SERVER-28865 Replace ReplicationExecutor's prng in ReplicationCoordinatorImpl with an explicit prng.

4 files changed, 25 insertions, 11 deletions

diff --git a/src/mongo/db/repl/replication_coordinator_impl.cpp b/src/mongo/db/repl/replication_coordinator_impl.cpp
index 64888f7fa89..e538718650c 100644
--- a/src/mongo/db/repl/replication_coordinator_impl.cpp
+++ b/src/mongo/db/repl/replication_coordinator_impl.cpp
@@ -320,7 +320,8 @@ ReplicationCoordinatorImpl::ReplicationCoordinatorImpl(
       _sleptLastElection(false),
       _canAcceptNonLocalWrites(!(settings.usingReplSets() || settings.isSlave())),
       _canServeNonLocalReads(0U),
-      _storage(storage) {
+      _storage(storage),
+      _random(prngSeed) {
 
     invariant(_service);
 
@@ -336,7 +337,7 @@ ReplicationCoordinatorImpl::ReplicationCoordinatorImpl(
     _externalState->setupNoopWriter(kNoopWriterPeriod);
 }
 
-ReplicationCoordinatorImpl::~ReplicationCoordinatorImpl() {}
+ReplicationCoordinatorImpl::~ReplicationCoordinatorImpl() = default;
 
 void ReplicationCoordinatorImpl::waitForStartUpComplete_forTest() {
     _waitForStartUpComplete();
@@ -370,7 +371,8 @@ Date_t ReplicationCoordinatorImpl::getElectionTimeout_forTest() const {
 }
 
 Milliseconds ReplicationCoordinatorImpl::getRandomizedElectionOffset_forTest() {
-    return _getRandomizedElectionOffset();
+    stdx::lock_guard<stdx::mutex> lk(_mutex);
+    return _getRandomizedElectionOffset_inlock();
 }
 
 boost::optional<Date_t> ReplicationCoordinatorImpl::getPriorityTakeover_forTest() const {
@@ -3786,5 +3788,9 @@ ReplicationCoordinatorImpl::_cancelElectionIfNeeded_inTopoLock() {
     return _electionFinishedEvent;
 }
 
+int64_t ReplicationCoordinatorImpl::_nextRandomInt64_inlock(int64_t limit) {
+    return _random.nextInt64(limit);
+}
+
 }  // namespace repl
 }  // namespace mongo
diff --git a/src/mongo/db/repl/replication_coordinator_impl.h b/src/mongo/db/repl/replication_coordinator_impl.h
index fd37c6a7a47..04c22e2e829 100644
--- a/src/mongo/db/repl/replication_coordinator_impl.h
+++ b/src/mongo/db/repl/replication_coordinator_impl.h
@@ -49,6 +49,7 @@
 #include "mongo/db/storage/snapshot_name.h"
 #include "mongo/executor/task_executor.h"
 #include "mongo/platform/atomic_word.h"
+#include "mongo/platform/random.h"
 #include "mongo/platform/unordered_map.h"
 #include "mongo/platform/unordered_set.h"
 #include "mongo/stdx/condition_variable.h"
@@ -808,7 +809,7 @@ private:
      * Return a randomized offset amount that is scaled in proportion to the size of the
      * _electionTimeoutPeriod. Used to add randomization to an election timeout.
      */
-    Milliseconds _getRandomizedElectionOffset();
+    Milliseconds _getRandomizedElectionOffset_inlock();
 
     /**
      * Helper for _handleHeartbeatResponse.
@@ -1233,6 +1234,11 @@ private:
      */
     Status _waitUntilClusterTimeForRead(OperationContext* opCtx, LogicalTime clusterTime);
 
+    /**
+     * Returns a pseudorandom number no less than 0 and less than limit (which must be positive).
+     */
+    int64_t _nextRandomInt64_inlock(int64_t limit);
+
     //
     // All member variables are labeled with one of the following codes indicating the
     // synchronization rules for accessing them.
@@ -1430,6 +1436,9 @@ private:
     // Cached copy of the current config protocol version.
     AtomicInt64 _protVersion;  // (S)
 
+    // Source of random numbers used in setting election timeouts, etc.
+    PseudoRandom _random;  // (M)
+
     // This setting affects the Applier prefetcher behavior.
     mutable stdx::mutex _indexPrefetchMutex;
     ReplSettings::IndexPrefetchConfig _indexPrefetchConfig =
diff --git a/src/mongo/db/repl/replication_coordinator_impl_elect.cpp b/src/mongo/db/repl/replication_coordinator_impl_elect.cpp
index a1074150b03..be7d6ea79d7 100644
--- a/src/mongo/db/repl/replication_coordinator_impl_elect.cpp
+++ b/src/mongo/db/repl/replication_coordinator_impl_elect.cpp
@@ -173,7 +173,7 @@ void ReplicationCoordinatorImpl::_onFreshnessCheckComplete() {
             break;
         case FreshnessChecker::FreshnessTie:
             if ((_selfIndex != 0) && !_sleptLastElection) {
-                const auto ms = Milliseconds(_replExecutor->nextRandomInt64(1000) + 50);
+                const auto ms = Milliseconds(_nextRandomInt64_inlock(1000) + 50);
                 const Date_t nextCandidateTime = now + ms;
                 log() << "possible election tie; sleeping " << ms << " until "
                       << dateToISOStringLocal(nextCandidateTime);
@@ -245,7 +245,7 @@ void ReplicationCoordinatorImpl::_onElectCmdRunnerComplete() {
               << " votes, but needed at least " << _rsConfig.getMajorityVoteCount();
         // Suppress ourselves from standing for election again, giving other nodes a chance
         // to win their elections.
-        const auto ms = Milliseconds(_replExecutor->nextRandomInt64(1000) + 50);
+        const auto ms = Milliseconds(_nextRandomInt64_inlock(1000) + 50);
         const Date_t now(_replExecutor->now());
         const Date_t nextCandidateTime = now + ms;
         log() << "waiting until " << nextCandidateTime << " before standing for election again";
diff --git a/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp b/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp
index 68e1a61c4ac..99b39be285f 100644
--- a/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp
+++ b/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp
@@ -69,7 +69,7 @@ MONGO_FP_DECLARE(blockHeartbeatReconfigFinish);
 
 using executor::RemoteCommandRequest;
 
-Milliseconds ReplicationCoordinatorImpl::_getRandomizedElectionOffset() {
+Milliseconds ReplicationCoordinatorImpl::_getRandomizedElectionOffset_inlock() {
     long long electionTimeout = durationCount<Milliseconds>(_rsConfig.getElectionTimeoutPeriod());
     long long randomOffsetUpperBound =
         electionTimeout * _externalState->getElectionTimeoutOffsetLimitFraction();
@@ -79,8 +79,7 @@ Milliseconds ReplicationCoordinatorImpl::_getRandomizedElectionOffset() {
         return Milliseconds(0);
     }
 
-    int64_t randomOffset = _replExecutor->nextRandomInt64(randomOffsetUpperBound);
-    return Milliseconds(randomOffset);
+    return Milliseconds{_nextRandomInt64_inlock(randomOffsetUpperBound)};
 }
 
 void ReplicationCoordinatorImpl::_doMemberHeartbeat(executor::TaskExecutor::CallbackArgs cbData,
@@ -302,7 +301,7 @@ stdx::unique_lock<stdx::mutex> ReplicationCoordinatorImpl::_handleHeartbeatRespo
 
                 // Add randomized offset to calculated priority takeover delay.
                 Milliseconds priorityTakeoverDelay = _rsConfig.getPriorityTakeoverDelay(_selfIndex);
-                Milliseconds randomOffset = _getRandomizedElectionOffset();
+                Milliseconds randomOffset = _getRandomizedElectionOffset_inlock();
                 _priorityTakeoverWhen = _replExecutor->now() + priorityTakeoverDelay + randomOffset;
                 log() << "Scheduling priority takeover at " << _priorityTakeoverWhen;
                 _priorityTakeoverCbh = _scheduleWorkAt(
@@ -827,7 +826,7 @@ void ReplicationCoordinatorImpl::_cancelAndRescheduleElectionTimeout_inlock() {
         return;
     }
 
-    Milliseconds randomOffset = _getRandomizedElectionOffset();
+    Milliseconds randomOffset = _getRandomizedElectionOffset_inlock();
     auto now = _replExecutor->now();
     auto when = now + _rsConfig.getElectionTimeoutPeriod() + randomOffset;
     invariant(when > now);