On 4/11/06, Alan Robertson <[EMAIL PROTECTED]> wrote:
> Andrew Beekhof wrote:
> > On 4/6/06, Alan Robertson <[EMAIL PROTECTED]> wrote:
> >> Huang Zhen wrote:
> >>> The GUI client and CIM need management library to implement batch
> >>> operation.
> >>> It means that user can add a set of operations but not need do it 
> >>> actually,
> >>> then commit all these operations in one command.
> >>>
> >>> To do this we need cib's help.
> >>> With current cib API all changes are sent cib daemon immeidately.
> >>> Like
> >>> "
> >>>     rc = cib_conn->cmds->update(
> >>>             cib_conn, "crm_config", fragment, &output, 0);
> >>> "
> >>>
> >>> Can we implement a kind of copied cib to do the batch operation?
> >>>
> >>> The management library will create a copy of cib when user begins the
> >>> batch operation.
> >>> Then all changes would be sent to this copy.
> >>> When user wants to commit, send the xml in local cib copy to cib daemon.
> >>>
> >>> We can reuse the current cib code. Two functions may be enough:
> >>> copy_cib_conn() and commit_cib_conn().
> >>>
> >>> And your suggestions on this batch operations?
> >>
> >> To me, this sounds reasonable.  Andrew?
> >
> > You can do that already... but what if the CIB changed in the meantime?
> >
> > And before someone says locking...
> > What if a transition is in progress?
> > What if a failure occurs while the CIB is being modified?
> > Should I deny the updates from resource actions?
>
> Resource actions should only affect the status section.
>
> Ditto for transitions.

so?  they're updating the whole CIB.

and what about node attribute changes for things like ipfail?
they don't have to go in the status section.

>
> Any particular kind of failure did you have in mind above?
_______________________________________________________
Linux-HA-Dev: [email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/

Reply via email to