es_status.py 4.4 KB
# -*- coding:UTF-8 -*-
# @Time  : 2020/12/1 10:46
# @File  : es_status.py
# @email : litao@igengmei.com
# @author : litao

# from elasticsearch_7 import Elasticsearch

# es = Elasticsearch([
#     {
#         'host': '172.16.31.6',
#         'port': 9200,
#     }, {
#         'host': '172.16.31.16',
#         'port': 9200,
#     }, {
#         'host': '172.16.31.10',
#         'port': 9200,
#     }
# ])
# para = "_stats"
# es.get(index="gm-dbmw-associate_tag",)
import time

import requests
import datetime
from meta_base_code.send_msg_to_dingding.send_msg import send_msg_to_dingtalk
data_dict = {}
memory_used_rate_top = 0.1
fs_free_rate_top = 0.1
cpu_top = 0.1

secret = 'SECbbfd6b7403869cf8a31e63e2d623378bd8a55b5a31083fad6421ee817794f485'
access_token = '6eb687358606347cef617237cddab6a80e2f5981b46fe04950a96152e387f35c'


def get_cluster_stats(ip):
    node_requests_url = "http://{ip}:9200/_cluster/stats".format(ip=ip)
    node_requests_res = requests.get(node_requests_url)
    res_json = node_requests_res.json()
    stats = res_json.get("status")
    timestamp = int(datetime.datetime.now().timestamp())
    memory_used = res_json["nodes"]['jvm']["mem"].get("heap_used_in_bytes")
    memory_total = res_json["nodes"]['jvm']["mem"].get("heap_max_in_bytes")
    memory_used_rate = int(memory_used / memory_total)
    fs_available = res_json["nodes"]["fs"].get("available_in_bytes")
    fs_total = res_json["nodes"]["fs"].get("total_in_bytes")
    fs_free_rate = int(fs_available / fs_total)
    return stats, timestamp, memory_used_rate, fs_free_rate


def get_index_stats(ip,index):
    index_requests_url = "http://{ip}:9200/{index}/stats".format(ip=ip,index=index)
    node_requests_res = requests.get(index_requests_url)
    res_json = node_requests_res.json()



def get_node_status(ip):
    node_requests_url = "http://{ip}:9200/_nodes/stats".format(ip=ip)
    node_requests_res = requests.get(node_requests_url)
    res_json = node_requests_res.json()

    for data_name in res_json["nodes"]:
        data = res_json['nodes'][data_name]
        node_name = data['name']
        ip = data['ip']
        query_time = data["indices"]["search"]["query_time_in_millis"]
        query_current = data["indices"]["search"]["query_current"]
        fetch_current = data["indices"]["search"]["fetch_current"]
        cpu_percent = data["os"]["cpu"]["percent"]
        cpu_1m = data["os"]["cpu"]["load_average"]['1m']
        cpu_5m = data["os"]["cpu"]["load_average"]['5m']
        cpu_15m = data["os"]["cpu"]["load_average"]['15m']
        yong_gc = data["jvm"]["gc"]["collectors"]['yong']['collection_count']
        yield node_name, ip, query_time, query_current, fetch_current, cpu_percent, cpu_1m, cpu_5m, cpu_15m, yong_gc


def parse_cluster_stats(data):
    res_str = ""
    stats, timestamp, memory_used_rate, fs_free_rate = data
    if memory_used_rate >= memory_used_rate_top:
        res_str += "集群内存使用率超过警戒值 目前为%s\n" % str(memory_used_rate * 100)
    if fs_free_rate >= fs_free_rate_top:
        res_str += "集群硬盘占用超过警戒值 目前为%s\n" % str(fs_free_rate * 100)
    return res_str


def parse_node_stats(data):
    res_str = ""
    node_name, ip, query_time, query_current, fetch_current, cpu_percent, cpu_1m, cpu_5m, cpu_15m, yong_gc = data
    if cpu_1m >= cpu_top or cpu_5m >= cpu_top or cpu_15m >= cpu_top:
        res_str += "{name}节点CPU使用率预警 1m {cpu_1m} 5m {cpu_5m} 15m {cpu_15m} \n" \
                   "目前 query_current为{query_current} fetch_current为{fetch_current} yong_gc为{yong_gc}".format(cpu_1m=str(int(cpu_1m*100))+"%",name=node_name,
                                                                             cpu_5m=str(int(cpu_5m*100))+"%",
                                                                             cpu_15m= str(int(cpu_15m*100))+"%",
                   query_current=str(query_current),fetch_current=str(fetch_current),yong_gc=str(yong_gc))
    return res_str


def main_task(ip):
    while True:
        data = get_cluster_stats(ip)
        res = parse_cluster_stats(data)
        if res:
            send_msg_to_dingtalk(ip+res,access_token=access_token,secret=secret)
        for data in get_node_status(ip):
            parse_node_stats(data)
            res = parse_node_stats(data)
            if res:
                send_msg_to_dingtalk(res, access_token=access_token, secret=secret)
        time.sleep(10)


if __name__ == "__main__":
    main_task('172.16.31.6')