I'm not an expert on JVM atomicity, but i believe you are correct ... 
would you mind opening a bug and submitting a patch?

(i wonder if there are are there any other longs where we have the same problem)

: > public abstract class RequestHandlerBase implements SolrRequestHandler,
: > SolrInfoMBean {
: > 
: >   // statistics
: >   // TODO: should we bother synchronizing these, or is an off-by-one error
: >   // acceptable every million requests or so?
: >   long numRequests;
: >   long numErrors;
: Shouldn't numRequests and numErrors be volatile variables?  Writes are not
: atomic on longs unless they are volatile.
: 
: -Sean




-Hoss

Reply via email to