Продолжаем с классикой - "Calvin: Fast Distributed Transactions
for Partitioned Database Systems":
https://www.cs.yale.edu/homes/thomson/publications/calvin-sigmod12.pdfСУБД которая реализована по принципам из статьи - YDB.
TL;DR
The essence of Calvin lies in separating the system into three separate layers of processing:
• The sequencing layer (or “sequencer”) intercepts transactional inputs and places them into a global transactional input
sequence—this sequence will be the order of transactions to
which all replicas will ensure serial equivalence during their
execution. The sequencer therefore also handles the replication and logging of this input sequence.
• The scheduling layer (or “scheduler”) orchestrates transaction execution using a deterministic locking scheme to guarantee equivalence to the serial order specified by the sequencing layer while allowing transactions to be executed concurrently by a pool of transaction execution threads. (Although
they are shown below the scheduler components in Figure 1,
these execution threads conceptually belong to the scheduling layer.)
• The storage layer handles all physical data layout. Calvin
transactions access data using a simple CRUD interface; any
storage engine supporting a similar interface can be plugged
into Calvin fairly easily.
Такой дизайн накладывает ограничение:
All transactions are therefore required to declare their full read/write
sets in advance;
Но есть трюки что бы его почти обойти.
