20. Handle Distributed Transactions | Two-Phase Commit (2PC), Three-Phase Commit (3PC), SAGA Pattern

➡️ Notes link: Shared in the Member Community Post (If you are Member of this channel, then pls check the Member community post, i have shared the Notes link there)

➡️ Join this channel to get access to member only perks:

Topics i have covered:

- What is Transaction
- ACID Properties of the Transaction
- Understand Distributed Transaction Problem Statement
- Two-Phase Commit approach
- Interview Questions of Two Phase Commit
- Three-Phase Commit Approach
- Interview Questions of Three Phase Commit
- SAGA Pattern

Chapters:
00:00 - Introduction
00:57 - What is Transaction
02:00 - ACID Properties of Transaction
08:05 - Problem with Distributed Transactions
12:06 - Two-Phase Commit Approach
17:49 - Interview Questions of Two Phase Commit
27:34 - Three-Phase Commit Approach
37:45 - SAGA Pattern Approach

Other Playlist:

You can also connect with me 1:1 on Topmate:

Join this channel to get access to perks:

#systemdesign #softwareengineer

[krizh]

This is the single best 2PC and 3PC video I've come across! Really helpful for my exam tomorrow

[Deepz]

consistency in ACID essentially means that the data should be a in a consistent state before and after the transaction i.e., it should satisfy all DB constraints.

[valentinussofa]

I need more videos that talk about distributed systems like this. Thank you so much. 🙏

[saurabhtiwari]

Got best gift on my birthday. Thank u bhaiya a ton. 🎉🎉🎉

[DurgaShiva]

Finally the wait is over.. GOAT of HLD is back, , thankyou for such content, u rock !!

[doingsneakypeakylike]

Great explanation like always. Thanks!

[aartikelkar]

Thank you for the informative videos.
I have a doubt at 35:00. You said that all participants know about each other. So in 2 PC, suppose TC sends commit msg to one participant and then fails, meaning only 1 out of two participants got the commit msg. Then in this case, 2PC should not be blocking the transaction as the Participants are aware of each other's status.
Thanks

[dragonhot]

Suggestion:
please try laser pen and more pictorial reprresentation and will be great if used tab to write down things properly or just keep a slide of it

[timefly]

One doubt in SAGA desing pattern @Shrayansh
Since each local transaction does commit and move on next service and then complete its transaction and do commit. Suppose last microservice get failed,
you said it generate some sort of event which let all the previous transaction to be rolled back. But how come it happen since for all the previous transaction we have done commit. so those are already persisted to the db memory.

[shubhamsoni]

5:36
One transaction should not be locking another transaction. Ideally, only row-level locks should be used to maximize concurrency. If both transactions are accessing the same row, an Exclusive lock should be applied to ensure data integrity. However, if the isolation level is set to Serializable, transactions might still lock each other to maintain consistency and prevent issues like phantom reads.

[arishtajain]

Hi, great video overall, though I find a gap in the explanation.
In the traditional Two-Phase Commit (2PC) protocol, participants do not communicate directly with each other; they only communicate with the Transaction Coordinator (TC). This is why, 3 PC introduces additional complexity (coordinating between participants).

If participant start communicating between each other in both, then 2PC and 3 PC are not much different.

[Akashdeepkashyap]

Hi Shreyansh, Great video. One doubt, in 2pc if the commit msg is lost, since services can talk to each other it can ask other services if they have received the commit msgs if yes than continue with commit or abort. Why go for 3PC?

[KulbirMalik-pxnf]

Hi Shreyansh, just a small doubt . How does pre-commit solve the problem in 3 phase, it can be done the same way in 2 phase only . If participants have information of each other and in 2 phase algorithm only they didn't got any COMMIT message ( coordinator went down before 2nd phase only ), they can abort there only right ?

[shreyashosur]

Hi Shreyansh,

You have explained it very well.
I have few doubts here. I browsed these doubts all over the internet and I did not get proper answer in GPT as well.

1. What if the paricipants have multiple replicas ? How co-ordinator will make sure the prepare and commit request will always hit the same replica ?
2. I am thinking out this doubt in Java/Spring Implementation. I know transactions can be created using Transaction Annotation and a same thread will work until the transaction is committed.
Once participant creates local transaction, it must be in a new transaction thread.
So how this transaction thread is stopped from committing and how it commits after receiving the external request(http thread) from co-ordinator since both threads are different. In simple terms, how the two different threads are communicated ? one is new http thread(prepare or commit) and already created transaction thread ?

I hope the understanding of doubt is clear. Please let me know if there is a problem. I will elaborate maybe.
I am badly searching for the doubts.

[abhinavpandey]

Shreyansh one thing like can u please pin some of the common doubts asked so that u also need not to answer every time and we also need not to see every chat. Just a suggestion. Also good content 🎉

[adityaxmittal]

if participants can communicate with each other, then why is there a blocking phase in 2PC? Since a participant can check with others regarding the final decision and act accordingly

[manish.adhikari]

Thankyou for the nice explanation.
Could you please elaborate more on how the participating MSs enquire each other about decision taken by TC?
Furthermore, if participating MSs can enquire each other if they have received decision from TC then isn't it a good idea in 2PC that the MSs can abort the transaction and release the lock on the corresponding rows, thus reducing the blocking behaviour?
And when the TC comes up again it can check the incompleted transactions (like the above scenario where even before taking decision the TC went down) and discard them.

[amiransari]

great stuff, Can you make a video to do it practically if it is possible, thanks

[SantoshGupta-jlcj]

Your explanation is not precise...mixed up and confused to me with 2ph and 3ph

[rohitkumarratnesh]

What if pre-commit msg sent from co-ordinator fails? Isn't this same as commit msg from co-ordinator fails in 2PC and participant is waiting for what to do ? If participants can talk to each other whether other participants received commit msg or not in both 2PC and 3PC, how 3 PC solves the problem of 2PC(commit msg lost)?