Building a Large-Scale Distributed Storage System Based on Raft

cncf.io

163 points·jinqueeny··5 comments

We also introduced another role (Raft-learner) to the Raft group and make a columnar extension. More details can be found here: https://pingcap.com/blog/delivering-real-time-analytics-and-...

I'm not quite following the ensuring secure execution of split operations section. There's an incomplete sentence:

>The split process is coupled with network isolation, which can lead to very complicated.

Maybe the second idea should be first, demonstrate the complexity of splits and then presenting the solution?

Thank you for pointing out, this place is indeed a typo, what I want to say here is: because in our implementation, the split operation of a raft group is treated as a special raft log, because raft is a quourum-based algorithm, so in the case of network isolation, there may be followers that cannot receive the split command. I added a practical example after the paragraph. The solution to this problem is written behind, using the epoch + logic clock to ensure that the system is able to confirm the newest leader for specific raft group at any time.

Typically, you want to be able to put multiple ranges on the same node, (for example if you keep an extra index with a reverse mapping). If both the forward mapping and the reverse mapping are located on the same node, you only have a 'local' transaction.

Do you mean 1PC transaction?

Reply on news.ycombinator.com