Hey, thanks that solves this issue. Most documentation I find is marked as "deprecated", the storage documenation does not clearly reflect this. Anway, thanks for clearing this issue!

Once upon a time, m.skrzetuski(a)gmail.com <m.skrzetuski(a)gmail.com&gt; said: > I'd give up on the ISO domain. I started like you and then read the docs which said that ISO domain is deprecated. > I'd upload all files to a data domain. Note that that only works if your data domain is NFS... iSCSI data domains will let you upload ISOs, but connecting them to a VM fails.

Once upon a time, Nir Soffer <nsoffer(a)redhat.com&gt; said: > On Tue, Jan 7, 2020 at 4:02 PM Chris Adams <cma(a)cmadams.net&gt; wrote: > > Once upon a time, m.skrzetuski(a)gmail.com <m.skrzetuski(a)gmail.com&gt; said: > > > I'd give up on the ISO domain. I started like you and then read the docs > > which said that ISO domain is deprecated. > > > I'd upload all files to a data domain. > > > > Note that that only works if your data domain is NFS... iSCSI data > > domains will let you upload ISOs, but connecting them to a VM fails. > > ISO on iSCSI/FC domains works fine for starting a VM from ISO, which is the > main use case. Okay - it didn't the last time I tried it (I just got errors). Thanks.

On Fri, Jan 17, 2020 at 6:41 AM Strahil Nikolov <hunter86_bg(a)yahoo.com&gt; wrote: > On January 17, 2020 12:10:56 AM GMT+02:00, Chris Adams <cma(a)cmadams.net&gt; > wrote: > >Once upon a time, Nir Soffer <nsoffer(a)redhat.com&gt; said: > >> On Tue, Jan 7, 2020 at 4:02 PM Chris Adams <cma(a)cmadams.net&gt; wrote: > >> > Once upon a time, m.skrzetuski(a)gmail.com <m.skrzetuski(a)gmail.com&gt; > >said: > >> > > I'd give up on the ISO domain. I started like you and then read > >the docs > >> > which said that ISO domain is deprecated. > >> > > I'd upload all files to a data domain. > >> > > >> > Note that that only works if your data domain is NFS... iSCSI data > >> > domains will let you upload ISOs, but connecting them to a VM > >fails. > >> > >> ISO on iSCSI/FC domains works fine for starting a VM from ISO, which > >is the > >> main use case. > > > >Okay - it didn't the last time I tried it (I just got errors). Thanks. > > I have opened and RFE for ISO checksumming, as currently the uploader can > silently corrupt your DVD. > Can you share the bug number? > With gluster, I have an option to check the ISO checksum and > verify/replace the file, but with Block-based storage that will be quite > difficult. > Checksumming is a general feature not related to ISO uploads. But checksumming tools do not understand sparseness so you should really use a tool designed for compare disk images, like "qemu-img compare". Here is an example: 1. Create fedora 30 image for testing: $ virt-builder fedora-30 -o fedora-30.raw ... $ qemu-img info fedora-30.raw image: fedora-30.raw file format: raw virtual size: 6 GiB (6442450944 bytes) disk size: 1.15 GiB 2. Create a checksum of the image $ time shasum fedora-30.raw 991c2efee723e04b7d41d75f70d19bade02b400d fedora-30.raw real 0m14.641s user 0m12.653s sys 0m1.749s 3. Create compressed qcow2 image with same content $ qemu-img convert -f raw -O qcow2 -c fedora-30.raw fedora-30.qcow2 ... $ qemu-img info fedora-30.qcow2 image: fedora-30.qcow2 file format: qcow2 virtual size: 6 GiB (6442450944 bytes) disk size: 490 MiB cluster_size: 65536 Format specific information: compat: 1.1 lazy refcounts: false refcount bits: 16 corrupt: false This is typical file format used for publishing disk images. The contents of this image are the same as the raw version from the guest point of view. 3. Compare image content $ time qemu-img compare fedora-30.raw fedora-30.qcow2 Images are identical. real 0m4.680s user 0m4.273s sys 0m0.553s 3 times faster to compare 2 images with different format compared with creating a checksum of single image. Now lets see how we can use this to verify uploads. 4. Upload the qcow2 compressed image to a new raw disk (requires ovirt 4.4 alpah3): $ python3 upload_disk.py --engine-url https://engine/ --username admin@internal --password-file password \ --cafile ca.pem --sd-name nfs1-export2 --disk-format raw --disk-sparse fedora-30.qcow2 5. Download image to raw format: $ python3 download_disk.py --engine-url https://engine/ --username admin@internal --password-file password \ --cafile ca.pem --format raw f40023a5-ddc4-4fcf-b8e2-af742f372104 fedora-30.download.raw 6. Comparing original and downloaded images $ qemu-img compare fedora-30.qcow2 fedora-30.download.raw Images are identical. Back to the topic of ISO uploads to block storage. Block volumes in oVirt are always aligned to 128 MiB, so when you upload an image which is not aligned to 128 MiB, oVirt creates a bigger block device. The contents of the device after the image content are not defined, unless you zero this area during upload. The current upload_disk.py example does not zero the end of the device since the guest do not care about it, but this makes verifying uploads harder. The best way to handle this issue is to truncate the ISO image up to the next multiple of 128 MiB before uploading it: $ ls -l Fedora-Server-dvd-x86_64-30-1.2.iso -rw-rw-r--. 1 nsoffer nsoffer 3177185280 Nov 8 23:09 Fedora-Server-dvd-x86_64-30-1.2.iso $ python3 -c 'n = 3177185280 + 128 * 1024**2 - 1; print(n - (n % (128 * 1024**2)))' 3221225472 $ truncate -s 3221225472 Fedora-Server-dvd-x86_64-30-1.2.iso The contents of the iso image is the same as it will be on the block device after the upload, and uploading this image will zero the end of the device. If we upload this image, we can check the upload using qemu img compare. $ python3 upload_disk.py --engine-url https://engine/ --username admin@internal --password-file password \ --cafile ca.pem --sd-name iscsi-1 --disk-format raw Fedora-Server-dvd-x86_64-30-1.2.iso $ python3 download_disk.py --engine-url https://engine/ --username admin@internal --password-file password \ --cafile ca.pem --format raw 5f0b5347-bbbc-4521-9ca0-8fc17670bab0 iso.raw $ qemu-img compare iso.raw Fedora-Server-dvd-x86_64-30-1.2.iso Images are identical. This is not easy to use and requires knowledge about oVirt internals (128 MiB alignment), so I guess we need to make this simpler. Nir