4 ms·

I am curious, what kind of write characteristics can manage to saturate a 255s timeout on a storage device that does 10k+ iops and gigabytes per second throughp

I am curious, what kind of write characteristics can manage to saturate a 255s timeout on a storage device that does 10k+ iops and gigabytes per second throughput? Normally writes slowing down leads to backpressure because the syscalls issueing them take longer to return. I can imagine some crazy random access, small-record, vectored IO from large thread pools. But that's not exactly common because most software that is IO-heavy tries really hard to avoid these things.

lathiat 8y ago
According to the bug here: https://bugs.launchpad.net/ubuntu/bionic/+source/linux/+bug/1758466 https://bugs.launchpad.net/ubuntu/bionic/+source/linux/+bug/... It was actually filed about EBS block disks using NVME (on these new instances, there is a hardware card that presents network EBS volumes as a PCI-E NVME device). In certain failure cases since this is a network block store, they can fail for some period of time exceeding this timeout. The idea of this change is to ensure once they come back the machine is left in a usable state. Of course, I would not expect to see this on Local NVME disks which is what they announced - that you can now get such instances with local disk as well as EBS.
- ohre1Eda 8y ago
  Ah yeah, the article was about local NVMe, so the concern is probably not relevant here.
  - lathiat 8y ago
    Unrelated: love your pwgen username
benmmurphy 8y ago
EBS probably has really terrible tail latencies.