Right. Yarn support is truly necessary feature. I haven't heard about Apache Slider, but it looks very useful. Thanks for your suggestion.
Sincerely, Jihoon 2014-11-11 16:21 GMT+09:00 Hyunsik Choi <[email protected]>: > Yes, index support and concurrent execution of multiple stages are > definitely important features. We should do them. Especially, I'm very > interested in indexing feature. > > In addition, we should do Yarn support. Now, we can employ Apache > Slider in order to launch a Tajo cluster as a long-running Yarn > application in a Yarn cluster. > > Best regards, > Hyunsik > > On Mon, Nov 10, 2014 at 6:11 PM, Jihoon Son <[email protected]> wrote: > > Hi Hyunsik, > > thanks for your mail. > > > > On 0.9.1 release, if we had more discussion on our mailing list, it would > > be better because our team members, (including who are not working with > you > > in offline) could concentrate on the issues which would be included in > > 0.9.1 release. > > The above my mail was just a suggestion. Since anyone didn't reply to > that, > > I couldn't know others' opinion. > > Anyway, you seem to agree with my suggestion except TAJO-982. > > If you want to have a validation on that issue, we can reschedule it to > > future releases. > > > > On other issues, you might talk about our next major release. > > Your suggestion is excellent. > > It makes Tajo more valuable, and specifically for storage layer > suggestion, > > it gives Tajo a great potential! > > But, I'd like to suggest that we need to resolve performance improvement > > issues as well as practical usage issues and new features. > > For example, TAJO-1130 (Concurrent execution of independent execution > > blocks) will give us significant performance improvement by exploiting a > > large degree of parallelism. > > In addition, we can also consider the index feature which I'm working on. > > > > Welcome any opinions. > > Please feel free to give your opinion. > > > > Sincerely, > > Jihoon > > > > 2014-11-11 9:35 GMT+09:00 Hyunsik Choi <[email protected]>: > > > >> Recently, Hyoungjun worked hbase storage, and he has refactored > >> storage module to be more general. With this work, we can support more > >> storage types. Especially, I believe that JDBC-based storages (i.e., > >> RDBMS as a storage) would be helpful to complement Tajo's nature which > >> does not support interactive insert/update storage. Also, Tajo makes > >> good use of indexing feature of RDBMSs. > >> > >> Best regards, > >> Hyunsik > >> > >> > >> On Mon, Oct 13, 2014 at 11:27 PM, Jihoon Son <[email protected]> > wrote: > >> > Hi Hyunsik, > >> > > >> > More frequent release sounds great! > >> > Here are the notable issues that are currently scheduled for 0.9.1. > >> > > >> > - TAJO-1095: implement JSON file scanner > >> > - TAJO-1026: implement query history persistency manager > >> > - TAJO-982: improve fetcher to get multiple shuffle outputs > through a > >> > request > >> > - TAJO-269: change the de/serialization protocol of logical plan of > >> > query unit attempt from Json to ProtocolBuffer > >> > - TAJO-235, TAJO-233: support oracle and postgresql catalog store > >> > > >> > IMO, if we resolve the above issues, it would be sufficient for the > next > >> > release. > >> > > >> > Thanks! > >> > Jihoon > >> > > >> > > >> > 2014-10-13 17:58 GMT+09:00 Hyunsik Choi <[email protected]>: > >> > > >> >> Hi folks, > >> >> > >> >> I'd like to discuss the next roadmap after 0.9.0 release. Currently, > I > >> >> already separated 0.9.0 branch from master branch. So, we can feel > >> >> free to proceed our work in master branch. > >> >> > >> >> Because the current Tajo is very stable, it would be great if we > >> >> release Tajo more frequently. It will gain more attention from this > >> >> field. > >> >> > >> >> So, I propose that we release 0.9.1 after about one month. I'd like > to > >> >> hear you guys' opinions. > >> >> > >> >> Also, let's talk about the roadmap of 0.9.1. > >> >> > >> >> - Hyunsik > >> >> > >> >
