[Linux-ha-dev] Re: [Linux-ha-cvs] Linux-HA CVS: lib by sunjd from

[Sun Jiang Dong]

Alan Robertson wrote:

> Huang Zhen wrote:
> 
>> Lars Marowsky-Bree wrote:
>>
>>> On 2004-09-13T10:45:41,
>>>    linux-ha-cvs at lists.community.tummy.com said:
>>>
>>>
>>>> linux-ha CVS committal
>>>>
>>>> Author  : sunjd
>>>> Host    : Project : linux-ha
>>>> Module  : lib
>>>>
>>>> Dir     : linux-ha/lib/plugins/lrm
>>>>
>>>>
>>>> Modified Files:
>>>>     raexecocf.c
>>>>
>>>> Log Message:
>>>> * Export OCF_RESOURCE_INSTANCE before executing an OCF RA * Export 
>>>> OCF global environment variables before getting metadata
>>>>
>>>> ===================================================================
>>>> RCS file: /home/cvs/linux-ha/linux-ha/lib/plugins/lrm/raexecocf.c,v
>>>> retrieving revision 1.30
>>>> retrieving revision 1.31
>>>> diff -u -3 -r1.30 -r1.31
>>>> --- raexecocf.c    13 Sep 2004 08:30:51 -0000    1.30
>>>> +++ raexecocf.c    13 Sep 2004 10:45:41 -0000    1.31
>>>> @@ -66,11 +66,13 @@
>>>> #endif
>>>>
>>>> /* The begin of exported function list */
>>>> -static int execra(const char * rsc_type,
>>>> +static int execra(const char * rsc_id,
>>>> +          const char * rsc_type,
>>>>           const char * provider,
>>>>           const char * op_type,
>>>>            GHashTable * params);
>>>
>>>
>>>
>>>
>>> Maybe it would be easier by now if we passed in the full lrm_op_t
>>> here...?
>>>
>>> (Just thinking out loud)
>>>
>>>
>>> Sincerely,
>>>     Lars Marowsky-Brée <lmb at suse.de>
>>>
>> We want the plugins of RA are independent with LRM. So third party can 
>> add other RA class easily.
> 
> 
> My first reaction to this is that either I disagree or I don't understand.
> 
> These plugins are largely for the LRM.  By their nature, they're fairly 
> specialized for the LRM - aren't aren't likely to be used with some 
> other software package besides the LRM.
> 
> And, third parties can easily add new RA classes whether the information 
> is passed in arguments or the same information is passed in a 
> structure.  This attribute of the solution is unaffected by this.
> 
> Where it does come up to be different is this:
> 
> If you have third-parties supplying these resource agents, they are more 
> likely to have to be recompiled if the structure changes for reasons 
> unrelated to the interface to the plugins.  So, the interface between 
> the LRM and the RAs will change more often (in effect) if you put the 
> whole structure in there.  But as the number of elements being passed as 
> parameters goes up, the probability of this kind of false interface 
> churn goes down, and the balance shifts to the side of preferring to 
> pass the whole structure.
> 
> On the other hand, it would be a easier to use, including easier to 
> initially write 3rd party plugins if the interface had fewer parameters 
> to it.
> 
> It is rarely an improvement to replace passing a structure with the list 
> of variables in that structure passed explicitly.
> 
> I suspect that we are unlikely to get many (if any) 3rd party plugins.  
> In this case, it's far more likely for us to write them than for someone 
> else to do so and package them outside of Linux-HA.
> 
It's a good suggestion.
Formerly we just want to make the data structure dependency is single 
direction: lrm -> RA plugin, as the single dependency direction on function.


-- 
BRs,

Sun Jiang Dong