Circuit-breakers count all faults they handle as an aggregate, not separately

[vany0114]

I have a policy wrap with WaitAndRetry and CircuitBreaker policies, so, from my understanding the circuit only breaks when the exception is the same consecutively.

var policies = new [] {
Policy
                .Handle<SqlException>(ex => ex.Number == 40613)
                .Or<SqlException>(ex => ex.Number == 40197)
                .Or<SqlException>(ex => ex.Number == 40501)
                .Or<SqlException>(ex => ex.Number == 49918)
                .WaitAndRetryAsync(...),
Policy
                .Handle<SqlException>(ex => ex.Number == 40613)
                .Or<SqlException>(ex => ex.Number == 40197)
                .Or<SqlException>(ex => ex.Number == 40501)
                .Or<SqlException>(ex => ex.Number == 49918)
                .CircuitBreakerAsync(...)
};

await Policy.Wrap(policies).ExecuteAsync(DoSomehitngAsync);

private async Task<bool> DoSomehitngAsync()
{
            await Task.Delay(TimeSpan.FromMilliseconds(500));

            if (++_retriesCounter <= _threshold - 1)
                throw GetRandomException();

            return true;
}

So, the GetRandomException method returns a SqlException randomly (40613, 40197, 40501 or 49918). So, my circuit breaker policy is configured to break the circuit after 3 exceptions, but I would expect that only happen if I throw, let's say the 40613 SqlException in a row 3 times, but currently, it's breaking the circuit despite I'm throwing random exceptions.

I don't know if I'm misunderstanding the usage or I have something wrong with the policies.

[reisenberger]

The circuit-breaker will break after N consecutive actions executed through the policy have thrown 'a' handled exception - any of the exceptions handled by the policy. This is intentional behaviour.

The documentation here used to say 'a' handled exception: I've just updated it to say 'any' handled exception, to make that clearer. Thanks for the question.

---

If you wanted to construct a system that counts different faults (or groups of faults) separately for the purposes of circuit-breaking, you could achieve that by defining separate circuit-breaker policy instances for each fault (or group of faults) you wanted counted separately, and then nesting those circuit-breakers in a PolicyWrap.