On 4/7/2014 3:00 AM, Ralf Matulat wrote:
we are currently facing a new problem while reindexing one of our SOLR
4.4 instances:
We are using SOLR 4.4 getting data via DIH out of a MySQL Server.
The data is constantly growing.
We have reindexed our data a lot of times without any trouble.
The problem can be reproduced.
There is another server, configured exactly the same way (via git)
which was reindex 3 days ago against the same MySQL Server without
problems.
But: That server has more RAM and more powerfull CPUs as the one
making headaches today.
The error log says:
java.lang.ArrayIndexOutOfBoundsException
at
org.apache.lucene.util.packed.Packed64SingleBlock$Packed64SingleBlock4.get(Packed64SingleBlock.java:336)
at
org.apache.lucene.util.packed.GrowableWriter.get(GrowableWriter.java:56)
at
org.apache.lucene.util.packed.AbstractPagedMutable.get(AbstractPagedMutable.java:88)
at org.apache.lucene.util.fst.NodeHash.addNew(NodeHash.java:151)
at org.apache.lucene.util.fst.NodeHash.rehash(NodeHash.java:169)
at org.apache.lucene.util.fst.NodeHash.add(NodeHash.java:133)
at
org.apache.lucene.util.fst.Builder.compileNode(Builder.java:197)
at
org.apache.lucene.util.fst.Builder.freezeTail(Builder.java:289)
at org.apache.lucene.util.fst.Builder.add(Builder.java:394)
This looks a little bit like a problem that recently surfaced in
automated testing. That particular problem was caused by IBM's J9 Java
(based on JDK7) miscompiling a low-level lucene function.
Are you using a JVM from a vendor other than Oracle? At the moment, the
JVM recommendation is Oracle Java 7u25. When 7u60 comes out (expected
in May 2014), that will most likely be the recommended version. Are
there other differences between these two systems, like the garbage
collector being used, 32-bit vs. 64-bit, different max heap size,
running in a different servlet container, etc?
Are there any other errors, such as an OutOfMemory error?
I could be completely wrong with my guess.
Thanks,
Shawn
