Repl internals
Download as PPT, PDF2 likes451 views
Replication allows data to be copied across multiple servers for redundancy and failover. The replication process involves an oplog which records all data modifications. Slave servers copy the oplog from the master to initially sync their data and then stream ongoing changes from the master's oplog. Replica sets improve on replica pairs by allowing multiple redundant servers where an election process selects a new primary if the existing primary fails.
![Replication Internals MongoBerlin 2010-10-4 Mathias Stearn [email_address]](https://image.slidesharecdn.com/replinternals-101012152116-phpapp02/85/Repl-internals-1-320.jpg)
![Questions? Ask away! #mongodb on irc.freenode.net [email_address] [email_address] Thanks for coming!](https://image.slidesharecdn.com/replinternals-101012152116-phpapp02/85/Repl-internals-11-320.jpg)
Repl internals
- 1. Replication Internals MongoBerlin 2010-10-4 Mathias Stearn [email_address]
- 2. Replication is Good! Use it! mongod --master mongod --slave --source host:port 30 second demo…
- 3. Basic Phases of a Slave If not empty skip to step 4 (--fastsync) Note master's position in OpLog Copy data from master ($snapshot) Replay OpLog to catch up Continue watching OpLog for new changes ??? Profit
- 4. The OpLog “ local.oplog.$main” for Master/Slave Only on Master “ local.oplog.rs” for ReplicaSets On all nodes Uses a capped collection (--oplogSize) “ Wraps” when full (LRI) Make sure it's large enough You can query it like any other collection
- 5. OpLog Entries { "ts" : { "t" : 1286179802000, "i" : 1 }, "op" : "i", "ns" : "test.foo", "o" : { "_id":1, "a" : 1 } } Operations: I : insert U : update (modified to be idempotent) D : delete (one per object) C : command N : No-op (ignore these)
- 6. OpLog demo Insert and update an object Look at OpLog Use local db.oplog.$main.find({op:{$ne:'n'}}) sort({$natural:-1}) Show data on slave Add new slave and show data
- 7. Replication Optimizations All of these QueryOptions are used internally Exaust – Sends all data w/o GetMores Tailable – Leaves cursor open AwaitData – Query blocks for data (push) OplogReplay – Optimizes scan for entries newer than supplied timestamp so we don't need an index
- 8. Replica Pairs Deprecated by ReplSets in 1.6 Don't use!
- 9. Replica Sets Evolution of Replica Pairs Reflects “lessons learned” Any number of nodes All keep a copy of the OpLog At any time one node is Primay If Primary fails, remaining nodes elect a new Primary with newest data
- 10. Replica Set Gotchas When Primary fails, short period of time with no Primary Election requires a majority of nodes up Works best with 3 failure domains Writes not commited to quorum can be rolled-back Use getLastError with “W” Cascading not support yet (planned)
- 11. Questions? Ask away! #mongodb on irc.freenode.net [email_address] [email_address] Thanks for coming!