SERVER-5706 if only slaveDelay'd nodes are available, use them

SERVER-4750 prevent syncing to secondaries with a slavedelay greater than yourself

Conflicts:

	db/repl/rs_initialsync.cpp
	jstests/replsets/slavedelay3.js

2 files changed, 62 insertions, 9 deletions

diff --git a/db/repl/rs_initialsync.cpp b/db/repl/rs_initialsync.cpp
index 112d7396bc5..1677bfc38c6 100644
--- a/db/repl/rs_initialsync.cpp
+++ b/db/repl/rs_initialsync.cpp
@@ -80,7 +80,7 @@ namespace mongo {
     }
 
     const Member* ReplSetImpl::getMemberToSyncTo() {
-        Member *closest = 0;
+
         bool buildIndexes = true;
 
         // wait for 2N pings before choosing a sync target
@@ -95,16 +95,31 @@ namespace mongo {
             buildIndexes = myConfig().buildIndexes;
         }
 
+        Member *closest = 0;
+
         // find the member with the lowest ping time that has more data than me
-        for (Member *m = _members.head(); m; m = m->next()) {
-            if (m->hbinfo().up() &&
-                // make sure members with buildIndexes sync from other members w/indexes
-                (!buildIndexes || (buildIndexes && m->config().buildIndexes)) &&
-                (m->state() == MemberState::RS_PRIMARY ||
-                 (m->state() == MemberState::RS_SECONDARY && m->hbinfo().opTime > lastOpTimeWritten)) &&
-                (!closest || m->hbinfo().ping < closest->hbinfo().ping)) {
-                closest = m;
+
+        // Make two attempts.  The first attempt, we ignore those nodes with
+        // slave delay higher than our own.  The second attempt includes such
+        // nodes, in case those are the only ones we can reach.
+        for (int attempts = 0; attempts < 2; ++attempts) {
+            for (Member *m = _members.head(); m; m = m->next()) {
+                if (m->hbinfo().up() &&
+                    // make sure members with buildIndexes sync from other members w/indexes
+                    (!buildIndexes || (buildIndexes && m->config().buildIndexes)) &&
+                    (m->state() == MemberState::RS_PRIMARY ||
+                     (m->state() == MemberState::RS_SECONDARY && 
+                      m->hbinfo().opTime > lastOpTimeWritten)) &&
+                    (!closest || m->hbinfo().ping < closest->hbinfo().ping)) {
+
+                    if ( attempts == 0 && 
+                         myConfig().slaveDelay < m->config().slaveDelay ) {
+                        break; // skip this one in the first attempt
+                    }
+                    closest = m;
+                }
             }
+            if (closest) break; // no need for second attempt
         }
 
         {
diff --git a/jstests/replsets/slavedelay3.js b/jstests/replsets/slavedelay3.js
new file mode 100644
index 00000000000..e89fe965f46
--- /dev/null
+++ b/jstests/replsets/slavedelay3.js
@@ -0,0 +1,38 @@
+load("jstests/replsets/rslib.js");
+
+var name = 'slavedelay3';
+var replTest = new ReplSetTest({ name: name, nodes: 3 });
+var nodes = replTest.startSet();
+var config = replTest.getReplSetConfig();
+// ensure member 0 is primary
+config.members[0].priority = 2;
+config.members[1].priority = 0;
+config.members[1].slaveDelay = 5;
+
+replTest.initiate(config);
+replTest.awaitReplication();
+replTest.bridge();
+
+
+var master = replTest.getMaster().getDB(name);
+ var slaveConns = replTest.liveNodes.slaves;
+ var slave = [];
+ for (var i in slaveConns) {
+     var d = slaveConns[i].getDB(name);
+     d.getMongo().setSlaveOk();
+     slave.push(d);
+ }
+
+waitForAllMembers(master);
+
+
+
+replTest.partition(0,2);
+replTest.awaitReplication();
+
+master.foo.insert({x:1});
+
+// make sure the record still appears in the remote slave
+assert.soon( function() { return slave[1].foo.findOne() != null; } );
+
+replTest.stopSet();
\ No newline at end of file