Debounce the explicit gc.collect() storm on the watch-check path

Several places call gc.collect() after a check to keep C-level memory
(pyppeteer buffers, libxml2 documents, PIL, brotli) from accumulating.
Individually each is reasonable; run concurrently by many fetch workers they
become a storm. Every gc.collect() is a full stop-the-world pass that walks the
whole heap holding the GIL, so at FETCH_WORKERS=50 the process spends most of
its time stopped in the collector - which also starves each worker's asyncio
loop, leaving its CDP websocket unread.

Measured on 153 real puppeteer checks of a live site at FETCH_WORKERS=50, with
an `//div` include filter so the lxml document tree is realistic:

                     collects  objects freed  gc time  checks/sec  CPU/check     RSS
  one per call site       790      2,594,098    91.3s       0.766     1.373s  279.6MB
  debounced to 1s          48      2,219,010     7.3s       1.433     0.731s  275.3MB
  none at all               0              0     0.0s       1.503     0.685s  287.7MB

Debouncing keeps 86% of the reclamation for 6% of the collections: 1.9x the
throughput, half the CPU per check, gc down from 30.8% to 6.4% of wall time,
and a lower resident plateau than collecting every time. It works because the
collector is process-wide - any worker's collection breaks every other worker's
cycles too, so with many workers the calls are overwhelmingly redundant
duplicates rather than independently necessary.

Removing them entirely is slightly faster still, but it was the only
configuration whose RSS had not plateaued by the end of the run, so it is not
the default. EXPLICIT_GC_MIN_INTERVAL=0 restores the previous behaviour.

Collecting a younger generation was measured and rejected: gen 0 freed 1,136
objects against the full pass's 2,594,098, because objects surviving a 10-30s
fetch have already been promoted out of gen 0.

Two call sites are additionally fixed because they could never reclaim anything:

- Watch.py brotli: brotli.Compressor is not gc-tracked, so the collector cannot
  see it - `del` frees it by refcount. Over 60 x 2.2MB compressions, RSS growth
  was +0.9MB with neither mechanism, +0.2MB with gc.collect() alone, and +0.0MB
  with malloc_trim() alone or with both. malloc_trim is the load-bearing line
  and is kept; the collect cost ~31ms of stop-the-world per snapshot save for no
  reclamation.

- puppeteer quit(): runs twice per check (run()'s finally, then the worker's
  safety net) and nulls self.page/self.browser in its own finally blocks, so the
  second call closes nothing and breaks no cycles yet still paid for a full
  collection - 88 calls across 51 checks. Now only collects when it actually
  closed something.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

.github/workflows/test-stack-reusable-workflow.yml

@@ -205,7 +205,7 @@ jobs:
           # keeps the run short. Each file is wrapped in its own log group and the failing one is
           # named explicitly, because everything after it is SKIPPED rather than run, and that is
           # otherwise easy to misread as "the whole browser suite broke".
-          for t in tests/fetchers/test_content.py tests/test_errorhandling.py tests/visualselector/test_fetch_data.py tests/fetchers/test_custom_js_before_content.py; do
+          for t in tests/fetchers/test_content.py tests/test_errorhandling.py tests/visualselector/test_fetch_data.py tests/fetchers/test_custom_js_before_content.py tests/fetchers/test_renavigation.py; do
             echo "::group::pytest $t"
             if ! docker run --rm -e "FLASK_SERVER_NAME=cdio" -e "PLAYWRIGHT_DRIVER_URL=ws://sockpuppetbrowser:3000" --network changedet-network --hostname=cdio test-changedetectionio \
                  bash -c "cd changedetectionio;pytest -vv --capture=tee-sys --showlocals --tb=long --live-server-host=0.0.0.0 --live-server-port=5004 $t"; then
@@ -253,7 +253,7 @@ jobs:
       - name: Pyppeteer - Specific tests in built container
         run: |
           # Fail fast, but name the file that failed - see the note in the playwright job above
-          for t in tests/fetchers/test_content.py tests/test_errorhandling.py tests/visualselector/test_fetch_data.py tests/fetchers/test_custom_js_before_content.py; do
+          for t in tests/fetchers/test_content.py tests/test_errorhandling.py tests/visualselector/test_fetch_data.py tests/fetchers/test_custom_js_before_content.py tests/fetchers/test_renavigation.py; do
             echo "::group::pytest $t"
             if ! docker run --rm -e "FLASK_SERVER_NAME=cdio" -e "FAST_PUPPETEER_CHROME_FETCHER=True" -e "PLAYWRIGHT_DRIVER_URL=ws://sockpuppetbrowser:3000" --network changedet-network --hostname=cdio test-changedetectionio \
                  bash -c "cd changedetectionio;pytest --live-server-host=0.0.0.0 --live-server-port=5004 $t"; then

Dockerfile

@@ -101,6 +101,31 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     libxrender-dev \
     && apt-get clean && rm -rf /var/lib/apt/lists/*
 
+# Actually generate the locales. Installing the `locales` package above only
+# ships /etc/locale.gen - it does not build any locale, so the image had just
+# C, C.utf8 and POSIX. That made the `ENV LC_ALL=en_US.UTF-8` below unsatisfiable:
+# locale.setlocale() in flask_app.py failed, fell back to C, and the
+# format_number_locale / format_int_locale Jinja filters silently lost their
+# thousands separators - 1234567.89 rendered as "1234567.89" rather than
+# "1,234,567.89" in the restock/price overview, which is the very thing the
+# `locales` package was added for.
+#
+# More than en_US is generated so that operators can override LC_ALL / LANG and
+# get formatting for their own region (de_DE gives 1.234.567,89, fr_FR gives
+# 1 234 567,89). Costs ~21MB and ~16s of build time.
+#
+# This list mirrors the UI translations in changedetectionio/translations - one
+# glibc locale per language we ship a translation for, so any language a user
+# can pick in the UI also has a working locale. Keep the two in sync when adding
+# a translation. The territory for each bare language code comes from CLDR's
+# likely-subtags (cs -> cs_CZ, ja -> ja_JP, ko -> ko_KR, uk -> uk_UA, zh ->
+# zh_CN, zh_Hant_TW -> zh_TW), NOT from uppercasing the language code.
+RUN for l in cs_CZ de_DE en_GB en_US es_ES fr_FR id_ID it_IT ja_JP ko_KR \
+             pl_PL pt_BR ru_RU tr_TR uk_UA zh_CN zh_TW; do \
+      sed -i "s/^# *${l}.UTF-8 UTF-8/${l}.UTF-8 UTF-8/" /etc/locale.gen; \
+    done \
+    && locale-gen
+
 
 # https://stackoverflow.com/questions/58701233/docker-logs-erroneously-appears-empty-until-container-stops
 ENV PYTHONUNBUFFERED=1

changedetectionio/content_fetchers/playwright.py

@@ -296,6 +296,20 @@ class fetcher(Fetcher):
 
             self.page = await context.new_page()
 
+            # Track the LATEST main-frame document response for the whole fetch, not just the one
+            # goto() returns. This app compares the text of the page the browser ends up on, and a
+            # site that gates with an interstitial (503/429 + meta-refresh) navigates to the real
+            # page *during* the extra_wait below - judging the fetch on the first response fails a
+            # watch whose content is present and fine. Same for plain client-side redirects.
+            # Shared with action_goto_url() so only one 'response' listener exists on the page.
+            from changedetectionio.browser_steps.browser_steps import track_latest_navigation_response
+            # Must be an identity check - the tracker hands back the same (initially empty, so
+            # falsy) dict the listener writes into, and `or {}` would quietly swap in a different
+            # one that never gets updated.
+            latest_navigation_response = track_latest_navigation_response(self.page)
+            if latest_navigation_response is None:
+                latest_navigation_response = {}
+
             # Listen for all console events and handle errors
             self.page.on("console", lambda msg: logger.debug(f"Playwright console: Watch URL: {url} {msg.type}: {msg.text} {msg.args}"))
 
@@ -336,6 +350,25 @@ class fetcher(Fetcher):
             extra_wait = int(os.getenv("WEBDRIVER_DELAY_BEFORE_CONTENT_READY", 5)) + self.render_extract_delay
             await self.page.wait_for_timeout(extra_wait * 1000)
 
+            # A meta-refresh or client-side redirect usually lands during that wait, so judge the
+            # fetch on the document we are actually about to extract rather than the first one.
+            latest = latest_navigation_response.get('response')
+            if latest is not None and latest is not response:
+                logger.debug(f"Page navigated again while waiting, judging the fetch on {latest.url} "
+                             f"(status {latest.status}) instead of the first response for {url}")
+                response = latest
+                try:
+                    self.headers = await response.all_headers()
+                except Exception as e:
+                    logger.debug(f"Could not refresh headers from the final document: {e}")
+
+            # Don't extract while a navigation is mid-flight, that is what produces
+            # "Execution context was destroyed, most likely because of a navigation"
+            try:
+                await self.page.wait_for_load_state('load', timeout=extra_wait * 1000)
+            except Exception as e:
+                logger.debug(f"Page did not reach a settled load state, continuing anyway: {e}")
+
             try:
                 self.status_code = response.status
             except Exception as e:

changedetectionio/content_fetchers/puppeteer.py

@@ -1,5 +1,4 @@
 import asyncio
-import gc
 import json
 import os
 import websockets.exceptions
@@ -10,6 +9,7 @@ from loguru import logger
 from changedetectionio.content_fetchers import SCREENSHOT_MAX_HEIGHT_DEFAULT, visualselector_xpath_selectors, \
     SCREENSHOT_SIZE_STITCH_THRESHOLD, SCREENSHOT_DEFAULT_QUALITY, XPATH_ELEMENT_JS, INSTOCK_DATA_JS, \
     SCREENSHOT_MAX_TOTAL_HEIGHT, FAVICON_FETCHER_JS
+from changedetectionio import gc_debounce
 from changedetectionio.content_fetchers.base import Fetcher, get_playwright_bypass_csp, manage_user_agent
 from changedetectionio.content_fetchers.exceptions import PageUnloadable, Non200ErrorCodeReceived, EmptyReply, BrowserFetchTimedOut, \
     BrowserConnectError
@@ -236,6 +236,7 @@ class fetcher(Fetcher):
 
     async def quit(self, watch=None):
         watch_uuid = watch.get('uuid') if watch else 'unknown'
+        closed_something = bool(getattr(self, 'page', None) or getattr(self, 'browser', None))
 
         # Close page
         try:
@@ -263,8 +264,16 @@ class fetcher(Fetcher):
 
         logger.info(f"[{watch_uuid}] Cleanup puppeteer complete")
 
-        # Force garbage collection to release resources
-        gc.collect()
+        # Only collect if this call actually closed something.
+        #
+        # quit() runs twice per check - from run()'s finally, then again from the worker's
+        # safety net - and it sets self.page/self.browser to None in its own finally
+        # blocks. The second call therefore closes nothing, creates no garbage and breaks
+        # no cycles, but still paid for a full stop-the-world collection: measured at 88
+        # calls across 51 checks, roughly half of them reclaiming nothing. The pyppeteer
+        # page/connection/session graph is genuinely cyclic, so the first call still runs.
+        if closed_something:
+            gc_debounce.collect('puppeteer.quit')
 
     async def fetch_page(self,
                          current_include_filters,
@@ -399,43 +408,69 @@ class fetcher(Fetcher):
         # Enable Network domain to detect when first bytes arrive
         await self.page._client.send('Network.enable')
 
-        # Now set up the frame navigation handlers
-        async def handle_frame_navigation(event=None):
-            # Wait n seconds after the frameStartedLoading, not from any frameStartedLoading/frameStartedNavigating
-            logger.debug(f"Frame navigated: {event}")
-            w = extra_wait - 2 if extra_wait > 4 else 2
-            logger.debug(f"Waiting {w} seconds before calling Page.stopLoading...")
-            await asyncio.sleep(w)
+        # Navigate (bounded), then wait the configured "wait n seconds before extracting text"
+        # delay, then stop whatever is still loading, then extract. The delay is measured from
+        # when navigation finished, not from when it started, because the point of it is to let
+        # JS-rendered content appear *after* load - anchoring it to the start would quietly give a
+        # slow-loading page almost no settle time.
+        #
+        # Only that delay is a user-facing setting. The navigation bound above is a safety net with
+        # a sane default, not a tuning knob, so there is still one number for users to think about.
+        #
+        # There is no way to know a page is "finished" - plenty of sites navigate as part of their
+        # normal design, and some sit forever on a subresource that never answers. So the delay
+        # restarts if the MAIN frame replaces its document (a redirect or interstitial gets the
+        # same settle time the first document got), iframes do not restart it, and it is capped so
+        # a page that re-navigates in a loop cannot extend it indefinitely.
+        max_content_ready_resets = int(os.getenv("BROWSER_CONTENT_READY_MAX_RESETS", 2))
 
-            # Check if page still exists (might have been closed due to error during sleep)
-            if not self.page or not hasattr(self.page, '_client'):
-                logger.debug("Page already closed, skipping stopLoading")
-                return
+        async def wait_for_content_ready_then_stop_loading():
+            main_frame_id = self.page.mainFrame._id
+            renavigated = asyncio.Event()
 
-            logger.debug("Issuing stopLoading command...")
-            await self.page._client.send('Page.stopLoading')
-            logger.debug("stopLoading command sent!")
+            def _on_main_frame_navigation(event):
+                if event.get('frameId') == main_frame_id:
+                    renavigated.set()
 
-        async def setup_frame_handlers_on_first_response(event):
-            # Only trigger for the main document response
-            if event.get('type') == 'Document':
-                logger.debug("First response received, setting up frame handlers for forced page stop load.")
-                self.page._client.on('Page.frameStartedNavigating', lambda e: asyncio.create_task(handle_frame_navigation(e)))
-                self.page._client.on('Page.frameStartedLoading', lambda e: asyncio.create_task(handle_frame_navigation(e)))
-                self.page._client.on('Page.frameStoppedLoading', lambda e: logger.debug(f"Frame stopped loading: {e}"))
-                logger.debug("First response received, setting up frame handlers for forced page stop load DONE SETUP")
-                # De-register this listener - we only need it once
-                self.page._client.remove_listener('Network.responseReceived', setup_frame_handlers_on_first_response)
+            self.page._client.on('Page.frameStartedLoading', _on_main_frame_navigation)
+            self.page._client.on('Page.frameStoppedLoading', lambda e: logger.debug(f"Frame stopped loading: {e}"))
+            try:
+                resets = 0
+                while True:
+                    renavigated.clear()
+                    try:
+                        await asyncio.wait_for(renavigated.wait(), timeout=extra_wait)
+                        # Main frame started a new document
+                        resets += 1
+                        if resets > max_content_ready_resets:
+                            logger.debug(f"Main frame keeps re-navigating, not restarting the content-ready wait again")
+                            break
+                        logger.debug(f"Main frame started a new document, restarting the {extra_wait}s "
+                                     f"content-ready wait ({resets}/{max_content_ready_resets})")
+                    except asyncio.TimeoutError:
+                        # Quiet for the whole delay - the page is as ready as it is going to get
+                        break
+            finally:
+                self.page._client.remove_listener('Page.frameStartedLoading', _on_main_frame_navigation)
 
-        # Listen for first response to trigger frame handler setup
-        self.page._client.on('Network.responseReceived', setup_frame_handlers_on_first_response)
+            # Stop whatever is still in flight so the DOM and screenshot come from what rendered,
+            # rather than waiting on a subresource that may never answer
+            try:
+                logger.debug(f"Content-ready wait of {extra_wait}s elapsed, issuing Page.stopLoading before extracting")
+                await self.page._client.send('Page.stopLoading')
+                logger.debug("stopLoading command sent!")
+            except Exception as e:
+                logger.debug(f"Page.stopLoading skipped, page is most likely already gone: {e}")
 
-        # Chrome 153+ refuses to commit a navigation when an error status arrives with a
-        # zero-length body, so goto() raises net::ERR_HTTP_RESPONSE_CODE_FAILURE instead of handing
-        # back the response. The response was received fine, we just never get it as a return value,
-        # so keep the main-frame response from the 'response' event and use that instead - the
-        # status check below then reports a real "Error - 404" instead of a raw net:: string.
-        # Kept as the latest matching response so a redirect chain reports its final hop.
+        # Track the LATEST main-frame document response for the whole fetch, not just the one that
+        # goto() happens to return. This app compares the text of the page the browser ends up on,
+        # and plenty of sites navigate again after the first response:
+        #  - an interstitial answering 503/429 with a meta-refresh into the real 200 page, where
+        #    judging the first response fails a watch whose content is sitting right there
+        #  - a plain client-side redirect to another host (slated.com -> get.slated.com)
+        # It also covers Chrome 153+, which refuses to commit a navigation when an error status
+        # arrives with a zero-length body: goto() raises net::ERR_HTTP_RESPONSE_CODE_FAILURE rather
+        # than returning the response, but the response itself still arrives on this event.
         navigation_response = {}
 
         def _keep_navigation_response(response):
@@ -445,25 +480,62 @@ class fetcher(Fetcher):
 
         self.page.on('response', _keep_navigation_response)
 
+        # pyppeteer's navigation watcher is bound to the loaderId of the navigation it started. If
+        # the page replaces that document (redirect/interstitial) the 'load' it waits for never
+        # arrives for that loaderId, so goto() never returns - and with timeout=0 it would block
+        # until the hard PUPPETEER_MAX_PROCESSING_TIMEOUT_SECONDS kill, burning a worker slot for
+        # minutes on a page that is fully loaded. Bound it, then fall back to the document we can
+        # see. Verified against slated.com and getastra.com, which hang indefinitely otherwise.
+        nav_timeout = int(os.getenv("BROWSER_NAVIGATION_TIMEOUT_SECONDS", 30))
+
         response = None
         attempt=0
         try:
             while not response:
                 logger.debug(f"Attempting page fetch {url} attempt {attempt}")
-                asyncio.create_task(handle_frame_navigation())
-                try:
-                    response = await self.page.goto(url, timeout=0)
-                except Exception as e:
-                    if 'ERR_HTTP_RESPONSE_CODE_FAILURE' not in str(e) or not navigation_response:
-                        raise
-                    response = navigation_response['response']
-                    logger.debug(f"Navigation was aborted by the browser (empty body on an error status), "
-                                 f"recovered status {response.status} from the response event")
-                await asyncio.sleep(1 + extra_wait)
-                # Check if page still exists before sending command
-                if self.page and hasattr(self.page, '_client'):
-                    await self.page._client.send('Page.stopLoading')
+                # Race goto() against the main frame actually firing 'load'. In the re-navigation
+                # case goto() can never resolve, but the replacement document does fire 'load' -
+                # usually within a few seconds - so this returns then instead of sitting out the
+                # whole nav_timeout. Whichever arrives first means "the document is loaded".
+                main_frame_loaded = asyncio.Event()
+                main_frame_id = self.page.mainFrame._id
 
+                def _on_lifecycle(event):
+                    if event.get('name') == 'load' and event.get('frameId') == main_frame_id:
+                        main_frame_loaded.set()
+
+                self.page._client.on('Page.lifecycleEvent', _on_lifecycle)
+                goto_task = asyncio.ensure_future(self.page.goto(url, timeout=0))
+                load_task = asyncio.ensure_future(main_frame_loaded.wait())
+                try:
+                    done, _pending = await asyncio.wait({goto_task, load_task},
+                                                        timeout=nav_timeout,
+                                                        return_when=asyncio.FIRST_COMPLETED)
+
+                    if goto_task in done:
+                        try:
+                            response = goto_task.result()
+                        except Exception as e:
+                            if 'ERR_HTTP_RESPONSE_CODE_FAILURE' not in str(e) or not navigation_response:
+                                raise
+                            response = navigation_response['response']
+                            logger.debug(f"Navigation was aborted by the browser (empty body on an error status), "
+                                         f"recovered status {response.status} from the response event")
+                    else:
+                        # Either the replacement document loaded, or we ran out of patience
+                        response = navigation_response.get('response')
+                        if not response:
+                            raise BrowserFetchTimedOut(msg=f"Browser did not finish navigating to {url} within "
+                                                           f"{nav_timeout}s and no main-frame response was seen.")
+                        why = ("the page replaced the document it started on" if load_task in done
+                               else f"navigation did not settle within {nav_timeout}s")
+                        logger.warning(f"Continuing with the document actually loaded ({why}) - "
+                                       f"status {response.status} for {response.url}")
+                finally:
+                    self.page._client.remove_listener('Page.lifecycleEvent', _on_lifecycle)
+                    for t in (goto_task, load_task):
+                        if not t.done():
+                            t.cancel()
                 if response:
                     break
                 if not response:
@@ -472,6 +544,19 @@ class fetcher(Fetcher):
                     logger.warning(f"Content Fetcher > Response object was none (as in, the response from the browser was empty, not just the content) exiting attempt {attempt}")
                     raise EmptyReply(url=url, status_code=None)
                 attempt+=1
+
+            # Navigation is done; now honour "wait n seconds before extracting text" and then
+            # force-stop whatever is still loading, so extraction always gets what rendered.
+            # Awaited inline rather than fired off as a task, so nothing can outlive the fetch.
+            await wait_for_content_ready_then_stop_loading()
+
+            # That wait is where a meta-refresh interstitial typically swaps in the real page, so
+            # re-check which document we are actually on before judging the status code.
+            latest = navigation_response.get('response')
+            if latest is not None and latest is not response:
+                logger.debug(f"Page navigated again while waiting, judging the fetch on {latest.url} "
+                             f"(status {latest.status}) instead of {response.url} (status {response.status})")
+                response = latest
         finally:
             self.page.remove_listener('response', _keep_navigation_response)
 
@@ -534,8 +619,7 @@ class fetcher(Fetcher):
         self.screenshot = await capture_full_page(page=self.page, screenshot_format=self.screenshot_format, watch_uuid=watch_uuid, lock_viewport_elements=self.lock_viewport_elements)
 
         # Force garbage collection - pyppeteer base64 decode creates temporary buffers
-        import gc
-        gc.collect()
+        gc_debounce.collect('puppeteer.after_screenshot')
         self.xpath_data = await self.page.evaluate(XPATH_ELEMENT_JS, {
             "visualselector_xpath_selectors": visualselector_xpath_selectors,
             "max_height": MAX_TOTAL_HEIGHT

changedetectionio/gc_debounce.py

@@ -0,0 +1,73 @@
+"""
+Debounced explicit garbage collection for the watch-check hot path.
+
+Several places call gc.collect() after a check to keep C-level memory (pyppeteer buffers,
+libxml2 documents, PIL, brotli) from accumulating. Individually each is reasonable. Run
+concurrently by many fetch workers they become a storm: with FETCH_WORKERS=50 and roughly
+five call sites per check, the process spends most of its time stopped in the collector,
+because every gc.collect() is a full stop-the-world pass that walks the entire heap while
+holding the GIL.
+
+Measured on 153 real puppeteer checks of a live site, FETCH_WORKERS=50, an `//div` filter
+so the lxml document tree is realistic:
+
+                       collects  objects freed   gc time  checks/sec  CPU/check  RSS plateau
+  one per call site         790      2,594,098     91.3s       0.766     1.373s      279.6MB
+  debounced to 1s            48      2,219,010      7.3s       1.433     0.731s      275.3MB
+  none at all                 0              0      0.0s       1.503     0.685s      287.7MB
+
+Debouncing keeps 86% of the reclamation for 6% of the collections, and resident memory
+ends up LOWER than collecting every time. It works because the collector is process-wide:
+any worker's collection breaks every other worker's cycles too, so with many workers the
+calls are overwhelmingly redundant duplicates rather than independently necessary.
+
+Removing them entirely was also measured. It is slightly faster still, but it was the only
+configuration whose RSS had not plateaued by the end of the run, so it is not offered.
+
+Collecting a younger generation was measured and rejected: gen 0 freed 1,136 objects
+against the full pass's 2,594,098, because objects that survive a 10-30 second fetch have
+already been promoted out of gen 0. It is cheap because it does almost nothing.
+
+Environment:
+  EXPLICIT_GC_MIN_INTERVAL   seconds between explicit collections, process-wide.
+                             Default 1.0. Set 0 to collect at every call site as before.
+  EXPLICIT_GC_COLLECT        set false to disable explicit collection entirely. A
+                             measurement switch for attributing a slowdown, not a
+                             recommended setting - expect resident memory to drift.
+"""
+
+import gc
+import os
+import threading
+import time
+
+from changedetectionio.strtobool import strtobool
+
+ENABLED = strtobool(os.getenv('EXPLICIT_GC_COLLECT', 'true'))
+MIN_INTERVAL = float(os.getenv('EXPLICIT_GC_MIN_INTERVAL', '1.0') or 0)
+
+_last_collect = 0.0
+_lock = threading.Lock()
+
+
+def collect(where=None):
+    """Explicit collection for the per-check hot path, rate-limited process-wide.
+
+    `where` is a short label for the call site, kept so callers read clearly and so a
+    future caller can log it. Returns the number of objects collected, or 0 when the call
+    was debounced or disabled - matching gc.collect()'s return, so this is a drop-in
+    replacement for it.
+    """
+    global _last_collect
+
+    if not ENABLED:
+        return 0
+
+    if MIN_INTERVAL > 0:
+        now = time.monotonic()
+        with _lock:
+            if now - _last_collect < MIN_INTERVAL:
+                return 0
+            _last_collect = now
+
+    return gc.collect()

changedetectionio/model/Watch.py

@@ -30,6 +30,7 @@ from changedetectionio.validate_url import is_safe_valid_url
 
 from changedetectionio.strtobool import strtobool
 from changedetectionio.jinja2_custom import render as jinja_render
+from changedetectionio import gc_debounce
 from . import watch_base
 from .persistence import EntityPersistenceMixin
 import os
@@ -107,9 +108,21 @@ def _brotli_save(contents, filepath, mode=None, fallback_uncompressed=False):
 
         logger.debug(f"Finished brotli compression - From {original_size} to {total_compressed_size} bytes.")
 
-        # Cleanup: Delete compressor, force Python GC, then force C-level memory release
+        # Cleanup: drop the compressor, then force C-level memory back to the OS.
+        #
+        # There is deliberately no gc.collect() here. brotli.Compressor is not gc-tracked,
+        # so the collector can never reclaim it - `del` frees it immediately by refcount.
+        # Measured over 60 x 2.2MB compressions, RSS growth was:
+        #
+        #   neither              +0.9MB
+        #   gc.collect() only    +0.2MB
+        #   malloc_trim() only   +0.0MB   <- does all of the work
+        #   both (previous)      +0.0MB   <- the collect contributed nothing
+        #
+        # malloc_trim below is the load-bearing line: brotli's retention is glibc holding
+        # freed arenas, which only a trim returns. The collect cost ~31ms of stop-the-world
+        # per snapshot save for no reclamation.
         del compressor
-        gc.collect()
 
         # Force release of C-level memory back to OS (since brotli is a C library)
         try:
@@ -689,7 +702,7 @@ class model(EntityPersistenceMixin, watch_base):
 
         # reimport
         bump = self.history
-        gc.collect()
+        gc_debounce.collect('watch.history_bump')
 
     # Save some text file to the appropriate path and bump the history
     # result_obj from fetch_site_status.run()

changedetectionio/static/js/search-modal.js

@@ -7,7 +7,6 @@
     // The Search button is rendered in the left rail and the mobile drawer.
     const openSearchButtons = document.querySelectorAll('.js-open-search-modal');
     const closeSearchButton = document.getElementById('close-search-modal');
-    const searchForm = document.getElementById('search-form');
     const searchInput = document.getElementById('search-modal-input');
 
     if (!searchModal || openSearchButtons.length === 0) {
@@ -63,6 +62,15 @@
 
     // Close modal when clicking the backdrop
     searchModal.addEventListener('click', function(e) {
+      // Only real pointer clicks can land on the backdrop. Keyboard-synthesised clicks
+      // report detail 0 and coordinates of 0,0, which the geometry test below reads as
+      // "outside the dialog" - and implicit form submission (Enter in the input) fires
+      // exactly such a click at the Search button. That closed the modal and blanked
+      // the input mid-dispatch, so the submit that followed hit an empty `required`
+      // field and was rejected: Enter appeared to just dismiss the form.
+      if (e.detail === 0) {
+        return;
+      }
       const rect = searchModal.getBoundingClientRect();
       const isInDialog = (
         rect.top <= e.clientY &&
@@ -93,43 +101,8 @@
       }
     });
 
-    // Handle Enter key in search input
-    if (searchInput) {
-      searchInput.addEventListener('keydown', function(e) {
-        if (e.key === 'Enter') {
-          e.preventDefault();
-          if (searchForm) {
-            // Trigger form submission programmatically
-            searchForm.dispatchEvent(new Event('submit'));
-          }
-        }
-      });
-    }
-
-    // Handle form submission
-    if (searchForm) {
-      searchForm.addEventListener('submit', function(e) {
-        e.preventDefault();
-
-        // Get form data
-        const formData = new FormData(searchForm);
-        const searchQuery = formData.get('q');
-        const tags = formData.get('tags');
-
-        // Build URL
-        const params = new URLSearchParams();
-        if (searchQuery) {
-          params.append('q', searchQuery);
-        }
-        if (tags) {
-          params.append('tags', tags);
-        }
-
-        // Navigate to search results (always redirect to watchlist home)
-        // Use base_path if available (for sub-path deployments like /enlighten-richerx)
-        const basePath = typeof base_path !== 'undefined' ? base_path : '';
-        window.location.href = basePath + '/?' + params.toString();
-      });
-    }
+    // Submission is left to the browser: the form carries a server-rendered action
+    // (correct under a reverse-proxy sub-path) and Enter in the input triggers implicit
+    // submission via the footer's submit button, which also runs `required` validation.
   });
 })();

changedetectionio/static/styles/scss/parts/_hamburger_menu.scss

@@ -95,7 +95,9 @@
   overflow-y: auto;
   padding-top: 60px;
 
-
+  .action-label {
+    display: block !important;
+  }
   #cdio-logo {
     color: var(--color-text);
 

changedetectionio/templates/base.html

@@ -317,11 +317,18 @@
         <h2 class="modal-title" id="search-modal-title">{{ _('Search') }}</h2>
       </div>
       <div class="modal-body">
-        <form id="search-form" method="GET">
+        {# Plain GET submit to the watchlist - url_for() carries the reverse-proxy sub-path
+           (SCRIPT_NAME), so no client-side URL building is needed. #}
+        <form id="search-form" method="GET" action="{{ url_for('watchlist.index') }}">
           <div class="pure-control-group">
-            <label for="search-modal-input">{% if active_tag_uuid %}{{ _("URL or Title in '%(title)s'", title=active_tag.title) }}{% else %}{{ _('URL or Title') }}{% endif %}</label>
+            {# Matches watch_passes_search() in blueprint/watchlist/filters.py - title, URL and last error text #}
+            <label for="search-modal-input">{{ _('URL, title or error text') }}</label>
             <input id="search-modal-input" class="m-d" name="q" placeholder="{{ _('Enter search term...') }}" required type="text" value="" autofocus>
-            <input name="tags" type="hidden" value="{% if active_tag_uuid %}{{active_tag_uuid}}{% endif %}">
+            {# 'tag' (not 'tags') - that's the arg the watchlist filters on #}
+            {% if active_tag_uuid %}
+            <input name="tag" type="hidden" value="{{ active_tag_uuid }}">
+            <span class="pure-form-message-inline">{{ _("Searching in current group '%(title)s' only", title=active_tag.title) }}</span>
+            {% endif %}
           </div>
         </form>
       </div>

changedetectionio/tests/fetchers/test_renavigation.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python3
+
+"""The browser fetchers must judge a fetch on the document they end up extracting.
+
+Plenty of sites answer the first request with an interstitial carrying an error status and a
+client-side redirect, then serve the real page. Judging the fetch on the first navigation fails a
+watch whose content is present and fine (reported against fotokoch.de: 503 + meta refresh -> 200),
+and on the pyppeteer fetcher a replaced document used to hang goto() until the hard processing
+timeout because its navigation watcher is bound to the loaderId it started on.
+"""
+
+import os
+from flask import url_for
+from ..util import wait_for_all_checks
+
+
+def _cdio(url):
+    # The browser runs in another container in CI and reaches the test server as 'cdio'
+    return url.replace('localhost.localdomain', 'cdio').replace('localhost', 'cdio')
+
+
+def test_interstitial_redirect_is_followed(client, live_server, measure_memory_usage, datastore_path):
+    assert os.getenv('PLAYWRIGHT_DRIVER_URL'), "Needs PLAYWRIGHT_DRIVER_URL set for this test"
+
+    res = client.post(
+        url_for("settings.settings_page"),
+        data={
+            "application-empty_pages_are_a_change": "",
+            "requests-time_between_check-minutes": 180,
+            'application-fetch_backend': "html_webdriver",
+        },
+        follow_redirects=True
+    )
+    assert b"Settings updated." in res.data
+
+    test_url = _cdio(url_for('test_interstitial', key='renav', _external=True))
+
+    res = client.post(
+        url_for("imports.import_page"),
+        data={"urls": test_url},
+        follow_redirects=True
+    )
+    assert b"1 Imported" in res.data
+    wait_for_all_checks(client)
+
+    # The interstitial answered 503, so judging the first navigation would have failed the watch
+    uuid = next(iter(live_server.app.config['DATASTORE'].data['watching']))
+    watch = live_server.app.config['DATASTORE'].data['watching'][uuid]
+    assert not watch.get('last_error'), \
+        f"Watch was judged on the interstitial instead of the page it landed on: {watch.get('last_error')}"
+
+    res = client.get(url_for("watchlist.index"))
+    assert b'Error - 503' not in res.data
+
+    assert watch.history_n >= 1, "Fetch succeeded but no snapshot was stored"
+    snapshot = watch.get_history_snapshot(list(watch.history.keys())[-1])
+    assert 'The real page content is here' in snapshot
+    assert 'Browser check in progress' not in snapshot
+
+    client.post(url_for("ui.form_delete", uuid="all"), follow_redirects=True)

changedetectionio/tests/test_search.py

@@ -1,5 +1,6 @@
 from flask import url_for
 from .util import set_original_response, set_modified_response, live_server_setup
+import re
 import time
 
 
@@ -71,3 +72,43 @@ def test_search_in_tag_limit(client, live_server, measure_memory_usage, datastor
     assert urls[0].split(' ')[0].encode('utf-8') in res.data, urls[0].encode('utf-8')
     assert urls[1].split(' ')[0].encode('utf-8') not in res.data, urls[0].encode('utf-8')
 
+
+
+def test_search_modal_form_action(client, live_server, measure_memory_usage, datastore_path):
+    # The search modal submits as a plain GET form, so its action has to carry the
+    # reverse-proxy sub-path (SCRIPT_NAME), otherwise search jumps to the host root.
+    res = client.get(url_for("watchlist.index"))
+    assert b'<form id="search-form" method="GET" action="/">' in res.data
+
+    res = client.get("/", base_url="http://localhost/sub-path")
+    assert b'<form id="search-form" method="GET" action="/sub-path/">' in res.data
+
+
+def test_search_modal_tag_field_is_filterable(client, live_server, measure_memory_usage, datastore_path):
+    # The modal carries the active tag as a hidden field so a search stays scoped to the
+    # tag you were viewing. The field name has to be the one the watchlist filters on.
+    urls = ['https://localhost:12300?first-result=1 tag-one',
+            'https://localhost:5000?second-result=1 tag-two'
+            ]
+    res = client.post(
+        url_for("imports.import_page"),
+        data={"urls": "\r\n".join(urls)},
+        follow_redirects=True
+    )
+    assert b"2 Imported" in res.data
+
+    res = client.get(url_for("watchlist.index") + "?tag=tag-one")
+    form = re.search(rb'<form id="search-form".*?</form>', res.data, re.DOTALL)
+    assert form, "search modal form not rendered"
+    field = re.search(rb'<input name="([^"]+)" type="hidden" value="([^"]+)"', form.group(0))
+    assert field, f"no populated hidden tag field in {form.group(0)}"
+    name, value = field.group(1).decode(), field.group(2).decode()
+
+    # The scoping is spelled out in the modal, so narrowed results aren't a surprise
+    assert b'Searching in current group' in form.group(0)
+    assert b'tag-one' in form.group(0)
+
+    # 'localhost' matches both watches, so only the tag field can narrow it down
+    res = client.get(url_for("watchlist.index") + f"?q=localhost&{name}={value}")
+    assert urls[0].split(' ')[0].encode('utf-8') in res.data
+    assert urls[1].split(' ')[0].encode('utf-8') not in res.data, f"'{name}' is not filtered on by the watchlist"

changedetectionio/tests/util.py

@@ -249,6 +249,31 @@ def new_live_server_setup(live_server):
         import secrets
         return "Random content - {}\n".format(secrets.token_hex(64))
 
+    # Re-navigation gate: the first hit answers with an error status AND a client-side meta
+    # refresh, the second serves the real page. This mirrors sites that gate visitors they have
+    # not seen recently (reported against fotokoch.de, which answers 503 + meta refresh and then
+    # serves a 200). The browser follows the refresh, so the fetch has to be judged on the
+    # document we actually end up extracting rather than on the interstitial.
+    # Keyed on last-seen time rather than a hit count, so EVERY fresh check starts out gated -
+    # a counter would serve a clean 200 to the second check and let the test pass without the fix.
+    _interstitial_last_seen = {}
+
+    @live_server.app.route('/test-interstitial')
+    def test_interstitial():
+        key = request.args.get('key', 'default')
+        now = time.time()
+        seen_recently = (now - _interstitial_last_seen.get(key, 0)) < 10
+        _interstitial_last_seen[key] = now
+        if not seen_recently:
+            resp = make_response(
+                '<html><head><meta http-equiv="refresh" content="1"></head>'
+                '<body>Browser check in progress, you will be redirected</body></html>', 503)
+        else:
+            resp = make_response(
+                '<html><body><h1>The real page content is here</h1></body></html>', 200)
+        resp.headers['Content-Type'] = 'text/html'
+        return resp
+
     @live_server.app.route('/test-endpoint2')
     def test_endpoint2():
         return "<html><body>some basic content</body></html>"

changedetectionio/translations/cs/LC_MESSAGES/messages.po

@@ -4571,18 +4571,18 @@ msgid "Search"
 msgstr "Hledat"
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr ""
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr ""
+msgid "URL, title or error text"
+msgstr "URL, název nebo text chyby"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr ""
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "Hledá se pouze ve skupině '%(title)s'"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/de/LC_MESSAGES/messages.po

@@ -4625,18 +4625,18 @@ msgid "Search"
 msgstr "Suchen"
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr "URL oder Titel in '%(title)s'"
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr "URL oder Titel"
+msgid "URL, title or error text"
+msgstr "URL, Titel oder Fehlertext"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr "Suchbegriff eingeben..."
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "Suche nur in Gruppe '%(title)s'"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/en_GB/LC_MESSAGES/messages.po

@@ -4563,18 +4563,18 @@ msgid "Search"
 msgstr ""
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr ""
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr ""
+msgid "URL, title or error text"
+msgstr "URL, title or error text"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr ""
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "Searching in current group '%(title)s' only"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/en_US/LC_MESSAGES/messages.po

@@ -4563,18 +4563,18 @@ msgid "Search"
 msgstr ""
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr ""
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr ""
+msgid "URL, title or error text"
+msgstr "URL, title or error text"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr ""
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "Searching in current group '%(title)s' only"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/es/LC_MESSAGES/messages.po

@@ -4640,18 +4640,18 @@ msgid "Search"
 msgstr "Buscar"
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr "URL o título en '%(title)s'"
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr "URL o título"
+msgid "URL, title or error text"
+msgstr "URL, título o texto de error"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr "Introduzca el término de búsqueda..."
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "Buscando solo en el grupo '%(title)s'"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/fr/LC_MESSAGES/messages.po

@@ -4578,18 +4578,18 @@ msgid "Search"
 msgstr "Rechercher"
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr ""
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr ""
+msgid "URL, title or error text"
+msgstr "URL, titre ou texte d'erreur"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr ""
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "Recherche uniquement dans le groupe '%(title)s'"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/id/LC_MESSAGES/messages.po

@@ -4684,18 +4684,18 @@ msgid "Search"
 msgstr "Cari"
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr "URL atau Judul di '%(title)s'"
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr "URL atau Judul"
+msgid "URL, title or error text"
+msgstr "URL, judul atau teks kesalahan"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr "Masukkan istilah pencarian..."
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "Mencari hanya dalam grup '%(title)s'"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/it/LC_MESSAGES/messages.po

@@ -4565,18 +4565,18 @@ msgid "Search"
 msgstr "Cerca"
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr ""
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr ""
+msgid "URL, title or error text"
+msgstr "URL, titolo o testo dell'errore"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr ""
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "Ricerca solo nel gruppo '%(title)s'"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/ja/LC_MESSAGES/messages.po

@@ -4596,18 +4596,18 @@ msgid "Search"
 msgstr "検索"
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr "'%(title)s' 内のURLまたはタイトル"
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr "URLまたはタイトル"
+msgid "URL, title or error text"
+msgstr "URL、タイトル、エラーテキスト"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr "検索語を入力..."
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "グループ「%(title)s」内のみを検索"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/ko/LC_MESSAGES/messages.po

@@ -4573,18 +4573,18 @@ msgid "Search"
 msgstr "검색"
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr "'%(title)s'의 URL 또는 제목"
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr "URL 또는 제목"
+msgid "URL, title or error text"
+msgstr "URL, 제목 또는 오류 텍스트"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr "검색어를 입력하세요..."
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "'%(title)s' 그룹 내에서만 검색"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/messages.pot

@@ -8,7 +8,7 @@ msgid ""
 msgstr ""
 "Project-Id-Version: changedetection.io 0.60.4\n"
 "Report-Msgid-Bugs-To: EMAIL@ADDRESS\n"
-"POT-Creation-Date: 2026-09-12 15:41+0200\n"
+"POT-Creation-Date: 2026-09-12 17:18+0200\n"
 "PO-Revision-Date: YEAR-MO-DA HO:MI+ZONE\n"
 "Last-Translator: FULL NAME <EMAIL@ADDRESS>\n"
 "Language-Team: LANGUAGE <LL@li.org>\n"
@@ -4562,18 +4562,18 @@ msgid "Search"
 msgstr ""
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr ""
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
+msgid "URL, title or error text"
 msgstr ""
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr ""
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr ""
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/pl/LC_MESSAGES/messages.po

@@ -4745,18 +4745,18 @@ msgid "Search"
 msgstr "Wyszukiwanie"
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr "Adres URL lub tytuł w „%(title)s”"
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr "Adres URL lub tytuł"
+msgid "URL, title or error text"
+msgstr "URL, tytuł lub treść błędu"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr "Wpisz szukane hasło..."
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "Szukanie tylko w grupie '%(title)s'"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/pt_BR/LC_MESSAGES/messages.po

@@ -4615,18 +4615,18 @@ msgid "Search"
 msgstr "Pesquisar"
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr "URL ou Título em '%(title)s'"
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr "URL ou Título"
+msgid "URL, title or error text"
+msgstr "URL, título ou texto de erro"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr "Digite o termo de busca..."
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "Buscando apenas no grupo '%(title)s'"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/ru/LC_MESSAGES/messages.po

@@ -4693,18 +4693,18 @@ msgid "Search"
 msgstr "Поиск"
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr "URL-адрес или заголовок в формате «%(title)s»"
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr "URL или заголовок"
+msgid "URL, title or error text"
+msgstr "URL, заголовок или текст ошибки"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr "Введите поисковый запрос..."
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "Поиск только в группе '%(title)s'"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/tr/LC_MESSAGES/messages.po

@@ -4620,18 +4620,18 @@ msgid "Search"
 msgstr "Ara"
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr "'%(title)s' içinde URL veya Başlık"
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr "URL veya Başlık"
+msgid "URL, title or error text"
+msgstr "URL, başlık veya hata metni"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr "Arama terimini girin..."
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "Yalnızca '%(title)s' grubunda aranıyor"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/uk/LC_MESSAGES/messages.po

@@ -4597,18 +4597,18 @@ msgid "Search"
 msgstr "Пошук"
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr "URL або Назва у '%(title)s'"
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr "URL або Назва"
+msgid "URL, title or error text"
+msgstr "URL, заголовок або текст помилки"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr "Введіть пошуковий запит..."
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "Пошук лише в групі '%(title)s'"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/zh/LC_MESSAGES/messages.po

@@ -4571,18 +4571,18 @@ msgid "Search"
 msgstr "搜索"
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr "'%(title)s' 中的 URL 或标题"
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr "URL 或标题"
+msgid "URL, title or error text"
+msgstr "URL、标题或错误文本"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr "输入搜索关键词..."
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "仅在分组「%(title)s」中搜索"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/translations/zh_Hant_TW/LC_MESSAGES/messages.po

@@ -4574,18 +4574,18 @@ msgid "Search"
 msgstr "搜尋"
 
 #: changedetectionio/templates/base.html
-#, python-format
-msgid "URL or Title in '%(title)s'"
-msgstr "URL 或標題 在 '%(title)s'"
-
-#: changedetectionio/templates/base.html
-msgid "URL or Title"
-msgstr "URL 或標題"
+msgid "URL, title or error text"
+msgstr "URL、標題或錯誤文字"
 
 #: changedetectionio/templates/base.html
 msgid "Enter search term..."
 msgstr "輸入搜尋關鍵字 ..."
 
+#: changedetectionio/templates/base.html
+#, python-format
+msgid "Searching in current group '%(title)s' only"
+msgstr "僅在群組「%(title)s」中搜尋"
+
 #: changedetectionio/templates/edit/include_llm_intent.html
 msgid ""
 "<strong>On</strong> &ndash; every watch in this group uses the AI settings below, unless it fills in its own. "

changedetectionio/worker.py

@@ -6,6 +6,7 @@ from changedetectionio import html_tools
 from changedetectionio import worker_pool
 from changedetectionio.queuedWatchMetaData import PrioritizedItem
 from changedetectionio.pluggy_interface import apply_update_handler_alter, apply_update_finalize
+from changedetectionio import gc_debounce
 
 import asyncio
 import os
@@ -655,9 +656,8 @@ async def async_update_worker(worker_id, q, notification_q, app, datastore, exec
                         del update_handler
                         update_handler = None
 
-                # Force garbage collection
-                import gc
-                gc.collect()
+                # Force garbage collection (debounced process-wide, see gc_debounce)
+                gc_debounce.collect('worker.after_processing')
 
         except Exception as e:
             # Store the processing exception for plugin finalization hook
@@ -702,8 +702,7 @@ async def async_update_worker(worker_id, q, notification_q, app, datastore, exec
                         del contents
 
                     # Force garbage collection after all references are cleared
-                    import gc
-                    gc.collect()
+                    gc_debounce.collect('worker.cleanup_finally')
 
                     logger.debug(f"Worker {worker_id} completed watch {uuid} in {time.time()-fetch_start_time:.2f}s")
                 except Exception as cleanup_error: