The general feedback I've gotten when asking about this has been HDD-based clusters are not well-tested any longer due to resource allocation of the remaining HDD servers. I also feel this is quite unfortunate, as most large non-HPC clusters will likely still be HDD based for the significant cost advantages, potentially with DB/WAL on NVME/SSD.
Beyond just detecting issues such as those you have found in your testing (nice work), performance testing for regression/improvement is important, as well.
David
On Wed, Nov 30, 2022, at 00:58, Gregory Farnum wrote:
[ Moving to Ceph dev list instead of kernel ceph-devel. ]On Mon, Nov 28, 2022 at 6:51 PM Satoru Takeuchi<satoru.takeuchi@xxxxxxxxx> wrote:>> Hi,>> I shared my knowledge about the bugs which my team encountered and two> proposals to improve the official QA process in Ceph Virtual.>>> One of my proposals depends on the assumption that the official QA> doesn't use HDD.> It's because my team revealed many bugs in OSD creation/restart only> in HDD-based clusters.> Could you tell me whether my assumption is correct or not?It's possible to run the test on hard drives, and developers stillschedule them there sometimes, but doing so is quite expensive at thispoint (just because it takes a while to set up hard-drive based testnodes and takes longer to get a suite run completed against yourbranch) and so most of the lab is SSD/NVME based.I'm not sure if we run any HDD-based tests prior to releases or on aregular basis; I think Yuri can answer that question.-Greg>> Thanks,> Satoru>_______________________________________________Dev mailing list -- dev@xxxxxxxTo unsubscribe send an email to dev-leave@xxxxxxx
_______________________________________________ Dev mailing list -- dev@xxxxxxx To unsubscribe send an email to dev-leave@xxxxxxx
