UUID collision risk is usually discussed with giant numbers, but the engineering lesson is simple: correctly generated UUID v4 values almost never collide in normal systems. Bad generators, retries and data workflows cause more real problems.
Key takeaways
- UUID v4 has 122 random bits after version and variant bits.
- Correct randomness matters more than memorizing probability trivia.
- Still keep database uniqueness constraints.
Why UUID v4 collisions are so rare
UUID v4 has 122 bits available for randomness. That is an enormous space. The birthday problem means collision risk grows as you generate more values, but the practical risk remains tiny for ordinary application volumes when randomness is strong.
The important phrase is "when randomness is strong." A broken random number generator changes the whole story.
UUID v4 random space: about 2^122 possible valuesWhat causes duplicate IDs in real systems
Real duplicate incidents often come from boring causes: a test fixture copied into production, a migration run twice, a mocked generator returning the same value, or a fallback that used weak randomness.
That is why production tables should still enforce uniqueness. UUID probability is not a replacement for a unique index.
How UUID v7 changes the conversation
UUID v7 spends part of the identifier on timestamp ordering and uses random bits for the rest. A good implementation still makes collisions extremely unlikely, while offering better time sorting than random UUID v4.
If you generate a very high number of IDs in the same millisecond, implementation quality matters. Use reviewed libraries or database functions instead of improvising.
Practical safety checklist
Use platform crypto APIs, keep unique constraints in the database, avoid global mocked IDs in tests, and monitor insert failures. If a duplicate ever appears, treat it as a signal to inspect the generator and workflow rather than shrugging at probability.
| Risk | Better habit |
|---|---|
| Weak randomness | Use crypto-grade random bytes |
| No uniqueness constraint | Add unique indexes or primary keys |
| Copied fixtures | Generate stable but distinct fixture IDs |
| Improvised v7 logic | Use reviewed code for high-volume generation |