How implement rbd export/import using go-ceph?

[mysugar]

I did not found export func on rbd.Image stuct , any idea to implement this?

[agarwal-mudit]

Are you talking about this?

[nixpanic]

What kind of Export() interface do you expect? You already noticed you can use Read() and Write() to the image in #318.

[mysugar]

I mean the rbd export on linux cli

[root@nceph ~]# rbd info test/disk01
rbd image 'disk01':
	size 10GiB in 2560 objects
	order 22 (4MiB objects)
	block_name_prefix: rbd_data.10286b8b4567
	format: 2
	features: layering
	flags: 
	create_timestamp: Wed Jun 17 15:40:08 2020
[root@nceph ~]# rbd export test/disk01  /tmp/disk01.rbd
Exporting image: 100% complete...done.
[root@nceph ~]# du -sh /tmp/disk01.rbd 
639M	/tmp/disk01.rbd

If use Read() the size of /tmp/disk01.rbd will be 10G.

[nixpanic]

Ah, that probably uses rbd_read_iterate2() to skip writing the zero-filled areas of the image. We do not have that in go-ceph yet.

Or it could be used in combination with SeekHole and SeekData, but we do not have that yet either.

[phlogistonjohn]

We have issue #279 filed for rbd_read_iterate2. I don't think seek style interfaces are implemented (in librdb). Where they do exist, in cephfs, the Seek(Hole|Data) whence values don't actually seek the holes or data, they're just stubs (see issue #227).

In the short term I suggest doing a simple copy loop as @nixpanic suggests with read calls and perhaps do hole detection yourself (eg. read N-size blocks and if the block is all zeros seek forward in your destination file). In the future, when rbd_read_iterate2 or other hole sensitive calls are available use it as an optimization. It's not pretty but that's today's situation AFAICT.

[tareksha]

Hi @nixpanic, @phlogistonjohn, the export command in rbd cli does not only dump the sparse contents. It produces a file in a special format that also contains the image metadata and diffs.
https://docs.ceph.com/en/pacific/dev/rbd-export/
Is it withinbgo-ceph's scope to have a full implementation of export/import?