-based detection")
+            logger.error(f"Error getting a more precise mime type from 'puremagic' library ({str(e)}), using content-based detection")
 
         # Content-based detection (most reliable for text formats)
         # Check for HTML patterns first - if found, override magic's text/plain

changedetectionio/processors/text_json_diff/processor.py

@@ -556,6 +556,20 @@ class perform_site_check(difference_detection_processor):
             else:
                 logger.debug(f"check_unique_lines: UUID {watch.get('uuid')} had unique content")
 
+        # Note: Explicit cleanup is only needed here because text_json_diff handles
+        # large strings (100KB-300KB for RSS/HTML). The other processors work with
+        # small strings and don't need this.
+        #
+        # Python would clean these up automatically, but explicit `del` frees memory
+        # immediately rather than waiting for function return, reducing peak memory usage.
+        del content
+        if 'html_content' in locals() and html_content is not stripped_text:
+            del html_content
+        if 'text_content_before_ignored_filter' in locals() and text_content_before_ignored_filter is not stripped_text:
+            del text_content_before_ignored_filter
+        if 'text_for_checksuming' in locals() and text_for_checksuming is not stripped_text:
+            del text_for_checksuming
+
         return changed_detected, update_obj, stripped_text
 
     def _apply_diff_filtering(self, watch, stripped_text, text_before_filter):