SERVER-39484 IndexBuildsCoordinator receives notifications on rollback/step up to primary

This partially restores some of the replication state transition functions removed in
commit adab670.

Author: benety

src/mongo/db/index_builds_coordinator.cpp

@@ -160,6 +160,29 @@ void logFailure(Status status,
           << replState->collectionUUID << " ): " << status;
 }
 
+/**
+ * Iterates over index builds with the provided function.
+ */
+void forEachIndexBuild(
+    const std::vector<std::shared_ptr<ReplIndexBuildState>>& indexBuilds,
+    StringData logPrefix,
+    std::function<void(std::shared_ptr<ReplIndexBuildState> replState)> onIndexBuild) {
+    if (indexBuilds.empty()) {
+        return;
+    }
+
+    log() << logPrefix << "active index builds: " << indexBuilds.size();
+
+    for (auto replState : indexBuilds) {
+        std::string indexNamesStr;
+        str::joinStringDelim(replState->indexNames, &indexNamesStr, ',');
+        log() << logPrefix << replState->buildUUID << ": collection: " << replState->collectionUUID
+              << "; indexes: " << replState->indexNames.size() << " [" << indexNamesStr << "]";
+
+        onIndexBuild(replState);
+    }
+}
+
 }  // namespace
 
 const auto getIndexBuildsCoord =
@@ -398,6 +421,21 @@ void IndexBuildsCoordinator::abortIndexBuildByBuildUUID(OperationContext* opCtx,
     }
 }
 
+void IndexBuildsCoordinator::onStepUp(OperationContext* opCtx) {
+    log() << "IndexBuildsCoordinator::onStepUp - this node is stepping up to primary";
+
+    auto indexBuilds = _getIndexBuilds();
+    auto onIndexBuild = [](std::shared_ptr<ReplIndexBuildState> replState) {};
+    forEachIndexBuild(indexBuilds, "IndexBuildsCoordinator::onStepUp - "_sd, onIndexBuild);
+}
+
+void IndexBuildsCoordinator::onRollback(OperationContext* opCtx) {
+    log() << "IndexBuildsCoordinator::onRollback - this node is entering the rollback state";
+    auto indexBuilds = _getIndexBuilds();
+    auto onIndexBuild = [](std::shared_ptr<ReplIndexBuildState> replState) {};
+    forEachIndexBuild(indexBuilds, "IndexBuildsCoordinator::onRollback - "_sd, onIndexBuild);
+}
+
 void IndexBuildsCoordinator::recoverIndexBuilds() {
     // TODO: not yet implemented.
 }
@@ -1305,6 +1343,17 @@ StatusWith<std::shared_ptr<ReplIndexBuildState>> IndexBuildsCoordinator::_getInd
     return it->second;
 }
 
+std::vector<std::shared_ptr<ReplIndexBuildState>> IndexBuildsCoordinator::_getIndexBuilds() const {
+    std::vector<std::shared_ptr<ReplIndexBuildState>> indexBuilds;
+    {
+        stdx::unique_lock<Latch> lk(_mutex);
+        for (auto pair : _allIndexBuilds) {
+            indexBuilds.push_back(pair.second);
+        }
+    }
+    return indexBuilds;
+}
+
 ScopedStopNewDatabaseIndexBuilds::ScopedStopNewDatabaseIndexBuilds(
     IndexBuildsCoordinator* indexBuildsCoordinator, StringData dbName)
     : _indexBuildsCoordinatorPtr(indexBuildsCoordinator), _dbName(dbName.toString()) {

src/mongo/db/index_builds_coordinator.h

@@ -210,6 +210,18 @@ class IndexBuildsCoordinator {
                                     const UUID& buildUUID,
                                     const std::string& reason);
 
+    /**
+     * Invoked when the node enters the primary state.
+     * Unblocks index builds that have been waiting to commit/abort during the secondary state.
+     */
+    void onStepUp(OperationContext* opCtx);
+
+    /**
+     * Invoked when the node enters the rollback state.
+     * Unblocks index builds that have been waiting to commit/abort during the secondary state.
+     */
+    void onRollback(OperationContext* opCtx);
+
     /**
      * TODO: This is not yet implemented.
      */
@@ -418,6 +430,12 @@ class IndexBuildsCoordinator {
      */
     StatusWith<std::shared_ptr<ReplIndexBuildState>> _getIndexBuild(const UUID& buildUUID) const;
 
+    /**
+     * Returns a snapshot of active index builds. Since each index build state is reference counted,
+     * it is fine to examine the returned index builds without re-locking 'mutex'.
+     */
+    std::vector<std::shared_ptr<ReplIndexBuildState>> _getIndexBuilds() const;
+
     // Protects the below state.
     mutable Mutex _mutex = MONGO_MAKE_LATCH("IndexBuildsCoordinator::_mutex");
 

src/mongo/db/repl/SConscript

@@ -128,6 +128,7 @@ env.Library(
         '$BUILD_DIR/mongo/client/connection_pool',
         '$BUILD_DIR/mongo/client/fetcher',
         '$BUILD_DIR/mongo/db/concurrency/write_conflict_exception',
+        '$BUILD_DIR/mongo/db/index_builds_coordinator_interface',
         '$BUILD_DIR/mongo/db/service_context',
         '$BUILD_DIR/mongo/util/concurrency/thread_pool',
     ],
@@ -1197,6 +1198,7 @@ env.Library(
         'repl_server_parameters',
         '$BUILD_DIR/mongo/db/commands/mongod_fcv',
         '$BUILD_DIR/mongo/db/commands/test_commands_enabled',
+        '$BUILD_DIR/mongo/db/index_builds_coordinator_interface',
     ],
 )
 

src/mongo/db/repl/bgsync.cpp

@@ -45,6 +45,7 @@
 #include "mongo/db/concurrency/replication_state_transition_lock_guard.h"
 #include "mongo/db/concurrency/write_conflict_exception.h"
 #include "mongo/db/dbhelpers.h"
+#include "mongo/db/index_builds_coordinator.h"
 #include "mongo/db/repl/data_replicator_external_state_impl.h"
 #include "mongo/db/repl/oplog.h"
 #include "mongo/db/repl/oplog_interface_local.h"
@@ -633,6 +634,8 @@ void BackgroundSync::_runRollback(OperationContext* opCtx,
     // are visible before potentially truncating the oplog.
     storageInterface->waitForAllEarlierOplogWritesToBeVisible(opCtx);
 
+    IndexBuildsCoordinator::get(opCtx)->onRollback(opCtx);
+
     auto storageEngine = opCtx->getServiceContext()->getStorageEngine();
     if (!forceRollbackViaRefetch.load() && storageEngine->supportsRecoverToStableTimestamp()) {
         log() << "Rollback using 'recoverToStableTimestamp' method.";

src/mongo/db/repl/replication_coordinator_external_state_impl.cpp

@@ -53,6 +53,7 @@
 #include "mongo/db/dbdirectclient.h"
 #include "mongo/db/dbhelpers.h"
 #include "mongo/db/free_mon/free_mon_mongod.h"
+#include "mongo/db/index_builds_coordinator.h"
 #include "mongo/db/jsobj.h"
 #include "mongo/db/kill_sessions_local.h"
 #include "mongo/db/logical_clock.h"
@@ -468,6 +469,8 @@ OpTime ReplicationCoordinatorExternalStateImpl::onTransitionToPrimary(OperationC
 
     _dropAllTempCollections(opCtx);
 
+    IndexBuildsCoordinator::get(opCtx)->onStepUp(opCtx);
+
     notifyFreeMonitoringOnTransitionToPrimary();
 
     // It is only necessary to check the system indexes on the first transition to master.