Level 10 of 13 · about 2 weeks
Partitioning and distributed transactions
Splitting one dataset across many machines, indexing partitioned data, rebalancing as a cluster changes shape, why unreliable networks and clocks make distributed systems genuinely hard, two-phase commit, what consensus actually solves, and the CAP theorem used carefully rather than as a slogan.
Modules
Core: everyone needs it. Common: most jobs use it. Specialist: optional depth, safe to skip on a first read.
1
Partitioning strategies
splitting one large dataset across many machines, and what decides where each row lands
Core 5 parts · about 2 hours
2
Secondary indexes on partitioned data
why an index on a non-partitioning column needs its own partitioning decision, and what that decision costs at query time
Specialist 5 parts · about 2 hours
3
Rebalancing partitions
moving data between machines as a cluster grows or shrinks, without taking the system down to do it
Specialist 5 parts · about 2 hours
4
The trouble with distributed systems
why machines talking over a network cannot assume the network, the clock, or each other's state is reliable
Specialist 5 parts · about 2 hours
5
Distributed transactions and two-phase commit
making a transaction that touches several machines commit or abort on all of them together, and the cost of guaranteeing that
Specialist 5 parts · about 2 hours
6
Consensus, what it actually solves
getting a set of nodes to agree on one value even when some of them fail, and why that is harder than it sounds
Specialist 5 parts · about 2 hours
7
The CAP theorem, used carefully
what CAP actually claims, and why most popular summaries of it overreach
Specialist 5 parts · about 2 hours