To catch this particular issue, they would have had to overprovision webapp instances beyond their N*M limit, take down a subset of the originally provisioned w

To catch this particular issue, they would have had to overprovision webapp instances beyond their N*M limit, take down a subset of the originally provisioned webapp instances and reprovision new ones in order to have observed this particular issue with their HAProxy setup. It seems like it would take a superhuman level of foresight to catch this scenario for chaos engineering.