|
|
|
|
|
|
by drm237
4978 days ago
|
|
|
When Facebook opened their second data center, they basically had one be read/write and the other just read (if I remember correctly). You could do something similar and if a user causes a write or update, keep track in a cookie and use only the master for their reads and writes for a period of time equal (or longer) than the typical replication length. Then when the time expires, start using the slaves for their reads. I'm sure there are better ways than this, it's just one that I know has been used in production with some success. |
|
|