sysom1/sysom_server/sysom_rca/lib/rca_methods.py

# -*- coding: utf-8 -*-
from datetime import date, datetime, timedelta
import json
import time
import os,sys
import re
import traceback
import requests
from sysom_prio_item import sysom_item_prio
from sysom_monitor_item import sysommonit_items,sysommonit_items_obser
from clogger import logger

def sort_sysom_item_prio(sysom_items):
    try:
        sorted_l = []
        prio_dict = {}
        level_dict = {}
        prio_sort_list = []
        level_sort_list = []
        for item in sysom_items:
            prio_dict[item] = sysom_items[item]["priority"]
            level_dict[item] = sysom_items[item]["level"]
        level_sort_list = sorted(level_dict.items(), key=lambda d:d[1], reverse = True)
        prio_sort_list = sorted(prio_dict.items(), key=lambda d:d[1], reverse = False)
        prio_l = []
        for i in prio_sort_list:
            if i[1] not in prio_l:
                prio_l.append(i[1])
        for i in prio_l:
            for j in level_sort_list:
                if prio_dict[j[0]] == i:
                    sorted_l.append(j[0])
        return sorted_l


    except:
        traceback.print_exc()
        pass
    return list(sysom_items.keys())

sysom_item_sorted = sort_sysom_item_prio(sysom_item_prio)

def check_base_anormal(base_values_list):
    try:
        if base_values_list[1:] == base_values_list[:-1]:
            return False
    except:
        traceback.print_exc()
        pass
    return True

def rca_analysis_entry(dtw_retdict):
    retdict = {}
    retdict["summary"] = ""
    retdict["sum_dict"] = {}
    retdict["sum_dict"]["ref_item_sum"] = ""
    retdict["sum_dict"]["ref_item_list"] = ""
    retdict["sum_dict"]["fix_sum"] = ""
    retmsg = ""
    base_item_name = ""

    try:
        if "base_item_name" in dtw_retdict:
            base_item_name = dtw_retdict["base_item_name"]

        retmsg = "该时间点的%s指标异常分析结果：\n"%(base_item_name)

        top_item_l = []
        top_value_l = []
        if len(dtw_retdict["dist"]) > 0:
            for item in dtw_retdict["dist"]:
                if item[1] >= 16 or len(top_item_l) > 40:
                    break
                top_item_l.append(item[0])
                top_value_l.append(item[1])
        related_item_l = []
        if len(top_item_l) > 0:
            for item in top_item_l:
                if item in sysom_item_prio:
                    related_item_l.append(item)
        sorted_related_item_l = []
        for i in sysom_item_sorted:
            if i in related_item_l:
                sorted_related_item_l.append(i)


        retmsg_1 = ""
        items_num = len(sorted_related_item_l)
        if items_num > 0:
            retmsg_1 = "跟"
            cnt = 0
            for i in sorted_related_item_l:
                cnt += 1
                if cnt < items_num:
                    retmsg_1 = "%s%s领域的%s(%s)、"%(retmsg_1,sysom_item_prio[i]["field"],i,sysom_item_prio[i]["meaning"])
                else:
                    retmsg_1 = "%s%s领域的%s(%s)指标抖动有关"%(retmsg_1,sysom_item_prio[i]["field"],i,sysom_item_prio[i]["meaning"])
                    
            retmsg_1 = "%s；\n该指标异常可能的原因是：\n"%(retmsg_1)
            cnt = 0
            for i in sorted_related_item_l:
                if len(sysom_item_prio[i]["cause"]) > 0:
                    cnt += 1
                    retmsg_1 = "%s%s、 %s\n"%(retmsg_1,cnt,sysom_item_prio[i]["cause"])

            retmsg_1 = "%s可用"%(retmsg_1)
            cnt = 0
            for i in sorted_related_item_l:
                if len(sysom_item_prio[i]["tool"]) > 0 and sysom_item_prio[i]["tool"] not in retmsg_1:
                    cnt += 1
                    if cnt > 1:
                        retmsg_1 = "%s、%s"%(retmsg_1,sysom_item_prio[i]["tool"])
                    else:
                        retmsg_1 = "%s%s"%(retmsg_1,sysom_item_prio[i]["tool"])
            retmsg_1 = "%s工具进一步诊断。\n"%(retmsg_1)

        if check_base_anormal(dtw_retdict["base"]["value"]) == False:
            retmsg_1 = "未检测到该指标有异常抖动，请确认是否误报。\n"
            try:
                item_link = ""
                if base_item_name.split("-")[0] in sysommonit_items:
                    item_link = "/monitor/node_monitor"
                if base_item_name.split("-")[0] in sysommonit_items_obser:
                    item_link = "/app_observable/mysql"
                if len(item_link) > 0:
                    retmsg_1 = "%s[指标来源链接](%s)\n"%(retmsg_1,item_link)
            except:
                traceback.print_exc()
                pass

        retmsg = "%s%s"%(retmsg,retmsg_1)

        retdict["sum_dict"]["ref_item_sum"] = retmsg_1

        topn_s = ""
        dist_num = len(dtw_retdict["dist"])
        if dist_num > 0:
            if dist_num > 10:
                topn_s = "前10个"
            else:
                topn_s = "前%s个"%dist_num

        retmsg_2 = "根据抖动曲线匹配出相关的%s指标如下：\n"%topn_s
        cnt = 0
        for i in range(len(dtw_retdict["dist"])):
            retmsg_2 = "%s%s、 %s\n"%(retmsg_2,cnt+1,dtw_retdict["dist"][i][0])
            cnt += 1
            if cnt >= 10:
                break

        retmsg = "%s%s"%(retmsg,retmsg_2)

        retdict["sum_dict"]["ref_item_list"] = retmsg_2

        retmsg = "%s\n排查/修复建议：\n"%retmsg
        retmsg_3 = ""
        if items_num > 0:
            cnt = 0
            for item in sorted_related_item_l:
                if len(sysom_item_prio[item]["fix_advice"]) > 0:
                    cnt += 1
                    retmsg_3 = "%s%s、 %s\n"%(retmsg_3,cnt,sysom_item_prio[item]["fix_advice"])
        retmsg = "%s%s"%(retmsg,retmsg_3)
        retdict["summary"] = retmsg
        retdict["sum_dict"]["fix_sum"] = retmsg_3

    except:
        traceback.print_exc()
        pass
    return retdict