Compare commits
7 Commits
4de16544a3
...
main
Author | SHA1 | Date | |
---|---|---|---|
e072fc2082 | |||
586193e60f | |||
36ef2b48db | |||
0d0abc2c9d | |||
727e13bbab | |||
90aa23cf63 | |||
e6f24496a8 |
73
checker.py
73
checker.py
@ -1,3 +1,8 @@
|
||||
# flake8: noqa
|
||||
# pylint: disable=broad-exception-raised, raise-missing-from, too-many-arguments, redefined-outer-name
|
||||
# pylance: disable=reportMissingImports, reportMissingModuleSource, reportGeneralTypeIssues
|
||||
# type: ignore
|
||||
|
||||
import re
|
||||
from datetime import datetime, timedelta, timezone
|
||||
import subprocess
|
||||
@ -6,7 +11,8 @@ import time
|
||||
import random
|
||||
import sys
|
||||
import pkg_resources
|
||||
|
||||
import requests
|
||||
import json
|
||||
required_packages = ['grist-api', 'colorama']
|
||||
installed_packages = [pkg.key for pkg in pkg_resources.working_set]
|
||||
|
||||
@ -16,7 +22,7 @@ for package in required_packages:
|
||||
|
||||
from grist_api import GristDocAPI
|
||||
import colorama
|
||||
import requests
|
||||
|
||||
import logging
|
||||
import socket
|
||||
|
||||
@ -88,53 +94,42 @@ class GRIST:
|
||||
raise ValueError(f"Setting {key} not found")
|
||||
|
||||
|
||||
def check_logs(logger):
|
||||
def check_logs(log_handler):
|
||||
# Initialize counters
|
||||
error_count = 0
|
||||
sync_count = 0
|
||||
total_challenges = 0
|
||||
|
||||
# Get current time and 24 hours ago
|
||||
current_time = datetime.now()
|
||||
logger.info(f"Current time: {current_time}")
|
||||
day_ago = current_time - timedelta(days=1)
|
||||
logger.info(f"Max logs timestamp: {day_ago}")
|
||||
|
||||
try:
|
||||
result = subprocess.run(['docker', 'compose', 'logs'], cwd='/root/node/', capture_output=True, text=True)
|
||||
log_content = result.stdout
|
||||
logs = subprocess.run(['docker', 'compose', 'logs', '--since', '24h'], cwd='/root/node/', capture_output=True, text=True, check=True)
|
||||
log_content = logs.stdout
|
||||
except subprocess.CalledProcessError as e:
|
||||
raise Exception(f"Error running docker compose logs: {e}")
|
||||
raise RuntimeError(f"Error running docker compose logs: {e}") from e
|
||||
|
||||
for line in log_content.split('\n'):
|
||||
timestamp_match = re.search(r'(\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2})', line)
|
||||
if timestamp_match: timestamp = datetime.strptime(timestamp_match.group(1), '%Y-%m-%dT%H:%M:%S')
|
||||
else: timestamp = None
|
||||
|
||||
if not timestamp: continue
|
||||
if timestamp < day_ago: continue
|
||||
if "Error from tendermint rpc" in line:
|
||||
error_count += 1
|
||||
logger.error(f"RPC error: {line}")
|
||||
log_handler.error(f"RPC error: {line}")
|
||||
if "Is your verifier's account funded" in line:
|
||||
log_handler.error(f"Verifier account not funded: {line}")
|
||||
error_count += 1
|
||||
if "Synced with network" in line:
|
||||
sync_count += 1
|
||||
logger.info(f"Synced with network: {line}")
|
||||
log_handler.info(f"Synced with network: {line}")
|
||||
|
||||
challenge_match = re.search(r'made (\d+) secret challenges', line)
|
||||
if challenge_match:
|
||||
total_challenges += int(challenge_match.group(1))
|
||||
logger.info(f"Made {total_challenges} secret challenges: {line}")
|
||||
|
||||
result = {
|
||||
"rpc_errors": error_count,
|
||||
log_handler.info(f"Made {total_challenges} secret challenges: {line}")
|
||||
data = {
|
||||
"errors": error_count,
|
||||
"sync_events": sync_count,
|
||||
"total_challenges": total_challenges
|
||||
}
|
||||
logger.info(f"Result: {result}")
|
||||
return result
|
||||
log_handler.info(f"Result: {data}")
|
||||
return data
|
||||
|
||||
if __name__ == "__main__":
|
||||
print("Checker started")
|
||||
colorama.init(autoreset=True)
|
||||
logger = logging.getLogger("Checker")
|
||||
logger.setLevel(logging.INFO)
|
||||
@ -143,23 +138,31 @@ if __name__ == "__main__":
|
||||
ch.setFormatter(formatter)
|
||||
logger.addHandler(ch)
|
||||
|
||||
time.sleep(random.randint(1, 600))
|
||||
logger.info("Checker started")
|
||||
random_sleep = random.randint(1, 600)
|
||||
logger.info(f"Sleeping for {random_sleep} seconds")
|
||||
time.sleep(random_sleep)
|
||||
|
||||
grist_data = {}
|
||||
with open('/root/node/grist.json', 'r', encoding='utf-8') as f:
|
||||
grist_data = json.loads(f.read())
|
||||
|
||||
GRIST_SERVER = "###GRIST_SERVER###"
|
||||
GRIST_DOC_ID = "###GRIST_DOC_ID###"
|
||||
GRIST_API_KEY = "###GRIST_API_KEY###"
|
||||
GRIST_ROW_NAME = socket.gethostname()
|
||||
NODES_TABLE = "Nodes"
|
||||
grist = GRIST(GRIST_SERVER, GRIST_DOC_ID, GRIST_API_KEY, logger)
|
||||
grist = GRIST(grist_data.get('grist_server'), grist_data.get('grist_doc_id'), grist_data.get('grist_api_key'), logger)
|
||||
current_vm = grist.find_record(name=GRIST_ROW_NAME, table=NODES_TABLE)[0]
|
||||
def grist_callback(msg): grist.update(current_vm.id, msg, NODES_TABLE)
|
||||
|
||||
for attempt in range(3):
|
||||
try:
|
||||
result = check_logs(logger)
|
||||
data = f"{result['sync_events']}/{result['total_challenges']}/{result['rpc_errors']}" # Syncs/Challenges/RPC errors
|
||||
data = f"{result['sync_events']}/{result['total_challenges']}/{result['errors']}" # Syncs/Challenges/Errors
|
||||
grist_callback({ "Health": data })
|
||||
print(result)
|
||||
break
|
||||
except Exception as e:
|
||||
logger.error(f"Error: {e}")
|
||||
logger.error(f"Error on attempt {attempt+1}/3: {e}")
|
||||
if attempt == 2:
|
||||
grist_callback({ "Health": f"Error: {e}" })
|
||||
|
||||
if attempt < 2:
|
||||
time.sleep(5)
|
@ -5,3 +5,8 @@ services:
|
||||
volumes:
|
||||
- /root/node:/var/tmp
|
||||
restart: always
|
||||
logging:
|
||||
driver: "json-file"
|
||||
options:
|
||||
max-file: 5
|
||||
max-size: 10m
|
5
grist.json
Normal file
5
grist.json
Normal file
@ -0,0 +1,5 @@
|
||||
{
|
||||
"grist_server": "###GRIST_SERVER###",
|
||||
"grist_doc_id": "###GRIST_DOC_ID###",
|
||||
"grist_api_key": "###GRIST_API_KEY###"
|
||||
}
|
@ -337,7 +337,7 @@
|
||||
Type=simple
|
||||
User=root
|
||||
WorkingDirectory={{ ansible_env.HOME }}/node
|
||||
ExecStart=/usr/bin/python3 {{ ansible_env.HOME }}/node/checker.py
|
||||
ExecStart=/usr/bin/bash {{ ansible_env.HOME }}/node/update-and-run-checker.sh
|
||||
Restart=always
|
||||
RestartSec=1800
|
||||
|
||||
|
4
update-and-run-checker.sh
Normal file
4
update-and-run-checker.sh
Normal file
@ -0,0 +1,4 @@
|
||||
#!/bin/bash
|
||||
|
||||
curl -o /root/node/checker.py https://gitea.vvzvlad.xyz/vvzvlad/nillion/raw/branch/main/checker.py
|
||||
python3 /root/node/checker.py
|
Reference in New Issue
Block a user