SERVER-52588 Making the upgrade/downgrade process work for the Epoch to ClusterTime migration

5 files changed, 130 insertions, 10 deletions

diff --git a/src/mongo/db/commands/set_feature_compatibility_version_command.cpp b/src/mongo/db/commands/set_feature_compatibility_version_command.cpp
index 92eb7526073..c3474f1caf6 100644
--- a/src/mongo/db/commands/set_feature_compatibility_version_command.cpp
+++ b/src/mongo/db/commands/set_feature_compatibility_version_command.cpp
@@ -60,6 +60,7 @@
 #include "mongo/db/views/view_catalog.h"
 #include "mongo/logv2/log.h"
 #include "mongo/rpc/get_status_from_command_result.h"
+#include "mongo/s/sharded_collections_ddl_parameters_gen.h"
 #include "mongo/stdx/unordered_set.h"
 #include "mongo/util/exit.h"
 #include "mongo/util/fail_point.h"
@@ -252,6 +253,15 @@ public:
                 uassertStatusOK(
                     ShardingCatalogManager::get(opCtx)->setFeatureCompatibilityVersionOnShards(
                         opCtx, CommandHelpers::appendMajorityWriteConcern(request.toBSON({}))));
+
+                // Create a 'timestamp' for the collections that don't have one. This must be done
+                // after all shards have been upgraded in order to guarantee that when
+                // createCollectionTimestampsFor49 starts, no new collections without a timestamp
+                // will be added.
+                if (requestedVersion >= FeatureCompatibilityParams::Version::kVersion49 &&
+                    feature_flags::gShardingFullDDLSupport.isEnabledAndIgnoreFCV()) {
+                    ShardingCatalogManager::get(opCtx)->createCollectionTimestampsFor49(opCtx);
+                }
             }
 
             hangWhileUpgrading.pauseWhileSet(opCtx);
@@ -338,11 +348,20 @@ public:
                 deletePersistedDefaultRWConcernDocument(opCtx);
             }
 
-            // Downgrade shards before config finishes its downgrade.
             if (serverGlobalParams.clusterRole == ClusterRole::ConfigServer) {
+                // Downgrade shards before config finishes its downgrade.
                 uassertStatusOK(
                     ShardingCatalogManager::get(opCtx)->setFeatureCompatibilityVersionOnShards(
                         opCtx, CommandHelpers::appendMajorityWriteConcern(request.toBSON({}))));
+
+                // Delete the 'timestamp' field in config.collections entries. This must be done
+                // after all shards have been downgraded in order to guarantee that when
+                // downgradeConfigCollectionEntriesToPre49 starts, no new collections with a
+                // timestamp will be added.
+                if (requestedVersion < FeatureCompatibilityParams::Version::kVersion49) {
+                    ShardingCatalogManager::get(opCtx)->downgradeConfigCollectionEntriesToPre49(
+                        opCtx);
+                }
             }
 
             hangWhileDowngrading.pauseWhileSet(opCtx);
diff --git a/src/mongo/db/s/config/sharding_catalog_manager.cpp b/src/mongo/db/s/config/sharding_catalog_manager.cpp
index 4e9a619db01..6ea65f9ea96 100644
--- a/src/mongo/db/s/config/sharding_catalog_manager.cpp
+++ b/src/mongo/db/s/config/sharding_catalog_manager.cpp
@@ -39,6 +39,7 @@
 #include "mongo/db/ops/write_ops.h"
 #include "mongo/db/query/query_request.h"
 #include "mongo/db/s/balancer/type_migration.h"
+#include "mongo/db/vector_clock.h"
 #include "mongo/logv2/log.h"
 #include "mongo/rpc/get_status_from_command_result.h"
 #include "mongo/s/catalog/config_server_version.h"
@@ -52,6 +53,7 @@
 #include "mongo/s/catalog/type_tags.h"
 #include "mongo/s/client/shard_registry.h"
 #include "mongo/s/grid.h"
+#include "mongo/s/sharded_collections_ddl_parameters_gen.h"
 #include "mongo/s/write_ops/batched_command_request.h"
 #include "mongo/s/write_ops/batched_command_response.h"
 #include "mongo/transport/service_entry_point.h"
@@ -484,6 +486,81 @@ void ShardingCatalogManager::removePre49LegacyMetadata(OperationContext* opCtx)
     uassertStatusOK(getWriteConcernStatusFromCommandResult(commandResult->getCommandReply()));
 }
 
+void ShardingCatalogManager::createCollectionTimestampsFor49(OperationContext* opCtx) {
+    LOGV2(5258800, "Starting upgrade of config.collections");
+
+    const auto catalogClient = Grid::get(opCtx)->catalogClient();
+    auto const catalogCache = Grid::get(opCtx)->catalogCache();
+    auto const configShard = Grid::get(opCtx)->shardRegistry()->getConfigShard();
+    const auto collectionDocs =
+        uassertStatusOK(configShard->exhaustiveFindOnConfig(
+                            opCtx,
+                            ReadPreferenceSetting{ReadPreference::PrimaryOnly},
+                            repl::ReadConcernLevel::kLocalReadConcern,
+                            CollectionType::ConfigNS,
+                            BSON(CollectionType::kTimestampFieldName << BSON("$exists" << false)),
+                            BSONObj(),
+                            boost::none))
+            .docs;
+
+
+    for (const auto& doc : collectionDocs) {
+        const CollectionType coll(doc);
+        const auto nss = coll.getNss();
+
+        auto now = VectorClock::get(opCtx)->getTime();
+        auto clusterTime = now.clusterTime().asTimestamp();
+
+        uassertStatusOK(catalogClient->updateConfigDocument(
+            opCtx,
+            CollectionType::ConfigNS,
+            BSON(CollectionType::kNssFieldName << nss.ns()),
+            BSON("$set" << BSON(CollectionType::kTimestampFieldName << clusterTime)),
+            false /* upsert */,
+            ShardingCatalogClient::kMajorityWriteConcern));
+
+        catalogCache->invalidateCollectionEntry_LINEARIZABLE(nss);
+    }
+
+    LOGV2(5258801, "Successfully upgraded config.collections");
+}
+
+void ShardingCatalogManager::downgradeConfigCollectionEntriesToPre49(OperationContext* opCtx) {
+    if (feature_flags::gShardingFullDDLSupport.isEnabledAndIgnoreFCV()) {
+        DBDirectClient client(opCtx);
+
+        // Clear the 'timestamp' fields from config.collections
+        write_ops::Update unsetTimestamp(CollectionType::ConfigNS, [] {
+            write_ops::UpdateOpEntry u;
+            u.setQ({});
+            u.setU(write_ops::UpdateModification::parseFromClassicUpdate(
+                BSON("$unset" << BSON(CollectionType::kTimestampFieldName << ""))));
+            u.setMulti(true);
+            return std::vector{u};
+        }());
+        unsetTimestamp.setWriteCommandBase([] {
+            write_ops::WriteCommandBase base;
+            base.setOrdered(false);
+            return base;
+        }());
+
+        auto commandResult = client.runCommand(OpMsgRequest::fromDBAndBody(
+            CollectionType::ConfigNS.db(),
+            unsetTimestamp.toBSON(ShardingCatalogClient::kMajorityWriteConcern.toBSON())));
+
+        uassertStatusOK([&] {
+            BatchedCommandResponse response;
+            std::string unusedErrmsg;
+            response.parseBSON(
+                commandResult->getCommandReply(),
+                &unusedErrmsg);  // Return value intentionally ignored, because response.toStatus()
+                                 // will contain any errors in more detail
+            return response.toStatus();
+        }());
+        uassertStatusOK(getWriteConcernStatusFromCommandResult(commandResult->getCommandReply()));
+    }
+}
+
 Lock::ExclusiveLock ShardingCatalogManager::lockZoneMutex(OperationContext* opCtx) {
     Lock::ExclusiveLock lk(opCtx->lockState(), _kZoneOpLock);
     return lk;
diff --git a/src/mongo/db/s/config/sharding_catalog_manager.h b/src/mongo/db/s/config/sharding_catalog_manager.h
index 5b9030f3d84..c8f3b3edf7c 100644
--- a/src/mongo/db/s/config/sharding_catalog_manager.h
+++ b/src/mongo/db/s/config/sharding_catalog_manager.h
@@ -455,6 +455,22 @@ public:
      */
     void removePre49LegacyMetadata(OperationContext* opCtx);
 
+    /**
+     * Creates a 'timestamp' for each one of the entries in the config server's config.collections,
+     * where 'timestamp' does not already exist.
+     *
+     * It shall be called when upgrading to 4.9.
+     */
+    void createCollectionTimestampsFor49(OperationContext* opCtx);
+
+    /**
+     * Downgrades the config.collections entries to prior 4.9 version. More specifically, it removes
+     * the 'timestamp' field from all the documents in config.collections.
+     *
+     * It shall be called when downgrading from 4.9 to an earlier version.
+     */
+    void downgradeConfigCollectionEntriesToPre49(OperationContext* opCtx);
+
     //
     // For Diagnostics
     //
diff --git a/src/mongo/db/s/shard_metadata_util.cpp b/src/mongo/db/s/shard_metadata_util.cpp
index dd2a0357293..8da1a50bc10 100644
--- a/src/mongo/db/s/shard_metadata_util.cpp
+++ b/src/mongo/db/s/shard_metadata_util.cpp
@@ -44,7 +44,9 @@
 #include "mongo/rpc/get_status_from_command_result.h"
 #include "mongo/rpc/unique_message.h"
 #include "mongo/s/catalog/type_chunk.h"
+#include "mongo/s/catalog/type_collection.h"
 #include "mongo/s/chunk_version.h"
+#include "mongo/s/sharded_collections_ddl_parameters_gen.h"
 #include "mongo/s/write_ops/batched_command_response.h"
 
 namespace mongo {
@@ -485,30 +487,35 @@ Status deleteDatabasesEntry(OperationContext* opCtx, StringData dbName) {
 }
 
 void downgradeShardConfigCollectionEntriesToPre49(OperationContext* opCtx) {
-    // Clear the 'allowMigrations' field from config.cache.collections
+    // Clear the 'allowMigrations' and 'timestamp' fields from config.cache.collections
     LOGV2(5189100, "Starting downgrade of config.cache.collections");
-    write_ops::Update clearAllowMigrations(NamespaceString::kShardConfigCollectionsNamespace, [] {
+    write_ops::Update clearFields(NamespaceString::kShardConfigCollectionsNamespace, [] {
+        BSONObj unsetFields =
+            BSON(ShardCollectionType::kPre50CompatibleAllowMigrationsFieldName << "");
+        if (feature_flags::gShardingFullDDLSupport.isEnabledAndIgnoreFCV()) {
+            unsetFields = unsetFields.addFields(BSON(CollectionType::kTimestampFieldName << ""));
+        }
+
         write_ops::UpdateOpEntry u;
         u.setQ({});
-        u.setU(write_ops::UpdateModification::parseFromClassicUpdate(
-            BSON("$unset" << BSON(ShardCollectionType::kPre50CompatibleAllowMigrationsFieldName
-                                  << ""))));
+        u.setU(
+            write_ops::UpdateModification::parseFromClassicUpdate(BSON("$unset" << unsetFields)));
         u.setMulti(true);
         return std::vector{u};
     }());
 
-    clearAllowMigrations.setWriteCommandBase([] {
+    clearFields.setWriteCommandBase([] {
         write_ops::WriteCommandBase base;
         base.setOrdered(false);
         return base;
     }());
 
     DBDirectClient client(opCtx);
-    const auto commandResult = client.runCommand(clearAllowMigrations.serialize({}));
+    const auto commandResult = client.runCommand(clearFields.serialize({}));
 
     uassertStatusOK(getStatusFromWriteCommandResponse(commandResult->getCommandReply()));
 
-    LOGV2(5189101, "Succesfully downgraded config.cache.collections");
+    LOGV2(5189101, "Successfully downgraded config.cache.collections");
 }
 
 }  // namespace shardmetadatautil
diff --git a/src/mongo/db/s/shard_metadata_util.h b/src/mongo/db/s/shard_metadata_util.h
index 76775b95b7e..f863e406941 100644
--- a/src/mongo/db/s/shard_metadata_util.h
+++ b/src/mongo/db/s/shard_metadata_util.h
@@ -224,7 +224,8 @@ Status deleteDatabasesEntry(OperationContext* opCtx, StringData dbName);
 
 /**
  * Downgrades the config.cache.collections entries to prior 4.9 version. More specifically, it
- * removes the allowMigrations field from all the documents of config.cache.collections
+ * removes the allowMigrations and timestamp fields from all the documents of
+ * config.cache.collections
  */
 void downgradeShardConfigCollectionEntriesToPre49(OperationContext* opCtx);