On 28/03/2018, Mark Nelson wrote: [snip] > I'm not sure they have to be mutually exclusive. It could be that we > implement the obectstore log interface and bluestore more or less just sends > stuff to rocksdb anyway to piggyback the log write on the transaction and > avoid the extra seek. I guess the point though would be that everything > becomes more generic and the objecstore handles the logginess. Oh, sure. Whatever happens INSIDE bluestore is bluestore's business. That should just be subject to whatever works and gives the kind of performance we want. I just want to make sure we /have/ the interface so we don't face this problem again in two years. > I guess the thing that scares me the most is that my eyes glaze over when I > look at the pg log code and I'm super afraid to touch it. Maybe someone > else will be less scared of it? :D I am in /some/ sympathy with this feeling. One /could/ argue that that, in and of itself, is an argument that the whole PGLog code needs to be caressed by the Grand Refactoring Scalpel of Exquisite Compassion. But, yes. A group effort and all makes sense and calling on our bravest coders. So long as we're all on the same page and decide what the end goal /should/ be it'll be better than various independent skunkworksy hacks. -- Senior Software Engineer Red Hat Storage, Ann Arbor, MI, US IRC: Aemerson@OFTC, Actinic@Freenode 0x80F7544B90EDBFB9 E707 86BA 0C1B 62CC 152C 7C12 80F7 544B 90ED BFB9 -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
