github/krkn
1
0
Fork 0
mirror of https://github.com/krkn-chaos/krkn.git synced 2026-08-25 09:27:36 +00:00
Code Issues Packages Projects Releases Wiki Activity

updating health checks

Browse Source 
Signed-off-by: Paige Patton <prubenda@redhat.com>
This commit is contained in:
Paige Patton
2025-06-06 10:54:39 -04:00
committed by Naga Ravi Chaitanya Elluri
parent 5bdbf622c3
commit 4c462a8971
1 changed files with 44 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files
krkn/utils/HealthChecker.py
+44 -40
View File
@@ -13,16 +13,15 @@ class HealthChecker:
def make_request(self, url, auth=None, headers=None, verify=True):
response_data = {}
response = requests.get(url, auth=auth, headers=headers, verify=verify)
response = requests.get(url, auth=auth, headers=headers, verify=verify, timeout=3)
response_data["url"] = url
response_data["status"] = response.status_code == 200
response_data["status_code"] = response.status_code
return response_data
def run_health_check(self, health_check_config, health_check_telemetry_queue: queue.Queue):
def run_health_check(self, health_check_config, health_check_telemetry_queue: queue.Queue):
if health_check_config and health_check_config["config"] and any(config.get("url") for config in health_check_config["config"]):
health_check_start_time_stamp = datetime.now()
health_check_telemetry = []
health_check_tracker = {}
interval = health_check_config["interval"] if health_check_config["interval"] else 2
@@ -37,49 +36,54 @@ class HealthChecker:
if config["bearer_token"]:
bearer_token = "Bearer " + config["bearer_token"]
headers = {"Authorization": bearer_token}
if config["auth"]: auth = tuple(config["auth"].split(','))
response = self.make_request(url, auth, headers, verify_url)
if response["status_code"] != 200:
if config["url"] not in health_check_tracker:
start_timestamp = datetime.now()
health_check_tracker[config["url"]] = {
"status_code": response["status_code"],
"start_timestamp": start_timestamp
}
try:
response = self.make_request(url, auth, headers, verify_url)
except Exception:
response = {}
response['status_code'] = 500
if config["url"] not in health_check_tracker:
start_timestamp = datetime.now()
health_check_tracker[config["url"]] = {
"status_code": response["status_code"],
"start_timestamp": start_timestamp
}
if response["status_code"] != 200:
if response_tracker[config["url"]] != False: response_tracker[config["url"]] = False
if config["exit_on_failure"] and config["exit_on_failure"] == True and self.ret_value==0: self.ret_value = 2
else:
if config["url"] in health_check_tracker:
end_timestamp = datetime.now()
start_timestamp = health_check_tracker[config["url"]]["start_timestamp"]
previous_status_code = str(health_check_tracker[config["url"]]["status_code"])
duration = (end_timestamp - start_timestamp).total_seconds()
downtime_record = {
"url": config["url"],
"status": False,
"status_code": previous_status_code,
"start_timestamp": start_timestamp.isoformat(),
"end_timestamp": end_timestamp.isoformat(),
"duration": duration
}
health_check_telemetry.append(HealthCheck(downtime_record))
del health_check_tracker[config["url"]]
if response["status_code"] != health_check_tracker[config["url"]]["status_code"]:
end_timestamp = datetime.now()
start_timestamp = health_check_tracker[config["url"]]["start_timestamp"]
previous_status_code = str(health_check_tracker[config["url"]]["status_code"])
duration = (end_timestamp - start_timestamp).total_seconds()
change_record = {
"url": config["url"],
"status": False,
"status_code": previous_status_code,
"start_timestamp": start_timestamp.isoformat(),
"end_timestamp": end_timestamp.isoformat(),
"duration": duration
}
health_check_telemetry.append(HealthCheck(change_record))
if response_tracker[config["url"]] != True: response_tracker[config["url"]] = True
del health_check_tracker[config["url"]]
time.sleep(interval)
health_check_end_time_stamp = datetime.now()
for url, status in response_tracker.items():
if status == True:
duration = (health_check_end_time_stamp - health_check_start_time_stamp).total_seconds()
success_response = {
"url": url,
"status": True,
"status_code": 200,
"start_timestamp": health_check_start_time_stamp.isoformat(),
"end_timestamp": health_check_end_time_stamp.isoformat(),
"duration": duration
}
health_check_telemetry.append(HealthCheck(success_response))
for url in health_check_tracker.keys():
duration = (health_check_end_time_stamp - health_check_tracker[url]["start_timestamp"]).total_seconds()
success_response = {
"url": url,
"status": True,
"status_code": response["status_code"],
"start_timestamp": health_check_tracker[url]["start_timestamp"].isoformat(),
"end_timestamp": health_check_end_time_stamp.isoformat(),
"duration": duration
}
health_check_telemetry.append(HealthCheck(success_response))
health_check_telemetry_queue.put(health_check_telemetry)
else:
logging.info("health checks config is not defined, skipping them")