Replication Summarized in 9 Minutes | Systems Design Interview 0 to 1 with Ex-Google SWE

[@jaymistry4727]

Dude, in this stressful time of finding a job, your videos and jokes bring a smile on my face and I look forward to completing the series and hopefully landing a nice job soon! Thanks for the amazing series.

[@jordanhasnolife5163]

Best of luck my friend, you got it!

[@brandonwl]

You should do more summary videos. This was great!

[@xiyunliu9252]

+1

[@TheImplemented]

As for the multileader replication it is worth mentioning it is quite useful in multiple DC scenario, where you have a cluster of nodes in each (or some) DC and a leader per cluster. The clients will be writing to the nearest master node, and it will in turn replicate data to some of the nodes locally and to other master nodes.

[@kiranmathew9763]

7:41 in the correction note, you mention that it’s “low read latency” instead of “low read throughput”. It’s supposed to be “high read latency”, since longer wait time is observed.

[@jordanhasnolife5163]

yup nice catch

[@hinata4661]

7:02, How is leaderless replication throughput less then multi leader in theory when we are writing on to more than one node? Because of higher latency in leaderless to write to quorum number nodes?

[@jordanhasnolife5163]

Yep!

[@HimuBoddu]

Would'nt it be higher read latency at @7:40.

[@jordanhasnolife5163]

yep, oops!

[@rahullingala7311]

Amazing set of videos on replication!

[@vankram1552]

Skin looking good

[@jordanhasnolife5163]

Thanks I've been consulting a lot of people to help me moisturize, they keep wanting me to meet them in bath houses for some reason though

[@xiyunliu9252]

Your videos are the best. Thanks again! This summary is helpful! You shared the reason why we need single leader, multi leader and leaderless replication, those are clear. I have one question though, you discussed about how to deal with steal read in the early videos of replication. Those sounds like how to deal with write conflict, because I feel write conflict feels like we are trying to read the latest version of data. Could you help me understand more clear on that? How do we apply those with those different type of leader replication?

[@jordanhasnolife5163]

I'm sorry I'm not sure what you mean here. Can you elaborate please? If you just want to know how to deal with write conflicts, your options are: 1) avoid them by using the same leader for the same key 2) last write wins 3) Store siblings, and have a user resolve 4) Have the database eventually converge on a value via CRDTs or something like that

[@xiyunliu9252]

@@jordanhasnolife5163 Thanks Jordan! What I mean is I couldn't connect the section of "how to deal with steal read" with the whole picture of the replication. Your summary gives a great overview of dealing with write conflict, but not mentioned the dealing with steal read.

[@jordanhasnolife5163]

@@xiyunliu9252 I suppose there's not really much that can be done. If you need to never read stale data, you basically have to read from the leader (if it's single leader), or wait until the replica that you're reading from has the data that you think it should.

[@itssecret3780]

Why would replication help ensure that write throughput increase? Every node eventually will have all the writes, so even if user 1 is writing to node 1 and user 2 is writing to node 2, eventually node 2 and node 1 data will be exchanged, and so any write done by user 2 on node 2 will be done on node 1 also, so all nodes from the standpoint of write, will be consuming same cpu percentage. How is write throughput increase then, all writes are written to all replica eventually?

[@jordanhasnolife5163]

All of those writes are done in the background, and can be done asynchronously in batch

[@varundubey922]

feels like I am expanding my mind, thanks lol

[@jordanhasnolife5163]

They call me the expander (no one has ever called me this)

[@onurozer7218]

I like watching async in your video and audio when it comes to sharing your ipad with your face😂

[@jordanhasnolife5163]

I swear to God I record the video and audio in the same OBS instance so I have absolutely no clue how they always get out of sync

[@raghavendrag8036]

On a sidenote, may I ask what app on the Ipad do you use for the notes you show in the video?Trying to prepare notes myself.

[@jordanhasnolife5163]

OneNote

[@MuhammadUmarHayat-b2d]

is multileader replication effectively a sharding of single leader replication?

[@jordanhasnolife5163]

Nope, you can write to the same key in multi leader replication. With sharing, different primary keys are on different nodes.

[@vanessaserna3343]

The description box of the video tells me you're the last romantic

[@jordanhasnolife5163]

I'm here for a good time not a long time (nice comment though I loled)

[@chaitanyatanwar8151]

Thank you!

[@shota7517]

If you didn’t make out yet, you can ask me

[@shota7517]

No thanks

[@dibll]

Is the quorum concept applicable to Single Leader/Multi Leader replications too?

[@jordanhasnolife5163]

Nope - if you think aboutityou need the writeto go to a majority of nodes for a quorum to work, in single leader and multi leader replication writes are only sent to one node

[@Iwannayoucry]

@@jordanhasnolife5163 MongoDb uses Single Leader with quorum concept

[@radosawmul8963]

You know that if you hold down the pen after drawing something, like a circle, line or arrow, your Ipad will make it pretty, right? It would be cool to see at least for the predrawn pics 🤙 Now go create System design 3.0

[@jordanhasnolife5163]

Lol I'll give it a shot