On 01/05/2012 04:08 AM, MORITA Kazutaka wrote:
>> int update_epoch_store(uint32_t epoch)
>> > {
>> > - char new[1024];
>> > -
>> > - snprintf(new, sizeof(new), "%s%08u/", obj_path, epoch);
>> > - mkdir(new, def_dmode);
>> > + if (!strcmp(store.name, "simple")) {
>> > + char new[1024];
>> >
>> > + snprintf(new, sizeof(new), "%s%08u/", obj_path, epoch);
>> > + mkdir(new, def_dmode);
>> > + }
>> > return 0;
>> > }
> IIUC, the farm driver updates epoch in farm_begin_recover(). Why do
> the two drivers update it in the different places?
I tried to remove this function and push it to begin_recover. In the end
I found that not only recovery but also join & format phase depends on
this function to update the store layout.
So I am going to leave this function for later clean-up patch, for this
patch set, I'll concentrate on farm core.
Thanks,
Yuan
|
|