Avoid refreshing BftScanConnection too often when less than f scans are failing

[OriolMunoz-da]

Fixes too many refreshes happening even if BFT guarantees are still ensured (see #1141 (comment))

[OriolMunoz-da]

the alternative would be to drop the whole retry mechanism, but the problem then is that it's only refreshed every ~10m (which is a good-enough default otherwise)

[moritzkiefer-da]

Hm should this be f or 2f+1? I thought we want 2f+1 working connections so we can read from them and tolerate f failures.

[OriolMunoz-da]

actually that was stupid, refresh already does the check:

val defaultCallConfig = BftCallConfig.default(connections)
            // Most but not all calls will use the default config.
            // Fail early if there are not enough Scans for the default config
            if (!defaultCallConfig.enoughAvailableScans) {
              throw io.grpc.Status.FAILED_PRECONDITION
                .withDescription(
                  s"There are not enough Scans to satisfy f=${connections.f}. Will be retried. State: $newState"
                )
                .asRuntimeException()
            } else {
              connections
            }

[OriolMunoz-da]

fix'd

[OriolMunoz-da]

(where also the condition is: connections.size >= targetSuccess, i.e. connections.size >= f +1)

[moritzkiefer-da]

thanks!

[rautenrieth-da]

Thanks!

Would it make sense to add a test, something like the existing "periodically refresh the list of scans" in BftScanConnectionTest?