Memory management - Run HTML to text in sub process, a few more cleanups to playwright (#3110)

changedetectionio/content_fetchers/playwright.py

@@ -59,7 +59,10 @@ def capture_full_page(page):
         p.join()
         logger.debug(
             f"Screenshot (chunked/stitched) - Page height: {page_height} Capture height: {SCREENSHOT_MAX_TOTAL_HEIGHT} - Stitched together in {time.time() - start:.2f}s")
-
+        # Explicit cleanup
+        del screenshot_chunks
+        del p
+        del parent_conn, child_conn
         screenshot_chunks = None
         return screenshot
 
@@ -286,12 +289,28 @@ class fetcher(Fetcher):
                     pass
                 
                 # Clean up resources properly
-                context.close()
+                try:
-                context = None
+                    self.page.request_gc()
+                except:
+                    pass
 
+                try:
                     self.page.close()
+                except:
+                    pass
                 self.page = None
 
-                browser.close()
+                try:
-                borwser = None
+                    context.close()
+                except:
+                    pass
+                context = None
+
+                try:
+                    browser.close()
+                except:
+                    pass
+                browser = None
+
+
 

changedetectionio/html_tools.py

@@ -435,7 +435,9 @@ def cdata_in_document_to_text(html_content: str, render_anchor_tag_content=False
 
     return re.sub(pattern, repl, html_content)
 
-def html_to_text(html_content: str, render_anchor_tag_content=False, is_rss=False) -> str:
+
+def html_to_text_sub_worker(conn, html_content: str, render_anchor_tag_content=False, is_rss=False):
+
     from inscriptis import get_text
     from inscriptis.model.config import ParserConfig
 
@@ -470,9 +472,19 @@ def html_to_text(html_content: str, render_anchor_tag_content=False, is_rss=Fals
         html_content = re.sub(r'</title>', r'</h1>', html_content)
 
     text_content = get_text(html_content, config=parser_config)
+    conn.send(text_content)
+    conn.close()
 
-    return text_content
+# NOTE!! ANYTHING LIBXML, HTML5LIB ETC WILL CAUSE SOME SMALL MEMORY LEAK IN THE LOCAL "LIB" IMPLEMENTATION OUTSIDE PYTHON
+def html_to_text(html_content: str, render_anchor_tag_content=False, is_rss=False):
+    from multiprocessing import Process, Pipe
 
+    parent_conn, child_conn = Pipe()
+    p = Process(target=html_to_text_sub_worker, args=(child_conn, html_content, render_anchor_tag_content, is_rss))
+    p.start()
+    text = parent_conn.recv()
+    p.join()
+    return text
 
 # Does LD+JSON exist with a @type=='product' and a .price set anywhere?
 def has_ldjson_product_info(content):