PITR（point-in-time recovery）是在有备份的情况下,可以恢复到任何一个时间点。
之前文章写过,mongodb的备份工具mongodump,mongorestore,mongoexport,mongoimport
以上工具就不再详细介绍
这里介绍一下mongodump的一个选项--oplog
--oplog use oplog for taking a point-in-time snapshot
文档是这样介绍的，但是并不是任何时间点的snapshot哦，这个选项的意思是dump开始到dump结束这个时间段操作的oplog，oplog这个东西只有在replica sets里面有，因为主从节点之间就是通过这个东西同步的，所以建议大家即使是一个节点的mongodb也要配置一下--replSet选项，因为standalone模式下，local下面是没有oplog.rs这个表的。
PITR恢复其实主要是用到mongorestore的--oplogReplay --oplogLimit两个选项，一个是回放操作，另外一个oplogLimit可以选取任何一个点，这个点如何选择呢，接下来就要注意了，做个实例，希望大家理解。
这里replica sets的搭建就不要介绍了，直接主节点操作步骤如下：
创建两个测试表，一个test，一个tb1:
db.test.save({name:"apple"});
for(var i = 1; i <=100; i++) {db.tb1.insert({name:"hank"+i});}
然后做一下全备
mongodump -h 127.0.0.1 --port 5888 -o /home/mongo/
进入数据库做一些操作
db.test.drop();
db.tb1.remove({});
db.tb1.insert({name:"xxxxxxxxxxxxxxxxxxxx"});
现在，我们要恢复到操作之前，怎么做呢，第一个就是用我们刚才全备直接restore回去，另外如果想恢复到任何一个点的话，那么就需要oplog
导出oplog
mongodump -h 127.0.0.1 --port 5888 -d local -c oplog.rs -o /home/mongo/
mv oplog.rs.bson /home/mongo/local/oplog.bson --重命名导出的oplog
bsondump oplog.bson 查看oplog内容，这类我截取了部分内容，这里标红加粗的就是--oplogLimit 要跟的参数内容
这个参数就是db.test.drop();操作之前的点
{"ts":{"$timestamp":{"t":1488181183,"i":1}},"t":{"$numberLong":"4"},"h":{"$numberLong":"-6545284460632360892"},"v":2,"op":"c","ns":"test.$cmd","o":{"drop":"test"}}
{"ts":{"$timestamp":{"t":1488181203,"i":1}},"t":{"$numberLong":"4"},"h":{"$numberLong":"-1071465797953742786"},"v":2,"op":"d","ns":"test.tb1","o":{"_id":{"$oid":"58b3d7395089b35311fa9b1e"}}}
{"ts":{"$timestamp":{"t":1488181203,"i":2}},"t":{"$numberLong":"4"},"h":{"$numberLong":"7922879569243271327"},"v":2,"op":"d","ns":"test.tb1","o":{"_id":{"$oid":"58b3d7395089b35311fa9b1f"}}}
{"ts":{"$timestamp":{"t":1488181203,"i":3}},"t":{"$numberLong":"4"},"h":{"$numberLong":"5203619554553799738"},"v":2,"op":"d","ns":"test.tb1","o":{"_id":{"$oid":"58b3d7395089b35311fa9b20"}}}
{"ts":{"$timestamp":{"t":1488181203,"i":4}},"t":{"$numberLong":"4"},"h":{"$numberLong":"-4057525037056875752"},"v":2,"op":"d","ns":"test.tb1","o":{"_id":{"$oid":"58b3d7395089b35311fa9b21"}}}
{"ts":{"$timestamp":{"t":1488181203,"i":5}},"t":{"$numberLong":"4"},"h":{"$numberLong":"428652878437400693"},"v":2,"op":"d","ns":"test.tb1","o":{"_id":{"$oid":"58b3d7395089b35311fa9b22"}}}
比如我们要恢复到db.tb1.insert({name:"xxxxxxxxxxxxxxxxxxxx"});之前
那么执行以下命令
mongorestore -h 127.0.0.1 --port 5888 --drop --oplogReplay --oplogLimit "1488181183:1" /home/mongo/local
2017-02-27T17:07:11.670+0800 building a list of dbs and collections to restore from /home/mongo/local dir
2017-02-27T17:07:11.671+0800 replaying oplog
2017-02-27T17:07:11.681+0800 done
db.tb1.find()
{ "_id" : ObjectId("58b3d7e25bdf4a8fd701f07d"), "name" : "xxxxxxxxxxxxxxxxxxxx" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b1e"), "name" : "hank1" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b1f"), "name" : "hank2" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b20"), "name" : "hank3" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b21"), "name" : "hank4" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b22"), "name" : "hank5" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b23"), "name" : "hank6" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b24"), "name" : "hank7" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b25"), "name" : "hank8" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b26"), "name" : "hank9" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b27"), "name" : "hank10" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b28"), "name" : "hank11" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b29"), "name" : "hank12" }
这里可以看到恢复到了之前的状态，但是这里"name" : "xxxxxxxxxxxxxxxxxxxx"的记录还是在的
也就是说，这里只是回放了oplog操作，因为这条记录本身就有的，我们只是回放了之前插入的操作，所以这条记录还保留在表中
再比如，以下操作是
mongorestore -h 127.0.0.1 --port 5888 --oplogReplay --oplogLimit "1488181203:20" /home/mongo/local/
--oplogLimit=<seconds>[:ordinal]
这里是会恢复到1488181203这个操作的第20次
db.tb1.find()
{ "_id" : ObjectId("58b3d7e25bdf4a8fd701f07d"), "name" : "xxxxxxxxxxxxxxxxxxxx" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b32"), "name" : "hank21" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b33"), "name" : "hank22" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b34"), "name" : "hank23" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b35"), "name" : "hank24" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b36"), "name" : "hank25" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b37"), "name" : "hank26" }
{ "_id" : ObjectId("58b3d7395089b35311fa9b38"), "name" : "hank27" }
可以看到删除了前面20行记录
{"ts":{"$timestamp":{"t":1488187878,"i":1}},"t":{"$numberLong":"4"},"h":{"$numberLong":"2891167049592106048"},"v":2,"op":"c","ns":"test.$cmd","o"
:{"drop":"oplog"}}
1664 {"ts":{"$timestamp":{"t":1488187909,"i":1}},"t":{"$numberLong":"4"},"h":{"$numberLong":"4897030452135397992"},"v":2,"op":"c","ns":"test.$cmd","o"
:{"create":"hank"}}
1665 {"ts":{"$timestamp":{"t":1488187909,"i":2}},"t":{"$numberLong":"4"},"h":{"$numberLong":"-366641111972103000"},"v":2,"op":"i","ns":"test.hank","o"
:{"_id":{"$oid":"58b3f205ff3732609f8a099d"},"name":"apple"}}
1666 {"ts":{"$timestamp":{"t":1488187943,"i":1}},"t":{"$numberLong":"4"},"h":{"$numberLong":"448267919733245353"},"v":2,"op":"c","ns":"test.$cmd","o":
{"drop":"tb1"}}
drop恢复演示：
db.tb1.drop();
db.hank.drop();
删除后通过以下命令恢复
mongorestore -h 127.0.0.1 --port 5888 --oplogReplay --oplogLimit "1488187943:1" /home/mongo/local/
2017-02-27T17:41:02.631+0800 building a list of dbs and collections to restore from /home/mongo/local dir
2017-02-27T17:41:02.632+0800 replaying oplog
2017-02-27T17:41:03.107+0800 done
show tables; --可以看到2个表都已经恢复
hank
tb1
db.hank.find();
{ "_id" : ObjectId("58b3f205ff3732609f8a099d"), "name" : "apple" }
db.tb1.count();
82
记录也正常
另外如果有想通过备份+oplog做恢复的同学，一定注意，两次备份的时间间隔不要超过oplog的覆盖范围，
所以要权衡oplog的大小，要有足够的时间dump出oplog，发生故障，停止数据库的操作，以免超出oplog覆盖范围。
另外注意点，恢复点之后，新增的表，恢复之后，这个表不会被删除，还存在并且数据正常
比如：
这里我们的hank表在之前是没有的，那么我恢复到很早的一个点
show tables;
hank
tb1
db.hank.find();
{ "_id" : ObjectId("58b3f205ff3732609f8a099d"), "name" : "apple" }
mongorestore -h 127.0.0.1 --port 5888 --drop --oplogReplay --oplogLimit "1488181183:1" /home/mongo/local
这里恢复的这个时间点其实并没有hank表，但是恢复后还是存在，但是tb1恢复到了那个时间点。
show tables;
hank
tb1
db.tb1.count();
101
