initial commit

app.conf

+[kairos]
+IP=192.168.0.220
+
+[type]
+server=False
+camera=False
+
+[infra_alert_manager]
+ram_usage=False
+disk_usage=False
+service_status=True
+docker_container_status=True
+
+[service]
+service_list=docker
+include_active_status=False
+include_restart_status = True
+restart_max_count=3
+use_cloud_mail_gateway=True
+is_redis_as_docker=False
+env_name=DEV
+url= DEV
+IP=192.168.0.220
+
+[mqtt]
+mqtt_enable=True
+mqtt_broker=test.mosquitto.org
+mqtt_port=1883
+mqtt_topic=newtopic
+
+[ram]
+ram_threshold_value=90
+include_ram=False
+
+[disk]
+filesystem=""
+threshold_value=80
+include_all_filesystem=False
+
+[kafka]
+meta_file_path = /tmp/kafka-logs/
+meta_file_name = /tmp/kafka-logs/meta.properties
+IP=192.168.0.220
+port=9092
+
+
+[user_info]
+subject=Infra Alert
+sender_email = devopsilens@gmail.com
+receiver_email = jithu.tagore@knowledgelens.com
+password = aWxlbnNAMjAyMQ==
+
+[telegram]
+enable_alerts =False
+bot_token=5701133680:AAGeOaJ2F0Mi0AiBsamtjyZ03T4boEdVKa4
+bot_chatID=-1001656619117
+
+

app_weekly.conf

+[kairos]
+IP=192.168.0.220
+
+[type]
+server=False
+camera=False
+
+[infra_alert_manager]
+ram_usage=False
+disk_usage=False
+service_status=True
+docker_container_status=True
+
+[service]
+service_list=docker
+include_active_status=False
+include_restart_status = True
+restart_max_count=3
+use_cloud_mail_gateway=True
+is_redis_as_docker=False
+env_name=DEV
+url= DEV
+IP=192.168.0.220
+
+[mqtt]
+mqtt_enable=True
+mqtt_broker=test.mosquitto.org
+mqtt_port=1883
+mqtt_topic=newtopic
+
+[ram]
+ram_threshold_value=90
+include_ram=False
+
+[disk]
+filesystem=""
+threshold_value=80
+include_all_filesystem=False
+
+[kafka]
+meta_file_path = /tmp/kafka-logs/
+meta_file_name = /tmp/kafka-logs/meta.properties
+IP=192.168.0.220
+port=9092
+
+
+[user_info]
+subject=Infra Alert
+sender_email = devopsilens@gmail.com
+receiver_email = jithu.tagore@knowledgelens.com
+password = aWxlbnNAMjAyMQ==
+
+[telegram]
+enable_alerts =False
+bot_token=5701133680:AAGeOaJ2F0Mi0AiBsamtjyZ03T4boEdVKa4
+bot_chatID=-1001656619117
+
+

disk-usage.sh

+#!/bin/bash
+THRESHOLD_VALUE_ALERT=$1
+IFS=', ' read -r -a FILE_SYSTEMS <<< "$2"
+SHOW_ALL_FILE_SYSTEMS=$3
+disk_usage() {
+  if [ "$FILE_SYSTEMS" == '""' ]; then
+  partition_name=($(df -H | grep -vE 'overlay|shm' | awk '{ print $1 }' | awk '{if(NR>1)print}'))
+  used_storage_percentage=($(df -H | grep -vE 'overlay|shm' | awk '{ print $5 }'  | cut -d'%' -f1 | awk '{if(NR>1)print}'))
+  else
+  printf -v file_system_string '|%s' "${FILE_SYSTEMS[@]}"
+  file_system_string="${file_system_string:1}"
+  partition_name=($(df -H | grep -vE 'overlay|shm' | awk '{ print $1 }' |  grep -E ${file_system_string}))
+  used_storage_percentage=($(df -H | grep -vE 'overlay|shm' | awk '{ print $1,$5 }' | grep -E ${file_system_string} | awk '{ print $2 }' | cut -d'%' -f1))
+  fi
+  exceeded_storage_names=()
+  if [ "$SHOW_ALL_FILE_SYSTEMS" == True ]; then
+  exceeded_storage_names=( "${partition_name[@]}" )
+  else
+  for i in "${!used_storage_percentage[@]}"; do
+    if [ ${used_storage_percentage[$i]} -ge $THRESHOLD_VALUE_ALERT ]; then
+      exceeded_storage_names+=(${partition_name[$i]})
+    fi
+  done
+  fi
+  printf -v partition_name_string '|%s' "${exceeded_storage_names[@]}"
+  partition_name_string="${partition_name_string:1}"
+  df -H | grep -vE 'overlay|shm' | grep -E ${partition_name_string} >>usage.txt
+}
+
+touch usage.txt
+disk_usage
+exit 0

kafka-script/README.md

+**command:** python kafka_restart_script.py -p=9092 -ip=127.0.0.1 -S=kafka
+
+Update `meta_file_path` and `meta_file_name` variable in the kafka_restart_script.py
+
+Value for the above variables can be obtained in `/home/kafka/kafka/config/server.properties` file of that env, open the file and search for `log.dirs`, the value of `log.dirs` should be mentiones as `meta_file_path` 

kafka-script/kafka_restart_script.py

+import argparse
+import os
+import sys
+import time
+from datetime import datetime
+
+ap = argparse.ArgumentParser()
+meta_file_path = "/tmp/kafka-logs/"
+meta_file_name = "/tmp/kafka-logs/meta.properties"
+
+
+def kafka_specific_fun():
+    try:
+        date = datetime.now().strftime("%Y-%m-%d-%I:%M:%S-%p")
+        new_file_path = os.path.join(meta_file_path, f'meta.properties_{date}')
+        os.rename(meta_file_name, new_file_path)
+    except Exception as e:
+        print(f"Exception Occurred while renaming the meta.properties file {e.args}")
+
+
+def _restart_and_check_status_service(service_name):
+    try:
+        kafka_service_status = True if os.system(
+            f"sudo service {service_name} status") is 0 else False
+        print(kafka_service_status)
+        if not kafka_service_status:
+            os.system(f"sudo service {service_name} restart")
+            time.sleep(6)
+            kafka_service_status = True if os.system(
+                f"sudo service {service_name} status") is 0 else False
+        return kafka_service_status
+    except Exception as e:
+        print(f"Failed to restart the system service {service_name} as {e.args}")
+        return False
+
+
+if __name__ == "__main__":
+    ap.add_argument(
+        "--host",
+        "-ip",
+        required=False,
+        help="Check the Host Whether it is accessible or not...",
+    )
+    ap.add_argument(
+        "--service_name",
+        "-S",
+        required=False,
+        help="Checks the Service Status...",
+    )
+    ap.add_argument(
+        "--port",
+        "-p",
+        required=False,
+        help="Checks the Service Status...",
+    )
+
+    arguments = vars(ap.parse_args())
+
+    print(f"Given Host ---> {arguments['host']}")
+    print(f"Given service details ---> {arguments['service_name']}")
+    print(f"Given port details ---> {arguments.get('port')}")
+    # sys.exit()
+    HOST_UP = True if os.system("ping -c 1 " + f'{arguments["host"]}') == 0 else False
+
+    print(HOST_UP)
+    _service_status = False
+    port_number = arguments.get('port')
+    if HOST_UP:
+        for _ in range(0, 2):
+            _service_status = _restart_and_check_status_service(service_name=arguments['service_name'])
+            if _service_status:
+                _port_status = True if os.system(f"sudo netstat -nlp | grep :{port_number}") == 0 else False
+                if _port_status:
+                    sys.exit()
+                else:
+                    os.system(f"sudo service {arguments['service_name']} restart")
+                    _service_status = _restart_and_check_status_service(service_name=arguments['service_name'])
+        print(_service_status)
+        if not _service_status:
+            _port_status = True if os.system(f"sudo netstat -nlp | grep :{port_number}") == 0 else False
+            if _port_status:
+                os.system(f'kill -9 $(lsof -t -i:{port_number})')
+            elif arguments['service_name'] in {"kafka"} and not _port_status:
+                print("Changing meta.properties file name")
+                kafka_specific_fun()
+            _restart_and_check_status_service(service_name=arguments['service_name'])

ram_uasge.sh

+#!/bin/bash
+touch Ram_uasge.txt
+THRESHOLD=$1
+INCLUDE_RAM=$2
+RAM_TOTAL=$(free -g | grep Mem | awk '{print $2}')
+RAM_AVAIL=$(free -g | grep Mem | awk '{print $7}')
+RAM_USED=$(free -g | grep Mem | awk '{print $3}')
+THRESHOLD_RAM=$(echo $(( $RAM_TOTAL*$THRESHOLD/100 )))
+if [ "$INCLUDE_RAM" == True ]; then
+	free -g | awk '{ print $1 " " $2 " " $3 " " $4 " " $5 " " $6 }' | awk '{if(NR>1)print}' | grep Mem | awk '{ print $2 " " $3 "["int($3/$2*100)"%" "] " $4 "["int($4/$2*100)"%" "] " $5 " " $6}'  >Ram_uasge.txt
+else
+	if [ $RAM_USED -ge $THRESHOLD_RAM ]; then
+		free -g | awk '{ print $1 " " $2 " " $3 " " $4 " " $5 " " $6 }' | awk '{if(NR>1)print}' | grep Mem | awk '{ print $2 " " $3 "["int($3/$2*100)"%" "] " $4 "["int($4/$2*100)"%" "] " $5 " " $6}' >Ram_uasge.txt 
+	fi
+fi

requirements.txt

+prettytable==2.1.0
+requests==2.25.1
+psutil==5.8.0
+configparser==5.0.2
+python-telegram-bot
+paho-mqtt
\ No newline at end of file