【最佳实践】MongoDB导出导入数据
首先说一下这个3节点MongoDB集群各个维度的数据规模:
1、dataSize: 1.9T
2、storageSize: 600G
3、全量备份-加压缩开关:186G,耗时 8h
4、全量备份-不加压缩开关:1.8T,耗时 4h27m
具体导出的语法比较简单,此处不再赘述,本文重点描述导入的优化过程,最后给出导入的最佳实践。
■ 2023-09-13T20:00 第1次4并发导入测试
mongorestore --port=20000 -uadmin -p'passwd' --authenticationDatabase=admin --numInsertionWorkersPerCollection=4 --bypassDocumentValidation -d likingtest /u01/nfs/xxxxx_mongodb/10.1.1.1/20230913/likingtest >> 10.2.2.2.log 2>&1 &
tail -100f /u01/nfs/xxxxx_mongodb/10.1.1.1/20230913/10.2.2.2.log
以上导入:
2023-09-13T21:59:55.452+0800 The --db and --collection flags are deprecated for this use-case; please use --nsInclude instead, i.e. with --nsInclude=${DATABASE}.${COLLECTION}
2023-09-13T21:59:55.452+0800 building a list of collections to restore from /u01/nfs/xxxxx_mongodb/10.1.1.1/20230913/likingtest dir
2023-09-13T21:59:55.466+0800 reading metadata for likingtest.oprceConfiguration from /u01/nfs/xxxxx_mongodb/10.1.1.1/20230913/likingtest/oprceConfiguration.metadata.json
2023-09-13T21:59:55.478+0800 reading metadata for likingtest.oprceDataObj from /u01/nfs/xxxxx_mongodb/10.1.1.1/20230913/likingtest/oprceDataObj.metadata.json
2023-09-13T21:59:55.491+0800 reading metadata for likingtest.oprcesDataObjInit from /u01/nfs/xxxxx_mongodb/10.1.1.1/20230913/likingtest/oprcesDataObjInit.metadata.json
2023-09-13T21:59:55.503+0800 reading metadata for likingtest.role from /u01/nfs/xxxxx_mongodb/10.1.1.1/20230913/likingtest/role.metadata.json
2023-09-13T21:59:55.508+0800 reading metadata for likingtest.activityConfiguration from /u01/nfs/xxxxx_mongodb/10.1.1.1/20230913/likingtest/activityConfiguration.metadata.json
2023-09-13T21:59:55.511+0800 reading metadata for likingtest.history_task from /u01/nfs/xxxxx_mongodb/10.1.1.1/20230913/likingtest/history_task.metadata.json
2023-09-13T21:59:55.512+0800 reading metadata for likingtest.resOutRelDataSnapshot from /u01/nfs/xxxxx_mongodb/10.1.1.1/20230913/likingtest/resOutRelDataSnapshot.metadata.json
2023-09-13T21:59:55.520+0800 reading metadata for likingtest.snapshotResource from /u01/nfs/xxxxx_mongodb/10.1.1.1/20230913/likingtest/snapshotResource.metadata.json
2023-09-13T21:59:55.524+0800 reading metadata for likingtest.oprceDataObjDraft from /u01/nfs/xxxxx_mongodb/10.1.1.1/20230913/likingtest/oprceDataObjDraft.metadata.json
2023-09-13T21:59:55.526+0800 reading metadata for likingtest.oprceDataObjInit from /u01/nfs/xxxxx_mongodb/10.1.1.1/20230913/likingtest/oprceDataObjInit.metadata.json
2023-09-13T21:59:55.761+0800 restoring likingtest.snapshotResource from /u01/nfs/xxxxx_mongodb/10.1.1.1/20230913/likingtest/snapshotResource.bson
...
2023-09-13T22:00:01.451+0800 [........................] likingtest.oprceDataObj 408MB/1205GB (0.0%)
...
2023-09-13T21:59:58.323+0800 finished restoring likingtest.oprceDataObjDraft (1559 documents, 0 failures)
2023-09-13T22:00:01.034+0800 finished restoring likingtest.resOutRelDataSnapshot (34426 documents, 0 failures)
2023-09-13T22:00:01.559+0800 finished restoring likingtest.history_task (3629 documents, 0 failures)
2023-09-13T22:00:02.086+0800 finished restoring likingtest.activityConfiguration (974 documents, 0 failures)
2023-09-13T22:00:02.293+0800 finished restoring likingtest.oprceConfiguration (162 documents, 0 failures)
2023-09-13T22:00:02.529+0800 finished restoring likingtest.oprcesDataObjInit (4 documents, 0 failures)
2023-09-13T22:00:02.857+0800 finished restoring likingtest.role (10 documents, 0 failures)
2023-09-13T22:00:29.153+0800 [########################] likingtest.snapshotResource 2.04GB/2.04GB (100.0%)
2023-09-13T22:00:29.155+0800 finished restoring likingtest.snapshotResource (50320 documents, 0 failures)
...
2023-09-14T00:18:58.451+0800 [############............] likingtest.oprceDataObj 651GB/1205GB (54.0%)
2023-09-14T00:18:59.857+0800 [########################] likingtest.oprceDataObjInit 635GB/635GB (100.0%)
2023-09-14T00:18:59.888+0800 finished restoring likingtest.oprceDataObjInit (43776648 documents, 0 failures)
...
2023-09-14T02:05:58.904+0800 [########################] likingtest.oprceDataObj 1205GB/1205GB (100.0%)
2023-09-14T02:05:58.937+0800 finished restoring likingtest.oprceDataObj (53311330 documents, 0 failures)
2023-09-14T02:05:58.945+0800 no indexes to restore for collection likingtest.activityConfiguration
2023-09-14T02:05:58.945+0800 no indexes to restore for collection likingtest.history_task
2023-09-14T02:05:58.945+0800 restoring indexes for collection likingtest.oprcesDataObjInit from metadata
2023-09-14T02:05:58.976+0800 index: &idx.IndexDocument{Options:primitive.M{"name":"flowId_1_activityConfiguration.activityNameEn_1", "ns":"likingtest.oprcesDataObjInit", "v":2}, Key:primitive.D{primitive.E{Key:"flowId", Value:1}, primitive.E{Key:"activityConfiguration.activityNameEn", Value:1}}, PartialFilterExpression:primitive.D(nil)}
2023-09-14T02:05:58.976+0800 index: &idx.IndexDocument{Options:primitive.M{"name":"oprceInfo.oprceInstID_1_activityInfo.activityInstID_1_workitemInfo.workItemID_1", "ns":"likingtest.oprcesDataObjInit", "v":2}, Key:primitive.D{primitive.E{Key:"oprceInfo.oprceInstID", Value:1}, primitive.E{Key:"activityInfo.activityInstID", Value:1}, primitive.E{Key:"workitemInfo.workItemID", Value:1}}, PartialFilterExpression:primitive.D(nil)}
2023-09-14T02:05:58.976+0800 no indexes to restore for collection likingtest.role
2023-09-14T02:05:58.976+0800 no indexes to restore for collection likingtest.snapshotResource
2023-09-14T02:05:58.976+0800 no indexes to restore for collection likingtest.oprceDataObjDraft
2023-09-14T02:05:58.976+0800 restoring indexes for collection likingtest.oprceDataObjInit from metadata
2023-09-14T02:05:58.976+0800 index: &idx.IndexDocument{Options:primitive.M{"name":"oprceInfo.oprceInstID_1_activityInfo.activityInstID_1_workitemInfo.workItemID_1", "ns":"likingtest.oprceDataObjInit", "v":2}, Key:primitive.D{primitive.E{Key:"oprceInfo.oprceInstID", Value:1}, primitive.E{Key:"activityInfo.activityInstID", Value:1}, primitive.E{Key:"workitemInfo.workItemID", Value:1}}, PartialFilterExpression:primitive.D(nil)}
2023-09-14T02:05:58.976+0800 index: &idx.IndexDocument{Options:primitive.M{"name":"flowNo_1", "ns":"likingtest.oprceDataObjInit", "v":2}, Key:primitive.D{primitive.E{Key:"flowNo", Value:1}}, PartialFilterExpression:primitive.D(nil)}
2023-09-14T02:05:58.976+0800 no indexes to restore for collection likingtest.oprceConfiguration
2023-09-14T02:05:58.976+0800 no indexes to restore for collection likingtest.resOutRelDataSnapshot
2023-09-14T02:05:58.976+0800 restoring indexes for collection likingtest.oprceDataObj from metadata
2023-09-14T02:05:58.976+0800 index: &idx.IndexDocument{Options:primitive.M{"name":"flowId_1_activityConfiguration.activityNameEn_1", "ns":"likingtest.oprceDataObj", "v":2}, Key:primitive.D{primitive.E{Key:"flowId", Value:1}, primitive.E{Key:"activityConfiguration.activityNameEn",Value:1}}, PartialFilterExpression:primitive.D(nil)}
2023-09-14T02:05:58.976+0800 index: &idx.IndexDocument{Options:primitive.M{"name":"flowNo_1", "ns":"likingtest.oprceDataObj", "v":2}, Key:primitive.D{primitive.E{Key:"flowNo", Value:1}}, PartialFilterExpression:primitive.D(nil)}
2023-09-14T02:05:58.976+0800 index: &idx.IndexDocument{Options:primitive.M{"name":"oprceInfo.oprceInstID_1_activityInfo.activityInstID_1_workitemInfo.workItemID_1", "ns":"likingtest.oprceDataObj", "v":2}, Key:primitive.D{primitive.E{Key:"oprceInfo.oprceInstID", Value:1}, primitive.E{Key:"activityInfo.activityInstID", Value:1}, primitive.E{Key:"workitemInfo.workItemID", Value:1}}, PartialFilterExpression:primitive.D(nil)}
2023-09-14T02:05:58.976+0800 index: &idx.IndexDocument{Options:primitive.M{"name":"flowId_1_activityConfiguration.activityNameEn_1", "ns":"likingtest.oprceDataObjInit", "v":2}, Key:primitive.D{primitive.E{Key:"flowId", Value:1}, primitive.E{Key:"activityConfiguration.activityNameEn", Value:1}}, PartialFilterExpression:primitive.D(nil)}
2023-09-14T03:45:47.152+0800 97179062 document(s) restored successfully. 0 document(s) failed to restore.
可见:
1、配置并发参数 --numInsertionWorkersPerCollection=4 和 检查参数 bypassDocumentValidation 后,restore速度大大提升,1.2T 的一个大集合 oprceDataObj,由原来默认restore方式约 12h,降为:4h
2、restore完所有数据以后,最后再restore索引,restore索引还是需要一定的时间,本次耗时:1h40m【注:实际没有成功,索引并未生效】
3、新版本的 -d -c 参数需统一修改为:--nsInclude --nsFrom= --nsTo=
■ 2023-09-14T10:40 第2次8并发导入测试
mongorestore --port=20000 -uadmin -p'passwd' --authenticationDatabase=admin --numInsertionWorkersPerCollection=8 --bypassDocumentValidation -d likingtest /u01/nfs/xxxxx_mongodb/10.1.1.1/20230914/likingtest >> 10.2.2.2.log 2>&1 &
tail -100f /u01/nfs/xxxxx_mongodb/10.1.1.1/20230914/10.2.2.2.log
---
2023-09-14T10:40:45.492+0800 The --db and --collection flags are deprecated for this use-case; please use --nsInclude instead, i.e. with --nsInclude=${DATABASE}.${COLLECTION}
...
2023-09-14T10:40:48.493+0800 [........................] likingtest.oprceDataObj 112MB/1208GB (0.0%)
...
2023-09-14T12:57:34.859+0800 [########################] likingtest.oprceDataObj 1208GB/1208GB (100.0%)
2023-09-14T12:57:34.867+0800 finished restoring likingtest.oprceDataObj (53413481 documents, 0 failures)
可见:
1、配置并发参数 --numInsertionWorkersPerCollection=8 和 检查参数 --bypassDocumentValidation 后,restore速度再次大大提升,1.2T的一个大集合 oprceDataObj,由原来默认restore方式约 12h,降为:2h17m
2、本次恢复采用nfs备份恢复,一台8C的虚机,8并发恢复时cpu占用约40%,网络接收速度300MB/s左右,本地磁盘写入速度在30-200MB/s左右,可见网络带段不是瓶颈。可以预见,如果采用更高的主机配置,尤其是IO更好的磁盘,resotore时间必将更少。
■ 2023-09-14T16:10 第3次12并发导入测试
【注意】由于新版本mongorestore摒弃了-d -c参数,虽然可用但使用不够灵活,因此需使用新参数--nsInclude,对于该参数的使用,摸索了多次才找到使用的限制条件,即 directory 必须为数据库备份的根目录/上一级目录,而不是 数据库目录!即类似 dumpdir/20230914,而不是 dumpdir/20230914/database!这是一个巨大的坑,切记!当然,这个目录下一定不能有其他不可识别的文件,否则也会报错。
mongorestore --port=20000 -uadmin -p'passwd' --authenticationDatabase=admin --numInsertionWorkersPerCollection=12 --bypassDocumentValidation --nsInclude="likingtest.*" /u01/nfs/xxxxx_mongodb/10.1.1.1/20230914 > 20230914.10.2.2.2-3.log 2>&1 &
tail -100f /u01/nfs/xxxxx_mongodb/10.1.1.1/20230914.10.2.2.2-3.log
---
2023-09-14T16:10:19.245+0800 preparing collections to restore from
...
2023-09-14T18:18:18.996+0800 [########################] likingtest.oprceDataObj 1208GB/1208GB (100.0%)
2023-09-14T18:18:19.014+0800 finished restoring likingtest.oprceDataObj (53413481 documents, 0 failures)
可见:
1、并发由 8 增至 12 并无效率提升,结论是 6-8 个并发就可以,这一点与oracle的并发导入设置为 6 基本是最佳实践类似。
2、本次恢复采用nfs备份恢复,一台8C的虚机,12并发恢复时cpu占用约60%,网络接收速度300MB/s左右,本地磁盘写入速度在30-500MB/s左右,可见网络带段不是瓶颈。可以预见,如果采用更高的主机配置,尤其是IO更好的磁盘,resotore时间必将更少。
3、关于索引的restore,restore时首先恢复数据,最后再创建索引,比较大的集合的索引创建还是需要较多的时间:
currentOpTime: '2023-09-14T20:23:59.435+08:00',
...
command: {
createIndexes: 'oprceDataObj',
indexes: [
{
key: { flowId: 1, 'activityConfiguration.activityNameEn': 1 },
name: 'flowId_1_activityConfiguration.activityNameEn_1',
ns: 'likingtest.oprceDataObj'
},
{
key: { flowNo: 1 },
name: 'flowNo_1',
ns: 'likingtest.oprceDataObj'
},
{
key: {
'oprceInfo.oprceInstID': 1,
'activityInfo.activityInstID': 1,
'workitemInfo.workItemID': 1
},
name: 'oprceInfo.oprceInstID_1_activityInfo.activityInstID_1_workitemInfo.workItemID_1',
ns: 'likingtest.oprceDataObj'
}
],
.....
currentOpTime: '2023-09-14T20:23:59.489+08:00',
...
command: {
createIndexes: 'oprcesDataObjInit',
indexes: [
{
key: { flowId: 1, 'activityConfiguration.activityNameEn': 1 },
name: 'flowId_1_activityConfiguration.activityNameEn_1',
ns: 'likingtest.oprcesDataObjInit'
},
{
key: {
'oprceInfo.oprceInstID': 1,
'activityInfo.activityInstID': 1,
'workitemInfo.workItemID': 1
},
name: 'oprceInfo.oprceInstID_1_activityInfo.activityInstID_1_workitemInfo.workItemID_1',
ns: 'likingtest.oprcesDataObjInit'
}
],
......第二天再看,还没创建完索引:
currentOpTime: '2023-09-15T09:16:16.460+08:00',
effectiveUsers: [ { user: 'admin', db: 'admin' } ],
runBy: [ { user: '__system', db: 'local' } ],
threaded: true,
opid: 'shard1:11312917',
lsid: {
id: new UUID("e78379ff-9664-46b1-9e87-2bdd4abc5c5f"),
uid: Binary.createFromBase64("O0CMtIVItQN4IsEOsJdrPL8s7jv5xwh5a/A5Qfvs2A8=", 0)
},
secs_running: Long("53877"),
microsecs_running: Long("53877330742"),
op: 'command',
ns: 'likingtest.oprcesDataObjInit',
redacted: false,
command: {
createIndexes: 'oprcesDataObjInit',
......第二天满24h,还没创建完索引:
currentOpTime: '2023-09-15T18:55:16.877+08:00',
effectiveUsers: [ { user: 'admin', db: 'admin' } ],
runBy: [ { user: '__system', db: 'local' } ],
threaded: true,
opid: 'shard1:11312917',
lsid: {
id: new UUID("e78379ff-9664-46b1-9e87-2bdd4abc5c5f"),
uid: Binary.createFromBase64("O0CMtIVItQN4IsEOsJdrPL8s7jv5xwh5a/A5Qfvs2A8=", 0)
},
secs_running: Long("88617"),
microsecs_running: Long("88617747875"),
op: 'command',
ns: 'likingtest.oprcesDataObjInit',
redacted: false,
command: {
createIndexes: 'oprcesDataObjInit',
indexes: [
{
key: { flowId: 1, 'activityConfiguration.activityNameEn': 1 },
name: 'flowId_1_activityConfiguration.activityNameEn_1',
ns: 'likingtest.oprcesDataObjInit'
},
以上可见,mongorestore 导入数据库的数据效率目前是基本可控、可接受的,至少对于1.2T的大集合是可以接受的,但是最后的索引创建实在过于缓慢,且没有找到合适的解决办法:索引需多并发执行创建,且确保索引生效,本次索引创建最后并未生效
■ 2023-09-15T19:02 第4次10并发导入测试,不恢复索引
mongorestore --port=20000 -uadmin -p'passwd' --authenticationDatabase=admin --numInsertionWorkersPerCollection=10 --bypassDocumentValidation --nsInclude="likingtest.*" --nsFrom="likingtest.*" --nsTo="likingtest.*" --noIndexRestore /u01/nfs/xxxxx_mongodb/10.1.1.1/20230914 > 20230914.10.2.2.2-4.log 2>&1 &
tail -100f /u01/nfs/xxxxx_mongodb/10.1.1.1/20230914.10.2.2.2-4.log
2023-09-15T19:02:59.747+0800 preparing collections to restore from
...
2023-09-15T21:24:36.145+0800 [########################] likingtest.oprceDataObj 1208GB/1208GB (100.0%)
2023-09-15T21:24:36.161+0800 finished restoring likingtest.oprceDataObj (53413481 documents, 0 failures)
2023-09-15T21:24:36.165+0800 97367732 document(s) restored successfully. 0 document(s) failed to restore.
以上可见,耗时:2h22m
结论
1、restore 时需设置大数据量 collection 多并发导入:--numInsertionWorkersPerCollection=8
2、不恢复索引:--noIndexRestore
3、数据恢复后,后台创建索引:本站搜索"MongoDB 重建索引"
【最佳实践】MongoDB导出导入数据的更多相关文章
- mongoDB导出-导入数据
--导出数据集 C:\MongoDB\db\bin>mongoexport -d ttx-xwms-test -c things -o d:\mongo_data\things.txt C:\M ...
- mongodb导出导入数据
在使用mongodump导出单个表的时候,遇到了一个错误 # mongodump --host xxx --port 27017 --username 'admin' -p '123456' -d 数 ...
- MongoDB中导入数据命令的使用(mongoimport)
MongoDB中导入数据命令的使用(mongoimport) 制作人:全心全意 语法: mongoimport <options> <file> 介绍: 该命令可以将CSV,T ...
- GreenPlum/postgres copy命令导出/导入数据
一.COPY命令简单实用 1.copy在postgres与GreenPlum介绍 1.1 postgrespostgres的COPY命令可以快速的导出/导入数据到postgresql数据库中,支持常用 ...
- mongodb导出导入实例记录
mongodb导出导入实例记录 平时很用mongodb,所以用到了,就需要去网上搜索方法,干脆将自己的实际经历记录下来,方便日后使用. # 大致需求 源库:db_name_mongo 源IP:192. ...
- mongodb使用自带命令工具导出导入数据
记录 mongo 数据库用原生自带的命令工具使用 json 文件方式进行导入.导出的操作! 在一次数据更新中,同事把老数据进行了清空操作,但是新的逻辑数据由于某种原因(好像是她的电脑中病毒了),一直无 ...
- BCP 导出导入数据(SQL Server)
BCP指令工具可通过安装SQL Server获得. 1. 根据现有的数据库生成表的format文件(导入导出数据的时候需要) bcp db_test.dbo.Table1 format nul -c ...
- 使用BCP导出导入数据
bcp 实用工具可以在 Microsoft SQL Server 实例和用户指定格式的数据文件间大容量复制数据. 使用 bcp 实用工具可以将大量新行导入 SQL Server 表,或将表数据导出到数 ...
- oracle 导出导入数据
在window的运行中输出cmd,然后执行下面的一行代码, imp blmp/blmp@orcl full=y file=D:\blmp.dmp OK,问题解决.如果报找不到该blmp.dmp文件,就 ...
- 【Teradata Utility】使用SQL Assistant导出导入数据
1.导出 (1)选择菜单栏File,点击Export Results,输入导出数据的SQL: select * from etl_data.soure_table; (2)选择导出数据格式为txt或h ...
随机推荐
- SpringBoot 2 种方式快速实现分库分表,轻松拿捏!
大家好,我是小富- (一)好好的系统,为什么要分库分表? (二)分库分表的 21 条法则,hold 住! 本文是<分库分表ShardingSphere5.x原理与实战>系列的第三篇文章,本 ...
- .Net下验证MongoDB 的 Linq 模式联合查询是否可用
MongoDB.Driver 类库提供了 Linq 查询的支持.然而,在使用 Linq 进行联合查询时,是否能够正确转换为 MongoDB 底层的查询语句还有待验证.今天,我将进行实验来验证一下. 输 ...
- java使用SFTP连接服务器下载,上传文件
package mocha.framework.util; /* * @author Xiehj * @version 2019年10月28日 上午9:37:28 */ import java.io. ...
- Java架构师之路:从Java码农到年薪八十万,最牛Java架构师进阶路线
Java架构师之路:从Java码农到年薪八十万,最牛Java架构师进阶路线 摘要:本文将为Java开发工程师提供一条从Java码农到年薪八十万的进阶之路,探讨如何成为一名顶尖的Java架构师.我们将介 ...
- ch-manager.sh
[root@dev-clickhouse1 ~]# cat ch-manager.sh #!/bin/bash ch_arr=(ch1-shard1-main ch1-shard2-sub ch2-s ...
- 跟进 .NET 8 Blazor 之 ReuseTabs 支持 Query 属性绑定
ASP.NET 团队和社区在 .NET 8 继续全力投入 Blazor,为它带来了非常多的新特性,特别是在服务端渲染(SSR)方面,一定程度解决之前 WASM 加载慢,Server 性能不理想等局限性 ...
- xshell批量新建会话
1,xshell的会话形式就是文件夹和文件,我们只需要修改文件名称即可,但是一个个弄很麻烦,可以利用这个程序来批量新建(程序很烂,但能用) 基于python编写,文件处理 --> 文件下载 提取 ...
- vivo 容器集群监控系统优化之道
作者:vivo 互联网容器团队 - Han Rucheng 本文介绍了vivo容器团队基于 Prometheus等云原生监控生态来构建的容器集群监控体系,在业务接入容器监控的过程中遇到的挑战.困难,并 ...
- Netty源码学习2——NioEventLoop的执行
系列文章目录和关于我 零丶引入 在<Netty源码学习1--NioEventLoopGroup的初始化>中,我们学习了NioEventLoopGroup和NioEventLoop的初始化, ...
- VScode软件的安装以及C/C++环境配置的方法
今天和大家分享一下VScode软件的安装以及C/C++环境配置的方法.手把手教大家入门. 1,下载VScode编译器 (1) 官网下载链接:https://code.visualstudio.c ...