SERVER-32006 add a parameter to wait before secondaries perform noop write

6 files changed, 73 insertions, 10 deletions

diff --git a/src/mongo/db/read_concern.cpp b/src/mongo/db/read_concern.cpp
index 557c83c5f6a..4a11f36908f 100644
--- a/src/mongo/db/read_concern.cpp
+++ b/src/mongo/db/read_concern.cpp
@@ -102,6 +102,8 @@ private:
 };
 
 
+MONGO_EXPORT_SERVER_PARAMETER(waitForSecondaryBeforeNoopWriteMS, int, 10);
+
 /**
  *  Schedule a write via appendOplogNote command to the primary of this replica set.
  */
@@ -112,6 +114,20 @@ Status makeNoopWriteIfNeeded(OperationContext* opCtx, LogicalTime clusterTime) {
     auto& writeRequests = getWriteRequestsSynchronizer(opCtx->getClient()->getServiceContext());
 
     auto lastAppliedOpTime = LogicalTime(replCoord->getMyLastAppliedOpTime().getTimestamp());
+
+    // secondaries may lag primary so wait first to avoid unnecessary noop writes.
+    if (clusterTime > lastAppliedOpTime && replCoord->getMemberState().secondary()) {
+        auto deadline = Date_t::now() + Milliseconds(waitForSecondaryBeforeNoopWriteMS.load());
+        auto readConcernArgs =
+            repl::ReadConcernArgs(clusterTime, repl::ReadConcernLevel::kLocalReadConcern);
+        auto waitStatus = replCoord->waitUntilOpTimeForReadUntil(opCtx, readConcernArgs, deadline);
+        lastAppliedOpTime = LogicalTime(replCoord->getMyLastAppliedOpTime().getTimestamp());
+        if (!waitStatus.isOK()) {
+            LOG(1) << "Wait for clusterTime: " << clusterTime.toString()
+                   << " until deadline: " << deadline << " failed with " << waitStatus.toString();
+        }
+    }
+
     auto status = Status::OK();
     int remainingAttempts = 3;
     // this loop addresses the case when two or more threads need to advance the opLog time but the
diff --git a/src/mongo/db/repl/replication_coordinator.h b/src/mongo/db/repl/replication_coordinator.h
index e02a4efffae..403792a0f99 100644
--- a/src/mongo/db/repl/replication_coordinator.h
+++ b/src/mongo/db/repl/replication_coordinator.h
@@ -376,6 +376,16 @@ public:
                                           const ReadConcernArgs& settings) = 0;
 
     /**
+     * Waits until the deadline or until the optime of the current node is at least the opTime
+     * specified in 'settings'.
+     *
+     * Returns whether the wait was successful.
+     */
+    virtual Status waitUntilOpTimeForReadUntil(OperationContext* opCtx,
+                                               const ReadConcernArgs& settings,
+                                               boost::optional<Date_t> deadline) = 0;
+
+    /**
      * Retrieves and returns the current election id, which is a unique id that is local to
      * this node and changes every time we become primary.
      * TODO(spencer): Use term instead.
diff --git a/src/mongo/db/repl/replication_coordinator_impl.cpp b/src/mongo/db/repl/replication_coordinator_impl.cpp
index 27a10f50e1b..f50edc09f3c 100644
--- a/src/mongo/db/repl/replication_coordinator_impl.cpp
+++ b/src/mongo/db/repl/replication_coordinator_impl.cpp
@@ -1176,6 +1176,12 @@ Status ReplicationCoordinatorImpl::waitUntilOpTimeForRead(OperationContext* opCt
         return Status::OK();
     }
 
+    return waitUntilOpTimeForReadUntil(opCtx, readConcern, boost::none);
+}
+
+Status ReplicationCoordinatorImpl::waitUntilOpTimeForReadUntil(OperationContext* opCtx,
+                                                               const ReadConcernArgs& readConcern,
+                                                               boost::optional<Date_t> deadline) {
     if (getReplicationMode() != repl::ReplicationCoordinator::modeReplSet) {
         // For master/slave and standalone nodes, readAfterOpTime is not supported, so we return an
         // error. However, we consider all writes "committed" and can treat MajorityReadConcern as
@@ -1185,7 +1191,7 @@ Status ReplicationCoordinatorImpl::waitUntilOpTimeForRead(OperationContext* opCt
     }
 
     if (readConcern.getArgsClusterTime()) {
-        return _waitUntilClusterTimeForRead(opCtx, readConcern);
+        return _waitUntilClusterTimeForRead(opCtx, readConcern, deadline);
     } else {
         return _waitUntilOpTimeForReadDeprecated(opCtx, readConcern);
     }
@@ -1193,7 +1199,8 @@ Status ReplicationCoordinatorImpl::waitUntilOpTimeForRead(OperationContext* opCt
 
 Status ReplicationCoordinatorImpl::_waitUntilOpTime(OperationContext* opCtx,
                                                     bool isMajorityReadConcern,
-                                                    OpTime targetOpTime) {
+                                                    OpTime targetOpTime,
+                                                    boost::optional<Date_t> deadline) {
     if (!isMajorityReadConcern) {
         // This assumes the read concern is "local" level.
         // We need to wait for all committed writes to be visible, even in the oplog (which uses
@@ -1245,7 +1252,19 @@ Status ReplicationCoordinatorImpl::_waitUntilOpTime(OperationContext* opCtx,
         LOG(3) << "waitUntilOpTime: OpID " << opCtx->getOpID() << " is waiting for OpTime "
                << waiter << " until " << opCtx->getDeadline();
 
-        auto waitStatus = opCtx->waitForConditionOrInterruptNoAssert(condVar, lock);
+        auto waitStatus = Status::OK();
+        if (deadline) {
+            auto waitUntilStatus =
+                opCtx->waitForConditionOrInterruptNoAssertUntil(condVar, lock, *deadline);
+            if (!waitUntilStatus.isOK()) {
+                waitStatus = waitUntilStatus.getStatus();
+            }
+            // If deadline is set no need to wait until the targetTime time is reached.
+            return waitStatus;
+        } else {
+            waitStatus = opCtx->waitForConditionOrInterruptNoAssert(condVar, lock);
+        }
+
         if (!waitStatus.isOK()) {
             return waitStatus;
         }
@@ -1254,8 +1273,9 @@ Status ReplicationCoordinatorImpl::_waitUntilOpTime(OperationContext* opCtx,
     return Status::OK();
 }
 
-Status ReplicationCoordinatorImpl::_waitUntilClusterTimeForRead(
-    OperationContext* opCtx, const ReadConcernArgs& readConcern) {
+Status ReplicationCoordinatorImpl::_waitUntilClusterTimeForRead(OperationContext* opCtx,
+                                                                const ReadConcernArgs& readConcern,
+                                                                boost::optional<Date_t> deadline) {
     auto clusterTime = *readConcern.getArgsClusterTime();
     invariant(clusterTime != LogicalTime::kUninitialized);
 
@@ -1267,7 +1287,7 @@ Status ReplicationCoordinatorImpl::_waitUntilClusterTimeForRead(
     const bool isMajorityReadConcern =
         readConcern.getLevel() == ReadConcernLevel::kMajorityReadConcern;
 
-    return _waitUntilOpTime(opCtx, isMajorityReadConcern, targetOpTime);
+    return _waitUntilOpTime(opCtx, isMajorityReadConcern, targetOpTime, deadline);
 }
 
 // TODO: remove when SERVER-29729 is done
diff --git a/src/mongo/db/repl/replication_coordinator_impl.h b/src/mongo/db/repl/replication_coordinator_impl.h
index 69b57830d69..ff6053ce5a7 100644
--- a/src/mongo/db/repl/replication_coordinator_impl.h
+++ b/src/mongo/db/repl/replication_coordinator_impl.h
@@ -167,6 +167,10 @@ public:
     virtual OpTime getMyLastAppliedOpTime() const override;
     virtual OpTime getMyLastDurableOpTime() const override;
 
+    virtual Status waitUntilOpTimeForReadUntil(OperationContext* opCtx,
+                                               const ReadConcernArgs& readConcern,
+                                               boost::optional<Date_t> deadline) override;
+
     virtual Status waitUntilOpTimeForRead(OperationContext* opCtx,
                                           const ReadConcernArgs& readConcern) override;
 
@@ -1111,7 +1115,10 @@ private:
     /**
      * Waits until the optime of the current node is at least the 'opTime'.
      */
-    Status _waitUntilOpTime(OperationContext* opCtx, bool isMajorityReadConcern, OpTime opTime);
+    Status _waitUntilOpTime(OperationContext* opCtx,
+                            bool isMajorityReadConcern,
+                            OpTime opTime,
+                            boost::optional<Date_t> deadline = boost::none);
 
     /**
      * Waits until the optime of the current node is at least the opTime specified in 'readConcern'.
@@ -1122,11 +1129,13 @@ private:
                                              const ReadConcernArgs& readConcern);
 
     /**
-     * Waits until the optime of the current node is at least the clusterTime specified in
-     * 'readConcern'. Supports local and majority readConcern.
+     * Waits until the deadline or until the optime of the current node is at least the clusterTime
+     * specified in 'readConcern'. Supports local and majority readConcern.
+     * If maxTimeMS and deadline are both specified, it waits for min(maxTimeMS, deadline).
      */
     Status _waitUntilClusterTimeForRead(OperationContext* opCtx,
-                                        const ReadConcernArgs& readConcern);
+                                        const ReadConcernArgs& readConcern,
+                                        boost::optional<Date_t> deadline);
 
     /**
      * Returns a pseudorandom number no less than 0 and less than limit (which must be positive).
diff --git a/src/mongo/db/repl/replication_coordinator_mock.cpp b/src/mongo/db/repl/replication_coordinator_mock.cpp
index 1db09263d55..58a3efd664d 100644
--- a/src/mongo/db/repl/replication_coordinator_mock.cpp
+++ b/src/mongo/db/repl/replication_coordinator_mock.cpp
@@ -229,6 +229,11 @@ Status ReplicationCoordinatorMock::waitUntilOpTimeForRead(OperationContext* opCt
     return Status::OK();
 }
 
+Status ReplicationCoordinatorMock::waitUntilOpTimeForReadUntil(OperationContext* opCtx,
+                                                               const ReadConcernArgs& settings,
+                                                               boost::optional<Date_t> deadline) {
+    return Status::OK();
+}
 
 OID ReplicationCoordinatorMock::getElectionId() {
     // TODO
diff --git a/src/mongo/db/repl/replication_coordinator_mock.h b/src/mongo/db/repl/replication_coordinator_mock.h
index 4472247de44..12bf54c1f6c 100644
--- a/src/mongo/db/repl/replication_coordinator_mock.h
+++ b/src/mongo/db/repl/replication_coordinator_mock.h
@@ -132,6 +132,9 @@ public:
     virtual Status waitUntilOpTimeForRead(OperationContext* opCtx,
                                           const ReadConcernArgs& settings) override;
 
+    virtual Status waitUntilOpTimeForReadUntil(OperationContext* opCtx,
+                                               const ReadConcernArgs& settings,
+                                               boost::optional<Date_t> deadline) override;
     virtual OID getElectionId();
 
     virtual OID getMyRID() const;