On 06/02/09 20:00, Liane Praza wrote:
> Sangeeta Misra wrote:
>> On 06/01/09 19:06, michael schuster wrote:
>>>
>>>>> Is it possible for ilbd daemon to wipe out the configuration   via 
>>>>> some scf function , *before* invoking  smf_maintain_instance() .
>>>>> Then the user could just manually do  svcadm enable <ilb service> 
>>>>> to get the system running ILB with and empty config?
>>>>>
>>>>
>>>> Alan
>>>> Any comment on the above? If yes,how can the daemon wipe out the 
>>>> config ?
>>>
>>> Sangeeta,
>>>
>>> IIRC we preferred the daemon to NOT wipe the existing config, for 
>>> debug purposes.
>>>
>> Ah I remember. So daemon invokes  smf_maintain_instance() to put the 
>> service in maintanance mode. THen we want the user to use some non- 
>> ilbadm command to wipe out the ilb config in SCF and then issue the 
>> svcadm enable  <ilb service manually>
>>
>> So any imput on what is the correct  scf command  to wipe out ilb 
>> config in SCF?
>
> Take a look at svccfg selectsnap and revert.  They're documented in 
> the manpage.  I believe there's a procedure for using them in the 
> System Administration Guide as well, but could be wrong.
>
Will do
> I'd really expect that reverting to the snapshot would be an 
> exceptional operation, not a normal one though.

Correct  - the above discussion was about what to do if ilbd experiences 
a fatal error. In such a case, we want the ilbd daemon to automatically 
make the ILB service go into maintanence mode. The user  would send us  
configuration, and a description of what happened, so we can work on the 
problem.

But while the user waits for the problem to be identified, they *may* 
wish to run the ILB service from scratch. Hence the question about 
wiping out the config  via SCF.

Sangeeta
> Most configuration errors are recovered from by identifying for the 
> user the part of their config that's broken and having them repair the 
> individual broken pieces, not by asking the user to undo all of their 
> customizations and start from scratch.
>
> liane


Reply via email to