Generalize recloning docs on initial oplog application SERVER-4270

Conflicts:

	db/repl.cpp
	db/repl.h
	db/repl/rs_sync.cpp

5 files changed, 48 insertions, 41 deletions

diff --git a/db/oplog.cpp b/db/oplog.cpp
index e248d5e11ec..151a4cd2269 100644
--- a/db/oplog.cpp
+++ b/db/oplog.cpp
@@ -480,6 +480,40 @@ namespace mongo {
         }
     }
 
+    bool shouldRetry(const BSONObj& o, const string& hn) {
+        OplogReader missingObjReader;
+
+        // we don't have the object yet, which is possible on initial sync.  get it.
+        log() << "replication info adding missing object" << endl; // rare enough we can log
+        uassert(15916, str::stream() << "Can no longer connect to initial sync source: " << hn, missingObjReader.connect(hn));
+
+        const char *ns = o.getStringField("ns");
+        // might be more than just _id in the update criteria
+        BSONObj query = BSONObjBuilder().append(o.getObjectField("o2")["_id"]).obj();
+        BSONObj missingObj;
+        try {
+            missingObj = missingObjReader.findOne(ns, query);
+        } catch(DBException& e) {
+            log() << "replication assertion fetching missing object: " << e.what() << endl;
+            throw;
+        }
+
+        if( missingObj.isEmpty() ) {
+            log() << "replication missing object not found on source. presumably deleted later in oplog" << endl;
+            log() << "replication object: " << o.getObjectField("o2").toString() << endl;
+            log() << "replication o object: " << o.getObjectField("o").toString() << endl;
+
+            return false;
+        }
+        else {
+            Client::Context ctx(ns);
+            DiskLoc d = theDataFileMgr.insert(ns, (void*) missingObj.objdata(), missingObj.objsize());
+            uassert(15917, "Got bad disk location when attempting to insert", !d.isNull());
+
+            return true;
+        }
+    }
+
     /** @param fromRepl false if from ApplyOpsCmd
         @return true if was and update should have happened and the document DNE.  see replset initial sync code.
      */
diff --git a/db/oplog.h b/db/oplog.h
index cc8b4d5d972..2401e98d5fb 100644
--- a/db/oplog.h
+++ b/db/oplog.h
@@ -212,6 +212,12 @@ namespace mongo {
      * take an op and apply locally
      * used for applying from an oplog
      * @param fromRepl really from replication or for testing/internal/command/etc...
+     * Returns if the op was an update that could not be applied (true on failure)
      */
     bool applyOperation_inlock(const BSONObj& op , bool fromRepl = true );
+
+    /**
+     * If applyOperation_inlock should be called again after an update fails.
+     */
+    bool shouldRetry(const BSONObj& op , const string& hn);
 }
diff --git a/db/repl.cpp b/db/repl.cpp
index b14034d308a..4d648204a92 100644
--- a/db/repl.cpp
+++ b/db/repl.cpp
@@ -851,7 +851,10 @@ namespace mongo {
 
     void ReplSource::applyOperation(const BSONObj& op) {
         try {
-            applyOperation_inlock( op );
+            bool failedUpdate = applyOperation_inlock( op );
+            if (failedUpdate && shouldRetry(op, hostName)) {
+                uassert(15914, "Failure retrying initial sync update", applyOperation_inlock(op));
+            }
         }
         catch ( UserException& e ) {
             log() << "sync: caught user assertion " << e << " while applying op: " << op << endl;;
@@ -1351,6 +1354,7 @@ namespace mongo {
                             setLastSavedLocalTs( nextLastSaved );
                         }
                     }
+
                     if( oplogReader.awaitCapable() && tailing )
                         okResultCode = 0; // don't sleep
                     syncedTo = nextOpTime;
diff --git a/db/repl.h b/db/repl.h
index 45036fa30d7..e44607919e8 100644
--- a/db/repl.h
+++ b/db/repl.h
@@ -134,7 +134,7 @@ namespace mongo {
     public:
         OplogReader oplogReader;
 
-        static void applyOperation(const BSONObj& op);
+        void applyOperation(const BSONObj& op);
         bool replacing; // in "replace mode" -- see CmdReplacePeer
         bool paired; // --pair in use
         string hostName;    // ip addr or hostname plus optionally, ":<port>"
diff --git a/db/repl/rs_sync.cpp b/db/repl/rs_sync.cpp
index 7f8522de44b..82c845e7145 100644
--- a/db/repl/rs_sync.cpp
+++ b/db/repl/rs_sync.cpp
@@ -58,7 +58,6 @@ namespace mongo {
 
         const string hn = source->h().toString();
         OplogReader r;
-        OplogReader missingObjReader;
         try {
             if( !r.connect(hn) ) {
                 log() << "replSet initial sync error can't connect to " << hn << " to read " << rsoplog << rsLog;
@@ -137,44 +136,8 @@ namespace mongo {
 
                     if( ts >= applyGTE ) { // optimes before we started copying need not be applied.
                         bool failedUpdate = syncApply(o);
-                        if( failedUpdate ) {
-                            // we don't have the object yet, which is possible on initial sync.  get it.
-                            log() << "replSet info adding missing object" << endl; // rare enough we can log
-                            if( !missingObjReader.connect(hn) ) { // ok to call more than once
-                                log() << "replSet initial sync fails, couldn't connect to " << hn << endl;
-                                return false;
-                            }
-                            const char *ns = o.getStringField("ns");
-                            BSONObj query = BSONObjBuilder().append(o.getObjectField("o2")["_id"]).obj(); // might be more than just _id in the update criteria
-                            BSONObj missingObj;
-                            try {
-                                missingObj = missingObjReader.findOne(
-                                    ns,
-                                    query );
-                            } catch(...) {
-                                log() << "replSet assertion fetching missing object" << endl;
-                                throw;
-                            }
-                            if( missingObj.isEmpty() ) {
-                                log() << "replSet missing object not found on source. presumably deleted later in oplog" << endl;
-                                log() << "replSet op: " << o.toString() << endl;
-                            }
-                            else {
-                                Client::Context ctx(ns);
-                                try {
-                                    DiskLoc d = theDataFileMgr.insert(ns, (void*) missingObj.objdata(), missingObj.objsize());
-                                    assert( !d.isNull() );
-                                } catch(...) {
-                                    log() << "replSet assertion during insert of missing object" << endl;
-                                    throw;
-                                }
-                                // now reapply the update from above
-                                bool failed = syncApply(o);
-                                if( failed ) {
-                                    log() << "replSet update still fails after adding missing object " << ns << endl;
-                                    assert(false);
-                                }
-                            }
+                        if( failedUpdate && shouldRetry(o, hn)) {
+                            uassert(15915, "replSet update still fails after adding missing object", syncApply(o));
                         }
                     }
                     _logOpObjRS(o);   /* with repl sets we write the ops to our oplog too */