I think Elliot's concern with the histogram improvements is that if it goes in hadoop-common, HBase is only going to get it after it gets rolled into a new hadoop-common release and HBase bumps the version on the hadoop-common dependency. I don't think this is so bad.
For now though, I'm willing to maintain an HBase patchset against this future improved hadoop-common until it's actually released. Thanks, Andrew On Fri, Jun 29, 2012 at 5:31 PM, Andrew Purtell <andrew.purt...@gmail.com>wrote: > Metrics2 is in both Hadoop 1 and 2. (?) > > But my thing was the latter point below anyhow. > > On Jun 29, 2012, at 1:53 PM, Elliott Clark <ecl...@stumbleupon.com> wrote: > > > I'm excited about the new accurate histogram, however I don't think that > > we'll get it when we move to metrics2. Any direct usage would break > > backward compat for all older hadoop installs. > > For backwards compatibility on hadoop versions we will either have to > > create a shim layer with reflection, or pull the code in. Since we > already > > have a MeticsHistogram I think pulling the code into hbase may be the > right > > choice. > > > > On Fri, Jun 29, 2012 at 1:15 PM, Andrew Purtell <apurt...@apache.org> > wrote: > > > >> On Fri, Jun 29, 2012 at 1:10 PM, Alex Baranau <alex.barano...@gmail.com > > > >> wrote: > >>> Meanwhile I suggested taking on this one: > >>> https://issues.apache.org/jira/browse/HBASE-4050 separately/first. > What > >> do > >>> you think? > >> > >> +1 > >> > >> In related discussion, it looks to me like Stack thinks pulling in the > >> histogram accuracy improvement work via metrics2 is a good idea. > >> > >> Independent of this something I would like to be able to do is log > >> metrics to a file as well as Ganglia concurrently, and metrics2 can do > >> that. > >> > >> > >> Best regards, > >> > >> - Andy > >> > >> Problems worthy of attack prove their worth by hitting back. - Piet > >> Hein (via Tom White) > >> >
