F****n
发帖数: 3271 | 1 If my master server failed how to handle data that have NOT been replicated
to the slave servers? |
g*****g
发帖数: 34805 | 2 Depends on how the master server fails, if the data is available in commit
log, restarting the server can fix it.
If it's harddisk failure, however, obviously data is lost.
replicated
【在 F****n 的大作中提到】
: If my master server failed how to handle data that have NOT been replicated
: to the slave servers?
|
F****n
发帖数: 3271 | 3 有啥应对办法
【在 g*****g 的大作中提到】
: Depends on how the master server fails, if the data is available in commit
: log, restarting the server can fix it.
: If it's harddisk failure, however, obviously data is lost.
:
: replicated
|
w**z
发帖数: 8232 | 4 你是怎么ack写的, master node, 写了，就ack?
【在 F****n 的大作中提到】
: 有啥应对办法
|
g**e
发帖数: 6127 | |
F****n
发帖数: 3271 | 6 Right, asynchronous replication.
I know it is not possible to have strict atomic consistency.
It is called "eventual consistency", right:)
But I would like to know what is the best practice to deal with (not solve)
this kind of situations.
【在 w**z 的大作中提到】
: 你是怎么ack写的, master node, 写了，就ack?
|
F****n
发帖数: 3271 | |
g*****g
发帖数: 34805 | 8 In principle, you need multiple nodes to ack on a transaction if the data is
super critical. Obviously there's a performance penalty here.
【在 F****n 的大作中提到】
: Right, asynchronous replication.
: I know it is not possible to have strict atomic consistency.
: It is called "eventual consistency", right:)
: But I would like to know what is the best practice to deal with (not solve)
: this kind of situations.
|
w**z
发帖数: 8232 | 9 as goodbug said, and C* way of handling, it is written in commit log every
10 secs and replicated to other nodes. if the hard drive doesn't fail, you
could lose up to 10 secs of writes. if your HD is gone, you will lose all
unreplicated data.
【在 F****n 的大作中提到】
: Right, asynchronous replication.
: I know it is not possible to have strict atomic consistency.
: It is called "eventual consistency", right:)
: But I would like to know what is the best practice to deal with (not solve)
: this kind of situations.
|
g*****g
发帖数: 34805 | 10 The fsync interval can be configured though.
every
【在 w**z 的大作中提到】
: as goodbug said, and C* way of handling, it is written in commit log every
: 10 secs and replicated to other nodes. if the hard drive doesn't fail, you
: could lose up to 10 secs of writes. if your HD is gone, you will lose all
: unreplicated data.
|
b*******s
发帖数: 5216 | 11 有什么好的js framework推荐？要能很容易配置restful的，和可以轻松scale out
谢谢
【在 g*****g 的大作中提到】
: The fsync interval can be configured though.
:
: every
|
g*****g
发帖数: 34805 | 12 我不懂js，node上面不是有啥Express之类的。
【在 b*******s 的大作中提到】
: 有什么好的js framework推荐？要能很容易配置restful的，和可以轻松scale out
: 谢谢
|
b*******s
发帖数: 5216 | 13 好，谢谢
【在 g*****g 的大作中提到】
: 我不懂js，node上面不是有啥Express之类的。
|
W***o
发帖数: 6519 | 14 do not use master to store data, if master fails, have the slaves elect a
new master
replicated
【在 F****n 的大作中提到】
: If my master server failed how to handle data that have NOT been replicated
: to the slave servers?
|
