---

On Sun, Jan 10, 2016 at 2:56 PM, Oved Ourfali <oourfali@redhat.com> wrote:

On Jan 10, 2016 3:37 PM, "Nir Soffer" <nsoffer@redhat.com> wrote:
>
> On Sun, Jan 10, 2016 at 2:57 PM, Oved Ourfali <oourfali@redhat.com> wrote:
> > Thanks for the summary!
> > See one comment inline.
> >
> > On Sun, Jan 10, 2016 at 2:49 PM, Yaniv Bronheim <ybronhei@redhat.com> wrote:
> >>
> >>
> >> (fromani, nsoffer, ybronhei, alitke)
> >>
> >>  - Removing xmlrpc for good - who should accept it? where do we stand with
> >> full jsonrpc client ? (we didn't get to any conclusions and said that we'll
> >> reraise this topic next week with pioter)
> >>
> >
> > With regards to that, in order to move to 3.6 cluster level, you MUST have
> > all hosts in jsonrpc protocol. So, we just need to make sure no piece of
> > code uses that explicitly, and if so move that to jsonrpc as well.
>
> I don't remember that this was discussed here, and storage never
> approved this change
> for 3.6.  We need to keep the xmlrpc option in 3.6, as a backup for
> jsonrpc issues.
>

No we don't. And we won't. We had it around for one version for this reason, but no need for more.
We had a bug about that, and communicated it to whomever is relevant.
We want to get rid of it entirely in 4.0, so 3.6 cluster won't work with it.
If you know of any issue then please fix it now, or open a bug about it.

> We just fixed couple of jsonrpc verbs that were returning True instead of
> the correct return value (caused by incorrect schema).
> - https://github.com/oVirt/vdsm/commit/0ca680700596564b4d6b0ef01ed4b0ae7c488de7
> - https://gerrit.ovirt.org/40402 VM.getDiskAlignment cannot be used in jsonrpc
>

That's great. As said earlier, if you know of others please open bugs.

> However this is not the topic of the discussion, we are discussion the next
> version (3.7/4.0). We are adding lot of new verbs as part of removing the spm,
> and we don't want to invest time in adding xmlrpc and vdsClient support.
>
> Example new verb merged recently:
> https://github.com/oVirt/vdsm/commit/bbbb72a192d8b54d21c8d65f6a10278404a966db
>
> In the new verbs, we cleaned up the api, so integer values are passed
> as integers,
> not as strings. Previously we use to require strings since xmlrpc did
> not support large
> numbers (> 2**31 - 1).
>
> So in the schema, we require now a uint:
>
> +##
> +# @CreateVolumeInfo:
> +#
> ...
> +# @virtual_size: The Volume size in bytes
> ...
> +# @initial_size: #optional If specified, the initial allocated size of volume
> +# in bytes. Allowed only when creating a thinly provisioned
> +# volume on block storage.
> +#
> +# Since: 4.18
> +##
> +{'type': 'CreateVolumeInfo',
> + 'data': {'sd_id': 'UUID', 'img_id': 'UUID', 'vol_id': 'UUID',
> + 'virtual_size': 'uint', 'vol_format': 'VolumeFormat',
> + 'disk_type': 'DiskType', 'description': 'str',
> + '*parent_img_id': 'UUID', '*parent_vol_id': 'UUID',
> + '*initial_size': 'uint'}}
>
> To support xmlrpc, we added this ugly code in bindingxmlrpc.py:
>
> + def sdm_create_volume(self, args):
> + validateArgTypes(args, [str, parse_json_obj])
> +
> + # Convert large integers to strings. The server's xmlrpc binding will
> + # restore them to their proper int types.
> + vol_info = args[1]
> + for param in 'virtual_size', 'initial_size':
> + if param in vol_info:
> + vol_info[param] = str(vol_info[param])
> +
> + res = self.s.sdm_create_volume(*args)
> + if res['status']['code']:
> + return res['status']['code'], res['status']['message']
> +
> + return 0, ''
> +
>
> To support vdsClient, we added this:
>
> + def sdm_create_volume(self, job_id, vol_info):
> + sdm = API.SDM()
> +
> + # As a workaround for the 32bit signed integer limitation of xmlrpc,
> + # allow large integers to be passed as strings. We convert them back
> + # to the correct type here.
> + for param in 'virtual_size', 'initial_size':
> + if param in vol_info:
> + vol_info[param] = int(vol_info[param])
> +
> + return sdm.create_volume(job_id, vol_info)
> +
>
> All this work is waste effort on our side.
>
> We should make a decision now - do we support xmlrpc in vdsm?
>
> I think we should not support it, after two version we support both
> xmlrpc and jsonrpc.

I agree. But again, 3.6 clusters will require jsonrpc. That's why in 4.0 xmlrpc will no longer be relevant.

>
> If we stop supporting xmlrpc, we must have a replacement for vdsClient,
>
> We need to have commnad line client, both for development, and for sos
> plugin. I think this patch, owned now by Piotr, is the best direction:
> https://gerrit.ovirt.org/35181/
>
> But we need to give this high priority, as having a command line client is
> a must for developing vdsm.
>
> Adding Aharon - supporting both xmlrpc and jsonrpc means we need to test
> everything twice,  I don't think Aharion will like to do that.
>
> Piotr, Francesco, Dan, Adam: your thoughts?
>

I would love to get rid of it but we need to make sure that none of the code is using

xmlrpc. I think that we need to prioritize tasks before we can stop using it.

It would be great to change format of the schema for new client and drop/replace

code which is still using xmlrpc.

I agree, it does not make sense to maintain both protocols.

Let's think for a minute how it will work with the customers.

If the move from RHEV 3 to 4 would be similar as from 2 to 3, when 4 is a complete new system and some migration path would be needed, I think it is ok to say that we stop supporting all the old hosts running RHEL6 that can do XML-RPC only. But 3.6 definitely should still support XML RPC, because some customers would continue running old hosts in old compatibility mode clusters. Because we support it.

https://access.redhat.com/support/policy/updates/rhev/

When working on a command line client, also keep in mind that customers would attempt using it on their production environment, even if it may create inconsistency with their engine database. It would be good to enable some kind of production mode, where the customer would have to hit a confirmation yes, if the command may affect the global setup (all storage operations). There is at least one customer that I know of that is using vdsClient in their production system (I protested as much as I could).

> >>
> >>  - Moving from nose to pytest - generally good approach to achieve. It
> >> requires some changes in current testlib.py code. must be an item for next
> >> major version (nir already managed to run most of the tests with it, and
> >> stated few gaps)
> >>
> >>  - Exception patches - still on progress, please review
> >> (https://gerrit.ovirt.org/48868)
> >>
> >>  - python3 effort to cover all asyncProc usage, and allowing utils import
> >> without having python3-cpopen - https://gerrit.ovirt.org/51421
> >> https://gerrit.ovirt.org/49441 . still under review
> >>
> >> We didn't take notes during that talk, so if I forgot to mention something
> >> I apologize. Feel free to reply and raise it
> >>
> >> Greetings,
> >>
> >> --
> >> Yaniv Bronhaim.
> >>
> >> _______________________________________________
> >> Devel mailing list
> >> Devel@ovirt.org
> >> http://lists.ovirt.org/mailman/listinfo/devel
> >
> >
> >
> > _______________________________________________
> > Devel mailing list
> > Devel@ovirt.org
> > http://lists.ovirt.org/mailman/listinfo/devel

--

--
 mku