Pedro Müller

DDIA Chapter 7: Transactions

Part II: Distributed Data

A transaction groups reads and writes into one logical unit that succeeds or fails together. “ACID” is slippery in marketing, so know what isolation you actually have.

ACID in practice

  • Atomicity: mid-flight faults abort the whole unit; safe to retry (with care)
  • Consistency: mostly application invariants the database helps enforce
  • Isolation: concurrent transactions should not step on each other (levels vary widely)
  • Durability: committed data survives crashes (and enough replicas)

Single-object atomicity is common; multi-object transactions across partitions are hard, so many distributed stores weaken or drop them.

Weak isolation you will actually meet

Read committed prevents dirty reads/writes, but not read skew. Snapshot isolation gives each transaction a consistent freeze of the database. It still allows lost updates, write skew, and phantoms unless you add more.

Lost-update defenses: atomic UPDATE... SET x = x + 1, SELECT FOR UPDATE, automatic lost-update detection, compare-and-set, or app-level merge on replicas.

Write skew and phantoms often need serializable isolation or carefully designed locks / materialized conflicts.

Serializability

Strongest isolation: transactions behave as if run one at a time. Implementations:

  • Actual serial execution: single thread; great when txs are short and in-memory
  • Two-phase locking (2PL): correct but latency-heavy under contention
  • Serializable snapshot isolation (SSI): optimistic; abort on conflict at commit; often the modern sweet spot

In code

Snippets below use Java, SQL, or plain text as labeled.

Lost update:

-- Both txs read 100 and write 120 → one increment lost
UPDATE accounts SET balance = 120 WHERE id = 1;

UPDATE accounts SET balance = balance + 20 WHERE id = 1;
SELECT balance FROM accounts WHERE id = 1 FOR UPDATE;

Compare-and-set:

UPDATE documents
SET content = $new, version = version + 1
WHERE id = $id AND version = $seen_version;

Write skew sketch:

Two doctors each see "the other is on call" and both go off duty.
Row locks on different rows are not enough, need serializable (or equivalent).

Remember

“ACID” on the box does not mean serializable. Read committed still allows lost updates and write skew. Use stronger isolation when you cannot reason about races by inspection.