An update on the timeline for Java 8. Oracle has announced the end of public updates to Java 7 in April of 2015. Java 8 has been in production settings for 6+ months and we have not seen issues with it like there were with Java 7 when it was first released (IIRC, no major Java bugs announced). Apache Lucene has just voted to move their trunk to Java 8. Apache Spark supports Java 8 lambdas. Considering Java 8 as the basis for a production runtime is no longer a theoretical discussion. It will still be some time before some of the subtasks that involve language level issues could be looked at, these depend on making Java 8 the minimum supported version, but others could conditionally enable code after runtime checks.
Considering Java 8 as a supported runtime, I have linked some related Hadoop JIRAs here.
HADOOP-10786: Until Hadoop fixes an internal detail of UserGroupInformation, UGI#reloginFromKeytab will silently fail. This is a blocker for deploying secure HBase. Our secure deployment model expects HBase daemons to login using keytabs and a silent failure to re-login will cause service failure as soon as the TGT expires. There is a patch available on the issue. We just need the Hadoop side to care enough about it to adjust minimum Java version policy on branch-2 and commit the change there. Or perhaps there could be an alternate fix that is compatible with Java 6, but I'm pessimistic about that and don't see the point of it either, 6 is EOL.
HADOOP-11098: The MaxDirectMemorySize default setting has changed to unlimited. The memory bean's getMax will return -1 to indicate this, which may cause one of our metrics to become invalid. More importantly we don't need to change the default if using off heap cache, so we should update relevant sections of hbase-env.sh and the online manual.