The authors of the article share their experience with IO issues in a cloud infrastructure, specifically with YugabyteDB, a cloud-native database. They initially thought they had replicated the issue by oversubscribing on disk limits, but further investigation revealed that the problem was caused by delays in the cloud provider's storage subsystem, which were not immediately apparent from the virtual machine's logs or IOPS/MBPS limits. The authors used Prometheus data and 0xtools to gather more detailed information about the issue, including kernel stack backtraces, and eventually proved that the problem lay outside of the virtual machine in the cloud infrastructure. With this evidence, the cloud provider was able to identify and fix the underlying issues, which had caused delays in their storage subsystem. The authors emphasize the importance of monitoring resources even in the cloud and not relying on first-line support assumptions.