path: root/src/mongo/db/repl/rs_rollback.cpp

/**
 *    Copyright (C) 2008-2017 MongoDB, Inc.
 *
 *    This program is free software: you can redistribute it and/or  modify
 *    it under the terms of the GNU Affero General Public License, version 3,
 *    as published by the Free Software Foundation.
 *
 *    This program is distributed in the hope that it will be useful,
 *    but WITHOUT ANY WARRANTY; without even the implied warranty of
 *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *    GNU Affero General Public License for more details.
 *
 *    You should have received a copy of the GNU Affero General Public License
 *    along with this program.  If not, see <http://www.gnu.org/licenses/>.
 *
 *    As a special exception, the copyright holders give permission to link the
 *    code of portions of this program with the OpenSSL library under certain
 *    conditions as described in each individual source file and distribute
 *    linked combinations including the program with the OpenSSL library. You
 *    must comply with the GNU Affero General Public License in all respects for
 *    all of the code used other than as permitted herein. If you modify file(s)
 *    with this exception, you may extend this exception to your version of the
 *    file(s), but you are not obligated to do so. If you do not wish to do so,
 *    delete this exception statement from your version. If you delete this
 *    exception statement from all source files in the program, then also delete
 *    it in the license file.
 */

#define MONGO_LOG_DEFAULT_COMPONENT ::mongo::logger::LogComponent::kReplicationRollback

#include "mongo/platform/basic.h"

#include "mongo/db/repl/rs_rollback.h"

#include <algorithm>
#include <memory>

#include "mongo/bson/bsonelement_comparator.h"
#include "mongo/bson/util/bson_extract.h"
#include "mongo/db/auth/authorization_manager.h"
#include "mongo/db/catalog/collection_catalog_entry.h"
#include "mongo/db/catalog/database_holder.h"
#include "mongo/db/catalog/document_validation.h"
#include "mongo/db/catalog/index_catalog.h"
#include "mongo/db/catalog/rename_collection.h"
#include "mongo/db/catalog/uuid_catalog.h"
#include "mongo/db/catalog_raii.h"
#include "mongo/db/client.h"
#include "mongo/db/commands.h"
#include "mongo/db/concurrency/write_conflict_exception.h"
#include "mongo/db/db_raii.h"
#include "mongo/db/dbhelpers.h"
#include "mongo/db/exec/working_set_common.h"
#include "mongo/db/logical_session_id.h"
#include "mongo/db/logical_time_validator.h"
#include "mongo/db/ops/delete.h"
#include "mongo/db/ops/update.h"
#include "mongo/db/ops/update_lifecycle_impl.h"
#include "mongo/db/ops/update_request.h"
#include "mongo/db/query/internal_plans.h"
#include "mongo/db/repl/bgsync.h"
#include "mongo/db/repl/drop_pending_collection_reaper.h"
#include "mongo/db/repl/oplog.h"
#include "mongo/db/repl/oplog_interface.h"
#include "mongo/db/repl/replication_coordinator.h"
#include "mongo/db/repl/replication_coordinator_impl.h"
#include "mongo/db/repl/replication_process.h"
#include "mongo/db/repl/roll_back_local_operations.h"
#include "mongo/db/repl/rollback_source.h"
#include "mongo/db/repl/rslog.h"
#include "mongo/db/s/shard_identity_rollback_notifier.h"
#include "mongo/db/session_catalog_mongod.h"
#include "mongo/db/transaction_participant.h"
#include "mongo/util/exit.h"
#include "mongo/util/fail_point_service.h"
#include "mongo/util/log.h"
#include "mongo/util/scopeguard.h"

namespace mongo {

using std::shared_ptr;
using std::unique_ptr;
using std::list;
using std::map;
using std::set;
using std::string;
using std::pair;

namespace repl {

using namespace rollback_internal;

bool DocID::operator<(const DocID& other) const {
    int comp = uuid.toString().compare(other.uuid.toString());
    if (comp < 0)
        return true;
    if (comp > 0)
        return false;

    const StringData::ComparatorInterface* stringComparator = nullptr;
    BSONElementComparator eltCmp(BSONElementComparator::FieldNamesMode::kIgnore, stringComparator);
    return eltCmp.evaluate(_id < other._id);
}

bool DocID::operator==(const DocID& other) const {
    // Since this is only used for tests, going with the simple impl that reuses operator< which is
    // used in the real code.
    return !(*this < other || other < *this);
}

void FixUpInfo::removeAllDocsToRefetchFor(UUID collectionUUID) {
    docsToRefetch.erase(docsToRefetch.lower_bound(DocID::minFor(collectionUUID)),
                        docsToRefetch.upper_bound(DocID::maxFor(collectionUUID)));
}

void FixUpInfo::removeRedundantOperations() {
    for (const auto& collectionUUID : collectionsToDrop) {
        removeAllDocsToRefetchFor(collectionUUID);
        indexesToDrop.erase(collectionUUID);
        indexesToCreate.erase(collectionUUID);
        collectionsToRename.erase(collectionUUID);
        collectionsToResyncMetadata.erase(collectionUUID);
    }
}

bool FixUpInfo::removeRedundantIndexCommands(UUID uuid, std::string indexName) {
    LOG(2) << "Attempting to remove redundant index operations from the set of indexes to create "
              "for collection "
           << uuid << ", for index '" << indexName << "'";

    // See if there are any indexes to create for this collection.
    auto indexes = indexesToCreate.find(uuid);

    // There are no indexes to create for this collection UUID, so there are no index creation
    // operations to remove.
    if (indexes == indexesToCreate.end()) {
        LOG(2)
            << "Collection " << uuid
            << " has no indexes to create. Not removing any index creation operations for index '"
            << indexName << "'.";
        return false;
    }

    // This is the set of all indexes to create for the given collection UUID. Keep a reference so
    // we can modify the original object.
    std::map<std::string, BSONObj>* indexesToCreateForColl = &(indexes->second);

    // If this index was not previously added to the set of indexes that need to be created for this
    // collection, then we do nothing.
    if (indexesToCreateForColl->find(indexName) == indexesToCreateForColl->end()) {
        LOG(2) << "Index '" << indexName << "' was not previously set to be created for collection "
               << uuid << ". Not removing any index creation operations.";
        return false;
    }

    // This index was previously added to the set of indexes to create for this collection, so we
    // remove it from that set.
    LOG(2) << "Index '" << indexName << "' was previously set to be created for collection " << uuid
           << ". Removing this redundant index creation operation.";
    indexesToCreateForColl->erase(indexName);
    // If there are now no remaining indexes to create for this collection, remove it from
    // the set of collections that we need to create indexes for.
    if (indexesToCreateForColl->empty()) {
        indexesToCreate.erase(uuid);
    }

    return true;
}

void FixUpInfo::recordRollingBackDrop(const NamespaceString& nss, OpTime opTime, UUID uuid) {

    // Records the collection that needs to be removed from the drop-pending collections
    // list in the DropPendingCollectionReaper.
    collectionsToRemoveFromDropPendingCollections.emplace(uuid, std::make_pair(opTime, nss));

    // Records the collection drop as a rename from the drop pending
    // namespace to its namespace before it was dropped.
    RenameCollectionInfo info;
    info.renameTo = nss;
    info.renameFrom = nss.makeDropPendingNamespace(opTime);

    // We do not need to check if there is already an entry in collectionsToRename
    // for this collection because it is not possible that a renameCollection occurs
    // on the same collection after it has been dropped. Thus, we know that this
    // will be the first RenameCollectionInfo entry for this collection and do not
    // need to change the renameFrom entry to account for multiple renames.
    collectionsToRename[uuid] = info;
}

Status FixUpInfo::recordDropTargetInfo(const BSONElement& dropTarget,
                                       const BSONObj& obj,
                                       OpTime opTime) {
    StatusWith<UUID> dropTargetUUIDStatus = UUID::parse(dropTarget);
    if (!dropTargetUUIDStatus.isOK()) {
        std::string message = str::stream() << "Unable to roll back renameCollection. Cannot parse "
                                               "dropTarget UUID. Returned status: "
                                            << redact(dropTargetUUIDStatus.getStatus())
                                            << ", oplog entry: " << redact(obj);
        error() << message;
        return dropTargetUUIDStatus.getStatus();
    }
    UUID dropTargetUUID = dropTargetUUIDStatus.getValue();

    // The namespace of the collection that was dropped is the same namespace
    // that we are trying to rename the collection to.
    NamespaceString droppedNs(obj.getStringField("to"));

    // Records the information necessary for undoing the dropTarget.
    recordRollingBackDrop(droppedNs, opTime, dropTargetUUID);

    return Status::OK();
}

Status rollback_internal::updateFixUpInfoFromLocalOplogEntry(FixUpInfo& fixUpInfo,
                                                             const BSONObj& ourObj,
                                                             bool isNestedApplyOpsCommand) {

    // Checks that the oplog entry is smaller than 512 MB. We do not roll back if the
    // oplog entry is larger than 512 MB.
    if (ourObj.objsize() > 512 * 1024 * 1024)
        throw RSFatalException(str::stream() << "Rollback too large, oplog size: "
                                             << ourObj.objsize());

    // If required fields are not present in the BSONObj for an applyOps entry, create these fields
    // and populate them with dummy values before parsing ourObj as an oplog entry.
    BSONObjBuilder bob;
    if (isNestedApplyOpsCommand) {
        if (!ourObj.hasField("ts")) {
            bob.appendTimestamp("ts");
        }
        if (!ourObj.hasField("h")) {
            long long dummyHash = 0;
            bob.append("h", dummyHash);
        }
    }
    bob.appendElements(ourObj);
    BSONObj fixedObj = bob.obj();

    // Parse the oplog entry.
    const OplogEntry oplogEntry(fixedObj);

    if (isNestedApplyOpsCommand) {
        LOG(2) << "Updating rollback FixUpInfo for nested applyOps oplog entry: "
               << redact(oplogEntry.toBSON());
    }

    // Extract the op's collection namespace and UUID.
    NamespaceString nss = oplogEntry.getNss();
    auto uuid = oplogEntry.getUuid();

    if (oplogEntry.getOpType() == OpTypeEnum::kNoop)
        return Status::OK();

    if (oplogEntry.getNss().isEmpty()) {
        throw RSFatalException(str::stream() << "Local op on rollback has no ns: "
                                             << redact(oplogEntry.toBSON()));
    }

    auto obj = oplogEntry.getOperationToApply();
    if (obj.isEmpty()) {
        throw RSFatalException(str::stream() << "Local op on rollback has no object field: "
                                             << redact(oplogEntry.toBSON()));
    }

    // If the operation being rolled back has a txnNumber, then the corresponding entry in the
    // session transaction table needs to be refetched.
    const auto& operationSessionInfo = oplogEntry.getOperationSessionInfo();
    auto txnNumber = operationSessionInfo.getTxnNumber();
    if (txnNumber) {
        auto sessionId = operationSessionInfo.getSessionId();
        invariant(sessionId);
        invariant(oplogEntry.getStatementId());

        auto transactionTableUUID = fixUpInfo.transactionTableUUID;
        if (transactionTableUUID) {
            BSONObjBuilder txnBob;
            txnBob.append("_id", sessionId->toBSON());
            auto txnObj = txnBob.obj();

            DocID txnDoc(txnObj, txnObj.firstElement(), transactionTableUUID.get());
            txnDoc.ns = NamespaceString::kSessionTransactionsTableNamespace.ns();

            fixUpInfo.docsToRefetch.insert(txnDoc);
            fixUpInfo.refetchTransactionDocs = true;
        } else {
            throw RSFatalException(
                str::stream() << NamespaceString::kSessionTransactionsTableNamespace.ns()
                              << " does not have a UUID, but local op has a transaction number: "
                              << redact(oplogEntry.toBSON()));
        }
    }

    if (oplogEntry.getOpType() == OpTypeEnum::kCommand) {

        // The first element of the object is the name of the command
        // and the collection it is acting on, e.x. {renameCollection: "test.x"}.
        BSONElement first = obj.firstElement();

        switch (oplogEntry.getCommandType()) {
            case OplogEntry::CommandType::kCreate: {
                // Example create collection oplog entry
                // {
                //     ts: ...,
                //     h: ...,
                //     op: "c",
                //     ns: "foo.$cmd",
                //     ui: BinData(...),
                //     o: {
                //            create: "abc", ...
                //        }
                //     ...
                // }

                fixUpInfo.collectionsToDrop.insert(*uuid);
                return Status::OK();
            }
            case OplogEntry::CommandType::kDrop: {
                // Example drop collection oplog entry
                // {
                //     ts: ...,
                //     h: ...,
                //     op: "c",
                //     ns: "foo.$cmd",
                //     ui: BinData(...),
                //     o: {
                //            drop: "abc"
                //        }
                //     ...
                // }
                NamespaceString collectionNamespace(nss.getSisterNS(first.valuestr()));

                // Registers the collection to be removed from the drop pending collection
                // reaper and to be renamed from its drop pending namespace to original namespace.
                fixUpInfo.recordRollingBackDrop(collectionNamespace, oplogEntry.getOpTime(), *uuid);

                return Status::OK();
            }
            case OplogEntry::CommandType::kDropIndexes: {
                // Example drop indexes objects
                //     o: {
                //            dropIndexes: "x",
                //            index: "x_1"
                //        }
                //     o2:{
                //            v: 2,
                //            key: { x: 1 },
                //            name: "x_1",
                //            ns: "foo.x"
                //        }

                string ns = nss.db().toString() + '.' + first.valuestr();

                string indexName;
                auto status = bsonExtractStringField(obj, "index", &indexName);
                if (!status.isOK()) {
                    severe()
                        << "Missing index name in dropIndexes operation on rollback, document: "
                        << redact(oplogEntry.toBSON());
                    throw RSFatalException(
                        "Missing index name in dropIndexes operation on rollback.");
                }

                BSONObj obj2 = oplogEntry.getObject2().get().getOwned();

                // Inserts the index name and the index spec of the index to be created into the map
                // of index name and index specs that need to be created for the given collection.
                fixUpInfo.indexesToCreate[*uuid].insert(
                    std::pair<std::string, BSONObj>(indexName, obj2));

                return Status::OK();
            }
            case OplogEntry::CommandType::kCreateIndexes: {
                // Example create indexes obj
                // o:{
                //       createIndex: x,
                //       v: 2,
                //       key: { x: 1 },
                //       name: "x_1",
                //   }

                string indexName;
                auto status = bsonExtractStringField(obj, "name", &indexName);
                if (!status.isOK()) {
                    severe()
                        << "Missing index name in createIndexes operation on rollback, document: "
                        << redact(oplogEntry.toBSON());
                    throw RSFatalException(
                        "Missing index name in createIndexes operation on rollback.");
                }

                // Checks if a drop was previously done on this index. If so, we remove it from the
                // indexesToCreate because a dropIndex and createIndex operation on the same
                // collection for the same index cancel each other out. We do not record the
                // createIndexes command in the fixUpInfo struct since applying both of these
                // commands will lead to the same final state as not applying either of the
                // commands. We only cancel out in the direction of [create] -> [drop] indexes
                // because it is possible that in the [drop] -> [create] direction, when we create
                // an index with the same name it may have a different index spec from that index
                // that was previously dropped.
                if (fixUpInfo.removeRedundantIndexCommands(*uuid, indexName)) {
                    return Status::OK();
                }

                // Inserts the index name to be dropped into the set of indexes that
                // need to be dropped for the collection.
                fixUpInfo.indexesToDrop[*uuid].insert(indexName);

                return Status::OK();
            }
            case OplogEntry::CommandType::kRenameCollection: {
                // Example rename collection obj
                // o:{
                //        renameCollection: "foo.x",
                //        to: "foo.y",
                //        stayTemp: false,
                //        dropTarget: BinData(...),
                //   }

                // dropTarget will be false if no collection is dropped during the rename.
                // The ui field will contain the UUID of the new collection that is created.

                BSONObj cmd = obj;

                std::string ns = first.valuestrsafe();
                if (ns.empty()) {
                    std::string message = str::stream()
                        << "Collection name missing from oplog entry: " << redact(obj);
                    log() << message;
                    return Status(ErrorCodes::UnrecoverableRollbackError, message);
                }

                // Checks if dropTarget is present. If it has a UUID value, we need to
                // make sure to un-drop the collection that was dropped in the process
                // of renaming.
                if (auto dropTarget = obj.getField("dropTarget")) {
                    auto status =
                        fixUpInfo.recordDropTargetInfo(dropTarget, obj, oplogEntry.getOpTime());
                    if (!status.isOK()) {
                        return status;
                    }
                }

                RenameCollectionInfo info;
                info.renameTo = NamespaceString(ns);
                info.renameFrom = NamespaceString(obj.getStringField("to"));

                // Checks if this collection has been renamed before within the same database.
                // If it has been, update the renameFrom field of the RenameCollectionInfo
                // that we will use to create the oplog entry necessary to rename the
                // collection back to its original state.
                auto collToRename = fixUpInfo.collectionsToRename.find(*uuid);
                if (collToRename != fixUpInfo.collectionsToRename.end()) {
                    info.renameFrom = (*collToRename).second.renameFrom;
                }
                fixUpInfo.collectionsToRename[*uuid] = info;

                // Because of the stayTemp field, we add any collections that have been renamed
                // to collectionsToResyncMetadata to ensure that the collection is properly set
                // as either a temporary or permanent collection.
                fixUpInfo.collectionsToResyncMetadata.insert(*uuid);

                return Status::OK();
            }
            case OplogEntry::CommandType::kDropDatabase: {
                // Example drop database oplog entry
                // {
                //     ts: ...,
                //     h: ...,
                //     op: "c",
                //     ns: "foo.$cmd",
                //     o:{
                //            "dropDatabase": 1
                //        }
                //     ...
                // }

                // Since we wait for all internal collection drops to be committed before recording
                // a 'dropDatabase' oplog entry, this will always create an empty database.
                // Creating an empty database doesn't mean anything, so we do nothing.
                return Status::OK();
            }
            case OplogEntry::CommandType::kCollMod: {
                for (auto field : obj) {
                    // Example collMod obj
                    // o:{
                    //       collMod : "x",
                    //       validationLevel : "off",
                    //       index: {
                    //                  name: "indexName_1",
                    //                  expireAfterSeconds: 600
                    //              }
                    //    }

                    const auto modification = field.fieldNameStringData();
                    if (modification == "collMod") {
                        continue;  // Skips the command name. The first field in the obj will be the
                                   // command name.
                    }

                    if (modification == "validator" || modification == "validationAction" ||
                        modification == "validationLevel" || modification == "usePowerOf2Sizes" ||
                        modification == "noPadding") {
                        fixUpInfo.collectionsToResyncMetadata.insert(*uuid);
                        continue;
                    }
                    // Some collMod fields cannot be rolled back, such as the index field.
                    string message = "Cannot roll back a collMod command: ";
                    severe() << message << redact(obj);
                    throw RSFatalException(message);
                }
                return Status::OK();
            }
            case OplogEntry::CommandType::kApplyOps: {
                // Example Apply Ops oplog entry
                //{
                //    op : "c",
                //    ns : admin.$cmd,
                //    o : {
                //             applyOps : [ {
                //                            op : "u", // must be idempotent!
                //                            ns : "test.x",
                //                            ui : BinData(...),
                //                            o2 : {
                //                                _id : 1
                //                            },
                //                            o : {
                //                                _id : 2
                //                            }
                //                        }]
                //         }
                // }
                if (first.type() != Array) {
                    std::string message = str::stream()
                        << "Expected applyOps argument to be an array; found " << redact(first);
                    severe() << message;
                    return Status(ErrorCodes::UnrecoverableRollbackError, message);
                }
                for (const auto& subopElement : first.Array()) {
                    if (subopElement.type() != Object) {
                        std::string message = str::stream()
                            << "Expected applyOps operations to be of Object type, but found "
                            << redact(subopElement);
                        severe() << message;
                        return Status(ErrorCodes::UnrecoverableRollbackError, message);
                    }
                    // In applyOps, the object contains an array of different oplog entries, we call
                    // updateFixUpInfoFromLocalOplogEntry here in order to record the information
                    // needed for rollback that is contained within the applyOps, creating a nested
                    // call.
                    auto subStatus =
                        updateFixUpInfoFromLocalOplogEntry(fixUpInfo, subopElement.Obj(), true);
                    if (!subStatus.isOK()) {
                        return subStatus;
                    }
                }
                return Status::OK();
            }
            default: {
                std::string message = str::stream() << "Can't roll back this command yet: "
                                                    << " cmdname = " << first.fieldName();
                severe() << message << " document: " << redact(obj);
                throw RSFatalException(message);
            }
        }
    }

    // If we are inserting/updating/deleting a document in the oplog entry, we will update
    // the doc._id field when we actually insert the docID into the docsToRefetch set.
    DocID doc = DocID(fixedObj, BSONElement(), *uuid);

    doc._id = oplogEntry.getIdElement();
    if (doc._id.eoo()) {
        std::string message = str::stream() << "Cannot roll back op with no _id. ns: " << nss.ns();
        severe() << message << ", document: " << redact(oplogEntry.toBSON());
        throw RSFatalException(message);
    }
    fixUpInfo.docsToRefetch.insert(doc);
    return Status::OK();
}

namespace {

/**
 * This must be called before making any changes to our local data and after fetching any
 * information from the upstream node. If any information is fetched from the upstream node after we
 * have written locally, the function must be called again.
 */
void checkRbidAndUpdateMinValid(OperationContext* opCtx,
                                const int rbid,
                                const RollbackSource& rollbackSource,
                                ReplicationProcess* replicationProcess) {
    // It is important that the steps are performed in order to avoid racing with upstream
    // rollbacks.
    // 1. Gets the last doc in their oplog.
    // 2. Gets their RBID and fail if it has changed.
    // 3. Sets our minValid to the previously fetched OpTime of the top of their oplog.
    const auto newMinValidDoc = rollbackSource.getLastOperation();
    if (newMinValidDoc.isEmpty()) {
        uasserted(40500, "rollback error newest oplog entry on source is missing or empty");
    }
    if (rbid != rollbackSource.getRollbackId()) {
        // Our source rolled back so the data we received is not necessarily consistent.
        uasserted(40508, "rollback rbid on source changed during rollback, canceling this attempt");
    }

    // We have items we are writing that aren't from a point-in-time. Thus, it is best not to come
    // online until we get to that point in freshness. In other words, we do not transition from
    // RECOVERING state to SECONDARY state until we have reached the minValid oplog entry.

    OpTime minValid = fassert(40492, OpTime::parseFromOplogEntry(newMinValidDoc));
    log() << "Setting minvalid to " << minValid;

    // This method is only used with storage engines that do not support recover to stable
    // timestamp. As a result, the timestamp on the 'appliedThrough' update does not matter.
    invariant(!opCtx->getServiceContext()->getStorageEngine()->supportsRecoverToStableTimestamp());
    replicationProcess->getConsistencyMarkers()->clearAppliedThrough(opCtx, {});
    replicationProcess->getConsistencyMarkers()->setMinValid(opCtx, minValid);

    if (MONGO_FAIL_POINT(rollbackHangThenFailAfterWritingMinValid)) {

        // This log output is used in jstests so please leave it.
        log() << "rollback - rollbackHangThenFailAfterWritingMinValid fail point "
                 "enabled. Blocking until fail point is disabled.";
        while (MONGO_FAIL_POINT(rollbackHangThenFailAfterWritingMinValid)) {
            invariant(!globalInShutdownDeprecated());  // It is an error to shutdown while enabled.
            mongo::sleepsecs(1);
        }
        uasserted(40502,
                  "failing rollback due to rollbackHangThenFailAfterWritingMinValid fail point");
    }
}

/**
 * Drops an index from the collection based on its name by removing it from the indexCatalog of the
 * collection.
 */
void dropIndex(OperationContext* opCtx,
               IndexCatalog* indexCatalog,
               const string& indexName,
               NamespaceString& nss) {
    bool includeUnfinishedIndexes = false;
    auto indexDescriptor =
        indexCatalog->findIndexByName(opCtx, indexName, includeUnfinishedIndexes);
    if (!indexDescriptor) {
        warning() << "Rollback failed to drop index " << indexName << " in " << nss.toString()
                  << ": index not found.";
        return;
    }
    WriteUnitOfWork wunit(opCtx);
    auto status = indexCatalog->dropIndex(opCtx, indexDescriptor);
    if (!status.isOK()) {
        severe() << "Rollback failed to drop index " << indexName << " in " << nss.toString()
                 << ": " << redact(status);
    }
    wunit.commit();
}

/**
 * Rolls back all createIndexes operations for the collection by dropping the
 * created indexes.
 */
void rollbackCreateIndexes(OperationContext* opCtx, UUID uuid, std::set<std::string> indexNames) {

    NamespaceString nss = UUIDCatalog::get(opCtx).lookupNSSByUUID(uuid);
    Lock::DBLock dbLock(opCtx, nss.db(), MODE_X);
    Collection* collection = UUIDCatalog::get(opCtx).lookupCollectionByUUID(uuid);

    // If we cannot find the collection, we skip over dropping the index.
    if (!collection) {
        LOG(2) << "Cannot find the collection with uuid: " << uuid.toString()
               << " in UUID catalog during roll back of a createIndexes command.";
        return;
    }

    // If we cannot find the index catalog, we skip over dropping the index.
    auto indexCatalog = collection->getIndexCatalog();
    if (!indexCatalog) {
        LOG(2) << "Cannot find the index catalog in collection with uuid: " << uuid.toString()
               << " during roll back of a createIndexes command.";
        return;
    }

    for (auto itIndex = indexNames.begin(); itIndex != indexNames.end(); itIndex++) {
        const string& indexName = *itIndex;

        log() << "Dropping index in rollback for collection: " << nss << ", UUID: " << uuid
              << ", index: " << indexName;

        dropIndex(opCtx, indexCatalog, indexName, nss);

        LOG(1) << "Dropped index in rollback for collection: " << nss << ", UUID: " << uuid
               << ", index: " << indexName;
    }
}

/**
 * Rolls back all the dropIndexes operations for the collection by re-creating
 * the indexes that were dropped.
 */
void rollbackDropIndexes(OperationContext* opCtx,
                         UUID uuid,
                         std::map<std::string, BSONObj> indexNames) {

    NamespaceString nss = UUIDCatalog::get(opCtx).lookupNSSByUUID(uuid);
    Lock::DBLock dbLock(opCtx, nss.db(), MODE_X);
    Collection* collection = UUIDCatalog::get(opCtx).lookupCollectionByUUID(uuid);
    // If we cannot find the collection, we skip over dropping the index.
    if (!collection) {
        LOG(2) << "Cannot find the collection with uuid: " << uuid.toString()
               << "in UUID catalog during roll back of a dropIndexes command.";
        return;
    }

    for (auto itIndex = indexNames.begin(); itIndex != indexNames.end(); itIndex++) {

        const string indexName = itIndex->first;
        BSONObj indexSpec = itIndex->second;
        // We replace the namespace field because it is possible that a
        // renameCollection command has occurred, changing the namespace of the
        // collection from what it initially was during the creation of this index.
        BSONObjBuilder updatedNss;
        updatedNss.append("ns", nss.ns());

        BSONObj updatedNssObj = updatedNss.obj();
        indexSpec = indexSpec.addField(updatedNssObj.firstElement());

        log() << "Creating index in rollback for collection: " << nss << ", UUID: " << uuid
              << ", index: " << indexName;

        createIndexForApplyOps(opCtx, indexSpec, nss, {}, OplogApplication::Mode::kRecovering);

        LOG(1) << "Created index in rollback for collection: " << nss << ", UUID: " << uuid
               << ", index: " << indexName;
    }
}

/**
 * Drops the given collection from the database.
 */
void dropCollection(OperationContext* opCtx,
                    NamespaceString nss,
                    Collection* collection,
                    Database* db) {
    if (RollbackImpl::shouldCreateDataFiles()) {
        Helpers::RemoveSaver removeSaver("rollback", "", nss.ns());

        // Performs a collection scan and writes all documents in the collection to disk
        // in order to keep an archive of items that were rolled back.
        auto exec = InternalPlanner::collectionScan(
            opCtx, nss.toString(), collection, PlanExecutor::YIELD_AUTO);
        BSONObj curObj;
        PlanExecutor::ExecState execState;
        while (PlanExecutor::ADVANCED == (execState = exec->getNext(&curObj, NULL))) {
            auto status = removeSaver.goingToDelete(curObj);
            if (!status.isOK()) {
                severe() << "Rolling back createCollection on " << nss
                         << " failed to write document to remove saver file: " << redact(status);
                throw RSFatalException(
                    "Rolling back createCollection. Failed to write document to remove saver "
                    "file.");
            }
        }

        // If we exited the above for loop with any other execState than IS_EOF, this means that
        // a FAILURE or DEAD state was returned. If a DEAD state occurred, the collection or
        // database that we are attempting to save may no longer be valid. If a FAILURE state
        // was returned, either an unrecoverable error was thrown by exec, or we attempted to
        // retrieve data that could not be provided by the PlanExecutor. In both of these cases
        // it is necessary for a full resync of the server.

        if (execState != PlanExecutor::IS_EOF) {
            if (execState == PlanExecutor::FAILURE &&
                WorkingSetCommon::isValidStatusMemberObject(curObj)) {
                Status errorStatus = WorkingSetCommon::getMemberObjectStatus(curObj);
                severe() << "Rolling back createCollection on " << nss << " failed with "
                         << redact(errorStatus) << ". A full resync is necessary.";
                throw RSFatalException(
                    "Rolling back createCollection failed. A full resync is necessary.");
            } else {
                severe() << "Rolling back createCollection on " << nss
                         << " failed. A full resync is necessary.";
                throw RSFatalException(
                    "Rolling back createCollection failed. A full resync is necessary.");
            }
        }
    }

    WriteUnitOfWork wunit(opCtx);

    // We permanently drop the collection rather than 2-phase drop the collection
    // here. By not passing in an opTime to dropCollectionEvenIfSystem() the collection
    // is immediately dropped.
    fassert(40504, db->dropCollectionEvenIfSystem(opCtx, nss));
    wunit.commit();
}

/**
 * Renames a collection out of the way when another collection during rollback
 * is attempting to use the same namespace.
 */
void renameOutOfTheWay(OperationContext* opCtx, RenameCollectionInfo info, Database* db) {

    // Finds the UUID of the collection that we are renaming out of the way.
    auto collection = db->getCollection(opCtx, info.renameTo);
    invariant(collection);

    // Creates the oplog entry to temporarily rename the collection that is
    // preventing the renameCollection command from rolling back to a unique
    // namespace.
    auto tmpNameResult = db->makeUniqueCollectionNamespace(opCtx, "rollback.tmp%%%%%");
    if (!tmpNameResult.isOK()) {
        severe() << "Unable to generate temporary namespace to rename collection " << info.renameTo
                 << " out of the way. " << tmpNameResult.getStatus().reason();
        throw RSFatalException(
            "Unable to generate temporary namespace to rename collection out of the way.");
    }
    const auto& tempNss = tmpNameResult.getValue();

    LOG(2) << "Attempted to rename collection from " << info.renameFrom << " to " << info.renameTo
           << " but " << info.renameTo << " exists already. Temporarily renaming collection "
           << info.renameTo << " with UUID " << collection->uuid().get() << " out of the way to "
           << tempNss;

    // Renaming the collection that was clashing with the attempted rename
    // operation to a different collection name.
    auto uuid = collection->uuid().get();
    auto renameStatus = renameCollectionForRollback(opCtx, tempNss, uuid);

    if (!renameStatus.isOK()) {
        severe() << "Unable to rename collection " << info.renameTo << " out of the way to "
                 << tempNss;
        throw RSFatalException("Unable to rename collection out of the way");
    }
}

/**
 * Rolls back a renameCollection operation on the given collection.
 */
void rollbackRenameCollection(OperationContext* opCtx, UUID uuid, RenameCollectionInfo info) {

    auto dbName = info.renameFrom.db();

    log() << "Attempting to rename collection with UUID: " << uuid << ", from: " << info.renameFrom
          << ", to: " << info.renameTo;
    Lock::DBLock dbLock(opCtx, dbName, MODE_X);
    auto db = DatabaseHolder::getDatabaseHolder().openDb(opCtx, dbName);
    invariant(db);

    auto status = renameCollectionForRollback(opCtx, info.renameTo, uuid);

    // If we try to roll back a collection to a collection name that currently exists
    // because another collection was renamed or created with the same collection name,
    // we temporarily rename the conflicting collection.
    if (status == ErrorCodes::NamespaceExists) {

        renameOutOfTheWay(opCtx, info, db);

        // Retrying to renameCollection command again now that the conflicting
        // collection has been renamed out of the way.
        status = renameCollectionForRollback(opCtx, info.renameTo, uuid);

        if (!status.isOK()) {
            severe() << "Rename collection failed to roll back twice. We were unable to rename "
                     << "collection " << info.renameFrom << " to " << info.renameTo << ". "
                     << status.toString();
            throw RSFatalException(
                "Rename collection failed to roll back twice. We were unable to rename "
                "the collection.");
        }
    } else if (!status.isOK()) {
        severe() << "Unable to roll back renameCollection command: " << status.toString();
        throw RSFatalException("Unable to rollback renameCollection command");
    }

    LOG(1) << "Renamed collection with UUID: " << uuid << ", from: " << info.renameFrom
           << ", to: " << info.renameTo;
}

Status _syncRollback(OperationContext* opCtx,
                     const OplogInterface& localOplog,
                     const RollbackSource& rollbackSource,
                     int requiredRBID,
                     ReplicationCoordinator* replCoord,
                     ReplicationProcess* replicationProcess) {
    invariant(!opCtx->lockState()->isLocked());

    FixUpInfo how;
    log() << "Starting rollback. Sync source: " << rollbackSource.getSource() << rsLog;
    how.rbid = rollbackSource.getRollbackId();
    uassert(
        40506, "Upstream node rolled back. Need to retry our rollback.", how.rbid == requiredRBID);

    // Find the UUID of the transactions collection. An OperationContext is required because the
    // UUID is not known at compile time, so the SessionCatalog needs to load the collection.
    how.transactionTableUUID = MongoDSessionCatalog::getTransactionTableUUID(opCtx);

    log() << "Finding the Common Point";
    try {

        auto processOperationForFixUp = [&how](const BSONObj& operation) {
            return updateFixUpInfoFromLocalOplogEntry(how, operation, false);
        };

        // Calls syncRollBackLocalOperations to run updateFixUpInfoFromLocalOplogEntry
        // on each oplog entry up until the common point.
        auto res = syncRollBackLocalOperations(
            localOplog, rollbackSource.getOplog(), processOperationForFixUp);
        if (!res.isOK()) {
            const auto status = res.getStatus();
            switch (status.code()) {
                case ErrorCodes::OplogStartMissing:
                case ErrorCodes::UnrecoverableRollbackError:
                    return status;
                default:
                    throw RSFatalException(status.toString());
            }
        }

        how.commonPoint = res.getValue().getOpTime();
        how.commonPointOurDiskloc = res.getValue().getRecordId();
        how.removeRedundantOperations();
    } catch (const RSFatalException& e) {
        return Status(ErrorCodes::UnrecoverableRollbackError,
                      str::stream()
                          << "need to rollback, but unable to determine common point between"
                             " local and remote oplog: "
                          << e.what());
    }

    OpTime commonPoint = how.commonPoint;
    OpTime lastCommittedOpTime = replCoord->getLastCommittedOpTime();
    OpTime committedSnapshot = replCoord->getCurrentCommittedSnapshotOpTime();

    log() << "Rollback common point is " << commonPoint;

    // Rollback common point should be >= the replication commit point.
    invariant(commonPoint.getTimestamp() >= lastCommittedOpTime.getTimestamp());
    invariant(commonPoint >= lastCommittedOpTime);

    // Rollback common point should be >= the committed snapshot optime.
    invariant(commonPoint.getTimestamp() >= committedSnapshot.getTimestamp());
    invariant(commonPoint >= committedSnapshot);

    try {
        ON_BLOCK_EXIT([&] {
            auto status = replicationProcess->incrementRollbackID(opCtx);
            fassert(40497, status);
        });
        syncFixUp(opCtx, how, rollbackSource, replCoord, replicationProcess);
    } catch (const RSFatalException& e) {
        return Status(ErrorCodes::UnrecoverableRollbackError, e.what());
    }

    if (MONGO_FAIL_POINT(rollbackHangBeforeFinish)) {
        // This log output is used in js tests so please leave it.
        log() << "rollback - rollbackHangBeforeFinish fail point "
                 "enabled. Blocking until fail point is disabled.";
        while (MONGO_FAIL_POINT(rollbackHangBeforeFinish)) {
            invariant(!globalInShutdownDeprecated());  // It is an error to shutdown while enabled.
            mongo::sleepsecs(1);
        }
    }

    return Status::OK();
}

}  // namespace

void rollback_internal::syncFixUp(OperationContext* opCtx,
                                  const FixUpInfo& fixUpInfo,
                                  const RollbackSource& rollbackSource,
                                  ReplicationCoordinator* replCoord,
                                  ReplicationProcess* replicationProcess) {
    unsigned long long totalSize = 0;

    // UUID -> doc id -> doc
    stdx::unordered_map<UUID, std::map<DocID, BSONObj>, UUID::Hash> goodVersions;
    auto& catalog = UUIDCatalog::get(opCtx);

    // Fetches all the goodVersions of each document from the current sync source.
    unsigned long long numFetched = 0;

    log() << "Starting refetching documents";

    for (auto&& doc : fixUpInfo.docsToRefetch) {
        invariant(!doc._id.eoo());  // This is checked when we insert to the set.

        UUID uuid = doc.uuid;
        NamespaceString nss = catalog.lookupNSSByUUID(uuid);

        try {
            LOG(2) << "Refetching document, collection: " << nss << ", UUID: " << uuid << ", "
                   << redact(doc._id);
            // TODO : Slow. Lots of round trips.
            numFetched++;

            BSONObj good;
            NamespaceString resNss;
            std::tie(good, resNss) =
                rollbackSource.findOneByUUID(nss.db().toString(), uuid, doc._id.wrap());

            // To prevent inconsistencies in the transactions collection, rollback fails if the UUID
            // of the collection is different on the sync source than on the node rolling back,
            // forcing an initial sync. This is detected if the returned namespace for a refetch of
            // a transaction table document is not "config.transactions," which implies a rename or
            // drop of the collection occured on either node.
            if (uuid == fixUpInfo.transactionTableUUID &&
                resNss != NamespaceString::kSessionTransactionsTableNamespace) {
                throw RSFatalException(
                    str::stream()
                    << "A fetch on the transactions collection returned an unexpected namespace: "
                    << resNss.ns()
                    << ". The transactions collection cannot be correctly rolled back, a full "
                       "resync is required.");
            }

            totalSize += good.objsize();

            // Checks that the total amount of data that needs to be refetched is at most
            // 300 MB. We do not roll back more than 300 MB of documents in order to
            // prevent out of memory errors from too much data being stored. See SERVER-23392.
            if (totalSize >= 300 * 1024 * 1024) {
                throw RSFatalException("replSet too much data to roll back.");
            }

            // Note good might be empty, indicating we should delete it.
            goodVersions[uuid].insert(std::pair<DocID, BSONObj>(doc, good));

        } catch (const DBException& ex) {
            // If the collection turned into a view, we might get an error trying to
            // refetch documents, but these errors should be ignored, as we'll be creating
            // the view during oplog replay.
            // Collection may be dropped on the sync source, in which case it will be dropped during
            // oplog replay. So it is safe to ignore NamespaceNotFound errors while trying to
            // refetch documents.
            if (ex.code() == ErrorCodes::CommandNotSupportedOnView ||
                ex.code() == ErrorCodes::NamespaceNotFound)
                continue;

            log() << "Rollback couldn't re-fetch from uuid: " << uuid << " _id: " << redact(doc._id)
                  << ' ' << numFetched << '/' << fixUpInfo.docsToRefetch.size() << ": "
                  << redact(ex);
            throw;
        }
    }

    log() << "Finished refetching documents. Total size of documents refetched: "
          << goodVersions.size();

    log() << "Checking the RollbackID and updating the MinValid if necessary";

    checkRbidAndUpdateMinValid(opCtx, fixUpInfo.rbid, rollbackSource, replicationProcess);

    invariant(!fixUpInfo.commonPointOurDiskloc.isNull());

    // Rolls back createIndexes commands by dropping the indexes that were created. It is
    // necessary to roll back createIndexes commands before dropIndexes commands because
    // it is possible that we previously dropped an index with the same name but a different
    // index spec. If we attempt to re-create an index that has the same name as an existing
    // index, the operation will fail. Thus, we roll back createIndexes commands first in
    // order to ensure that no collisions will occur when we re-create previously dropped
    // indexes.
    // We drop indexes before renaming collections so that if a collection name gets longer,
    // any indexes with names that are now too long will already be dropped.
    log() << "Rolling back createIndexes commands.";
    for (auto it = fixUpInfo.indexesToDrop.begin(); it != fixUpInfo.indexesToDrop.end(); it++) {

        UUID uuid = it->first;
        std::set<std::string> indexNames = it->second;

        rollbackCreateIndexes(opCtx, uuid, indexNames);
    }

    log() << "Dropping collections to roll back create operations";

    // Drops collections before updating individual documents. We drop these collections before
    // rolling back any other commands to prevent namespace collisions that may occur
    // when undoing renameCollection operations.
    for (auto uuid : fixUpInfo.collectionsToDrop) {

        // Checks that if the collection is going to be dropped, all commands that
        // were done on the collection to be dropped were removed during the function
        // call to removeRedundantOperations().
        invariant(!fixUpInfo.indexesToDrop.count(uuid));
        invariant(!fixUpInfo.indexesToCreate.count(uuid));
        invariant(!fixUpInfo.collectionsToRename.count(uuid));
        invariant(!fixUpInfo.collectionsToResyncMetadata.count(uuid));

        NamespaceString nss = UUIDCatalog::get(opCtx).lookupNSSByUUID(uuid);
        log() << "Dropping collection: " << nss << ", UUID: " << uuid;
        AutoGetDb dbLock(opCtx, nss.db(), MODE_X);

        Database* db = dbLock.getDb();
        if (db) {
            Collection* collection = UUIDCatalog::get(opCtx).lookupCollectionByUUID(uuid);
            dropCollection(opCtx, nss, collection, db);
            LOG(1) << "Dropped collection: " << nss << ", UUID: " << uuid;
        }
    }

    // Rolling back renameCollection commands.
    log() << "Rolling back renameCollection commands and collection drop commands.";

    for (auto it = fixUpInfo.collectionsToRename.begin(); it != fixUpInfo.collectionsToRename.end();
         it++) {

        UUID uuid = it->first;
        RenameCollectionInfo info = it->second;

        rollbackRenameCollection(opCtx, uuid, info);
    }

    log() << "Rolling back collections pending being dropped: Removing them from the list of "
             "drop-pending collections in the DropPendingCollectionReaper.";

    // Roll back any drop-pending collections. This must be done first so that the collection
    // exists when we attempt to resync its metadata or insert documents into it.
    for (const auto& collPair : fixUpInfo.collectionsToRemoveFromDropPendingCollections) {
        const auto& optime = collPair.second.first;
        const auto& collectionNamespace = collPair.second.second;
        LOG(1) << "Rolling back collection pending being dropped for OpTime: " << optime
               << ", collection: " << collectionNamespace;
        DropPendingCollectionReaper::get(opCtx)->rollBackDropPendingCollection(
            opCtx, optime, collectionNamespace);
    }

    // Full collection data and metadata resync.
    if (!fixUpInfo.collectionsToResyncMetadata.empty()) {

        // Retrieves collections from the sync source in order to obtain the collection
        // flags needed to roll back collMod operations. We roll back collMod operations
        // after create/renameCollection/drop commands in order to ensure that the
        // collections that we want to change actually exist. For example, if a collMod
        // occurs and then the collection is dropped. If we do not first re-create the
        // collection, we will not be able to retrieve the collection's catalog entries.
        for (auto uuid : fixUpInfo.collectionsToResyncMetadata) {
            NamespaceString nss = UUIDCatalog::get(opCtx).lookupNSSByUUID(uuid);

            log() << "Resyncing collection metadata for collection: " << nss << ", UUID: " << uuid;

            Lock::DBLock dbLock(opCtx, nss.db(), MODE_X);

            auto db = DatabaseHolder::getDatabaseHolder().openDb(opCtx, nss.db().toString());
            invariant(db);

            Collection* collection = UUIDCatalog::get(opCtx).lookupCollectionByUUID(uuid);
            invariant(collection);

            auto cce = collection->getCatalogEntry();

            auto infoResult = rollbackSource.getCollectionInfoByUUID(nss.db().toString(), uuid);

            if (!infoResult.isOK()) {
                // The collection was dropped by the sync source so we can't correctly change it
                // here. If we get to the roll-forward phase, we will drop it then. If the drop
                // is rolled back upstream and we restart, we expect to still have the
                // collection.

                log() << nss.ns() << " not found on remote host, so we do not roll back collmod "
                                     "operation. Instead, we will drop the collection soon.";
                continue;
            }

            auto info = infoResult.getValue();
            CollectionOptions options;

            // Updates the collection flags.
            if (auto optionsField = info["options"]) {
                if (optionsField.type() != Object) {
                    throw RSFatalException(str::stream() << "Failed to parse options " << info
                                                         << ": expected 'options' to be an "
                                                         << "Object, got "
                                                         << typeName(optionsField.type()));
                }

                // Removes the option.uuid field. We do not allow the options.uuid field
                // to be parsed while in parseForCommand mode in order to ensure that the
                // user cannot set the UUID of a collection.
                BSONObj optionsFieldObj = optionsField.Obj();
                optionsFieldObj = optionsFieldObj.removeField("uuid");

                auto status = options.parse(optionsFieldObj, CollectionOptions::parseForCommand);
                if (!status.isOK()) {
                    throw RSFatalException(str::stream() << "Failed to parse options " << info
                                                         << ": "
                                                         << status.toString());
                }
                // TODO(SERVER-27992): Set options.uuid.

            } else {
                // Use default options.
            }

            WriteUnitOfWork wuow(opCtx);

            // Set collection to whatever temp status is on the sync source.
            cce->setIsTemp(opCtx, options.temp);

            // Resets collection user flags such as noPadding and usePowerOf2Sizes.
            if (options.flagsSet || cce->getCollectionOptions(opCtx).flagsSet) {
                cce->updateFlags(opCtx, options.flags);
            }

            // Set any document validation options. We update the validator fields without
            // parsing/validation, since we fetched the options object directly from the sync
            // source, and we should set our validation options to match it exactly.
            auto validatorStatus = collection->updateValidator(
                opCtx, options.validator, options.validationLevel, options.validationAction);
            if (!validatorStatus.isOK()) {
                throw RSFatalException(
                    str::stream() << "Failed to update validator for " << nss.toString() << " ("
                                  << uuid
                                  << ") with "
                                  << redact(info)
                                  << ". Got: "
                                  << validatorStatus.toString());
            }

            wuow.commit();

            LOG(1) << "Resynced collection metadata for collection: " << nss << ", UUID: " << uuid
                   << ", with: " << redact(info)
                   << ", to: " << redact(cce->getCollectionOptions(opCtx).toBSON());
        }

        // Since we read from the sync source to retrieve the metadata of the
        // collection, we must check if the sync source rolled back as well as update
        // minValid if necessary.
        log() << "Rechecking the Rollback ID and minValid";
        checkRbidAndUpdateMinValid(opCtx, fixUpInfo.rbid, rollbackSource, replicationProcess);
    }

    // Rolls back dropIndexes commands by re-creating the indexes that were dropped.
    log() << "Rolling back dropIndexes commands.";
    for (auto it = fixUpInfo.indexesToCreate.begin(); it != fixUpInfo.indexesToCreate.end(); it++) {

        UUID uuid = it->first;
        std::map<std::string, BSONObj> indexNames = it->second;

        rollbackDropIndexes(opCtx, uuid, indexNames);
    }

    log() << "Deleting and updating documents to roll back insert, update and remove "
             "operations";
    unsigned deletes = 0, updates = 0;
    time_t lastProgressUpdate = time(0);
    time_t progressUpdateGap = 10;

    for (const auto& nsAndGoodVersionsByDocID : goodVersions) {

        // Keeps an archive of items rolled back if the collection has not been dropped
        // while rolling back createCollection operations.

        const auto& uuid = nsAndGoodVersionsByDocID.first;
        unique_ptr<Helpers::RemoveSaver> removeSaver;
        invariant(!fixUpInfo.collectionsToDrop.count(uuid));

        NamespaceString nss = catalog.lookupNSSByUUID(uuid);

        if (RollbackImpl::shouldCreateDataFiles()) {
            removeSaver = std::make_unique<Helpers::RemoveSaver>("rollback", "", nss.ns());
        }

        const auto& goodVersionsByDocID = nsAndGoodVersionsByDocID.second;
        for (const auto& idAndDoc : goodVersionsByDocID) {
            time_t now = time(0);
            if (now - lastProgressUpdate > progressUpdateGap) {
                log() << deletes << " delete and " << updates
                      << " update operations processed out of " << goodVersions.size()
                      << " total operations.";
                lastProgressUpdate = now;
            }
            const DocID& doc = idAndDoc.first;
            BSONObj pattern = doc._id.wrap();  // { _id : ... }
            try {

                // TODO: Lots of overhead in context. This can be faster.
                const NamespaceString docNss(doc.ns);
                Lock::DBLock docDbLock(opCtx, docNss.db(), MODE_X);
                OldClientContext ctx(opCtx, doc.ns.toString());
                Collection* collection = catalog.lookupCollectionByUUID(uuid);

                // Adds the doc to our rollback file if the collection was not dropped while
                // rolling back createCollection operations. Does not log an error when
                // undoing an insert on a no longer existing collection. It is likely that
                // the collection was dropped as part of rolling back a createCollection
                // command and the document no longer exists.

                if (collection && removeSaver) {
                    BSONObj obj;
                    bool found = Helpers::findOne(opCtx, collection, pattern, obj, false);
                    if (found) {
                        auto status = removeSaver->goingToDelete(obj);
                        if (!status.isOK()) {
                            severe() << "Rollback cannot write document in namespace " << nss.ns()
                                     << " to archive file: " << redact(status);
                            throw RSFatalException(str::stream()
                                                   << "Rollback cannot write document in namespace "
                                                   << nss.ns()
                                                   << " to archive file.");
                        }
                    } else {
                        error() << "Rollback cannot find object: " << pattern << " in namespace "
                                << nss.ns();
                    }
                }

                if (idAndDoc.second.isEmpty()) {
                    LOG(2) << "Deleting document with: " << redact(doc._id)
                           << ", from collection: " << doc.ns << ", with UUID: " << uuid;
                    // If the document could not be found on the primary, deletes the document.
                    // TODO 1.6 : can't delete from a capped collection. Need to handle that
                    // here.
                    deletes++;

                    if (collection) {
                        if (collection->isCapped()) {
                            // Can't delete from a capped collection - so we truncate instead.
                            // if this item must go, so must all successors.

                            try {
                                // TODO: IIRC cappedTruncateAfter does not handle completely
                                // empty. This will be slow if there is no _id index in
                                // the collection.

                                const auto clock = opCtx->getServiceContext()->getFastClockSource();
                                const auto findOneStart = clock->now();
                                RecordId loc = Helpers::findOne(opCtx, collection, pattern, false);
                                if (clock->now() - findOneStart > Milliseconds(200))
                                    warning() << "Roll back slow no _id index for " << nss.ns()
                                              << " perhaps?";
                                // Would be faster but requires index:
                                // RecordId loc = Helpers::findById(nsd, pattern);
                                if (!loc.isNull()) {
                                    try {
                                        writeConflictRetry(opCtx,
                                                           "cappedTruncateAfter",
                                                           collection->ns().ns(),
                                                           [&] {
                                                               WriteUnitOfWork wunit(opCtx);
                                                               collection->cappedTruncateAfter(
                                                                   opCtx, loc, true);
                                                               wunit.commit();
                                                           });
                                    } catch (const DBException& e) {
                                        if (e.code() == 13415) {
                                            // hack: need to just make cappedTruncate do this...
                                            writeConflictRetry(
                                                opCtx, "truncate", collection->ns().ns(), [&] {
                                                    WriteUnitOfWork wunit(opCtx);
                                                    uassertStatusOK(collection->truncate(opCtx));
                                                    wunit.commit();
                                                });
                                        } else {
                                            throw;
                                        }
                                    }
                                }
                            } catch (const DBException& e) {
                                // Replicated capped collections have many ways to become
                                // inconsistent. We rely on age-out to make these problems go away
                                // eventually.

                                warning() << "Ignoring failure to roll back change to capped "
                                          << "collection " << nss.ns() << " with _id "
                                          << redact(idAndDoc.first._id.toString(
                                                 /*includeFieldName*/ false))
                                          << ": " << redact(e);
                            }
                        } else {
                            deleteObjects(opCtx,
                                          collection,
                                          nss,
                                          pattern,
                                          true,   // justOne
                                          true);  // god
                        }
                    }
                } else {
                    LOG(2) << "Updating document with: " << redact(doc._id)
                           << ", from collection: " << doc.ns << ", UUID: " << uuid
                           << ", to: " << redact(idAndDoc.second);
                    // TODO faster...
                    updates++;

                    UpdateRequest request(nss);

                    request.setQuery(pattern);
                    request.setUpdates(idAndDoc.second);
                    request.setGod();
                    request.setUpsert();
                    UpdateLifecycleImpl updateLifecycle(nss);
                    request.setLifecycle(&updateLifecycle);

                    update(opCtx, ctx.db(), request);
                }
            } catch (const DBException& e) {
                log() << "Exception in rollback ns:" << nss.ns() << ' ' << pattern.toString() << ' '
                      << redact(e) << " ndeletes:" << deletes;
                throw;
            }
        }
    }

    log() << "Rollback deleted " << deletes << " documents and updated " << updates
          << " documents.";

    log() << "Truncating the oplog at " << fixUpInfo.commonPoint.toString() << " ("
          << fixUpInfo.commonPointOurDiskloc << "), non-inclusive";

    // Cleans up the oplog.
    {
        const NamespaceString oplogNss(NamespaceString::kRsOplogNamespace);
        Lock::DBLock oplogDbLock(opCtx, oplogNss.db(), MODE_IX);
        Lock::CollectionLock oplogCollectionLoc(opCtx->lockState(), oplogNss.ns(), MODE_X);
        OldClientContext ctx(opCtx, oplogNss.ns());
        Collection* oplogCollection = ctx.db()->getCollection(opCtx, oplogNss);
        if (!oplogCollection) {
            fassertFailedWithStatusNoTrace(
                40495,
                Status(ErrorCodes::UnrecoverableRollbackError,
                       str::stream() << "Can't find " << NamespaceString::kRsOplogNamespace.ns()));
        }
        // TODO: fatal error if this throws?
        oplogCollection->cappedTruncateAfter(opCtx, fixUpInfo.commonPointOurDiskloc, false);
    }

    Status status = AuthorizationManager::get(opCtx->getServiceContext())->initialize(opCtx);
    if (!status.isOK()) {
        severe() << "Failed to reinitialize auth data after rollback: " << redact(status);
        fassertFailedNoTrace(40496);
    }

    // If necessary, clear the memory of existing sessions.
    if (fixUpInfo.refetchTransactionDocs) {
        SessionCatalog::get(opCtx)->invalidateSessions(opCtx, boost::none);
    }

    if (auto validator = LogicalTimeValidator::get(opCtx)) {
        validator->resetKeyManagerCache();
    }

    // Reload the lastAppliedOpTime and lastDurableOpTime value in the replcoord and the
    // lastAppliedHash value in bgsync to reflect our new last op. The rollback common point does
    // not necessarily represent a consistent database state. For example, on a secondary, we may
    // have rolled back to an optime that fell in the middle of an oplog application batch. We make
    // the database consistent again after rollback by applying ops forward until we reach
    // 'minValid'.
    replCoord->resetLastOpTimesFromOplog(opCtx,
                                         ReplicationCoordinator::DataConsistency::Inconsistent);
}

Status syncRollback(OperationContext* opCtx,
                    const OplogInterface& localOplog,
                    const RollbackSource& rollbackSource,
                    int requiredRBID,
                    ReplicationCoordinator* replCoord,
                    ReplicationProcess* replicationProcess) {
    invariant(opCtx);
    invariant(replCoord);

    DisableDocumentValidation validationDisabler(opCtx);
    UnreplicatedWritesBlock replicationDisabler(opCtx);
    Status status = _syncRollback(
        opCtx, localOplog, rollbackSource, requiredRBID, replCoord, replicationProcess);

    log() << "Rollback finished. The final minValid is: "
          << replicationProcess->getConsistencyMarkers()->getMinValid(opCtx) << rsLog;

    return status;
}

void rollback(OperationContext* opCtx,
              const OplogInterface& localOplog,
              const RollbackSource& rollbackSource,
              int requiredRBID,
              ReplicationCoordinator* replCoord,
              ReplicationProcess* replicationProcess,
              stdx::function<void(int)> sleepSecsFn) {
    // Set state to ROLLBACK while we are in this function. This prevents serving reads, even from
    // the oplog. This can fail if we are elected PRIMARY, in which case we better not do any
    // rolling back. If we successfully enter ROLLBACK we will only exit this function fatally or
    // after transitioning to RECOVERING. We always transition to RECOVERING regardless of success
    // or (recoverable) failure since we may be in an inconsistent state. If rollback failed before
    // writing anything, SyncTail will quickly take us to SECONDARY since are are still at our
    // original MinValid, which is fine because we may choose a sync source that doesn't require
    // rollback. If it failed after we wrote to MinValid, then we will pick a sync source that will
    // cause us to roll back to the same common point, which is fine. If we succeeded, we will be
    // consistent as soon as we apply up to/through MinValid and SyncTail will make us SECONDARY
    // then.

    {
        Lock::GlobalWrite globalWrite(opCtx);
        auto status = replCoord->setFollowerMode(MemberState::RS_ROLLBACK);
        if (!status.isOK()) {
            log() << "Cannot transition from " << replCoord->getMemberState().toString() << " to "
                  << MemberState(MemberState::RS_ROLLBACK).toString() << causedBy(status);
            return;
        }
    }

    try {
        auto status = syncRollback(
            opCtx, localOplog, rollbackSource, requiredRBID, replCoord, replicationProcess);

        // Aborts only when syncRollback detects we are in a unrecoverable state.
        // WARNING: these statuses sometimes have location codes which are lost with uassertStatusOK
        // so we need to check here first.
        if (ErrorCodes::UnrecoverableRollbackError == status.code()) {
            severe() << "Unable to complete rollback. A full resync may be needed: "
                     << redact(status);
            fassertFailedNoTrace(40507);
        }

        // In other cases, we log the message contained in the error status and retry later.
        uassertStatusOK(status);
    } catch (const DBException& ex) {
        // UnrecoverableRollbackError should only come from a returned status which is handled
        // above.
        invariant(ex.code() != ErrorCodes::UnrecoverableRollbackError);

        warning() << "Rollback cannot complete at this time (retrying later): " << redact(ex)
                  << " appliedThrough= " << replCoord->getMyLastAppliedOpTime() << " minvalid= "
                  << replicationProcess->getConsistencyMarkers()->getMinValid(opCtx);

        // Sleep a bit to allow upstream node to coalesce, if that was the cause of the failure. If
        // we failed in a way that will keep failing, but wasn't flagged as a fatal failure, this
        // will also prevent us from hot-looping and putting too much load on upstream nodes.
        sleepSecsFn(5);  // 5 seconds was chosen as a completely arbitrary amount of time.
    } catch (...) {
        std::terminate();
    }

    // At this point we are about to leave rollback.  Before we do, wait for any writes done
    // as part of rollback to be durable, and then do any necessary checks that we didn't
    // wind up rolling back something illegal.  We must wait for the rollback to be durable
    // so that if we wind up shutting down uncleanly in response to something we rolled back
    // we know that we won't wind up right back in the same situation when we start back up
    // because the rollback wasn't durable.
    opCtx->recoveryUnit()->waitUntilDurable();

    // If we detected that we rolled back the shardIdentity document as part of this rollback
    // then we must shut down to clear the in-memory ShardingState associated with the
    // shardIdentity document.
    if (ShardIdentityRollbackNotifier::get(opCtx)->didRollbackHappen()) {
        severe() << "shardIdentity document rollback detected.  Shutting down to clear "
                    "in-memory sharding state.  Restarting this process should safely return it "
                    "to a healthy state";
        fassertFailedNoTrace(40498);
    }

    auto status = replCoord->setFollowerMode(MemberState::RS_RECOVERING);
    if (!status.isOK()) {
        severe() << "Failed to transition into " << MemberState(MemberState::RS_RECOVERING)
                 << "; expected to be in state " << MemberState(MemberState::RS_ROLLBACK)
                 << "; found self in " << replCoord->getMemberState() << causedBy(status);
        fassertFailedNoTrace(40499);
    }
}

}  // namespace repl
}  // namespace mongo