We currently have 2 tests that perform save/restore under io load with 1 NVMe disk attached.
- servicing_keepalive_with_unresponsive_io
- servicing_keepalive_with_io_queue_full
These are designed to deterministically fill the nvme io queue and cause long delays to io. We need some local stress testing mechanism that can perform similar servicing under io load in a loop. It would probably benefit from also having functionality to attach more than 1 disk backed by the nvme_test emulator.
We currently have 2 tests that perform save/restore under io load with 1 NVMe disk attached.
These are designed to deterministically fill the nvme io queue and cause long delays to io. We need some local stress testing mechanism that can perform similar servicing under io load in a loop. It would probably benefit from also having functionality to attach more than 1 disk backed by the nvme_test emulator.