Compare commits

...

3 Commits

Author SHA1 Message Date
dgtlmoon
9d8558fbc9 LXML memory leak workaround 2022-07-27 15:35:05 +02:00
dgtlmoon
a7a8ba58ed fix status 2022-07-27 13:11:14 +02:00
dgtlmoon
7823140442 WIP 2022-07-27 13:05:23 +02:00
3 changed files with 219 additions and 161 deletions

View File

@@ -105,9 +105,10 @@ def init_app_secret(datastore_path):
# running or something similar.
@app.template_filter('format_last_checked_time')
def _jinja2_filter_datetime(watch_obj, format="%Y-%m-%d %H:%M:%S"):
# Worker thread tells us which UUID it is currently processing.
for t in running_update_threads:
if t.current_uuid == watch_obj['uuid']:
for t in threading.enumerate():
if t.name == 'update_worker' and t.current_uuid == watch_obj['uuid']:
return '<span class="loader"></span><span> Checking now</span>'
if watch_obj['last_checked'] == 0:
@@ -1213,6 +1214,7 @@ def changedetection_app(config=None, datastore_o=None):
# @todo handle ctrl break
ticker_thread = threading.Thread(target=ticker_thread_check_time_launch_checks).start()
threading.Thread(target=ticker_thread_job_queue_processor).start()
threading.Thread(target=notification_runner).start()
@@ -1288,25 +1290,63 @@ def notification_runner():
# Trim the log length
notification_debug_log = notification_debug_log[-100:]
# Check the queue, when a job exists, start a fresh thread of update_worker
def ticker_thread_job_queue_processor():
from changedetectionio import update_worker
n_workers = int(os.getenv("FETCH_WORKERS", datastore.data['settings']['requests']['workers']))
while not app.config.exit.is_set():
time.sleep(0.3)
# Check that some threads are free
running = 0
for t in threading.enumerate():
if t.name == 'update_worker':
running += 1
if running >= n_workers:
continue
try:
uuid = update_q.get(block=False)
except queue.Empty:
# Go back to waiting for exit and/or another entry from the queue
continue
print ("Starting a thread fetch")
try:
# Launch the update_worker thread that will handle picking items off a queue and sending them off
# in the event that playwright or others have a memory leak, this should clean it up better than gc.collect()
# (By letting it exit entirely)
update_worker.update_worker(update_q, notification_q, app, datastore, uuid).start()
except Exception as e:
print ("Error launching update_worker for UUID {}.".format(uuid))
print (str(e))
print ("Running now {}", running)
# Thread runner to check every minute, look for new watches to feed into the Queue.
def ticker_thread_check_time_launch_checks():
import random
from changedetectionio import update_worker
recheck_time_minimum_seconds = int(os.getenv('MINIMUM_SECONDS_RECHECK_TIME', 20))
print("System env MINIMUM_SECONDS_RECHECK_TIME", recheck_time_minimum_seconds)
# Can go in its own function
# Always maintain the minimum number of threads, each thread will terminate when it has processed exactly 1 queued watch
# This is to be totally sure that they don't leak memory
# Spin up Workers that do the fetching
# Can be overriden by ENV or use the default settings
n_workers = int(os.getenv("FETCH_WORKERS", datastore.data['settings']['requests']['workers']))
for _ in range(n_workers):
new_worker = update_worker.update_worker(update_q, notification_q, app, datastore)
running_update_threads.append(new_worker)
new_worker.start()
while not app.config.exit.is_set():
# Get a list of watches by UUID that are currently fetching data
# Update our list of watches by UUID that are currently fetching data, used in the UI
running_uuids = []
for t in running_update_threads:
if t.current_uuid:

View File

@@ -4,8 +4,6 @@ from typing import List
from bs4 import BeautifulSoup
from jsonpath_ng.ext import parse
import re
from inscriptis import get_text
from inscriptis.model.config import ParserConfig
class FilterNotFoundInResponse(ValueError):
def __init__(self, msg):
@@ -183,9 +181,16 @@ def strip_ignore_text(content, wordlist, mode="content"):
def html_to_text(html_content: str, render_anchor_tag_content=False) -> str:
import multiprocessing
from inscriptis.model.config import ParserConfig
"""Converts html string to a string with just the text. If ignoring
rendering anchor tag content is enable, anchor tag content are also
included in the text
@NOTE: HORRIBLE LXML INDUCED MEMORY LEAK WORKAROUND HERE
https://www.reddit.com/r/Python/comments/j0gl8t/psa_pythonlxml_memory_leaks_and_a_solution/
:param html_content: string with html content
:param render_anchor_tag_content: boolean flag indicating whether to extract
@@ -207,8 +212,19 @@ def html_to_text(html_content: str, render_anchor_tag_content=False) -> str:
else:
parser_config = None
# get text and annotations via inscriptis
text_content = get_text(html_content, config=parser_config)
def parse_function(html_content, parser_config, results_queue):
from inscriptis import get_text
# get text and annotations via inscriptis
text_content = get_text(html_content, config=parser_config)
results_queue.put(text_content)
results_queue = multiprocessing.Queue()
parse_process = multiprocessing.Process(target=parse_function, args=(html_content, parser_config, results_queue))
parse_process.daemon = True
parse_process.start()
text_content = results_queue.get() # blocks until results are available
parse_process.terminate()
return text_content

View File

@@ -7,19 +7,20 @@ from changedetectionio.html_tools import FilterNotFoundInResponse
# A single update worker
#
# Requests for checking on a single site(watch) from a queue of watches
# (another process inserts watches into the queue that are time-ready for checking)
#
class update_worker(threading.Thread):
current_uuid = None
def __init__(self, q, notification_q, app, datastore, *args, **kwargs):
def __init__(self, q, notification_q, app, datastore, uuid, *args, **kwargs):
self.q = q
self.app = app
self.notification_q = notification_q
self.datastore = datastore
self.current_uuid = uuid
super().__init__(*args, **kwargs)
self.name = "update_worker"
def send_filter_failure_notification(self, uuid):
@@ -47,169 +48,170 @@ class update_worker(threading.Thread):
self.notification_q.put(n_object)
print("Sent filter not found notification for {}".format(uuid))
# Pick one job off the list, process it threaded, exist
def run(self):
# Go talk to the website
self.perform_site_update()
self.current_uuid = None # Done
self.q.task_done()
# Let the thread die after processing 1
# We will launch nice juicy fresh threads every time to prevent memory leaks in complex runner code (playwright etc)
print ("EXITING THREAD!")
self.app.config.exit.wait(1)
return
def perform_site_update(self):
from changedetectionio import fetch_site_status
if not self.current_uuid in list(self.datastore.data['watching'].keys()):
return
changed_detected = False
contents = ""
screenshot = False
update_obj= {}
xpath_data = False
now = time.time()
update_handler = fetch_site_status.perform_site_check(datastore=self.datastore)
try:
changed_detected, update_obj, contents, screenshot, xpath_data = update_handler.run(self.current_uuid)
# Re #342
# In Python 3, all strings are sequences of Unicode characters. There is a bytes type that holds raw bytes.
# We then convert/.decode('utf-8') for the notification etc
if not isinstance(contents, (bytes, bytearray)):
raise Exception("Error - returned data from the fetch handler SHOULD be bytes")
except PermissionError as e:
self.app.logger.error("File permission error updating", self.current_uuid, str(e))
except content_fetcher.ReplyWithContentButNoText as e:
# Totally fine, it's by choice - just continue on, nothing more to care about
# Page had elements/content but no renderable text
self.datastore.update_watch(uuid=self.current_uuid, update_obj={'last_error': "Got HTML content but no text found."})
except FilterNotFoundInResponse as e:
err_text = "Filter '{}' not found - Did the page change its layout?".format(str(e))
c = 0
if self.datastore.data['watching'].get(self.current_uuid, False):
c = self.datastore.data['watching'][self.current_uuid].get('consecutive_filter_failures', 5)
c += 1
while not self.app.config.exit.is_set():
# Send notification if we reached the threshold?
threshold = self.datastore.data['settings']['application'].get('filter_failure_notification_threshold_attempts', 0)
print("Filter for {} not found, consecutive_filter_failures: {}".format(self.current_uuid, c))
if threshold >0 and c >= threshold:
self.send_filter_failure_notification(self.current_uuid)
c = 0
self.datastore.update_watch(uuid=self.current_uuid, update_obj={'last_error': err_text,
'consecutive_filter_failures': c})
except content_fetcher.EmptyReply as e:
# Some kind of custom to-str handler in the exception handler that does this?
err_text = "EmptyReply - try increasing 'Wait seconds before extracting text', Status Code {}".format(e.status_code)
self.datastore.update_watch(uuid=self.current_uuid, update_obj={'last_error': err_text,
'last_check_status': e.status_code})
except content_fetcher.ScreenshotUnavailable as e:
err_text = "Screenshot unavailable, page did not render fully in the expected time - try increasing 'Wait seconds before extracting text'"
self.datastore.update_watch(uuid=self.current_uuid, update_obj={'last_error': err_text,
'last_check_status': e.status_code})
except content_fetcher.PageUnloadable as e:
err_text = "Page request from server didnt respond correctly"
self.datastore.update_watch(uuid=self.current_uuid, update_obj={'last_error': err_text,
'last_check_status': e.status_code})
except Exception as e:
self.app.logger.error("Exception reached processing watch UUID: %s - %s", self.current_uuid, str(e))
self.datastore.update_watch(uuid=self.current_uuid, update_obj={'last_error': str(e)})
else:
try:
uuid = self.q.get(block=False)
except queue.Empty:
pass
watch = self.datastore.data['watching'][self.current_uuid]
fname = "" # Saved history text filename
else:
self.current_uuid = uuid
# For the FIRST time we check a site, or a change detected, save the snapshot.
if changed_detected or not watch['last_checked']:
# A change was detected
fname = watch.save_history_text(contents=contents, timestamp=str(round(time.time())))
if uuid in list(self.datastore.data['watching'].keys()):
# Generally update anything interesting returned
update_obj['consecutive_filter_failures'] = 0
self.datastore.update_watch(uuid=self.current_uuid, update_obj=update_obj)
changed_detected = False
contents = ""
screenshot = False
update_obj= {}
xpath_data = False
now = time.time()
# A change was detected
if changed_detected:
n_object = {}
print (">> Change detected in UUID {} - {}".format(self.current_uuid, watch['url']))
try:
changed_detected, update_obj, contents, screenshot, xpath_data = update_handler.run(uuid)
# Re #342
# In Python 3, all strings are sequences of Unicode characters. There is a bytes type that holds raw bytes.
# We then convert/.decode('utf-8') for the notification etc
if not isinstance(contents, (bytes, bytearray)):
raise Exception("Error - returned data from the fetch handler SHOULD be bytes")
except PermissionError as e:
self.app.logger.error("File permission error updating", uuid, str(e))
except content_fetcher.ReplyWithContentButNoText as e:
# Totally fine, it's by choice - just continue on, nothing more to care about
# Page had elements/content but no renderable text
self.datastore.update_watch(uuid=uuid, update_obj={'last_error': "Got HTML content but no text found."})
except FilterNotFoundInResponse as e:
err_text = "Filter '{}' not found - Did the page change its layout?".format(str(e))
c = 0
if self.datastore.data['watching'].get(uuid, False):
c = self.datastore.data['watching'][uuid].get('consecutive_filter_failures', 5)
c += 1
# Notifications should only trigger on the second time (first time, we gather the initial snapshot)
if watch.history_n >= 2:
# Atleast 2, means there really was a change
self.datastore.update_watch(uuid=self.current_uuid, update_obj={'last_changed': round(now)})
# Send notification if we reached the threshold?
threshold = self.datastore.data['settings']['application'].get('filter_failure_notification_threshold_attempts', 0)
print("Filter for {} not found, consecutive_filter_failures: {}".format(uuid, c))
if threshold >0 and c >= threshold:
self.send_filter_failure_notification(uuid)
c = 0
watch_history = watch.history
dates = list(watch_history.keys())
# Theoretically it's possible that this could be just 1 long,
# - In the case that the timestamp key was not unique
if len(dates) == 1:
raise ValueError(
"History index had 2 or more, but only 1 date loaded, timestamps were not unique? maybe two of the same timestamps got written, needs more delay?"
)
prev_fname = watch_history[dates[-2]]
self.datastore.update_watch(uuid=uuid, update_obj={'last_error': err_text,
'consecutive_filter_failures': c})
except content_fetcher.EmptyReply as e:
# Some kind of custom to-str handler in the exception handler that does this?
err_text = "EmptyReply - try increasing 'Wait seconds before extracting text', Status Code {}".format(e.status_code)
self.datastore.update_watch(uuid=uuid, update_obj={'last_error': err_text,
'last_check_status': e.status_code})
except content_fetcher.ScreenshotUnavailable as e:
err_text = "Screenshot unavailable, page did not render fully in the expected time - try increasing 'Wait seconds before extracting text'"
self.datastore.update_watch(uuid=uuid, update_obj={'last_error': err_text,
'last_check_status': e.status_code})
except content_fetcher.PageUnloadable as e:
err_text = "Page request from server didnt respond correctly"
self.datastore.update_watch(uuid=uuid, update_obj={'last_error': err_text,
'last_check_status': e.status_code})
except Exception as e:
self.app.logger.error("Exception reached processing watch UUID: %s - %s", uuid, str(e))
self.datastore.update_watch(uuid=uuid, update_obj={'last_error': str(e)})
# Did it have any notification alerts to hit?
if len(watch['notification_urls']):
print(">>> Notifications queued for UUID from watch {}".format(self.current_uuid))
n_object['notification_urls'] = watch['notification_urls']
n_object['notification_title'] = watch['notification_title']
n_object['notification_body'] = watch['notification_body']
n_object['notification_format'] = watch['notification_format']
else:
try:
watch = self.datastore.data['watching'][uuid]
fname = "" # Saved history text filename
# No? maybe theres a global setting, queue them all
elif len(self.datastore.data['settings']['application']['notification_urls']):
print(">>> Watch notification URLs were empty, using GLOBAL notifications for UUID: {}".format(self.current_uuid))
n_object['notification_urls'] = self.datastore.data['settings']['application']['notification_urls']
n_object['notification_title'] = self.datastore.data['settings']['application']['notification_title']
n_object['notification_body'] = self.datastore.data['settings']['application']['notification_body']
n_object['notification_format'] = self.datastore.data['settings']['application']['notification_format']
else:
print(">>> NO notifications queued, watch and global notification URLs were empty.")
# For the FIRST time we check a site, or a change detected, save the snapshot.
if changed_detected or not watch['last_checked']:
# A change was detected
fname = watch.save_history_text(contents=contents, timestamp=str(round(time.time())))
# Only prepare to notify if the rules above matched
if 'notification_urls' in n_object:
# HTML needs linebreak, but MarkDown and Text can use a linefeed
if n_object['notification_format'] == 'HTML':
line_feed_sep = "</br>"
else:
line_feed_sep = "\n"
# Generally update anything interesting returned
update_obj['consecutive_filter_failures'] = 0
self.datastore.update_watch(uuid=uuid, update_obj=update_obj)
from changedetectionio import diff
n_object.update({
'watch_url': watch['url'],
'uuid': self.current_uuid,
'current_snapshot': contents.decode('utf-8'),
'diff': diff.render_diff(prev_fname, fname, line_feed_sep=line_feed_sep),
'diff_full': diff.render_diff(prev_fname, fname, True, line_feed_sep=line_feed_sep)
})
# A change was detected
if changed_detected:
n_object = {}
print (">> Change detected in UUID {} - {}".format(uuid, watch['url']))
self.notification_q.put(n_object)
# Notifications should only trigger on the second time (first time, we gather the initial snapshot)
if watch.history_n >= 2:
# Atleast 2, means there really was a change
self.datastore.update_watch(uuid=uuid, update_obj={'last_changed': round(now)})
except Exception as e:
# Catch everything possible here, so that if a worker crashes, we don't lose it until restart!
print("!!!! Exception in update_worker !!!\n", e)
self.app.logger.error("Exception reached processing watch UUID: %s - %s", self.current_uuid, str(e))
self.datastore.update_watch(uuid=self.current_uuid, update_obj={'last_error': str(e)})
watch_history = watch.history
dates = list(watch_history.keys())
# Theoretically it's possible that this could be just 1 long,
# - In the case that the timestamp key was not unique
if len(dates) == 1:
raise ValueError(
"History index had 2 or more, but only 1 date loaded, timestamps were not unique? maybe two of the same timestamps got written, needs more delay?"
)
prev_fname = watch_history[dates[-2]]
finally:
# Always record that we atleast tried
self.datastore.update_watch(uuid=self.current_uuid, update_obj={'fetch_time': round(time.time() - now, 3),
'last_checked': round(time.time())})
# Did it have any notification alerts to hit?
if len(watch['notification_urls']):
print(">>> Notifications queued for UUID from watch {}".format(uuid))
n_object['notification_urls'] = watch['notification_urls']
n_object['notification_title'] = watch['notification_title']
n_object['notification_body'] = watch['notification_body']
n_object['notification_format'] = watch['notification_format']
# No? maybe theres a global setting, queue them all
elif len(self.datastore.data['settings']['application']['notification_urls']):
print(">>> Watch notification URLs were empty, using GLOBAL notifications for UUID: {}".format(uuid))
n_object['notification_urls'] = self.datastore.data['settings']['application']['notification_urls']
n_object['notification_title'] = self.datastore.data['settings']['application']['notification_title']
n_object['notification_body'] = self.datastore.data['settings']['application']['notification_body']
n_object['notification_format'] = self.datastore.data['settings']['application']['notification_format']
else:
print(">>> NO notifications queued, watch and global notification URLs were empty.")
# Only prepare to notify if the rules above matched
if 'notification_urls' in n_object:
# HTML needs linebreak, but MarkDown and Text can use a linefeed
if n_object['notification_format'] == 'HTML':
line_feed_sep = "</br>"
else:
line_feed_sep = "\n"
from changedetectionio import diff
n_object.update({
'watch_url': watch['url'],
'uuid': uuid,
'current_snapshot': contents.decode('utf-8'),
'diff': diff.render_diff(prev_fname, fname, line_feed_sep=line_feed_sep),
'diff_full': diff.render_diff(prev_fname, fname, True, line_feed_sep=line_feed_sep)
})
self.notification_q.put(n_object)
except Exception as e:
# Catch everything possible here, so that if a worker crashes, we don't lose it until restart!
print("!!!! Exception in update_worker !!!\n", e)
self.app.logger.error("Exception reached processing watch UUID: %s - %s", uuid, str(e))
self.datastore.update_watch(uuid=uuid, update_obj={'last_error': str(e)})
finally:
# Always record that we atleast tried
self.datastore.update_watch(uuid=uuid, update_obj={'fetch_time': round(time.time() - now, 3),
'last_checked': round(time.time())})
# Always save the screenshot if it's available
if screenshot:
self.datastore.save_screenshot(watch_uuid=uuid, screenshot=screenshot)
if xpath_data:
self.datastore.save_xpath_data(watch_uuid=uuid, data=xpath_data)
# Always save the screenshot if it's available
if screenshot:
self.datastore.save_screenshot(watch_uuid=self.current_uuid, screenshot=screenshot)
if xpath_data:
self.datastore.save_xpath_data(watch_uuid=self.current_uuid, data=xpath_data)
self.current_uuid = None # Done
self.q.task_done()
# Give the CPU time to interrupt
time.sleep(0.1)
self.app.config.exit.wait(1)