On 6/5/2018 10:26 PM, TK Solr wrote:
I visualized the GC log with GCMV (GCVM?) and the graph shows Solr was
using less than half of the heap space at the peak.
This Solr doesn't get much query traffic and no indexing was running.
It's really a sudden death of JVM with no trace.
If you aren't concerned about what you see in a GC analysis, then the
heap may not be an issue. FYI, this is where I would have sent the log
once I got it:
http://gceasy.io/
This website does a VERY good job of detecting possible problems with
the heap and GC.
The only concern I have is that the Solr config files are that of Solr
5.x and they just upgraded to Solr 6.6. But I understand Solr 6
supports Solr 5 compatible mode. Has there been any issue in the
compatibility mode?
If the config was actually *designed* for 5.x, then it should have
little problem working in 6.x. If it was designed for an earlier
version and just happened to work in 5.x, then I would be less
optimistic about it working in 6.x. That said ... it is very unlikely
that anything in the index config files would cause crashes, even if
there is a compatibility problem.
The simple truth is that most Java software, including Solr, just
doesn't ever crash unless there's something VERY wrong.
Actual crashes do happen in the wild, they're just very rare. Extremely
severe memory starvation at the OS level can cause problems where
processes die without any logging, or the OS kills them explicitly. If
the java heap is properly sized for the system, that shouldn't be
possible. Since you're running Solr 6, you're running Java 8 minimum.
PermGen is gone in Java 8. Similar issues to what used to happen with
PermGen can still happen with the new piece called Metaspace, but if the
overall system config is good, that shouldn't be a problem.
Thanks,
Shawn
