try to fix rebalancing some more.

Msg5.cpp

@@ -1065,8 +1065,15 @@ bool Msg5::gotList2 ( ) {
 		// sanity check
 		//if ( KEYNEG(m_listPtrs[i]->getEndKey()) ) {
 		//	char *xx=NULL;*xx=0; }
-		if ( KEYCMP(m_listPtrs[i]->getEndKey(),m_minEndKey,m_ks)<0 ) 
+		if ( KEYCMP(m_listPtrs[i]->getEndKey(),m_minEndKey,m_ks)<0 ) {
 			KEYSET(m_minEndKey,m_listPtrs[i]->getEndKey(),m_ks);
+			// crap, if list is all negative keys, then the
+			// end key seems negative too! however in this
+			// case RdbScan::m_endKey seems positive so
+			// maybe we got a negative endkey in constrain?
+			//if (! (m_minEndKey[0] & 0x01) )
+			//	log("msg5: list had bad endkey");
+		}
 	}
 	// sanity check
 	//if ( KEYNEG( m_minEndKey) ) {char *xx=NULL;*xx=0; }
@@ -1393,7 +1400,7 @@ bool Msg5::gotList2 ( ) {
 	m_waitingForMerge = false;
 
 	// thread creation failed
-	if ( g_conf.m_useThreads )
+	if ( g_conf.m_useThreads && ! g_threads.m_disabled )
 		log(LOG_INFO,
 		    "net: Failed to create thread to merge lists. Doing "
 		    "blocking merge. Hurts performance.");

RdbBase.cpp

@@ -614,6 +614,17 @@ bool RdbBase::setFiles ( ) {
 			return false;
 	}
 
+	// everyone should start with file 0001.dat or 0000.dat
+	if ( m_numFiles > 0 && m_fileIds[0] > 1 ) {
+		log("db: missing file id 0001.dat for %s in coll %s. "
+		    "Fix this or it'll core later. Just rename the next file "
+		    "in line to 0001.dat/map. We probably cored at a "
+		    "really bad time during the end of a merge process.",
+		    m_dbname, m_coll );
+		char *xx=NULL; *xx=0;
+	}
+
+
 	m_dir.close();
 
 	// ensure files are sharded correctly
@@ -659,6 +670,7 @@ long RdbBase::addFile ( long id , bool isNew , long mergeNum , long id2 ,
 		    (long)MAX_RDB_FILES);
 		return -1;
 	}
+
 	// HACK: skip to avoid a OOM lockup. if RdbBase cannot dump
 	// its data to disk it can backlog everyone and memory will
 	// never get freed up.

RdbList.cpp

@@ -1631,9 +1631,12 @@ void RdbList::merge_r ( RdbList **lists         ,
 	// . we don't want any positive recs to go un annhilated
 	// . but don't worry about this check if start and end keys are equal
 	//if ( m_startKey != m_endKey && (m_endKey.n0 & 0x01) == 0x00 )
-	if ( KEYCMP(m_startKey,m_endKey,m_ks)!=0 && KEYNEG(m_endKey) )
+	if ( KEYCMP(m_startKey,m_endKey,m_ks)!=0 && KEYNEG(m_endKey) ) {
 		log(LOG_LOGIC,"db: rdblist: merge_r: Illegal endKey for "
-		    "merging.");
+		    "merging. fixing.");
+		// make it legal so it will be read first NEXT time
+		KEYSUB(m_endKey,1,m_ks);
+	}
 	// do nothing if no lists passed in
 	if ( numLists <= 0 ) return;
 	// inherit the key size of what we merge

RdbTree.cpp

@@ -3084,8 +3084,8 @@ void RdbTree::cleanTree ( ) { // char **bases ) {
 	if ( count == 0 ) return;
 	log(LOG_LOGIC,"db: Removed %li records from %s tree for invalid "
 	    "collection number %i.",count,m_dbname,collnum);
-	log(LOG_LOGIC,"db: Records not actually removed for safety. Except "
-	    "for those with negative colnums.");
+	//log(LOG_LOGIC,"db: Records not actually removed for safety. Except "
+	//    "for those with negative colnums.");
 	static bool s_print = true;
 	if ( ! s_print ) return;
 	s_print = false;

Rebalance.cpp

@@ -391,7 +391,7 @@ bool Rebalance::gotList ( ) {
 		return true;
 	}
 
-	char *last = NULL;
+	//char *last = NULL;
 
 	for ( ; ! m_list.isExhausted() ; m_list.skipCurrentRec() ) {
 		// get tht rec
@@ -399,9 +399,9 @@ bool Rebalance::gotList ( ) {
 		// get shard
 		long shard = getShardNum ( rdbId , rec );
 		// save last ptr
-		last = rec;
+		//last = rec;
+		m_list.getKey  ( rec , m_nextKey );
 		// debug!
-		//m_list.getKey  ( rec , m_nextKey );
 		//log("rebal: checking key %s",KEYSTR(m_nextKey,ks));
 		// count as scanned
 		m_scannedCount++;
@@ -445,12 +445,13 @@ bool Rebalance::gotList ( ) {
 	//log("rebal: done reading list");
 
 	//  update nextkey
-	if ( last ) {
+	//if ( last ) {
+	if ( ! m_list.isEmpty() ) {
 		// get the last key we scanned, all "ks" bytes of it.
 		// because some keys are compressed and we take the
 		// more significant compressed out bytes from m_list.m_*
 		// member vars
-		m_list.getKey  ( last , m_nextKey );
+		//m_list.getKey  ( last , m_nextKey );
 		// if it is not maxxed out, then incremenet it for the
 		// next scan round
 		if ( KEYCMP ( m_nextKey , KEYMAX() , ks ) != 0 )