Sign in Get started

Decision guide

Choosing a Transaction Isolation Level

Which isolation level should this transaction run at?

Side by side

CriterionRead CommittedRepeatable Read (snapshot)Serializable
Prevents lost updates in read-modify-writeNo (use atomic updates or locks)Postgres: yes (error + retry); MySQL: noYes
Prevents write skew / phantomsNoNoYes
Consistent multi-statement readsNoYesYes
Needs application retriesRarely (deadlocks only)Postgres: on update conflictsYes, routinely
Throughput under contentionHighestHighLower (aborts or lock waits)

When to choose each

Read Committed

Each statement sees data committed before it started. Postgres default.

Choose when

  • Single-statement reads and writes
  • Writes that use atomic updates, constraints or explicit locks
  • High-concurrency OLTP where you control locking yourself

Avoid when

  • Multi-statement reports that must add up
  • Read-modify-write logic without locks or version checks

Repeatable Read (snapshot)

The transaction reads one consistent snapshot. MySQL InnoDB default; snapshot isolation in Postgres.

Choose when

  • Multi-statement read-only reports and exports
  • Postgres read-modify-write where you can retry serialization failures

Avoid when

  • Rules spanning several rows (write skew still possible)
  • Long-running jobs on the primary (holds back VACUUM / purge)

Serializable

Results equal some serial order. Postgres: SSI with aborts; MySQL: shared locks on reads.

Choose when

  • Complex invariants across rows or tables (limits, rosters, quotas)
  • You can retry the whole transaction safely
  • Correctness matters more than peak throughput

Avoid when

  • Very hot data where retries would snowball
  • Transactions with side effects you can't repeat

Verdict

Default to Read Committed and make writes safe with atomic updates, constraints and explicit locks. Use Repeatable Read (read-only) for multi-statement reports. Reach for Serializable when an invariant spans several rows and every write path can be retried — and set the level explicitly in code that depends on it.

Learn more

Sources