Google SWE teaches systems design | EP14: Raft in 15 minutes

preview_player
Показать описание
I can't lie I'm pretty impressed with myself for this one - obviously I'm not proving anything in this video, but if you do want proofs look at the original Raft paper it's like 15 pages.

Recommended Viewing/Reading:
Рекомендации по теме
Комментарии
Автор

You are making these tough concepts quite easy with your videos and as well as with your reply on questions. Thank You.

shobhitarya
Автор

Hey Jordan, great video, you do a really good job of explaining the concept, the only feedback I have is maybe making the visuals in the presentations easier to understand. A wall of text is much less descriptive than a images, and possibly few slides showing how the actions performed by raft look per your current one slide. Other than that, one of the most concise explanations I've seen!💪

stanowskidev
Автор

Hey Jordan - got a quick question: When you mention things like replicated log / local log, do you actually mean write ahead log? Or they different than a write-ahead log? I understand we call them replicated log because they are replicated across many different server nodes.

ahnjmo
Автор

Enjoying your vids! At 1:03 you say we will have “eventual consistency”. This is a little confusing to me bc my understanding is that these consensus algorithms are used to achieve strong consistency without having to worry about manual failover. In that the message is not commited to the DB until there is a quorum among all replica nodes. So a client wouldn’t receive a success response until that quorum/commit happens which suggests strong consistency. Is my understanding incorrect?

neek
Автор

hey jordan! thanks for this video. at 4:19, you say that when "a leader receives responses from a majority of nodes accepting the write", what does this mean? is it similar to 2-phase commit where the leader is sending a prepare request to all the nodes, and once a majority responds back, it will then actually commit the transaction?

pipdot
Автор

Ah sorry for so many questions, I got one more: On 7:55, when a candidate term number is higher than the follower's term number and the follower's term number is made equal to the candidate's term number, does that mean the vote is casted for that candidate? If not, what's the importance of having the term number equal to the candidate's term number?

pipdot
Автор

I don't really get this.
What is the benefit of using raft?
If you only have a single leader then write throughput is limited to that one node. How is this better than using something like primary/secondary with failover.
Or is this just a way to implement primary/secondary failover. I'm dumbass 😐

yrfvnihfcvhjikfjn
Автор

I finally made the video likes reach the number of God: 69

adityahalder
Автор

Jordan from which leacture exactly can we expect the database choosing.. Actually little bit curious :) to get to hear the tradeoffs

utkarshgupta
Автор

Man, the narration is so boring it makes me want to sleep 😅

rkmr