On Sat, Apr 17, 2010 at 11:54 AM, Jonathan Gray <jg...@facebook.com> wrote: > Agreed that it's good to try to be agenda-less, but in the past we've always > taken the first couple hours to do a group discussion around some of the key > topics. Given there's a bunch of fairly major changes/testing going on these > days, I think there is a good bit of stuff that would benefit from group > discussion. After that, we can break up into smaller groups or individually > to start hacking away. Or for those not interested in the topics, you can > just hack from the start. >
The above sounds good, discussion of a few near future concerns. I think it important we not let it go on too long. All of the below look good. I've added a few comments. (Anyone else have suggestions on what should be discussed?) > More potential topics of discussion I had in mind: > > - Compaction, split, and flush policies/heuristics (HBASE-2453, HBASE-2462, > HBASE-2457, HBASE-2375, HBASE-1892, etc...) > - Define our desired behaviors related to versioning, deletes, and removal of > deletes in minor/major compactions. (HBASE-2453, HBASE-2457, HBASE-2243, etc) The new HBase ACID spec should get at least a passing airing (Its 'done'. All releases post 0.20.4 are required to adhere to it). > - Brainstorm on doing better distributed scenario testing (HBASE-2414) We also badly need to work on breaking down current daemons so their functions can be made more standalone and thus testable; e.g. load balancer in master, compacting code in regionservers, etc. > - Brainstorm on performance improvement ideas (top HDFS issues, better use of > HFile seeking, blooms, block pre-fetch, etc...) Would be cool to have a wiki > page w/ a list of these things. A wiki page of items to be discussed? > - Brainstorm on new functionality / updated road map. What priorities do the > various sponsoring companies have, what are nice to haves but not on anyones > schedule yet, etc. Again, this can seed a new (or updated) wiki page and/or > update the currently outdated road map wiki page. I moved aside the current 0.21 roadmap moving it to a page of its own rather than have it as a section on the roadmap page (http://wiki.apache.org/hadoop/HBase/Original021Roadmap#preview), and added comments on the state of roadmap items listed therein. A roadmap discussion could start with what is listed in the old roadmap, I'd suggest. I'd suggest that roadmap discussion should not go into deep depth because it has a tendency to consume mostly because it turns into a blue skying session and besides, 0.21 is fairly imminent and there isn't that much we could get into this release anyways. > - HBase PR. We could use a new web site (maven and otherwise), a centralized > blog, and also a refresh/cleanup of documentation. There's also agreement on > shipping w/ a few different configurations, which should be part of a new set > of getting started / new user docs. Would like to get everyones thoughts and > also come up with a schedule. On monday our petition to become an apache top level project goes before the apache board. If it passes, rolling out a site revamp might be timed to match our move to TLP. > - Ideas for future HUGs > > For anyone that will not be able to attend the hackathon we will post a > wrap-up afterwards with notes about all the discussions we had. Whatever > comes out of the hackathon should be posted into the proper jiras or mailing > list for full community discussion. > > Also, if anyone was not able to sign up for the HUG or Hackathon (both are > full now) and is a regular contributor, please contact me directly. > > Very awesome. Gonna be a great day of HBase! > Agreed. St.Ack > JG > > ________________________________________ > From: Andrew Purtell [apurt...@apache.org] > Sent: Saturday, April 17, 2010 10:28 AM > To: hbase-...@hadoop.apache.org > Cc: hbase-user@hadoop.apache.org > Subject: Hackathon agenda > > The Hackathon is basically agenda-less, but I'd like to propose a general > topic of discussion we should cover while we are all in the room together: > > - For HBASE-1964 (HBASE-2183, HBASE-2461, and related): injecting and/or > mocking exceptions thrown up from DFSClient. I think we want a toolkit for > that. Could be incorporated into the unit testing framework. Should be > possible to swap out a jar or something and make it active running on a real > cluster with real load. Should be possible to inject random exceptions with > adjustable probability. So what does HDFS have already? What do we need? If > we're adding something, does it make sense to put it into HBase or contribute > to HDFS? I think the latter. > > Let's gather a list of other topics, if any, that hackathon participants want > to see covered so we can make sure it will happen. > > - Andy > > > > >
