ovirt-imageio for s390x - seems to need 4k support

I posted a patch[1] to add s390x build for ovirt-imageio, but the build reveal that direct I/O using 512 bytes alignment does not work on the test slave (or maybe this is a platform issue?). See failing tests: https://jenkins.ovirt.org/job/ovirt-imageio_standard-check-patch/414/console... *00:05:34.273* [check-patch.fc28.s390x] func = <built-in method readinto of _io.FileIO object at 0x3ff850a2048>*00:05:34.273* [check-patch.fc28.s390x] args = (<mmap.mmap object at 0x3ff85238618>,)*00:05:34.273* [check-patch.fc28.s390x] e = IOError(22, 'Invalid argument') Increasing the default block size fixes these tests, but cause other test to fail, because the code is not ready for 4k yet: https://jenkins.ovirt.org/job/ovirt-imageio_standard-check-patch/415/console... So adding s390x build requires more work. I open a bug[2] for this work. We should have 4k support in imageio in 4.3, but I'm not sure we will have time to work on this before that, unless someone else wants to contribute :-) [1] https://gerrit.ovirt.org/c/94095/ [2] https://bugzilla.redhat.com/1624963 Nir

On 03.09.2018 19:38, Nir Soffer wrote:
I posted a patch[1] to add s390x build for ovirt-imageio, but the build reveal that direct I/O using 512 bytes alignment does not work on the test slave (or maybe this is a platform issue?).
See failing tests: https://jenkins.ovirt.org/job/ovirt-imageio_standard-check-patch/414/console...
*00:05:34.273* [check-patch.fc28.s390x] func = <built-in method readinto of _io.FileIO object at 0x3ff850a2048>*00:05:34.273* [check-patch.fc28.s390x] args = (<mmap.mmap object at 0x3ff85238618>,)*00:05:34.273* [check-patch.fc28.s390x] e = IOError(22, 'Invalid argument')
Increasing the default block size fixes these tests, but cause other test to fail, because the code is not ready for 4k yet: https://jenkins.ovirt.org/job/ovirt-imageio_standard-check-patch/415/console...
So adding s390x build requires more work. I open a bug[2] for this work.
We should have 4k support in imageio in 4.3, but I'm not sure we will have time to work on this before that, unless someone else wants to contribute :-)
[1] https://gerrit.ovirt.org/c/94095/ [2] https://bugzilla.redhat.com/1624963
Nir
Hi Nir, on 390x we have two kinds of disks, SCSI and DASD, where DASDs are formatted using 4K sectors. From the failure we can see that the test system is obviously installed on DASD. I plan to have a look but can't promise exactly when... -- Regards, Viktor Mihajlovski
participants (2)
-
Nir Soffer
-
Viktor VM Mihajlovski