----- Original Message -----
From: "Shubhendu Tripathi" <shtripat(a)redhat.com>
To: oliel(a)redhat.com, "Michael Pasternak" <mpastern(a)redhat.com>,
"Sahina Bose" <sabose(a)redhat.com>
Cc: "Dusmant Pati" <dpati(a)redhat.com>, "engine-devel"
<engine-devel(a)ovirt.org>
Sent: Monday, November 11, 2013 12:00:34 PM
Subject: Re: [Engine-devel] REST-API: Problem with additional DELETE action at collection
level
Self and Ori had a detailed discussion on the topic.
Points discussed -
1. Ori mentioned that Michael and Ori did not mean to have a separate
DELETE action for the purpose of commit in previous discussions.
That's right - no need for a new delete signature; existing signatures are enough.
2. Ori tried to understand the intention behind the separate commit
command in gluster and suggested that ideally gluster should remember
that a migration of data is started on the set of bricks, and if there
is a delete fired on the same bricks it should perform a commit action
because commit is nothing but a delete action
3. Ori suggested that in an ideal situation APIs needed would be
- start migrate
- stop migrate
- delete bricks
- retain bricks
4. As suggested by Ori, the remove bricks action should internally
decide if there was a data migration, started on the set of bricks. If
so, it should effectively fire a commit for the set of bricks. And if
there was not occurrence of data migration it should behave like a
normal force remove action.
Ori, please add your comments if I have missed out anything here.
I agree with what you wrote, let me sort of say it again in my words:
I don't like requiring two consecutive commands from the API user, to perform migrate.
If the user only does 'migrate', and doesn't do the second step, we are in a
state of
corruption. But Shubhendu explained to me that there's no way around this; the
administrator *must* take a look and decide; it is simply not possible to make this
decision in advance. So I accepted this heavy-heartedly.
Then we were left with the decision of how to model this two step operation. I tried to
look at it from the point of view of the user; what would be the simplest way to 'tell
him
the story?'
- If you want to migrate a brick, run 'migrate' on it.
- If you want to stop migration of the brick, run 'stopMigrate' on it.
- If you want to resume the migration of the brick, simply run 'migrate' on it
again.
After migration is done:
- if you want to remove the brick, DELETE it (regular delete).
- if you want to keep using the brick, run 'reactivate' on it.
And that's the whole story.
The advantages are:
* In simple English, not bound to Gluster terms, the administrator has to decide whether
to
get rid of the brick, remove it, when migration is done. So what would be more natural
than
to delete it? It makes sense, and there's no need for a new 'action'. On the
Gluster side,
if user tried to delete a brick which is undergoing migration - simply don't allow it.
This
is something you have to do anyway; a user might try to delete a brick which is
undergoing
migration and you have to stop him from doing so. So 0 extra work here.
* Instead of stopMigrate meaning two different things in two different contexts, we now
have stopMigration mean exactly what its name suggests: it stops the migration, and
'reactive'
mean exactly what it suggests - mark this brick as active again, allow it to be used.
Sahina, request your comments on the same.
Thanks and Regards,
Shubhendu
On 11/11/2013 10:30 AM, Shubhendu Tripathi wrote:
On 11/10/2013 11:06 PM, Moti Asayag wrote:
>
> ----- Original Message -----
>> From: "Shubhendu Tripathi" <shtripat(a)redhat.com>
>> To: "engine-devel" <engine-devel(a)ovirt.org>, "Michael
Pasternak"
>> <mpastern(a)redhat.com>, oliel(a)redhat.com
>> Sent: Friday, November 8, 2013 8:37:30 AM
>> Subject: [Engine-devel] REST-API: Problem with additional DELETE
>> action at collection level
>>
>> Hi All,
>>
>> There is a DELETE action defined at collection level for Gluster
>> Bricks with
>> signature -
>>
>> @DELETE
>> public Response remove ( GlusterBricks bricks );
>>
>> Recently we had needed a commit action to remove migrated bricks.
>> After multiple rounds of discussion on introducing a commit action
>> to remove
>> migrated bricks we introduced a DELETE action [1] which accepts a
>> boolean
>> parameter isForce.
>> If the parameter is set to true , forced deletion of bricks happens
>> without
>> any data migration.
>> And if the parameter is not set or set to false, the deletion is
>> meant for a
>> brick on which migration has already taken place.
>>
>> To achieve the above functionality we introduced another DELETE
>> action with
>> new signature as below and also marked the first action as deprecated -
>>
>> @DELETE
>> public Response remove ( Action action );
>>
>> The problem arises now as the new api works fine for all possible
>> scenarios
>> with below input structure -
>>
>> <action>
>> <force>true/false</force>
>> <bricks>
>> <brick>
>> <name> brick1-name</name>
>> <name>brick2-name</name>
>> </brick>
>> </bricks>
>> </action>
>>
>> BUT after these change backward compatibility is broken and the old
>> api does
>> not work.
>> If we try invoking old DELETE with bricks as input parameter as
>> below, it
>> still invokes the new api and gives an error saying " Invalid
>> parameter ".
>>
> Maybe I've missed it some where, but i wasn't able to find the new
> 'force' parameter
> in the rsdl, and specifically not in optionalArguments list of:
>
> - name:
> /api/clusters/{cluster:id}/glustervolumes/{glustervolume:id}/bricks|rel=delete
>
> and perhaps the correct approach will be to deprecate this signature
> and introduce a new
> one with the 'force' in the 'mandatoryArguments' section.
Moti, I have introduced another methods remove of DELETE type which
takes Action as input.
I pass list of bricks and force as parameter in the same Action.
Also, I tried marking the old method deprecated and introduced the new
one BUT as mentioned above, after introduction of new DELETE with
Action parameter old one STOPS working.
Is it OK to stop working for a method if its deprecated?? I don't feel
so. Please comment.
>
>> <bricks>
>> <brick id="brick1-id"/>
>> <brick id="brick2-id"/>
>> </bricks>
>>
>> Kindly suggest a solution around the same.
>>
>> PS: Both the actions are defined at collection level (
>> /api/clusters/<cluster-id>/glustervolumes/<volume-id>/bricks )
>>
>> [1]:
http://gerrit.ovirt.org/#/c/21043/
>>
>> Thanks and Regards,
>> Shubhendu
>>
>>
>>
>>
>>
>> _______________________________________________
>> Engine-devel mailing list
>> Engine-devel(a)ovirt.org
>>
http://lists.ovirt.org/mailman/listinfo/engine-devel
>>
_______________________________________________
Engine-devel mailing list
Engine-devel(a)ovirt.org
http://lists.ovirt.org/mailman/listinfo/engine-devel