The FUSE module does not (currentl) support O_DIRECT on files.
This prevents QEMU from starting guests using 'cache=none' for
their disks located on a GlusterFS filesystem. The same also
applies for a handful of other filesystems (notably tmpfs, or
any other FUSE filesystem).
This patch introduces a startup check in Nova compute service
which tries to create a file $instances_path/.direct_io.test
using the O_DIRECT flag. If this succeeds, then cache=none
will be used for all disks, otherwise it will fallback to
using cache=writethrough. While the latter does not have
performance which is as consistent as cache=none, it is still
host-crash safe and preserves data integrity with migration,
if the filesystem is cache coherant (cluster filesystems like
GlusterFS are, NFS by constrast is not).
By doing the dynamic check for O_DIRECT, we ensure that if
future FUSE modules gain O_DIRECT support, Nova will automatically
do the right thing.
* nova/tests/test_libvirt.py: Stub out os.open in
the _check_xml_and_disk_driver() to enable testing of
both O_DIRECT and non-O_DIRECT code paths
* nova/tests/test_virt_drivers.py: Set instances_path to
the current directory
* nova/virt/libvirt.xml.template: Replace hardcoded 'none'
string with the '$cachemode' variable for all disks.
Add missing 'cache' attribute for the config disk
* nova/virt/libvirt/connection.py: Check whether O_DIRECT
is supported on the "FLAGS.instances_path" directory
and use 'none' for cachemode if it is, 'writethrough'
otherwise
Bug: 959637
Change-Id: I60cbff1c3ad8299fe2aa37099390f9235f6724d0
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
d1a1feb91f