Merge pull request #1856 from dalf/checker_requires_redis

The checker requires Redis

docs/admin/engines/settings.rst

@@ -314,20 +314,21 @@ developer) account needs to be added to the *searxng-redis* group.
 
 .. admonition:: Tip for developers
 
-   To set up a local redis instance using sockets simply use::
-
-     $ ./manage redis.build
-     $ sudo -H ./manage redis.install
-     $ sudo -H ./manage redis.addgrp "${USER}"
-     # don't forget to logout & login to get member of group
-
-   The YAML setting for such a redis instance is:
+   To set up a local redis instance, first set the socket path of the Redis DB
+   in your YAML setting:
 
    .. code:: yaml
 
       redis:
         url: unix:///usr/local/searxng-redis/run/redis.sock?db=0
 
+   Then use the following commands to install the redis instance ::
+
+     $ ./manage redis.build
+     $ sudo -H ./manage redis.install
+     $ sudo -H ./manage redis.addgrp "${USER}"
+     # don't forget to logout & login to get member of group
+
 
 .. _settings outgoing:
 

searx/plugins/limiter.py

@@ -93,9 +93,8 @@ def init(app, settings):
     if not settings['server']['limiter']:
         return False
 
-    logger.debug("init limiter DB")  # pylint: disable=undefined-variable
-    if not redisdb.init():
-        logger.error("init limiter DB failed!!!")  # pylint: disable=undefined-variable
+    if not redisdb.client():
+        logger.error("The limiter requires Redis")  # pylint: disable=undefined-variable
         return False
 
     app.before_request(pre_request)

searx/search/checker/__init__.py

@@ -2,3 +2,5 @@
 
 from .impl import Checker
 from .background import initialize, get_result
+
+__all__ = ('Checker', 'initialize', 'get_result')

searx/search/checker/background.py

@@ -1,26 +1,28 @@
 # SPDX-License-Identifier: AGPL-3.0-or-later
 # lint: pylint
 # pylint: disable=missing-module-docstring
-# pyright: strict
+# pyright: basic
 
 import json
-import random
 import time
 import threading
 import os
 import signal
-from typing import Dict, Union, List, Any, Tuple
+from typing import Dict, Union, List, Any, Tuple, Optional
 from typing_extensions import TypedDict, Literal
 
+import redis.exceptions
+
 from searx import logger, settings, searx_debug
+from searx.shared.redisdb import client as get_redis_client
 from searx.exceptions import SearxSettingsException
 from searx.search.processors import PROCESSORS
 from searx.search.checker import Checker
-from searx.shared import schedule, storage  # pyright: ignore
+from searx.search.checker.scheduler import scheduler_function
 
 
-CHECKER_RESULT = 'CHECKER_RESULT'
-running = threading.Lock()
+REDIS_RESULT_KEY = 'SearXNG_checker_result'
+REDIS_LOCK_KEY = 'SearXNG_checker_lock'
 
 
 CheckerResult = Union['CheckerOk', 'CheckerErr', 'CheckerOther']
@@ -77,20 +79,24 @@ def _get_interval(every: Any, error_msg: str) -> Tuple[int, int]:
     return (every[0], every[1])
 
 
-def _get_every():
-    every = settings.get('checker', {}).get('scheduling', {}).get('every', (300, 1800))
-    return _get_interval(every, 'checker.scheduling.every is not a int or list')
-
-
 def get_result() -> CheckerResult:
-    serialized_result = storage.get_str(CHECKER_RESULT)
-    if serialized_result is not None:
-        return json.loads(serialized_result)
-    return {'status': 'unknown'}
+    client = get_redis_client()
+    if client is None:
+        # without Redis, the checker is disabled
+        return {'status': 'disabled'}
+    serialized_result: Optional[bytes] = client.get(REDIS_RESULT_KEY)
+    if serialized_result is None:
+        # the Redis key does not exist
+        return {'status': 'unknown'}
+    return json.loads(serialized_result)
 
 
 def _set_result(result: CheckerResult):
-    storage.set_str(CHECKER_RESULT, json.dumps(result))
+    client = get_redis_client()
+    if client is None:
+        # without Redis, the function does nothing
+        return
+    client.set(REDIS_RESULT_KEY, json.dumps(result))
 
 
 def _timestamp():
@@ -98,41 +104,29 @@ def _timestamp():
 
 
 def run():
-    if not running.acquire(blocking=False):  # pylint: disable=consider-using-with
-        return
     try:
-        logger.info('Starting checker')
-        result: CheckerOk = {'status': 'ok', 'engines': {}, 'timestamp': _timestamp()}
-        for name, processor in PROCESSORS.items():
-            logger.debug('Checking %s engine', name)
-            checker = Checker(processor)
-            checker.run()
-            if checker.test_results.successful:
-                result['engines'][name] = {'success': True}
-            else:
-                result['engines'][name] = {'success': False, 'errors': checker.test_results.errors}
+        # use a Redis lock to make sure there is no checker running at the same time
+        # (this should not happen, this is a safety measure)
+        with get_redis_client().lock(REDIS_LOCK_KEY, blocking_timeout=60, timeout=3600):
+            logger.info('Starting checker')
+            result: CheckerOk = {'status': 'ok', 'engines': {}, 'timestamp': _timestamp()}
+            for name, processor in PROCESSORS.items():
+                logger.debug('Checking %s engine', name)
+                checker = Checker(processor)
+                checker.run()
+                if checker.test_results.successful:
+                    result['engines'][name] = {'success': True}
+                else:
+                    result['engines'][name] = {'success': False, 'errors': checker.test_results.errors}
 
-        _set_result(result)
-        logger.info('Check done')
+            _set_result(result)
+            logger.info('Check done')
+    except redis.exceptions.LockError:
+        _set_result({'status': 'error', 'timestamp': _timestamp()})
+        logger.exception('Error while running the checker')
     except Exception:  # pylint: disable=broad-except
         _set_result({'status': 'error', 'timestamp': _timestamp()})
         logger.exception('Error while running the checker')
-    finally:
-        running.release()
-
-
-def _run_with_delay():
-    every = _get_every()
-    delay = random.randint(0, every[1] - every[0])
-    logger.debug('Start checker in %i seconds', delay)
-    time.sleep(delay)
-    run()
-
-
-def _start_scheduling():
-    every = _get_every()
-    if schedule(every[0], _run_with_delay):
-        run()
 
 
 def _signal_handler(_signum: int, _frame: Any):
@@ -147,27 +141,31 @@ def initialize():
         logger.info('Send SIGUSR1 signal to pid %i to start the checker', os.getpid())
         signal.signal(signal.SIGUSR1, _signal_handler)
 
-    # disabled by default
-    _set_result({'status': 'disabled'})
-
     # special case when debug is activate
-    if searx_debug and settings.get('checker', {}).get('off_when_debug', True):
+    if searx_debug and settings['checker']['off_when_debug']:
         logger.info('debug mode: checker is disabled')
         return
 
     # check value of checker.scheduling.every now
-    scheduling = settings.get('checker', {}).get('scheduling', None)
+    scheduling = settings['checker']['scheduling']
     if scheduling is None or not scheduling:
         logger.info('Checker scheduler is disabled')
         return
 
-    #
-    _set_result({'status': 'unknown'})
+    # make sure there is a Redis connection
+    if get_redis_client() is None:
+        logger.error('The checker requires Redis')
+        return
 
-    start_after = scheduling.get('start_after', (300, 1800))
-    start_after = _get_interval(start_after, 'checker.scheduling.start_after is not a int or list')
-    delay = random.randint(start_after[0], start_after[1])
-    logger.info('Start checker in %i seconds', delay)
-    t = threading.Timer(delay, _start_scheduling)
+    # start the background scheduler
+    every_range = _get_interval(scheduling.get('every', (300, 1800)), 'checker.scheduling.every is not a int or list')
+    start_after_range = _get_interval(
+        scheduling.get('start_after', (300, 1800)), 'checker.scheduling.start_after is not a int or list'
+    )
+    t = threading.Thread(
+        target=scheduler_function,
+        args=(start_after_range[0], start_after_range[1], every_range[0], every_range[1], run),
+        name='checker_scheduler',
+    )
     t.daemon = True
     t.start()

searx/search/checker/scheduler.lua

@@ -0,0 +1,36 @@
+-- SPDX-License-Identifier: AGPL-3.0-or-later
+--
+-- This script is not a string in scheduler.py, so editors can provide syntax highlighting.
+
+-- The Redis KEY is defined here and not in Python on purpose:
+-- only this LUA script can read and update this key to avoid lock and concurrency issues.
+local redis_key = 'SearXNG_checker_next_call_ts'
+
+local now = redis.call('TIME')[1]
+local start_after_from = ARGV[1]
+local start_after_to = ARGV[2]
+local every_from = ARGV[3]
+local every_to = ARGV[4]
+
+local next_call_ts = redis.call('GET', redis_key)
+
+if (next_call_ts == false or next_call_ts == nil) then
+    -- the scheduler has never run on this Redis instance, so:
+    -- 1/ the scheduler does not run now
+    -- 2/ the next call is a random time between start_after_from and start_after_to
+    local initial_delay = math.random(start_after_from, start_after_to)
+    redis.call('SET', redis_key, now + initial_delay)
+    return { false, delay }
+end
+
+-- next_call_ts is defined
+-- --> if now is lower than next_call_ts then we don't run the embedded checker
+-- --> if now is higher then we update next_call_ts and ask to run the embedded checker now.
+local call_now = next_call_ts <= now
+if call_now then
+    -- the checker runs now, define the timestamp of the next call:
+    -- this is a random delay between every_from and every_to
+    local periodic_delay = math.random(every_from, every_to)
+    next_call_ts = redis.call('INCRBY', redis_key, periodic_delay)
+end
+return { call_now, next_call_ts - now }

searx/search/checker/scheduler.py

@@ -0,0 +1,57 @@
+# SPDX-License-Identifier: AGPL-3.0-or-later
+# lint: pylint
+# pylint: disable=missing-module-docstring
+"""Lame scheduler which use Redis as a source of truth:
+* the Redis key SearXNG_checker_next_call_ts contains the next time the embedded checker should run.
+* to avoid lock, a unique Redis script reads and updates the Redis key SearXNG_checker_next_call_ts.
+* this Redis script returns a list of two elements:
+   * the first one is a boolean. If True, the embedded checker must run now in this worker.
+   * the second element is the delay in second to wait before the next call to the Redis script.
+
+This scheduler is not generic on purpose: if more feature are required, a dedicate scheduler must be used
+(= a better scheduler should not use the web workers)
+"""
+
+import logging
+import time
+import importlib
+from typing import Callable
+
+from searx.shared.redisdb import client as get_redis_client
+from searx.redislib import lua_script_storage
+
+
+logger = logging.getLogger('searx.search.checker')
+
+
+def scheduler_function(start_after_from: int, start_after_to: int, every_from: int, every_to: int, callback: Callable):
+    """Run the checker periodically. The function never returns.
+
+    Parameters:
+    * start_after_from and start_after_to: when to call "callback" for the first on the Redis instance
+    * every_from and every_to: after the first call, how often to call "callback"
+
+    There is no issue:
+    * to call this function is multiple workers
+    * to kill workers at any time as long there is one at least one worker
+    """
+    scheduler_now_script = importlib.resources.read_text(__package__, "scheduler.lua")
+    while True:
+        # ask the Redis script what to do
+        # the script says
+        # * if the checker must run now.
+        # * how to long to way before calling the script again (it can be call earlier, but not later).
+        script = lua_script_storage(get_redis_client(), scheduler_now_script)
+        call_now, wait_time = script(args=[start_after_from, start_after_to, every_from, every_to])
+
+        # does the worker run the checker now?
+        if call_now:
+            # run the checker
+            try:
+                callback()
+            except Exception:  # pylint: disable=broad-except
+                logger.exception("Error calling the embedded checker")
+            # only worker display the wait_time
+            logger.info("Next call to the checker in %s seconds", wait_time)
+        # wait until the next call
+        time.sleep(wait_time)

searx/settings.yml

@@ -78,7 +78,7 @@ server:
 
 redis:
   # https://redis-py.readthedocs.io/en/stable/connections.html#redis.client.Redis.from_url
-  url: unix:///usr/local/searxng-redis/run/redis.sock?db=0
+  url: false
 
 ui:
   # Custom static path - leave it blank if you didn't change

searx/settings_defaults.py

@@ -174,7 +174,7 @@ SCHEMA = {
         'default_http_headers': SettingsValue(dict, {}),
     },
     'redis': {
-        'url': SettingsValue(str, 'unix:///usr/local/searxng-redis/run/redis.sock?db=0'),
+        'url': SettingsValue((None, False, str), False),
     },
     'ui': {
         'static_path': SettingsDirectoryValue(str, os.path.join(searx_dir, 'static')),
@@ -225,7 +225,8 @@ SCHEMA = {
     'plugins': SettingsValue(list, []),
     'enabled_plugins': SettingsValue((None, list), None),
     'checker': {
-        'off_when_debug': SettingsValue(bool, True),
+        'off_when_debug': SettingsValue(bool, True, None),
+        'scheduling': SettingsValue((None, dict), None, None),
     },
     'categories_as_tabs': SettingsValue(dict, CATEGORIES_AS_TABS),
     'engines': SettingsValue(list, []),

searx/shared/__init__.py

@@ -1,39 +1,6 @@
 # SPDX-License-Identifier: AGPL-3.0-or-later
+# lint: pylint
+"""Initialization of a *shared* storage.
+"""
 
-import logging
-import importlib
-
-logger = logging.getLogger('searx.shared')
-
-__all__ = ['SharedDict', 'schedule']
-
-try:
-    uwsgi = importlib.import_module('uwsgi')
-except:
-    # no uwsgi
-    from .shared_simple import SimpleSharedDict as SharedDict, schedule
-
-    logger.info('Use shared_simple implementation')
-else:
-    try:
-        uwsgi.cache_update('dummy', b'dummy')
-        if uwsgi.cache_get('dummy') != b'dummy':
-            raise Exception()
-    except:
-        # uwsgi.ini configuration problem: disable all scheduling
-        logger.error(
-            'uwsgi.ini configuration error, add this line to your uwsgi.ini\n'
-            'cache2 = name=searxngcache,items=2000,blocks=2000,blocksize=4096,bitmap=1'
-        )
-        from .shared_simple import SimpleSharedDict as SharedDict
-
-        def schedule(delay, func, *args):
-            return False
-
-    else:
-        # uwsgi
-        from .shared_uwsgi import UwsgiCacheSharedDict as SharedDict, schedule
-
-        logger.info('Use shared_uwsgi implementation')
-
-storage = SharedDict()
+from . import redisdb

searx/shared/redisdb.py

@@ -26,26 +26,31 @@ import redis
 from searx import get_setting
 
 
-logger = logging.getLogger('searx.shared.redis')
-_client = None
+OLD_REDIS_URL_DEFAULT_URL = 'unix:///usr/local/searxng-redis/run/redis.sock?db=0'
+"""This was the default Redis URL in settings.yml."""
+
+_CLIENT = None
+logger = logging.getLogger('searx.shared.redisdb')
 
 
-def client():
-    global _client  # pylint: disable=global-statement
-    if _client is None:
-        # not thread safe: in the worst case scenario, two or more clients are
-        # initialized only one is kept, the others are garbage collected.
-        _client = redis.Redis.from_url(get_setting('redis.url'))
-    return _client
+def client() -> redis.Redis:
+    return _CLIENT
 
 
-def init():
+def initialize():
+    global _CLIENT  # pylint: disable=global-statement
+    redis_url = get_setting('redis.url')
     try:
-        c = client()
-        logger.info("connected redis DB --> %s", c.acl_whoami())
-        return True
-    except redis.exceptions.ConnectionError as exc:
+        if redis_url:
+            _CLIENT = redis.Redis.from_url(redis_url)
+            logger.info("connected redis: %s", redis_url)
+            return True
+    except redis.exceptions.ConnectionError:
         _pw = pwd.getpwuid(os.getuid())
-        logger.error("[%s (%s)] can't connect redis DB ...", _pw.pw_name, _pw.pw_uid)
-        logger.error("  %s", exc)
+        logger.exception("[%s (%s)] can't connect redis DB ...", _pw.pw_name, _pw.pw_uid)
+        if redis_url == OLD_REDIS_URL_DEFAULT_URL:
+            logger.info(
+                "You can safely ignore the above Redis error if you don't use Redis."
+                "You can remove this error by setting redis.url to false in your settings.yml."
+            )
     return False

searx/shared/shared_abstract.py

@@ -1,22 +0,0 @@
-# SPDX-License-Identifier: AGPL-3.0-or-later
-# pyright: strict
-from abc import ABC, abstractmethod
-from typing import Optional
-
-
-class SharedDict(ABC):
-    @abstractmethod
-    def get_int(self, key: str) -> Optional[int]:
-        pass
-
-    @abstractmethod
-    def set_int(self, key: str, value: int):
-        pass
-
-    @abstractmethod
-    def get_str(self, key: str) -> Optional[str]:
-        pass
-
-    @abstractmethod
-    def set_str(self, key: str, value: str):
-        pass

searx/shared/shared_simple.py

@@ -1,40 +0,0 @@
-# SPDX-License-Identifier: AGPL-3.0-or-later
-
-import threading
-from typing import Optional
-
-from . import shared_abstract
-
-
-class SimpleSharedDict(shared_abstract.SharedDict):
-
-    __slots__ = ('d',)
-
-    def __init__(self):
-        self.d = {}
-
-    def get_int(self, key: str) -> Optional[int]:
-        return self.d.get(key, None)
-
-    def set_int(self, key: str, value: int):
-        self.d[key] = value
-
-    def get_str(self, key: str) -> Optional[str]:
-        return self.d.get(key, None)
-
-    def set_str(self, key: str, value: str):
-        self.d[key] = value
-
-
-def schedule(delay, func, *args):
-    def call_later():
-        t = threading.Timer(delay, wrapper)
-        t.daemon = True
-        t.start()
-
-    def wrapper():
-        call_later()
-        func(*args)
-
-    call_later()
-    return True

searx/shared/shared_uwsgi.py

@@ -1,64 +0,0 @@
-# SPDX-License-Identifier: AGPL-3.0-or-later
-
-import time
-from typing import Optional
-import uwsgi  # pyright: ignore # pylint: disable=E0401
-from . import shared_abstract
-
-
-_last_signal = 10
-
-
-class UwsgiCacheSharedDict(shared_abstract.SharedDict):
-    def get_int(self, key: str) -> Optional[int]:
-        value = uwsgi.cache_get(key)
-        if value is None:
-            return value
-        else:
-            return int.from_bytes(value, 'big')
-
-    def set_int(self, key: str, value: int):
-        b = value.to_bytes(4, 'big')
-        uwsgi.cache_update(key, b)
-
-    def get_str(self, key: str) -> Optional[str]:
-        value = uwsgi.cache_get(key)
-        if value is None:
-            return value
-        else:
-            return value.decode('utf-8')
-
-    def set_str(self, key: str, value: str):
-        b = value.encode('utf-8')
-        uwsgi.cache_update(key, b)
-
-
-def schedule(delay, func, *args):
-    """
-    Can be implemented using a spooler.
-    https://uwsgi-docs.readthedocs.io/en/latest/PythonDecorators.html
-
-    To make the uwsgi configuration simple, use the alternative implementation.
-    """
-    global _last_signal
-
-    def sighandler(signum):
-        now = int(time.time())
-        key = 'scheduler_call_time_signal_' + str(signum)
-        uwsgi.lock()
-        try:
-            updating = uwsgi.cache_get(key)
-            if updating is not None:
-                updating = int.from_bytes(updating, 'big')
-                if now - updating < delay:
-                    return
-            uwsgi.cache_update(key, now.to_bytes(4, 'big'))
-        finally:
-            uwsgi.unlock()
-        func(*args)
-
-    signal_num = _last_signal
-    _last_signal += 1
-    uwsgi.register_signal(signal_num, 'worker', sighandler)
-    uwsgi.add_timer(signal_num, delay)
-    return True

searx/webapp.py

@@ -120,6 +120,7 @@ from searx.locales import (
 # renaming names from searx imports ...
 from searx.autocomplete import search_autocomplete, backends as autocomplete_backends
 from searx.languages import language_codes as languages
+from searx.shared.redisdb import initialize as redis_initialize
 from searx.search import SearchWithPlugins, initialize as search_initialize
 from searx.network import stream as http_stream, set_context_network_name
 from searx.search.checker import get_result as checker_get_result
@@ -1384,6 +1385,7 @@ werkzeug_reloader = flask_run_development or (searx_debug and __name__ == "__mai
 if not werkzeug_reloader or (werkzeug_reloader and os.environ.get("WERKZEUG_RUN_MAIN") == "true"):
     locales_initialize()
     _INFO_PAGES = infopage.InfoPageSet()
+    redis_initialize()
     plugin_initialize(app)
     search_initialize(enable_checker=True, check_network=True, enable_metrics=settings['general']['enable_metrics'])
 

utils/searxng.sh

@@ -295,7 +295,7 @@ In your instance, redis DB connector is configured at:
 
     ${redis_url}
 "
-    if searxng.instance.exec python -c "from searx.shared import redisdb; redisdb.init() or exit(42)"; then
+    if searxng.instance.exec python -c "from searx.shared import redisdb; redisdb.initialize() or exit(42)"; then
         info_msg "SearXNG instance is able to connect redis DB."
         return
     fi
@@ -684,7 +684,7 @@ To install uWSGI use::
         die 42 "SearXNG's uWSGI app not available"
     fi
 
-    if ! searxng.instance.exec python -c "from searx.shared import redisdb; redisdb.init() or exit(42)"; then
+    if ! searxng.instance.exec python -c "from searx.shared import redisdb; redisdb.initialize() or exit(42)"; then
         rst_para "\
 The configured redis DB is not available: If your server is public to the
 internet, you should setup a bot protection to block excessively bot queries.

utils/searxng_check.py

@@ -29,6 +29,6 @@ if os.path.isfile(OLD_SETTING):
 from searx.shared import redisdb
 from searx import get_setting
 
-if not redisdb.init():
+if not redisdb.initialize():
     warnings.warn("can't connect to redis DB at: %s" % get_setting('redis.url'), RuntimeWarning, stacklevel=2)
     warnings.warn("--> no bot protection without redis DB", RuntimeWarning, stacklevel=2)