2017-11-15 20:59 GMT+01:00 Max Reitz <mreitz(a)redhat.com&gt;: > Well, you can't just add support to qemu-img alone either. Every image > format supported by qemu-img is one supported by qemu as a whole and > thus needs a proper block driver that needs to support random accesses > as well. I was talking about qemu-img convert, just to convert an XVA image to something different, in a single pass, without having to extract the tar.

On Wed, Nov 15, 2017 at 09:07:12PM +0100, Max Reitz wrote: > On 2017-11-15 21:06, Gandalf Corvotempesta wrote: >> 2017-11-15 20:59 GMT+01:00 Max Reitz <mreitz(a)redhat.com&gt;: >>> Well, you can't just add support to qemu-img alone either. Every image >>> format supported by qemu-img is one supported by qemu as a whole and >>> thus needs a proper block driver that needs to support random accesses >>> as well. >> >> I was talking about qemu-img convert, just to convert an XVA image to >> something different, in a single pass, without having to extract the >> tar. > > I know, but that doesn't work. qemu-img convert uses the normal > general-purpose block drivers for that. In any case there's no need as qemu/qemu-img can already access files inside an uncompressed tarball using the offset/size support added to the raw driver exactly for this purpose: https://lists.gnu.org/archive/html/qemu-devel/2016-10/msg03945.html

Gandalf, is there an XVA file publically available (pref. not too big) that we can look at?

This will give you one offset/size/filename tuple for each file. The plugin will then simply need to calculate which file to access to resolve each virtual file range (or substitute zeroes for missing files).

On 2017-11-15 21:41, Gandalf Corvotempesta wrote: > 2017-11-15 21:29 GMT+01:00 Richard W.M. Jones <rjones(a)redhat.com&gt;: >> Gandalf, is there an XVA file publically available (pref. not >> too big) that we can look at? > > I can try to provide one, but it's simple: > > # tar tvf 20160630_124823_aa72_.xva.gz | head -n 50 > ---------- 0/0 42353 1970-01-01 01:00 ova.xml > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000000 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000000.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000001 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000001.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000003 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000003.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000004 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000004.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000005 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000005.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000006 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000006.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000007 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000007.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000009 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000009.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000010 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000010.checksum > > > You can ignore the ova.xml and just use the "Ref:175" directory. > Inside the XVA you'll fine one "Ref" directory for each virtual disk > (ref number is different for each disk) > Inside each Ref directory, you'll get tons of 1MB file, corrisponding > to the RAW image. > You have to merge these files in a single raw file with just an > exception: numbers are not sequential. > as you can see above, we have: 00000000, 00000001, 00000003 > > The 00000002 is missing, because it's totally blank. XenServer doesn't > export any empty block, thus it will skip the corrisponding 1MB file. > When building the raw image, you have to replace empty blocks with 1MB > full of zeros. > > This is (in addition to the tar extract) the most time-consuming part. > Currently I'm rebuilding a 250GB image, with tons of files to be > merge. > If qemu-img can be patched to automatically convert this kind of > format, I can save about 3 hours (30 minutes for extracting the > tarball, and about 2 hours to merge 250-300GB image) Hmmm... OK, so as Rich said, you can use the raw driver with an offset and a size. And you can use the null-co driver to emulate holes. And qemu-img convert supports concatenation. Taking all of that together, it should be possible to create an input "file" per 1 MB block and concatenate all of those with qemu-img convert. The only issue is that it'll get you a *really* long command line. (And you'll still have to gunzip the archive before, of course.) I'll try to write a script, but of course it's hart without an example... Max

I'm thinking on how to prove you a sample XVA I have to create (and populate) a VM because an empty image will result in an empty XVA And a VM is 300-400Mb as minimum Il 15 nov 2017 10:30 PM, "Max Reitz" <mreitz(a)redhat.com&gt; ha scritto: > On 2017-11-15 21:41, Gandalf Corvotempesta wrote: > > 2017-11-15 21:29 GMT+01:00 Richard W.M. Jones <rjones(a)redhat.com&gt;: > >> Gandalf, is there an XVA file publically available (pref. not > >> too big) that we can look at? > > > > I can try to provide one, but it's simple: > > > > # tar tvf 20160630_124823_aa72_.xva.gz | head -n 50 > > ---------- 0/0 42353 1970-01-01 01:00 ova.xml > > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000000 > > ---------- 0/0 40 1970-01-01 01:00 > Ref:175/00000000.checksum > > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000001 > > ---------- 0/0 40 1970-01-01 01:00 > Ref:175/00000001.checksum > > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000003 > > ---------- 0/0 40 1970-01-01 01:00 > Ref:175/00000003.checksum > > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000004 > > ---------- 0/0 40 1970-01-01 01:00 > Ref:175/00000004.checksum > > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000005 > > ---------- 0/0 40 1970-01-01 01:00 > Ref:175/00000005.checksum > > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000006 > > ---------- 0/0 40 1970-01-01 01:00 > Ref:175/00000006.checksum > > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000007 > > ---------- 0/0 40 1970-01-01 01:00 > Ref:175/00000007.checksum > > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000009 > > ---------- 0/0 40 1970-01-01 01:00 > Ref:175/00000009.checksum > > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000010 > > ---------- 0/0 40 1970-01-01 01:00 > Ref:175/00000010.checksum > > > > > > You can ignore the ova.xml and just use the "Ref:175" directory. > > Inside the XVA you'll fine one "Ref" directory for each virtual disk > > (ref number is different for each disk) > > Inside each Ref directory, you'll get tons of 1MB file, corrisponding > > to the RAW image. > > You have to merge these files in a single raw file with just an > > exception: numbers are not sequential. > > as you can see above, we have: 00000000, 00000001, 00000003 > > > > The 00000002 is missing, because it's totally blank. XenServer doesn't > > export any empty block, thus it will skip the corrisponding 1MB file. > > When building the raw image, you have to replace empty blocks with 1MB > > full of zeros. > > > > This is (in addition to the tar extract) the most time-consuming part. > > Currently I'm rebuilding a 250GB image, with tons of files to be > > merge. > > If qemu-img can be patched to automatically convert this kind of > > format, I can save about 3 hours (30 minutes for extracting the > > tarball, and about 2 hours to merge 250-300GB image) > > Hmmm... OK, so as Rich said, you can use the raw driver with an offset > and a size. And you can use the null-co driver to emulate holes. And > qemu-img convert supports concatenation. > > Taking all of that together, it should be possible to create an input > "file" per 1 MB block and concatenate all of those with qemu-img > convert. The only issue is that it'll get you a *really* long command > line. > > (And you'll still have to gunzip the archive before, of course.) > > I'll try to write a script, but of course it's hart without an example... > > Max > >

https://stacklet.com/downloads/XenServer-XVA-Template-Debian-7.8-Lightwei...

On 2017-11-15 22:50, Gandalf Corvotempesta wrote: > https://stacklet.com/downloads/XenServer-XVA-Template-Debian-7.8-Lightwei... Hmmm, that gives me a timeout when downloading. And another image I got gives me an "argument list too long" -- which I had feared already... It has a size of 3 MB and Linux "only" allows 2...

https://xanclic.moe/convert-xva.rb -- does this work? (It seems to works on the two example images I found...) An example is in the code, you use it like this: $ ./convert-xva.rb ~/Downloads/stats-appliance-2.36.020502.xva Ref:73

On 2017-11-16 11:08, Gandalf Corvotempesta wrote: > 2017-11-15 23:55 GMT+01:00 Max Reitz <mreitz(a)redhat.com&gt;: >> https://xanclic.moe/convert-xva.rb -- does this work? >> (It seems to works on the two example images I found...) >> >> An example is in the code, you use it like this: >> >> $ ./convert-xva.rb ~/Downloads/stats-appliance-2.36.020502.xva Ref:73 > > > It doesn't work on huge images: > > # time convert-xva.rb 20171115_193814_efff_.xva Ref:10 -O qcow2 disk1.qcow2 > /usr/bin/convert-xva.rb:119:in `exec': Argument list too long - > qemu-img (Errno::E2BIG) > from /usr/bin/convert-xva.rb:119:in `<main>' > > real 9m41.414s > user 0m19.280s > sys 0m3.260s Well, there is this: https://github.com/XanClic/qemu.rb/blob/master/examples/convert-xva-onlin... That should get around that restriction, but currently it is stupidly slow (because it just issues all 1 MB copy operations simultaneously), and it doesn't use qemu-img convert. Instead, it uses a real qemu with block jobs. Ideally, you'd use it like this: $ ./convert-xva-online.rb ~/Downloads/stats-appliance-2.36.020502.xva Ref:73 81936 M $ qemu-img create -f qcow2 [whatever options you need] disk1.qcow2 \ 81936M Formatting 'disk1.qcow2', fmt=qcow2 size=85916123136 encryption=off cluster_size=65536 lazy_refcounts=off refcount_bits=16 $ ./convert-xva-online.rb ~/Downloads/stats-appliance-2.36.020502.xva \ Ref:73 \ '{"driver":"qcow2", "detect-zeroes":"unmap", "discard":"unmap", "file":{"driver":"file","filename":"disk1.qcow2"}}' Adding block devices... Starting block jobs... 100.00 % of jobs completed (Maybe I can get it faster.) Max

<fs> ll /

2017-11-16 11:01 GMT+01:00 Richard W.M. Jones <rjones(a)redhat.com&gt;: > As mentioned before you can use this to do a qemu-img convert using > captive nbdkit: > > $ nbdkit -U - \ > perl script=./xva-reader.pl file=./debian8cloud.xva size=4294967296 \ > --run 'qemu-img convert -f raw $nbd -O qcow2 /var/tmp/output.qcow2 -p' What if XVA is hosting 2 or more partitions ? Which partition are you converting with this command ?

On Wed, 15 Nov 2017 21:41:20 +0100 Gandalf Corvotempesta <gandalf.corvotempesta(a)gmail.com&gt; wrote: > 2017-11-15 21:29 GMT+01:00 Richard W.M. Jones <rjones(a)redhat.com&gt;: > > Gandalf, is there an XVA file publically available (pref. not > > too big) that we can look at? > > I can try to provide one, but it's simple: > > # tar tvf 20160630_124823_aa72_.xva.gz | head -n 50 > ---------- 0/0 42353 1970-01-01 01:00 ova.xml > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000000 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000000.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000001 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000001.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000003 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000003.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000004 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000004.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000005 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000005.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000006 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000006.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000007 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000007.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000009 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000009.checksum > ---------- 0/0 1048576 1970-01-01 01:00 Ref:175/00000010 > ---------- 0/0 40 1970-01-01 01:00 Ref:175/00000010.checksum > > > You can ignore the ova.xml and just use the "Ref:175" directory. > Inside the XVA you'll fine one "Ref" directory for each virtual disk > (ref number is different for each disk) > Inside each Ref directory, you'll get tons of 1MB file, corrisponding > to the RAW image. > You have to merge these files in a single raw file with just an > exception: numbers are not sequential. > as you can see above, we have: 00000000, 00000001, 00000003 > > The 00000002 is missing, because it's totally blank. XenServer doesn't > export any empty block, thus it will skip the corrisponding 1MB file. > When building the raw image, you have to replace empty blocks with 1MB > full of zeros. > > This is (in addition to the tar extract) the most time-consuming part. > Currently I'm rebuilding a 250GB image, with tons of files to be > merge. > If qemu-img can be patched to automatically convert this kind of > format, I can save about 3 hours (30 minutes for extracting the > tarball, and about 2 hours to merge 250-300GB image) > Hi, I like what Richard and Max came up with. Pretty nifty solutions. While there's nothing wrong with them I decided to take my own shot at it. Since the blocks in tar file are pieces of raw image there is no conversion happening. All in all it's just moving bytes from one place to another. That means there shouldn't be a need for any heavy machinery, right? :) Attached is a shell script that uses dd to do the byte-shuffling. I'm pretty sure this could even be safely parallelized by running multiple instances of dd at the same time (e.g. with xargs). But I did not try that.