mirror of
https://github.com/searxng/searxng.git
synced 2024-05-23 05:48:05 +00:00
86b4d2f2d0
We have been using a static type checker (pyright) for a long time, but its check was not yet a prerequisite for passing the quality gate. It was checked in the CI, but the error messages were only logged. As is always the case in life, with checks that you have to do but which have no consequences; you neglect them :-) We didn't activate the checks back then because we (even today) have too much monkey patching in our code (not only in the engines, httpx and others objects are also affected). We want to replace monkey patching with clear interfaces for a long time, the basis for this is increased typing and we can only achieve this if we make type checking an integral part of the quality gate. This PR activates the type check; in order to pass the check, a few typings were corrected in the code, but most type inconsistencies were deactivated via inline comments. This was particularly necessary in places where the code uses properties that stick to the objects (monkey patching). The sticking of properties only happens in a few places, but the access to these properties extends over the entire code, which is why there are many `# type: ignore` markers in the code ... which we will hopefully be able to remove again successively in the future. Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
59 lines
2.5 KiB
Python
59 lines
2.5 KiB
Python
# SPDX-License-Identifier: AGPL-3.0-or-later
|
|
# 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)
|
|
"""
|
|
|
|
from typing import Callable
|
|
import logging
|
|
import time
|
|
from pathlib import Path
|
|
|
|
from searx.redisdb import client as get_redis_client
|
|
from searx.redislib import lua_script_storage
|
|
|
|
|
|
logger = logging.getLogger('searx.search.checker')
|
|
|
|
SCHEDULER_LUA = Path(__file__).parent / "scheduler.lua"
|
|
|
|
|
|
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 = SCHEDULER_LUA.open().read()
|
|
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)
|