replication issue on slaves

[Michael Tsadikov <mi...@myheritage.com>]

Hi

We're using good old master-slave replication (not SolrCloud yet)

Since we've upgraded to solr 4.3, replication causes the slave JVM to
hiccup (probably heap & GC issues) *during the download phase*, which
seemed strange to me because download was only supposed to copy files from
the master to a local directory and not do anything else... (I would
understand if this had happened during reopening of index readers...)

I noticed in the log below that it uses NRTCachingDirectory, and it got me
thinking that maybe the slave is misconfigured since we don't need NRT at
all.

Aug 13, 2013 5:55:47 AM org.apache.solr.handler.SnapPuller fetchLatestIndex
INFO: Starting download to *NRTCachingDirectory*
(org.apache.lucene.store.MMapDirectory@/srv/solr/shard-3-0/data/index.20130813055547333
lockFactory=org.apache.lucene.store.NativeFSLockFactory@35463dc3;
maxCacheMB=48.0 maxMergeSizeMB=4.0) fullCopy=false
Aug 13, 2013 5:59:22 AM org.apache.solr.handler.SnapPuller fetchLatestIndex
INFO: Total time taken for download : 214 secs

Our solrconfig contains the following on both master and slaves, but I
thought it is irrelevant to slaves:

    <updateLog>
      <str name="dir">${solr.ulog.dir:}</str>
    </updateLog>

     <autoCommit>
       <maxTime>3600000</maxTime>
       <openSearcher>true</openSearcher>
     </autoCommit>

How can I make replication behave in slaves like it did in pre-4.0 versions?

Thanks,
Michael