Public bug reported: Description =========== When we spawn instances without cache enabled (cache='none') on a file system there a check in nova code that test if file system support direct IO: https://github.com/openstack/nova/blob/master/nova/privsep/utils.py#L34 Because this test use 512b alignment size it seems to failed on newer block device that have logical block size > 512b like nvme:
parted /dev/nvme0n1 print | grep "Sector size" Sector size (logical/physical): 4096B/4096B reason should be that alignement size of direct io must be a multiple of logical block size of underlying device (not of fs block size) as explain here: http://man7.org/linux/man-pages/man2/open.2.html O_DIRECT ... Under Linux 2.4, transfer sizes, and the alignment of the user buffer and the file offset must all be multiples of the logical block size of the filesystem. Since Linux 2.6.0, alignment to the logical block size of the underlying storage (typically 512 bytes) suffices Because this test failed, it fallbacks value of cache to "writethrough" which have following consequences: 1) qemu run without direct io even device/fs support but with higher block size 2) qemu failed to start because cache=writethrough may conflict with other dev paramer like "io=native": with the following message: 2018-08-22 20:50:41.226 80512 ERROR oslo_messaging.rpc.server File "/usr/lib/python2.7/dist-packages/libvirt.py", line 1065, in createWithFlags 2018-08-22 20:50:41.226 80512 ERROR oslo_messaging.rpc.server if ret == -1: raise libvirtError ('virDomainCreateWithFlags() failed', dom=self) 2018-08-22 20:50:41.226 80512 ERROR oslo_messaging.rpc.server libvirtError: unsupported configuration: native I/O needs either no disk cache or directsync cache mode, QEMU will fallback to aio=threads Steps to reproduce ================== to reproduce spawn issue: having instances on fs with block device with logical block size > 512b (typically nvme with 4096 8192 sector size) nova.conf with: images_type=raw preallocate_images=space Solution ======== Can we consider increasing align_size from 512b to 8192b as it will work on most cases? Is there any other reason to keep 512b ? Set it to 4096 or 8192 fix the issue in my environment. Environment =========== I met the issue on newton, but same check with 512b exists on master. ** Affects: nova Importance: Undecided Status: New ** Summary changed: - Spawn may failed when cache=none on block device with logical block size > 512 + Spawn may fail when cache=none on block device with logical block size > 512 -- You received this bug notification because you are a member of Yahoo! Engineering Team, which is subscribed to OpenStack Compute (nova). https://bugs.launchpad.net/bugs/1801702 Title: Spawn may fail when cache=none on block device with logical block size > 512 Status in OpenStack Compute (nova): New Bug description: Description =========== When we spawn instances without cache enabled (cache='none') on a file system there a check in nova code that test if file system support direct IO: https://github.com/openstack/nova/blob/master/nova/privsep/utils.py#L34 Because this test use 512b alignment size it seems to failed on newer block device that have logical block size > 512b like nvme: parted /dev/nvme0n1 print | grep "Sector size" Sector size (logical/physical): 4096B/4096B reason should be that alignement size of direct io must be a multiple of logical block size of underlying device (not of fs block size) as explain here: http://man7.org/linux/man-pages/man2/open.2.html O_DIRECT ... Under Linux 2.4, transfer sizes, and the alignment of the user buffer and the file offset must all be multiples of the logical block size of the filesystem. Since Linux 2.6.0, alignment to the logical block size of the underlying storage (typically 512 bytes) suffices Because this test failed, it fallbacks value of cache to "writethrough" which have following consequences: 1) qemu run without direct io even device/fs support but with higher block size 2) qemu failed to start because cache=writethrough may conflict with other dev paramer like "io=native": with the following message: 2018-08-22 20:50:41.226 80512 ERROR oslo_messaging.rpc.server File "/usr/lib/python2.7/dist-packages/libvirt.py", line 1065, in createWithFlags 2018-08-22 20:50:41.226 80512 ERROR oslo_messaging.rpc.server if ret == -1: raise libvirtError ('virDomainCreateWithFlags() failed', dom=self) 2018-08-22 20:50:41.226 80512 ERROR oslo_messaging.rpc.server libvirtError: unsupported configuration: native I/O needs either no disk cache or directsync cache mode, QEMU will fallback to aio=threads Steps to reproduce ================== to reproduce spawn issue: having instances on fs with block device with logical block size > 512b (typically nvme with 4096 8192 sector size) nova.conf with: images_type=raw preallocate_images=space Solution ======== Can we consider increasing align_size from 512b to 8192b as it will work on most cases? Is there any other reason to keep 512b ? Set it to 4096 or 8192 fix the issue in my environment. Environment =========== I met the issue on newton, but same check with 512b exists on master. To manage notifications about this bug go to: https://bugs.launchpad.net/nova/+bug/1801702/+subscriptions -- Mailing list: https://launchpad.net/~yahoo-eng-team Post to : yahoo-eng-team@lists.launchpad.net Unsubscribe : https://launchpad.net/~yahoo-eng-team More help : https://help.launchpad.net/ListHelp
