David, Sorry for the late answer. :)
I suggest that we should go with the simplest approach that could work for now and change things in the future. As Lars suggested we can specify these dependencies using a configuration property (whirr.role-order). We need to be sure that the post-configure scripts are executed in the expected order (I believe this means changing ConfigureClusterAction). On Thu, Feb 3, 2011 at 12:48 AM, David Alves <[email protected]> wrote: > Hi > > Ok, I've start reading code :) > About WHIRR-221, some questions (the first batch, please bear with me > if the questions are too basic or obvious): > I know it somewhat breaks encapsulation between roles but some roles > have dependencies right? > e.g. - HBase depends on nn,dn and zookeeper. > Couldn't these dependencies be explicit per role? > We could then calculate start order based on the dependency graph > starting root nodes first and progressing up to the leaves, booting multiple > non-dependent roles per level. > > Anyway and pertaining to what Tibor Kiss commented in JIRA on this > issue is there a requirement to also serialize ConfigureClusterAction? > > Regards > David > > > > On Feb 2, 2011, at 10:04 AM, Andrei Savu wrote: > >> See bellow. >> >> On Wed, Feb 2, 2011 at 11:06 AM, David Alves <[email protected]> wrote: >> >>> Hi Andrei >>> >>> Great!! >>> I had already started browsing the project, looks good! >>> I think I'll start with WHIRR-221 just to get my feet wet and then >>> move on to WHIRR-214. >>> >> >> Awesome! >> >> >>> Some questions: >>> - Testing - beyond unit tests the wiki suggests iTests must be run >>> on a cloup provider. Do you have amazon (or other provider's) credits? do >>> you use your company(ies) credits? >>> >> >> I am using my credit card now, it's cheap enough but unfortunately slow. I'm >> trying to build a local Eucalyptus based private cloud at my university, >> they have some unused hardware. >> >> >>> - About WHIRR-214 the issue mentions adding/removing zookeeper >>> nodes, but zookeeper doesn't have this feature as of yet. >>> Are you thinking of a static zk ensemble (dynamic only for the other >>> services)? >>> >> >> You could resize a ZK ensemble by shutting down all the nodes and >> reconfiguring them. This is less than ideal but it could work. >> >> >>> >>> Thanks >>> david >>> >>> >>> >>> On Feb 1, 2011, at 10:08 PM, Andrei Savu wrote: I had already started >>> browsing the project, looks good! >>> >>> >>>> Hi David, >>>> >>>> Welcome :) >>>> >>>> You should start by reading: >>>> >>>> https://cwiki.apache.org/confluence/display/WHIRR/How+To+Contribute >>>> >>>> checking the sources from svn (or git), setting the IDE and running the >>> test >>>> suite. >>>> >>>> Here are two interesting unassigned issues that you could work on: >>>> >>>> https://issues.apache.org/jira/browse/WHIRR-221 >>>> https://issues.apache.org/jira/browse/WHIRR-214 >>>> >>>> What research project are you working on? >>>> >>>> Feel free to ask more questions :) >>>> >>>> On Tue, Feb 1, 2011 at 11:52 PM, David Alves <[email protected]> >>> wrote: >>>> >>>>> Hi >>>>> >>>>> I have some cloud/hadoop/hbase experience and I'd like to >>>>> contribute to whirr, if possible. >>>>> The project sounds very interesting and very inline with my >>> current >>>>> research. >>>>> Any outstanding unassigned issues I can start with, with a little >>>>> mentorship? >>>>> >>>>> Thanks >>>>> David >>>> >>>> >>>> >>>> >>>> -- >>>> Andrei Savu -- andreisavu.ro >>> >>> >> >> >> -- >> Andrei Savu -- andreisavu.ro > > -- Andrei Savu -- andreisavu.ro
