yugabyteDB's master, role in distributed transactions

Do we need master control for distributed transactions in yugabyteDB? For example, the state synchronization of sub-transactions performed by each tserver, does it need to be synchronized to the client via master?
From the following document: Transactional I/O path | YugabyteDB Docs
It seems that no master intervention is required.

That’s right. Regular DML do not need the master.
It is used for new connection and for DDL, as it holds the catalog, but transaction have their tables distributed in the tservers

Thank you very much.
When I ask a question on the kimi platform, the intelligent answer is as follows, that should be wrong:
Thus, the Master component does play an important role in distributed transactions, especially in the tracking and management of transaction state. However, the specific transaction commit and orchestration processes, including two-phase commit (2PC) of transactions, etc., are handled by other components and mechanisms to ensure atomicity and consistency of transactions.

By the way, the new connection here, which connections does it refer to? rpc connection? Do connections have anything to do with transactions as well?

SQL connections. It’s the PostgreSQL catalog that is on the master

1 Like

okay, I got it now.
Thanks a lot.