Flink任务优化分享

Flink任务优化分享

1.背景介绍

线上计算任务在某版本上线之后发现每日的任务时长都需要三个多小时才能完成,计算时间超过了预估时间,通过Dolphinscheduler的每日调度任务看,在数据层 dwd 的数据分段任务存在严重的性能问题,每天的计算耗时将近40分钟,并且随着数据量的上涨,时间越来越长,因此这个计算节点需要着重优化。

2.改进思路及实施

现在的大数据计算任务是用 flink 执行的,因此优化的入手点就是从 Flink History Server 上看任务的执行计划,找到耗时较多的节点以及是否有节点因为sql逻辑被重复执行,导致耗时较高。

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-GwHDHsdd-1690266698352)(/Users/apple/Documents/yangxin/develop/image-20230413173034642.png)]

如图所示,可以发现计算任务走了三个分叉,从sql最后的输出来看,只有两个insert表操作,所以这里至少有一条分叉是不必要的;然后就是找到分叉点的原因,为什么会导致任务分成了三个分支,这个就需要执行计划慢慢去理,界面上可以点开每个节点看到他的执行计算优化之后的结果,然后来判断这一节点对应了sql的哪一步。着重需要判断的就是产生分支的那个节点

Sort(orderBy=[tenant_id ASC, room_id ASC, msg_start_time ASC]
) -> 
Calc(select=[__etl_time__, date_id, tenant_id, brand_id, channel, channel_app_id, channel_session_type, msg_id, msg_start_time, msg_end_time, msg_from_id, msg_from_orig_id, msg_from_nk, msg_from_role, msg_to_ids, msg_to_users, msg_type, msg_content, msg_detail, group_chat_info, dialogue_id, room_id, operation_flags, recording_properties, asr_properties, metric_properties, tags, tag_properties, dialogue_properties, lastMsgEndTime, nextMsgStartTime, is_cut_by_msg_time, is_fit_specific_event, pre_is_fit_specific_event, fit_specific_row,CAST(FROM_UNIXTIME(w0$o0)) AS start_time,CAST(FROM_UNIXTIME(w0$o1)) AS end_time,CAST(w1$o0) AS fit_specific_rows,GenIsFitConsecutiveRowsAndTime(channel_session_type, tenant_id, CAST(w1$o0), CAST(CAST(FROM_UNIXTIME(w0$o0))), CAST(CAST(FROM_UNIXTIME(w0$o1))), is_fit_specific_event) AS is_fit_specific_flag,(is_cut_by_msg_time = _UTF-16LE'1':VARCHAR(2147483647) CHARACTER SET "UTF-16LE") AS $39, // (GenIsFitConsecutiveRowsAndTime(channel_session_type, tenant_id, CAST(w1$o0), CAST(CAST(FROM_UNIXTIME(w0$o0))), CAST(CAST(FROM_UNIXTIME(w0$o1))), is_fit_specific_event) = 1) AS $40]
) -> 
OverAggregate(partitionBy=[tenant_id, room_id],orderBy=[msg_start_time ASC],window#0=[LAG(is_fit_specific_flag) AS w0$o0RANG BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW],select=[__etl_time__, date_id, tenant_id, brand_id, channel, channel_app_id, channel_session_type, msg_id, msg_start_time, msg_end_time, msg_from_id, msg_from_orig_id, msg_from_nk, msg_from_role, msg_to_ids, msg_to_users, msg_type, msg_content, msg_detail, group_chat_info, dialogue_id, room_id, operation_flags, recording_properties, asr_properties, metric_properties, tags, tag_properties, dialogue_properties, lastMsgEndTime, nextMsgStartTime, is_cut_by_msg_time, is_fit_specific_event, pre_is_fit_specific_event, fit_specific_row,start_time, end_time, fit_specific_rows, is_fit_specific_flag, $39,    // 是否按时间切 $40,    // 当前条是否为 1w0$o0 -> pre_is_fit_specific_flag // 前一条是否满足特殊规则]
) -> (Calc(select=[date_id, tenant_id, channel_session_type, msg_id, msg_start_time, room_id, tags,IF(($39 OR (w0$o0 IS NULL AND $40) OR ((w0$o0 <> is_fit_specific_flag) IS TRUE AND w0$o0 IS NOT NULL)), 1, 0) AS is_cut_flag,CAST(tenant_id) AS $8,CAST(msg_start_time) AS $9,GenCutPointTypeByFeature(channel_session_type, tenant_id, tags) AS $10]) -> OverAggregate(partitionBy=[tenant_id, room_id],orderBy=[msg_start_time ASC],window#0=[COUNT(is_cut_flag) AS w0$o0,$SUM0(is_cut_flag) AS w0$o1RANG BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW],select=[date_id, tenant_id, channel_session_type, msg_id, msg_start_time, room_id, tags, is_cut_flag, $8, -> tenant_id$9, -> msg_start_time$10, -> 特征切分点: START/END/NOw0$o0, -> countw0$o1 -> sum]) -> Calc(select=[CONCAT_WS(_UTF-16LE'-', $8, room_id, date_id, CAST(CASE((w0$o0 > 0:BIGINT), w0$o1, null:INTEGER))) AS dialogue_id1,channel_session_type, tenant_id, msg_id, $9 AS $f4, tags, $10 AS cutPointType]), #####################CREATE TEMPORARY VIEW keep_cutpoint_view ASSELECT dialogue_id1, smoothRes.smoothResultVoMapFROM (SELECT dialogue_id1,smoothCutPoint(channel_session_type, tenant_id, dialogue_id1, msg_id, msg_start_time, tags, cutPointType) AS smoothResFROM gen_cut_type_by_feature_viewGROUP BY dialogue_id1);#####################Calc(select=[__etl_time__, date_id, tenant_id, brand_id, channel, channel_app_id, channel_session_type, msg_id, msg_start_time, msg_end_time, msg_from_id, msg_from_orig_id, msg_from_nk, msg_from_role, msg_to_ids, msg_to_users, msg_type, msg_content, msg_detail, group_chat_info, dialogue_id, room_id, operation_flags, recording_properties, asr_properties, metric_properties, tags, tag_properties, dialogue_properties, lastMsgEndTime, nextMsgStartTime, is_cut_by_msg_time, is_fit_specific_event, pre_is_fit_specific_event, fit_specific_row, start_time, end_time, fit_specific_rows, is_fit_specific_flag, w0$o0 AS pre_is_fit_specific_flag, CASE(w0$o0 IS NULL, is_fit_specific_flag, (w0$o0 <> is_fit_specific_flag), 1, 0) AS is_cut_by_specific, IF(($39 OR (w0$o0 IS NULL AND $40) OR ((w0$o0 <> is_fit_specific_flag) IS TRUE AND w0$o0 IS NOT NULL)), 1, 0) AS is_cut_flag, IF((IF(($39 OR (w0$o0 IS NULL AND $40) OR ((w0$o0 <> is_fit_specific_flag) IS TRUE AND w0$o0 IS NOT NULL)), 1, 0) = 1), _UTF-16LE'start', null:VARCHAR(2147483647) CHARACTER SET "UTF-16LE") AS $42,CAST(tenant_id) AS $43, GenCutPointTypeByFeature(channel_session_type, tenant_id, tags) AS $44]) -> OverAggregate(partitionBy=[tenant_id, room_id],orderBy=[msg_start_time ASC],window#0=[COUNT(is_cut_flag) AS w0$o0,$SUM0(is_cut_flag) AS w0$o1RANG BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW], select=[__etl_time__, date_id, tenant_id, brand_id, channel, channel_app_id, channel_session_type, msg_id, msg_start_time, msg_end_time, msg_from_id, msg_from_orig_id, msg_from_nk, msg_from_role, msg_to_ids, msg_to_users, msg_type, msg_content, msg_detail, group_chat_info, dialogue_id, room_id, operation_flags, recording_properties, asr_properties, metric_properties, tags, tag_properties, dialogue_properties, lastMsgEndTime, nextMsgStartTime, is_cut_by_msg_time, is_fit_specific_event, pre_is_fit_specific_event, fit_specific_row, start_time, end_time, fit_specific_rows, is_fit_specific_flag, pre_is_fit_specific_flag, is_cut_by_specific, is_cut_flag, $42, -> cut_point_type$43, $44, w0$o0, w0$o1]) -> Calc(select=[date_id, tenant_id, brand_id, channel, channel_app_id, channel_session_type, msg_id, msg_start_time, msg_end_time, msg_from_id, msg_from_orig_id, msg_from_nk, msg_from_role, msg_to_ids, msg_to_users, msg_type, msg_content, msg_detail, group_chat_info, room_id, operation_flags, recording_properties, asr_properties, metric_properties, tags, tag_properties, dialogue_properties, CONCAT_WS(_UTF-16LE'-', $43, room_id, date_id, CAST(CASE((w0$o0 > 0:BIGINT), w0$o1, null:INTEGER))) AS dialogue_id1, $44 AS cutPointType])
)######################
根据时间+特殊规则先生成一轮 dialogu_id1
特征的切分点命中被提前下推到这个阶段执行BUG&优化点:
1. 平滑使用了 groupBy 再 join 会主表, 导致计算流走了分支, 导致部分计算逻辑重复执行了, 这一部分可以考虑用 over 聚合来做gen_fit_sprcific_flag_view -> gen_cut_flag_view -> gen_dialogue_id_by_cut_flag_view -> gen_cut_type_by_feature_view -> keep_cutpoint_view2. CASE WHEN pre_is_fit_specific_flag IS NULL THEN is_fit_specific_flagWHEN pre_is_fit_specific_flag <> is_fit_specific_flag THEN 1WHEN pre_is_fit_specific_flag = is_fit_specific_flag THEN 0ELSE 0END AS is_cut_by_specific逻辑不对, 导致 GenIsFitConsecutiveRowsAndTime 被重复执行
3. IF(is_cut_flag = 1, 'start', CAST(NULL AS STRING)) AS cut_point_type, 需要判断一下是否还有必要5. 
======================
Sort(orderBy=[dialogue_id1 ASC]
) -> 
SortAggregate(isMerge=[false], groupBy=[dialogue_id1], select=[dialogue_id1, smoothCutPoint(channel_session_type, tenant_id, dialogue_id1, msg_id, $f4, tags, cutPointType) AS smoothRes]
) -> 
Calc(select=[dialogue_id1, smoothRes.smoothResultVoMap AS smoothResultVoMap]
) -> (Correlate(invocation=[GetCutPointBySplit($cor7.smoothResultVoMap)],correlate=[table(GetCutPointBySplit($cor7.smoothResultVoMap))],select=[dialogue_id1,smoothResultVoMap,msgId,cutPointMap],rowType=[RecordType(VARCHAR(2147483647) dialogue_id1, (VARCHAR(2147483647), (VARCHAR(2147483647), VARCHAR(2147483647)) MAP) MAP smoothResultVoMap,VARCHAR(2147483647) msgId,(VARCHAR(2147483647), VARCHAR(2147483647)) MAP cutPointMap)], joinType=[INNER]) -> Calc(select=[dialogue_id1, msgId, ITEM(cutPointMap, _UTF-16LE'isKeep') AS isKeep]),Correlate(invocation=[GetCutPointBySplit($cor9.smoothResultVoMap)],correlate=[table(GetCutPointBySplit($cor9.smoothResultVoMap))],select=[dialogue_id1,smoothResultVoMap,msgId,cutPointMap], rowType=[RecordType(VARCHAR(2147483647) dialogue_id1, (VARCHAR(2147483647), (VARCHAR(2147483647), VARCHAR(2147483647)) MAP) MAP smoothResultVoMap, VARCHAR(2147483647) msgId, (VARCHAR(2147483647), VARCHAR(2147483647)) MAP cutPointMap)], joinType=[INNER]) -> Calc(select=[dialogue_id1, msgId, ITEM(cutPointMap, _UTF-16LE'isKeep') AS isKeep]), Correlate(invocation=[GetCutPointBySplit($cor8.smoothResultVoMap)], correlate=[table(GetCutPointBySplit($cor8.smoothResultVoMap))], select=[dialogue_id1,smoothResultVoMap,msgId,cutPointMap], rowType=[RecordType(VARCHAR(2147483647) dialogue_id1, (VARCHAR(2147483647), (VARCHAR(2147483647), VARCHAR(2147483647)) MAP) MAP smoothResultVoMap, VARCHAR(2147483647) msgId, (VARCHAR(2147483647), VARCHAR(2147483647)) MAP cutPointMap)], joinType=[INNER]) ->Calc(select=[dialogue_id1, msgId, ITEM(cutPointMap, _UTF-16LE'isKeep') AS isKeep])
)######################
计算平滑逻辑优化点:
1. smoothCutPoint 很大的性能问题, 改成基于 over 聚合的 udaf, 优化掉 GROUPBY + LATERAL TABLE + JOIN

对应sql如下:

--根据配置文件特征数据对数据进行特征切分标记
CREATE TEMPORARY VIEW gen_cut_type_by_feature_view AS
SELECT *,GenCutPointTypeByFeature(channel_session_type, tenant_id, tags) AS cutPointType
FROM gen_dialogue_id_by_cut_flag_view;CREATE TEMPORARY VIEW keep_cutpoint_view AS
SELECT dialogue_id1, smoothRes.smoothResultVoMap
FROM (SELECT dialogue_id1,smoothCutPoint(channel_session_type, tenant_id, dialogue_id1, msg_id, msg_start_time, tags, cutPointType) AS smoothResFROM gen_cut_type_by_feature_viewGROUP BY dialogue_id1
);CREATE TEMPORARY VIEW keep_cutpoint_breakup AS
SELECT dialogue_id1, smoothResultVoMap, msgId, cutPointMap, cutPointMap['isKeep'] AS isKeep
FROM keep_cutpoint_view, LATERAL TABLE(GetCutPointBySplit(smoothResultVoMap)) AS T(msgId, cutPointMap);CREATE TEMPORARY VIEW keep_cutpoint_join AS
SELECT t1.*,t2.isKeep, IF(t2.isKeep = '0', 'no', t1.cutPointType) AS curCutPointType, msg_start_time
FROM gen_cut_type_by_feature_view t1
LEFT JOIN keep_cutpoint_breakup t2 ON t1.dialogue_id1 = t2.dialogue_id1 AND t1.msg_id = t2.msgId;CREATE TEMPORARY VIEW gen_dialogue_id_by_feature_view0 AS
SELECTdate_id,tenant_id,brand_id,channel,channel_app_id,channel_session_type,msg_id,msg_start_time,msg_end_time,msg_from_id,msg_from_nk,msg_from_orig_id,msg_from_role,msg_to_ids,msg_to_users,msg_type,msg_content,msg_detail,group_chat_info,room_id,operation_flags,recording_properties,asr_properties,metric_properties,tags,tag_properties,dialogue_properties,dialogue_id1,cutPointType,curCutPointType,preCutPointType,isKeep,CONCAT_WS('-',CAST(tenant_id AS STRING),room_id,date_id,CAST(SUM(IF(preCutPointType IS NULL OR preCutPointType = 'end' OR curCutPointType = 'start', 1, 0)) OVER (PARTITION BY tenant_id, room_id, date_id ORDER BY msg_start_time)AS STRING)) AS dialogue_id
FROM (SELECTdate_id,tenant_id,brand_id,channel,channel_app_id,channel_session_type,msg_id,msg_start_time,msg_end_time,msg_from_id,msg_from_nk,msg_from_orig_id,msg_from_role,msg_to_ids,msg_to_users,msg_type,msg_content,msg_detail,group_chat_info,room_id,operation_flags,recording_properties,asr_properties,metric_properties,tags,tag_properties,dialogue_properties,dialogue_id1,cutPointType,curCutPointType,isKeep,LAG(curCutPointType) OVER ( PARTITION BY dialogue_id1 ORDER BY msg_start_time) AS preCutPointTypeFROM keep_cutpoint_join);

之前sql对于这个分段平滑逻辑的实现是,先根据idalogue_id group by数据,使用udaf去得到聚合结果,然后在通过msg_id将聚合结果join回原来的明细数据里,这种做法就会产生分岔,不仅性能差,而且会重复执行计算节点导致耗时上升。这种做法在后边的相关性聚合也是差不多的,这样一分析问题就找到了,就是要把聚合结果join回主表这种做法换一种更高效的方式实现,具体改进思路就是将原来这种方式改成基于 over 聚合的 udaf, 优化掉 GROUPBY + LATERAL TABLE + JOIN

优化之后的sql:

--根据配置文件特征数据对数据进行特征切分标记
CREATE TEMPORARY VIEW gen_cut_type_by_feature_view AS
SELECT *,GenCutPointTypeByFeature(channel_session_type, tenant_id, tags) AS cutPointType
FROM gen_dialogue_id_by_cut_flag_view;CREATE TEMPORARY VIEW keep_cutpoint_view AS
SELECT *,smooth_result[msg_id]['is_keep'] AS isKeep,CASE WHEN smooth_result[msg_id]['is_keep'] = '0' THEN 'no' ELSE cutPointType END AS curCutPointType
FROM(SELECT *,smoothCutPoint(channel_session_type, tenant_id, dialogue_id1, msg_id, msg_start_time, tags, cutPointType) OVER ( PARTITION BY dialogue_id1) AS smooth_resultFROM gen_cut_type_by_feature_view);CREATE TEMPORARY VIEW gen_dialogue_id_by_feature_view0 AS
SELECTdate_id,tenant_id,brand_id,channel,channel_app_id,channel_session_type,msg_id,msg_start_time,msg_end_time,msg_from_id,msg_from_nk,msg_from_orig_id,msg_from_role,msg_to_ids,msg_to_users,msg_type,msg_content,msg_detail,group_chat_info,room_id,operation_flags,recording_properties,asr_properties,metric_properties,tags,tag_properties,dialogue_properties,CONCAT_WS('-',CAST(tenant_id AS STRING),room_id,date_id,CAST(SUM( CASE WHEN dialogue_id1 <> preDialogueId OR preCutPointType IS NULL OR preCutPointType = 'end' OR curCutPointType = 'start' THEN 1 ELSE 0 END) OVER (PARTITION BY tenant_id, room_id, date_id ORDER BY msg_start_time)AS STRING)) AS dialogue_id
FROM (SELECTdate_id,tenant_id,brand_id,channel,channel_app_id,channel_session_type,msg_id,msg_start_time,msg_end_time,msg_from_id,msg_from_nk,msg_from_orig_id,msg_from_role,msg_to_ids,msg_to_users,msg_type,msg_content,msg_detail,group_chat_info,room_id,operation_flags,recording_properties,asr_properties,metric_properties,tags,tag_properties,dialogue_properties,dialogue_id1,cutPointType,curCutPointType,LAG(dialogue_id1) OVER ( PARTITION BY tenant_id, room_id, date_id ORDER BY msg_start_time) AS preDialogueId,LAG(curCutPointType) OVER ( PARTITION BY tenant_id, room_id, date_id ORDER BY msg_start_time) AS preCutPointTypeFROM keep_cutpoint_view);

相关性的优化也是一样的思路,改成基于 over 聚合的 udaf,减少聚合结果join回原表的这种操作

相关性sql对比:

CREATE TEMPORARY VIEW dialogue_relevant_view AS
SELECT`tenant_id`,`brand_id`,`channel`,`channel_app_id`,`channel_session_type`,`date_id`,dialogue_id as dialogue_id,res.relevant_config_version as relevant_config_version ,res.relevant_config as relevant_config ,res.metrics as metrics ,res.dialogue_relevant as dialogue_relevant
FROM (select dialogue_relevant_udaf(channel_session_type, tenant_id, msg_id, msg_start_time, msg_end_time, msg_from_role,tags) as res,`tenant_id`,`brand_id`,`channel`,`channel_app_id`,`channel_session_type`,`date_id`,`dialogue_id`from gen_dialogue_id_by_feature_viewgroup by `tenant_id`,`brand_id`,`channel`,`channel_app_id`,`channel_session_type`,`date_id`,`dialogue_id`);CREATE TEMPORARY VIEW dialogue_view_all AS
selectNOW() as `__etl_time__`,a.date_id,a.tenant_id,a.brand_id,a.channel,a.channel_app_id,a.channel_session_type,a.msg_id,a.msg_start_time,a.msg_end_time,a.msg_from_id,a.msg_from_nk,a.msg_from_orig_id,a.msg_from_role,a.msg_to_ids,a.msg_to_users,a.msg_type,a.msg_content,a.msg_detail,a.group_chat_info,a.room_id,a.operation_flags,a.recording_properties,a.asr_properties,a.metric_properties,a.tags,a.tag_properties,map_put(map_put(a.dialogue_properties , 'dialogue_relevant' , b.dialogue_relevant),'relevant_config',b.relevant_config)  as dialogue_properties,a.dialogue_id1,a.cutPointType,a.curCutPointType,a.preCutPointType,a.isKeep,a.dialogue_id
from  gen_dialogue_id_by_feature_view  aleft join dialogue_relevant_view bon  a.tenant_id = b.tenant_id anda.brand_id = b.brand_id anda.channel = b.channel anda.channel_app_id = b.channel_app_id anda.channel_session_type = b.channel_session_type anda.dialogue_id = b.dialogue_id;#####################################CREATE TEMPORARY VIEW dialogue_view AS
selectdate_id,tenant_id,brand_id,channel,channel_app_id,channel_session_type,msg_id,msg_start_time,msg_end_time,msg_from_id,msg_from_nk,msg_from_orig_id,msg_from_role,msg_to_ids,msg_to_users,msg_type,msg_content,msg_detail,group_chat_info,room_id,operation_flags,recording_properties,asr_properties,metric_properties,tags,tag_properties,dialogue_properties,dialogue_id,dialogue_relevant_udaf(channel_session_type, tenant_id, msg_id, msg_start_time, msg_end_time, msg_from_role,tags) OVER (PARTITION BY `tenant_id`,`brand_id`,`channel`,`channel_app_id`,`channel_session_type`,`date_id`,`dialogue_id`) AS res
from gen_dialogue_id_by_feature_view ;

3.优化结果

优化之后的执行计划清爽很多,执行速度也有了明显提升,从原来的将近40分钟的计算时长,减少到7分钟,提升巨大

在这里插入图片描述

本文来自互联网用户投稿,该文观点仅代表作者本人,不代表本站立场。本站仅提供信息存储空间服务,不拥有所有权,不承担相关法律责任。如若转载,请注明出处:http://www.mzph.cn/news/9008.shtml

如若内容造成侵权/违法违规/事实不符,请联系多彩编程网进行投诉反馈email:809451989@qq.com,一经查实,立即删除!

相关文章

【python】flask查询更新指定的某一条记录

PackageRecord.query.filter_by(idpackage_id).update(json_data) 这段代码的问题在于它不能正确地更新指定的记录。这是因为 update() 方法是 SQLAlchemy 提供的一种批量更新的方法&#xff0c;他通过接收一个字典对象来更新记录。但是在你的代码中&#xff0c;json_data 应该…

20230721在WIN10下安装openssl并解密AES-128加密的ts视频切片

20230721在WIN10下安装openssl并解密AES-128加密的ts视频切片 2023/7/21 22:58 1、前言&#xff1a; AES-128加密的ts视频切片【第一个】&#xff0c;打开有时间限制的&#xff01; https://app1ce7glfm1187.h5.xiaoeknow.com/v2/course/alive/l_64af6130e4b03e4b54da1681?typ…

研发机器配网方案(针对禁止外网电脑的组网方案)

背景&#xff1a;公司是研发型小公司&#xff0c;难免会使用A某D和K某l 等国内免费软件&#xff0c;这两个是业界律师函发得最多的软件。最简单的方案是离网使用&#xff0c;但是离网使用比较麻烦的是要进行文件传输&#xff0c;需要使用U盘拷贝&#xff0c;另外研发型企业一般…

ChatGPT:人机交互新境界,AI智能引领未来

一、ChatGPT&#xff1a;智能交流的新标杆 ChatGPT是基于GPT技术的最新版本&#xff0c;拥有深度学习模型的基因。它通过在大量数据上进行预训练&#xff0c;可以理解和生成自然语言&#xff0c;从而实现了与人类更加自然流畅的对话和交流。 二、ChatGPT的技术背景和工作原理 …

【动态规划上分复盘】这是你熟悉的地下城游戏吗?

欢迎 前言一、动态规划五步曲二、地下城游戏题目分析思路&#xff1a;动态规划具体代码如下 总结 前言 本文讲解关于动态规划思路的两道题目。 一、动态规划五步曲 1.确定状态表示&#xff08;确定dp数组的含义&#xff09;2.确定状态转移方程&#xff08;确定dp的递推公式&a…

Docker:Docker-Compose

Docker:Docker-Compose 一、Docker-Compose 介绍1.1 Docker-Compose 概述二、Docker-Compose 安装2.1 Docker Compose 环境安装2.2 下载2.3 安装三、Docker-Compose 使用3.1 YAML 文件格式及编写注意事项3.2 使用 YAML 时需要注意下面事项3.3 Docker-Compose配置常用字段3.4 D…

Python中pyecharts模块

pyecharts模块 官网&#xff1a;pyecharts官网 pyecharts框架画廊 如果想要做出数据可视化效果图, 可以借助pyecharts模块来完成概况 : Echarts 是个由百度开源的数据可视化&#xff0c;凭借着良好的交互性&#xff0c;精巧的图表设计&#xff0c;得到了众多开发者的认可. 而…

java面试整理

一、 JVM部分 JVM内存溢出(一)之排查初体验_少负 | 气节的博客-CSDN博客 JVM内存溢出(二)之双亲委派机制_少负 | 气节的博客-CSDN博客 JVM内存溢出(三)之JVM8内存模型_少负 | 气节的博客-CSDN博客 JVM内存溢出(四)之垃圾回收器_少负 | 气节的博客-CSDN博客 JVM内存溢出(五…

解决 excel 文件头问题 sap 不识别

一个任务需要上传 excel 到 sap 使用 openpyxl 生成的不识别 excel 保存后可以识别 经过分析&#xff0c;可能是文件头的问题&#xff0c;因为看起来没有区别。 excel 的文件头为 [Content_Types].xml openpyxl 生成的文件头为 docProps/app.xml 修改为文件头有点过于繁琐…

巅峰极客2023 hellosql

随便输一个payload&#xff0c;有waf 这题只有两个回显&#xff0c;分别是太酷啦和nonono&#xff0c;不显示报错、登录成功等各种信息&#xff0c;目前只能想到用时间盲注。 抓包fuzz&#xff0c;194都是被过滤的 不止这些&#xff0c;手工测出来if、sleep、benchmark、*、rp…

办公室安全升级,如何保障人身财产安全?

视频监控&#xff0c;一种常见的安全措施&#xff0c;以监视和记录办公室内的活动。这项技术为企业提供了许多优势&#xff0c;包括保障员工和财产安全、帮助调查犯罪事件、提高业务管理效率以及应对突发事件。 因此&#xff0c;在合理范围内应用视频监控&#xff0c;将为企业提…

【算法题】2763. 所有子数组中不平衡数字之和

题目&#xff1a; 一个长度为 n 下标从 0 开始的整数数组 arr 的 不平衡数字 定义为&#xff0c;在 sarr sorted(arr) 数组中&#xff0c;满足以下条件的下标数目&#xff1a; 0 < i < n - 1 &#xff0c;和 sarr[i1] - sarr[i] > 1 这里&#xff0c;sorted(arr) 表…

【Python提取Excel表格中符合条件的数据】

使用Python提取Excel表格中符合条件的数据 在数据处理和分析的过程中&#xff0c;我们经常需要从Excel表格中提取特定条件下的数据。Python的pandas库为我们提供了方便的方法来进行数据查询和过滤。 表格内容如下&#xff1a; 序号xy11.52.823.24.732.13.644.31.954.13.2 我…

go压力测试

压力测试 1.1.1. Go怎么写测试用例 开发程序其中很重要的一点是测试&#xff0c;我们如何保证代码的质量&#xff0c;如何保证每个函数是可运行&#xff0c;运行结果是正确的&#xff0c;又如何保证写出来的代码性能是好的&#xff0c;我们知道单元测试的重点在于发现程序设计…

Oralce数据库 手工重新创建控制文件

控制文件对于Oralce数据库的作用&#xff0c;就好像微软操作系统中注册表的作用一样。控制文件是一个比较小的二进制文件&#xff0c;记录着数据库的结构信息。如果数据库控制文件发生孙华的话&#xff0c;则Oracle将无法正常启动。通常情况下&#xff0c;在创建数据库时会自动…

SAP从放弃到入门系列之批次派生-Batch Derivation-Part1

文章目录 一、概述二、系统配置三、主数据3.1 分类主数据3.2 派生规则设置3.2.1发送物料3.2.2 接收物料 四、 测试数据&#xff08;生产订单&#xff09;五、 最后 Batch Derivation翻译成批次派生&#xff08;衍生&#xff09;或批次继承都是问题不大&#xff0c;继承和派生个…

LeetCode104. 二叉树的最大深度

104. 二叉树的最大深度 文章目录 [104. 二叉树的最大深度](https://leetcode.cn/problems/maximum-depth-of-binary-tree/)一、题目二、题解方法一&#xff1a;递归方法二&#xff1a;迭代 一、题目 给定一个二叉树&#xff0c;找出其最大深度。 二叉树的深度为根节点到最远叶…

机器学习深度学习——线性回归的基本元素

回归用来表示输入输出之间的关系。 用实际例子来解释一下线性回归&#xff1a;根据房屋的面积、房龄来估算房屋价格。为了实现这个预测放假的模型&#xff0c;需要收集一个真实的数据集&#xff0c;该数据集包括了房屋的销售价格、面积和房龄。 在机器学习中&#xff0c;这个数…

lambda匿名函数

问题:什么是lambda函数?它有什么好处?举例说明 解答 含义 在Python中,不通过def来声明函数名字,而是通过lambda关键字来定义的函数称为匿名函数,即函数没有具体的名称,你可以理解为一句话写一个函数 Lambda表达式是Python中一类特殊的定义函数的形式,从语义上讲,它…

(学习笔记-IP)Ping的工作原理

Ping是基于ICMP协议工作的&#xff0c;ICMP报文封装在IP包里面&#xff0c;它工作在网络层&#xff0c;是IP协议的助手。 ICMP包头的类型字段&#xff0c;大致可分为两大类&#xff1a; 一类是用于诊断的查询消息&#xff0c;也就是查询报文类型一类是通知出错原因的错误消息&…