Testing Postgres race conditions with synchronization barriers (lirbank.com)

Thats not postgresql problem, thats your code

IMHO you should never write code like that, you can either do UPDATE employees SET salary = salary + 500 WHERE employee_id = 101;

Or if its more complex just use STORED PROCEDURE, there is no point of using database if you gonna do all transactional things in js

It'd be interesting to see a version of this that tries all the different interleavings of PostgreSQL operations between the two (or N) tasks. https://crates.io/crates/loom does something like this for Rust code that uses synchronization primitives.

[dead]

Javascript developers learn kindergarten basics of transactions and SQL. LOL. Is it the camp "we don't need a degree to be programmers"?

Postgres has SERIALIZABLE transaction isolation level. Just use it and then you never have to worry about any of these race conditions.

And if for some reason you refuse to, then this "barrier" or "hooks" approach to testing will in practice not help. It requires you to already know the potential race conditions, but if you are already aware of them then you will already write your code to avoid them. It is the non-obvious race conditions that should scare you.

To find these, you should use randomized testing that runs many iterations of different interleavings of transaction steps. You can build such a framework that will hook directly into your individual DB query calls. Then you don't have to add any "hooks" at all.

But even that won't find all race condition bugs, because it is possible to have race conditions surface even within a single database query.

You really should just use SERIALIZABLE and save yourself all the hassle and effort and spending hours writing all these tests.

That whole article should have been:

Use transactions table (just a name, like orders)

On it have an Insert trigger.

It should make a single update with simple “update … set balance += amount where accoundId = id”. This will be atomic thanks to db engine itself.

Also add check constraint >= 0 for balance so it would never become negative even if you have thousands of simultaneous payments. If it becomes negative, it will throw, insert trigger will rethrow, no insert will happen, your backend code will catch it.

—

That’s it: insert-trigger and check constraint.

No need for explicit locking, no stored procedures, no locks in you backend also, nada. Just a simple insert row. No matter the load and concurrent users it will work like magic. Blazingly fast too.

That’s why there is ACID in DBs.

—

Shameless plug: learn your tool. Don’t approach Postgresql/Mssql/whathaveyousql like you’re a backend engineer. DB is not a txt file.

to avoid these conditions i have usually inserted a row into a lock table used for this purpose to create a lock with a unique key for that row with a few minute timer, the once the transaction is complete it will delete the lock row. This way, simultaneous users will only get the first lock, all other requests would fail, and then if the timer expired, we would assume the transaction never completed and it could try again after a few minutes

I’ve idly toyed with this problem as well, I think there’s a good opportunity to build a nice framework in Python with monkeypatching (or perhaps in other languages using DB/ORM middleware) so you don’t need to modify the code under test.

I think you can do better than explicit barrier() calls. My hunch is the test middleware layer can intercept calls and impose a deterministic ordering.

(There are a few papers around looking into more complex OS level frameworks to systematically search for concurrency bugs, but these would be tough to drop into the average web app.)

We hit exactly this kind of race condition in our Go + Postgres SaaS when handling concurrent waitlist signups. Two requests would read the current count, both pass the limit check, and both insert — exceeding the waitlist cap.

Ended up using SELECT FOR UPDATE on the waitlist row before the count check. Simple but effective. The barrier testing approach described here would have caught this much earlier in development instead of discovering it under load.

One thing I'd add: in Go, it's tempting to handle this at the application level with mutexes, but that breaks the moment you have multiple instances. Pushing the serialization down to Postgres is almost always the right call for correctness.

[dead]

I'm confused.

> The simplest case: no transaction, just a SELECT and an UPDATE with a barrier between them:

There is no context where you do not have a transaction. Postgres requires them.

It's likely that the library the author is using is doing automatic implicit transactions, but it seems like the author needs to understand their tools a bit better.

This post confuses me a little. With my tests I try not to "reach inside" systems unless it's quite a specific integration test. Especially databases. In this case I feel like we're just... testing known PostgreSQL behavior?

Or to put another way; as others have observed, this could be solved with atomic updates and in some case SERIALIZABLE. These are right tools for balance operations - and if they’re used I’m not sure they need testing in this manner?

[dead]

In worst case scenario you can’t still get flaky test, right? Single thread runtime that will allow the queries to interleave sometimes and sometimes work correctly - talking about variant without "FOR UPDATE".