Thanks for the info Andy. I'm using this:
java version "1.6.0_20" OpenJDK Runtime Environment (IcedTea6 1.9.8) (rhel-1.22.1.9.8.el5_6-x86_64) OpenJDK 64-Bit Server VM (build 19.0-b09, mixed mode) On CentOS 5.6. I switched from (Sun) 1.6.0_u14 as an experiment a while back and I've never had reason to switch back. For reference, it's a 12-node cluster with ~6.4 billion HBase records, ~8-10 million inserts and a number of Hive queries per day. Not big as far as HBase clusters go, but it does see some use. Thanks, Sandy > -----Original Message----- > From: Andrew Purtell [mailto:[email protected]] > Sent: Monday, September 12, 2011 15:45 > To: [email protected] > Subject: Re: scanner deadlock? > > > From: Sandy Pratt <[email protected]> > > TLDR: OpenJDK ~= Oracle JDK, so why not use it? > > > This advice is given out of an abundance of caution. Some have been burned > in production by bad JVM versions in the past. Oracle's 1.6.0_u18 is a > particularly egregious example, it will segfault all over the place under > load. 1.6.0_u14 was widely used once. Now u21+ seems a good option given > the JIT and GC bugfixes that went in. IIRC, with u20 or previous you need to > enable -XX:+UseMembar to avoid a JIT bug that will cause object monitors to > miss wakeups every once in a while. I currently use Oracle 1.6.0_u26. > > Given the history of JVM issues, we do not automatically trust any given > version. None of the committers run HBase in production on OpenJDK that I > know of, so its suitability is unknown. I'm glad to hear it works for you. > What > specific version are you using? > > Best regards, > > > - Andy > > Problems worthy of attack prove their worth by hitting back. - Piet Hein (via > Tom White)
