SERVER-57769 Allow tenant migration recipients to advance cluster time with noops

(cherry picked from commit fca4506a72e32e2346ddaa7e0a7d9d59f2138d8e)

9 files changed, 37 insertions, 18 deletions

diff --git a/src/mongo/db/SConscript b/src/mongo/db/SConscript
index b6036db8e53..6060fad1faa 100644
--- a/src/mongo/db/SConscript
+++ b/src/mongo/db/SConscript
@@ -1068,6 +1068,7 @@ env.Library(
         'read_concern_mongod.idl',
     ],
     LIBDEPS_PRIVATE=[
+        "$BUILD_DIR/mongo/db/repl/tenant_migration_access_blocker",
         "$BUILD_DIR/mongo/idl/server_parameter",
         "$BUILD_DIR/mongo/s/grid",
         "catalog_raii",
diff --git a/src/mongo/db/commands/run_aggregate.cpp b/src/mongo/db/commands/run_aggregate.cpp
index ab8ff389252..ea677237211 100644
--- a/src/mongo/db/commands/run_aggregate.cpp
+++ b/src/mongo/db/commands/run_aggregate.cpp
@@ -493,7 +493,7 @@ void _adjustChangeStreamReadConcern(OperationContext* opCtx) {
     }
 
     // Wait for read concern again since we changed the original read concern.
-    uassertStatusOK(waitForReadConcern(opCtx, readConcernArgs, true));
+    uassertStatusOK(waitForReadConcern(opCtx, readConcernArgs, StringData(), true));
     setPrepareConflictBehaviorForReadConcern(
         opCtx, readConcernArgs, PrepareConflictBehavior::kIgnoreConflicts);
 }
diff --git a/src/mongo/db/read_concern.cpp b/src/mongo/db/read_concern.cpp
index 07944dd56df..fe9fd208c04 100644
--- a/src/mongo/db/read_concern.cpp
+++ b/src/mongo/db/read_concern.cpp
@@ -42,9 +42,10 @@ void setPrepareConflictBehaviorForReadConcern(OperationContext* opCtx,
 
 Status waitForReadConcern(OperationContext* opCtx,
                           const repl::ReadConcernArgs& readConcernArgs,
+                          StringData dbName,
                           bool allowAfterClusterTime) {
     static auto w = MONGO_WEAK_FUNCTION_DEFINITION(waitForReadConcern);
-    return w(opCtx, readConcernArgs, allowAfterClusterTime);
+    return w(opCtx, readConcernArgs, dbName, allowAfterClusterTime);
 }
 
 Status waitForLinearizableReadConcern(OperationContext* opCtx, int readConcernTimeout) {
diff --git a/src/mongo/db/read_concern.h b/src/mongo/db/read_concern.h
index 6168b393954..e12915870cf 100644
--- a/src/mongo/db/read_concern.h
+++ b/src/mongo/db/read_concern.h
@@ -37,6 +37,7 @@ class Status;
 template <typename T>
 class StatusWith;
 enum class PrepareConflictBehavior;
+class StringData;
 namespace repl {
 class ReadConcernArgs;
 class SpeculativeMajorityReadInfo;
@@ -64,6 +65,7 @@ void setPrepareConflictBehaviorForReadConcern(OperationContext* opCtx,
  */
 Status waitForReadConcern(OperationContext* opCtx,
                           const repl::ReadConcernArgs& readConcernArgs,
+                          StringData dbName,
                           bool allowAfterClusterTime);
 
 /*
diff --git a/src/mongo/db/read_concern_mongod.cpp b/src/mongo/db/read_concern_mongod.cpp
index 8cc004b8f64..f35d31c80cd 100644
--- a/src/mongo/db/read_concern_mongod.cpp
+++ b/src/mongo/db/read_concern_mongod.cpp
@@ -42,6 +42,7 @@
 #include "mongo/db/repl/optime.h"
 #include "mongo/db/repl/repl_client_info.h"
 #include "mongo/db/repl/speculative_majority_read_info.h"
+#include "mongo/db/repl/tenant_migration_access_blocker_registry.h"
 #include "mongo/db/s/sharding_state.h"
 #include "mongo/db/server_options.h"
 #include "mongo/db/storage/recovery_unit.h"
@@ -104,10 +105,21 @@ private:
     std::map<Timestamp, std::shared_ptr<Notification<Status>>> _writeRequests;
 };
 
+bool hasActiveTenantMigrationRecipient(OperationContext* opCtx, StringData dbName) {
+    if (dbName.empty()) {
+        return false;
+    }
+
+    auto mtab = TenantMigrationAccessBlockerRegistry::get(opCtx->getServiceContext())
+                    .getTenantMigrationAccessBlockerForDbName(
+                        dbName, TenantMigrationAccessBlocker::BlockerType::kRecipient);
+    return bool(mtab);
+}
+
 /**
  *  Schedule a write via appendOplogNote command to the primary of this replica set.
  */
-Status makeNoopWriteIfNeeded(OperationContext* opCtx, LogicalTime clusterTime) {
+Status makeNoopWriteIfNeeded(OperationContext* opCtx, LogicalTime clusterTime, StringData dbName) {
     repl::ReplicationCoordinator* const replCoord = repl::ReplicationCoordinator::get(opCtx);
     invariant(replCoord->isReplEnabled());
 
@@ -139,18 +151,14 @@ Status makeNoopWriteIfNeeded(OperationContext* opCtx, LogicalTime clusterTime) {
     // one that waits for the notification gets the later clusterTime, so when the request finishes
     // it needs to be repeated with the later time.
     while (clusterTime > lastAppliedOpTime) {
-        // standalone replica set, so there is no need to advance the OpLog on the primary.
-        if (serverGlobalParams.clusterRole == ClusterRole::None) {
+        // Standalone replica set, so there is no need to advance the OpLog on the primary. The only
+        // exception is after a tenant migration because the target time may be from the donor
+        // replica set and is not guaranteed to be in the recipient's oplog.
+        if (serverGlobalParams.clusterRole == ClusterRole::None &&
+            !hasActiveTenantMigrationRecipient(opCtx, dbName)) {
             return Status::OK();
         }
 
-        bool isConfig = (serverGlobalParams.clusterRole == ClusterRole::ConfigServer);
-
-        if (!isConfig && !ShardingState::get(opCtx)->enabled()) {
-            return {ErrorCodes::ShardingStateNotInitialized,
-                    "Failed noop write because sharding state has not been initialized"};
-        }
-
         if (!remainingAttempts--) {
             std::stringstream ss;
             ss << "Requested clusterTime " << clusterTime.toString()
@@ -282,6 +290,7 @@ void setPrepareConflictBehaviorForReadConcernImpl(OperationContext* opCtx,
 
 Status waitForReadConcernImpl(OperationContext* opCtx,
                               const repl::ReadConcernArgs& readConcernArgs,
+                              StringData dbName,
                               bool allowAfterClusterTime) {
     // If we are in a direct client within a transaction, then we may be holding locks, so it is
     // illegal to wait for read concern. This is fine, since the outer operation should have handled
@@ -373,7 +382,7 @@ Status waitForReadConcernImpl(OperationContext* opCtx,
                                       << "; current clusterTime: " << clusterTime.toString()};
             }
 
-            auto status = makeNoopWriteIfNeeded(opCtx, *targetClusterTime);
+            auto status = makeNoopWriteIfNeeded(opCtx, *targetClusterTime, dbName);
             if (!status.isOK()) {
                 LOGV2(20990,
                       "Failed noop write at clusterTime: {targetClusterTime} due to {error}",
diff --git a/src/mongo/db/s/migration_source_manager.cpp b/src/mongo/db/s/migration_source_manager.cpp
index 62aa6843250..13deb093da8 100644
--- a/src/mongo/db/s/migration_source_manager.cpp
+++ b/src/mongo/db/s/migration_source_manager.cpp
@@ -271,7 +271,8 @@ Status MigrationSourceManager::startClone() {
         auto const readConcernArgs = repl::ReadConcernArgs(
             replCoord->getMyLastAppliedOpTime(), repl::ReadConcernLevel::kLocalReadConcern);
 
-        auto waitForReadConcernStatus = waitForReadConcern(_opCtx, readConcernArgs, false);
+        auto waitForReadConcernStatus =
+            waitForReadConcern(_opCtx, readConcernArgs, StringData(), false);
         if (!waitForReadConcernStatus.isOK()) {
             return waitForReadConcernStatus;
         }
diff --git a/src/mongo/db/service_entry_point_mongod.cpp b/src/mongo/db/service_entry_point_mongod.cpp
index fd48254d9af..f6b22b26899 100644
--- a/src/mongo/db/service_entry_point_mongod.cpp
+++ b/src/mongo/db/service_entry_point_mongod.cpp
@@ -75,8 +75,10 @@ public:
     void waitForReadConcern(OperationContext* opCtx,
                             const CommandInvocation* invocation,
                             const OpMsgRequest& request) const override {
-        Status rcStatus = mongo::waitForReadConcern(
-            opCtx, repl::ReadConcernArgs::get(opCtx), invocation->allowsAfterClusterTime());
+        Status rcStatus = mongo::waitForReadConcern(opCtx,
+                                                    repl::ReadConcernArgs::get(opCtx),
+                                                    request.getDatabase(),
+                                                    invocation->allowsAfterClusterTime());
 
         if (!rcStatus.isOK()) {
             if (ErrorCodes::isExceededTimeLimitError(rcStatus.code())) {
diff --git a/src/mongo/embedded/read_concern_embedded.cpp b/src/mongo/embedded/read_concern_embedded.cpp
index d6fc2a0cbe7..708b41bc279 100644
--- a/src/mongo/embedded/read_concern_embedded.cpp
+++ b/src/mongo/embedded/read_concern_embedded.cpp
@@ -42,6 +42,7 @@ void setPrepareConflictBehaviorForReadConcernImpl(
 
 Status waitForReadConcernImpl(OperationContext* opCtx,
                               const repl::ReadConcernArgs& readConcernArgs,
+                              StringData dbName,
                               bool allowAfterClusterTime) {
     if (readConcernArgs.getLevel() == repl::ReadConcernLevel::kLinearizableReadConcern) {
         return {ErrorCodes::NotImplemented, "linearizable read concern not supported on embedded"};
diff --git a/src/mongo/embedded/service_entry_point_embedded.cpp b/src/mongo/embedded/service_entry_point_embedded.cpp
index 0ff33bc12c0..428d15e2d8a 100644
--- a/src/mongo/embedded/service_entry_point_embedded.cpp
+++ b/src/mongo/embedded/service_entry_point_embedded.cpp
@@ -82,8 +82,10 @@ public:
     void waitForReadConcern(OperationContext* opCtx,
                             const CommandInvocation* invocation,
                             const OpMsgRequest& request) const override {
-        auto rcStatus = mongo::waitForReadConcern(
-            opCtx, repl::ReadConcernArgs::get(opCtx), invocation->allowsAfterClusterTime());
+        auto rcStatus = mongo::waitForReadConcern(opCtx,
+                                                  repl::ReadConcernArgs::get(opCtx),
+                                                  request.getDatabase(),
+                                                  invocation->allowsAfterClusterTime());
         uassertStatusOK(rcStatus);
     }