SERVER-39679 Get oldest transaction time when snapshotting

14 files changed, 354 insertions, 111 deletions

diff --git a/src/mongo/db/db.cpp b/src/mongo/db/db.cpp
index a3dc258684c..5df53cc0ef3 100644
--- a/src/mongo/db/db.cpp
+++ b/src/mongo/db/db.cpp
@@ -132,6 +132,7 @@
 #include "mongo/db/storage/storage_engine_lock_file.h"
 #include "mongo/db/storage/storage_options.h"
 #include "mongo/db/system_index.h"
+#include "mongo/db/transaction_participant.h"
 #include "mongo/db/ttl.h"
 #include "mongo/db/wire_version.h"
 #include "mongo/executor/network_connection_hook.h"
@@ -522,9 +523,16 @@ ExitCode _initAndListen(int listenPort) {
 
         startFreeMonitoring(serviceContext);
 
+        auto replCoord = repl::ReplicationCoordinator::get(startupOpCtx.get());
+        invariant(replCoord);
+        if (replCoord->isReplEnabled()) {
+            storageEngine->setOldestActiveTransactionTimestampCallback(
+                TransactionParticipant::getOldestActiveTimestamp);
+        }
+
         if (serverGlobalParams.clusterRole == ClusterRole::ShardServer) {
             // Note: For replica sets, ShardingStateRecovery happens on transition to primary.
-            if (!repl::ReplicationCoordinator::get(startupOpCtx.get())->isReplEnabled()) {
+            if (!replCoord->isReplEnabled()) {
                 if (ShardingState::get(startupOpCtx.get())->enabled()) {
                     uassertStatusOK(ShardingStateRecovery::recover(startupOpCtx.get()));
                 }
@@ -553,7 +561,7 @@ ExitCode _initAndListen(int listenPort) {
                                       stdx::make_unique<LogicalTimeValidator>(keyManager));
         }
 
-        repl::ReplicationCoordinator::get(startupOpCtx.get())->startup(startupOpCtx.get());
+        replCoord->startup(startupOpCtx.get());
         if (getReplSetMemberInStandaloneMode(serviceContext)) {
             log() << startupWarningsLog;
             log() << "** WARNING: mongod started without --replSet yet document(s) are present in "
diff --git a/src/mongo/db/repl/rs_rollback.cpp b/src/mongo/db/repl/rs_rollback.cpp
index 91e4bef476c..8a61b54f0b3 100644
--- a/src/mongo/db/repl/rs_rollback.cpp
+++ b/src/mongo/db/repl/rs_rollback.cpp
@@ -1423,7 +1423,7 @@ void rollback_internal::syncFixUp(OperationContext* opCtx,
         log() << "Forcing the stable timestamp to the common point: "
               << fixUpInfo.commonPoint.getTimestamp();
         opCtx->getServiceContext()->getStorageEngine()->setStableTimestamp(
-            fixUpInfo.commonPoint.getTimestamp(), boost::none, force);
+            fixUpInfo.commonPoint.getTimestamp(), force);
 
         // We must not take a stable checkpoint until it is guaranteed to include all writes from
         // before the rollback (i.e. the stable timestamp is at least the local top of oplog). In
diff --git a/src/mongo/db/repl/storage_interface_impl.cpp b/src/mongo/db/repl/storage_interface_impl.cpp
index 166a1a08a1c..da758476930 100644
--- a/src/mongo/db/repl/storage_interface_impl.cpp
+++ b/src/mongo/db/repl/storage_interface_impl.cpp
@@ -1073,7 +1073,7 @@ Status StorageInterfaceImpl::upgradeNonReplicatedUniqueIndexes(OperationContext*
 }
 
 void StorageInterfaceImpl::setStableTimestamp(ServiceContext* serviceCtx, Timestamp snapshotName) {
-    serviceCtx->getStorageEngine()->setStableTimestamp(snapshotName, boost::none);
+    serviceCtx->getStorageEngine()->setStableTimestamp(snapshotName);
 }
 
 void StorageInterfaceImpl::setInitialDataTimestamp(ServiceContext* serviceCtx,
diff --git a/src/mongo/db/storage/kv/kv_engine.h b/src/mongo/db/storage/kv/kv_engine.h
index cdbd34697f9..621314e1a3d 100644
--- a/src/mongo/db/storage/kv/kv_engine.h
+++ b/src/mongo/db/storage/kv/kv_engine.h
@@ -39,6 +39,7 @@
 #include "mongo/db/catalog/collection_options.h"
 #include "mongo/db/storage/kv/kv_prefix.h"
 #include "mongo/db/storage/record_store.h"
+#include "mongo/db/storage/storage_engine.h"
 
 namespace mongo {
 
@@ -307,9 +308,7 @@ public:
     /**
      * See `StorageEngine::setStableTimestamp`
      */
-    virtual void setStableTimestamp(Timestamp stableTimestamp,
-                                    boost::optional<Timestamp> maximumTruncationTimestamp,
-                                    bool force) {}
+    virtual void setStableTimestamp(Timestamp stableTimestamp, bool force) {}
 
     /**
      * See `StorageEngine::setInitialDataTimestamp`
@@ -322,6 +321,12 @@ public:
     virtual void setOldestTimestampFromStable() {}
 
     /**
+     * See `StorageEngine::setOldestActiveTransactionTimestampCallback`
+     */
+    virtual void setOldestActiveTransactionTimestampCallback(
+        StorageEngine::OldestActiveTransactionTimestampCallback callback){};
+
+    /**
      * See `StorageEngine::setOldestTimestamp`
      */
     virtual void setOldestTimestamp(Timestamp newOldestTimestamp, bool force) {}
diff --git a/src/mongo/db/storage/kv/kv_storage_engine.cpp b/src/mongo/db/storage/kv/kv_storage_engine.cpp
index b10de937bb7..655935a3043 100644
--- a/src/mongo/db/storage/kv/kv_storage_engine.cpp
+++ b/src/mongo/db/storage/kv/kv_storage_engine.cpp
@@ -716,10 +716,8 @@ void KVStorageEngine::setJournalListener(JournalListener* jl) {
     _engine->setJournalListener(jl);
 }
 
-void KVStorageEngine::setStableTimestamp(Timestamp stableTimestamp,
-                                         boost::optional<Timestamp> maximumTruncationTimestamp,
-                                         bool force) {
-    _engine->setStableTimestamp(stableTimestamp, maximumTruncationTimestamp, force);
+void KVStorageEngine::setStableTimestamp(Timestamp stableTimestamp, bool force) {
+    _engine->setStableTimestamp(stableTimestamp, force);
 }
 
 void KVStorageEngine::setInitialDataTimestamp(Timestamp initialDataTimestamp) {
@@ -736,6 +734,11 @@ void KVStorageEngine::setOldestTimestamp(Timestamp newOldestTimestamp) {
     _engine->setOldestTimestamp(newOldestTimestamp, force);
 }
 
+void KVStorageEngine::setOldestActiveTransactionTimestampCallback(
+    StorageEngine::OldestActiveTransactionTimestampCallback callback) {
+    _engine->setOldestActiveTransactionTimestampCallback(callback);
+}
+
 bool KVStorageEngine::isCacheUnderPressure(OperationContext* opCtx) const {
     return _engine->isCacheUnderPressure(opCtx);
 }
diff --git a/src/mongo/db/storage/kv/kv_storage_engine.h b/src/mongo/db/storage/kv/kv_storage_engine.h
index d5d4f1dade5..c1c64a51fa8 100644
--- a/src/mongo/db/storage/kv/kv_storage_engine.h
+++ b/src/mongo/db/storage/kv/kv_storage_engine.h
@@ -146,9 +146,7 @@ public:
 
     virtual void cleanShutdown();
 
-    virtual void setStableTimestamp(Timestamp stableTimestamp,
-                                    boost::optional<Timestamp> maximumTruncationTimestamp,
-                                    bool force = false) override;
+    virtual void setStableTimestamp(Timestamp stableTimestamp, bool force = false) override;
 
     virtual void setInitialDataTimestamp(Timestamp initialDataTimestamp) override;
 
@@ -156,6 +154,9 @@ public:
 
     virtual void setOldestTimestamp(Timestamp newOldestTimestamp) override;
 
+    virtual void setOldestActiveTransactionTimestampCallback(
+        StorageEngine::OldestActiveTransactionTimestampCallback) override;
+
     virtual bool isCacheUnderPressure(OperationContext* opCtx) const override;
 
     virtual void setCachePressureForTest(int pressure) override;
diff --git a/src/mongo/db/storage/storage_engine.h b/src/mongo/db/storage/storage_engine.h
index 1f3f111340b..74f6687976c 100644
--- a/src/mongo/db/storage/storage_engine.h
+++ b/src/mongo/db/storage/storage_engine.h
@@ -36,6 +36,7 @@
 #include "mongo/bson/bsonobj.h"
 #include "mongo/bson/timestamp.h"
 #include "mongo/db/storage/temporary_record_store.h"
+#include "mongo/util/functional.h"
 #include "mongo/util/mongoutils/str.h"
 
 namespace mongo {
@@ -57,6 +58,15 @@ class StorageEngineMetadata;
 class StorageEngine {
 public:
     /**
+     * When the storage engine needs to know how much oplog to preserve for the sake of active
+     * transactions, it executes a callback that returns either the oldest active transaction
+     * timestamp, or boost::none if there is no active transaction, or an error if it fails.
+     */
+    using OldestActiveTransactionTimestampResult = StatusWith<boost::optional<Timestamp>>;
+    using OldestActiveTransactionTimestampCallback =
+        std::function<OldestActiveTransactionTimestampResult(Timestamp stableTimestamp)>;
+
+    /**
      * The interface for creating new instances of storage engines.
      *
      * A storage engine provides an instance of this class (along with an associated
@@ -416,19 +426,8 @@ public:
      * Sets the highest timestamp at which the storage engine is allowed to take a checkpoint. This
      * timestamp must not decrease unless force=true is set, in which case we force the stable
      * timestamp, the oldest timestamp, and the commit timestamp backward.
-     *
-     * The maximumTruncationTimestamp (and newer) must not be truncated from the oplog in order to
-     * recover from the `stableTimestamp`.  `boost::none` implies there are no additional
-     * constraints to what may be truncated.
-     *
-     * For proper truncation of the oplog, this method requires min(stableTimestamp,
-     * maximumTruncationTimestamp) to be monotonically increasing (where `min(stableTimestamp,
-     * boost::none) => stableTimestamp`). Otherwise truncation can race and remove a document
-     * before a call to this method protects it.
      */
-    virtual void setStableTimestamp(Timestamp stableTimestamp,
-                                    boost::optional<Timestamp> maximumTruncationTimestamp,
-                                    bool force = false) {}
+    virtual void setStableTimestamp(Timestamp stableTimestamp, bool force = false) {}
 
     /**
      * Tells the storage engine the timestamp of the data at startup. This is necessary because
@@ -455,6 +454,14 @@ public:
     virtual void setOldestTimestamp(Timestamp timestamp) {}
 
     /**
+     * Sets a callback which returns the timestamp of the oldest oplog entry involved in an
+     * active MongoDB transaction. The storage engine calls this function to determine how much
+     * oplog it must preserve.
+     */
+    virtual void setOldestActiveTransactionTimestampCallback(
+        OldestActiveTransactionTimestampCallback callback){};
+
+    /**
      * Indicates whether the storage engine cache is under pressure.
      *
      * Retrieves a cache pressure value in the range [0, 100] from the storage engine, and compares
diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.cpp
index d9d07d00cd1..1655591d191 100644
--- a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.cpp
+++ b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.cpp
@@ -260,6 +260,20 @@ private:
     AtomicWord<bool> _shuttingDown{false};
 };
 
+std::string toString(const StorageEngine::OldestActiveTransactionTimestampResult& r) {
+    if (r.isOK()) {
+        if (r.getValue()) {
+            // Timestamp.
+            return r.getValue().value().toString();
+        } else {
+            // boost::none.
+            return "null";
+        }
+    } else {
+        return r.getStatus().toString();
+    }
+}
+
 class WiredTigerKVEngine::WiredTigerCheckpointThread : public BackgroundJob {
 public:
     explicit WiredTigerCheckpointThread(WiredTigerKVEngine* wiredTigerKVEngine,
@@ -285,28 +299,31 @@ public:
                                       wiredTigerGlobalOptions.checkpointDelaySecs)));
             }
 
+            // Might have been awakened by another thread shutting us down.
+            if (_shuttingDown.load()) {
+                break;
+            }
+
             const Timestamp stableTimestamp = _wiredTigerKVEngine->getStableTimestamp();
             const Timestamp initialDataTimestamp = _wiredTigerKVEngine->getInitialDataTimestamp();
 
             // The amount of oplog to keep is primarily dictated by a user setting. However, in
             // unexpected cases, durable, recover to a timestamp storage engines may need to play
             // forward from an oplog entry that would otherwise be truncated by the user
-            // setting. Furthermore with prepared transactions, oplog entries can refer to
-            // previous oplog entries.
+            // setting. Furthermore, the entries in prepared or large transactions can refer to
+            // previous entries in the same transaction.
             //
-            // Live (replication) rollback will replay oplogs from exactly the stable
-            // timestamp. With prepared transactions, it may require some additional entries prior
-            // to the stable timestamp. These requirements are summarized in
-            // `getOplogNeededForRollback`. Truncating the oplog at this point is sufficient for
-            // in-memory configurations, but could cause an unrecoverable scenario if the node
-            // crashed and has to play from the last stable checkpoint.
+            // Live (replication) rollback will replay oplogs from exactly the stable timestamp.
+            // With prepared or large transactions, it may require some additional entries prior to
+            // the stable timestamp. These requirements are summarized in getOplogNeededForRollback.
+            // Truncating the oplog at this point is sufficient for in-memory configurations, but
+            // could cause an unrecoverable scenario if the node crashed and has to play from the
+            // last stable checkpoint.
             //
             // By recording the oplog needed for rollback "now", then taking a stable checkpoint,
             // we can safely assume that the oplog needed for crash recovery has caught up to the
             // recorded value. After the checkpoint, this value will be published such that actors
             // which truncate the oplog can read an updated value.
-            const Timestamp oplogNeededForRollback =
-                _wiredTigerKVEngine->getOplogNeededForRollback();
             try {
                 // Three cases:
                 //
@@ -331,21 +348,22 @@ public:
                         << stableTimestamp.toString()
                         << " InitialDataTimestamp: " << initialDataTimestamp.toString();
                 } else {
-                    // 'stableTimestamp' is the smallest possible value at which WT will take a
-                    // stable checkpoint. A newer stable timestamp may be used by WT if one is
-                    // concurrently set.
-                    LOG_FOR_RECOVERY(2) << "Performing stable checkpoint. StableTimestamp: "
-                                        << stableTimestamp;
+                    auto oplogNeededForRollback = _wiredTigerKVEngine->getOplogNeededForRollback();
+
+                    LOG_FOR_RECOVERY(2)
+                        << "Performing stable checkpoint. StableTimestamp: " << stableTimestamp
+                        << ", OplogNeededForRollback: " << toString(oplogNeededForRollback);
 
                     UniqueWiredTigerSession session = _sessionCache->getSession();
                     WT_SESSION* s = session->getSession();
                     invariantWTOK(s->checkpoint(s, "use_timestamp=true"));
 
-                    // Now that the checkpoint is durable, publish the oplog needed to recover
-                    // from it.
-                    {
+                    if (oplogNeededForRollback.isOK()) {
+                        // Now that the checkpoint is durable, publish the oplog needed to recover
+                        // from it.
                         stdx::lock_guard<stdx::mutex> lk(_oplogNeededForCrashRecoveryMutex);
-                        _oplogNeededForCrashRecovery.store(oplogNeededForRollback.asULL());
+                        _oplogNeededForCrashRecovery.store(
+                            oplogNeededForRollback.getValue().asULL());
                     }
                 }
             } catch (const WriteConflictException&) {
@@ -639,11 +657,14 @@ WiredTigerKVEngine::WiredTigerKVEngine(const std::string& canonicalName,
         _journalFlusher->go();
     }
 
+    // Until the Replication layer installs a real callback, prevent truncating the oplog.
+    setOldestActiveTransactionTimestampCallback(
+        [](Timestamp) { return StatusWith(boost::make_optional(Timestamp::min())); });
+
     if (!_readOnly && !_ephemeral) {
         if (!_recoveryTimestamp.isNull()) {
             setInitialDataTimestamp(_recoveryTimestamp);
-            // The `maximumTruncationTimestamp` is not persisted, so choose a conservative value.
-            setStableTimestamp(_recoveryTimestamp, Timestamp::min(), false);
+            setStableTimestamp(_recoveryTimestamp, false);
         }
 
         _checkpointThread =
@@ -1046,6 +1067,12 @@ void WiredTigerKVEngine::syncSizeInfo(bool sync) const {
     }
 }
 
+void WiredTigerKVEngine::setOldestActiveTransactionTimestampCallback(
+    StorageEngine::OldestActiveTransactionTimestampCallback callback) {
+    stdx::lock_guard<stdx::mutex> lk(_oldestActiveTransactionTimestampCallbackMutex);
+    _oldestActiveTransactionTimestampCallback = std::move(callback);
+};
+
 RecoveryUnit* WiredTigerKVEngine::newRecoveryUnit() {
     return new WiredTigerRecoveryUnit(_sessionCache.get());
 }
@@ -1528,9 +1555,7 @@ MONGO_FAIL_POINT_DEFINE(WTPreserveSnapshotHistoryIndefinitely);
 
 }  // namespace
 
-void WiredTigerKVEngine::setStableTimestamp(Timestamp stableTimestamp,
-                                            boost::optional<Timestamp> maximumTruncationTimestamp,
-                                            bool force) {
+void WiredTigerKVEngine::setStableTimestamp(Timestamp stableTimestamp, bool force) {
     if (stableTimestamp.isNull()) {
         return;
     }
@@ -1578,26 +1603,9 @@ void WiredTigerKVEngine::setStableTimestamp(Timestamp stableTimestamp,
     invariant(static_cast<std::size_t>(size) < sizeof(stableTSConfigString));
     invariantWTOK(_conn->set_timestamp(_conn, stableTSConfigString));
 
+    // After publishing a stable timestamp to WT, we can record the updated stable timestamp value
+    // for the necessary oplog to keep.
     _stableTimestamp.store(stableTimestamp.asULL());
-
-    // After publishing a stable timestamp to WT, we can publish the updated value for the
-    // necessary oplog to keep. Calls to this method require the min(stableTimestamp,
-    // maximumTruncationTimestamp) to be monotonically increasing. This allows us to safely record
-    // and publish a single value without additional concurrency control.
-    if (maximumTruncationTimestamp) {
-        // Until we discover otherwise, assume callers expect to obey the contract for proper
-        // oplog truncation.
-        DEV invariant(_oplogNeededForRollback.load() <=
-                      std::min(maximumTruncationTimestamp->asULL(), stableTimestamp.asULL()));
-        _oplogNeededForRollback.store(
-            std::min(maximumTruncationTimestamp->asULL(), stableTimestamp.asULL()));
-    } else {
-        // If there is no maximumTruncationTimestamp at this stable timestamp, WT is free to
-        // truncate the oplog to any value behind the last stable timestamp, once it is
-        // checkpointed.
-        _oplogNeededForRollback.store(stableTimestamp.asULL());
-    }
-
     if (_checkpointThread && !_checkpointThread->hasTriggeredFirstStableCheckpoint()) {
         _checkpointThread->triggerFirstStableCheckpoint(
             prevStable, Timestamp(_initialDataTimestamp.load()), stableTimestamp);
@@ -1840,8 +1848,29 @@ boost::optional<Timestamp> WiredTigerKVEngine::getLastStableRecoveryTimestamp()
     return boost::none;
 }
 
-Timestamp WiredTigerKVEngine::getOplogNeededForRollback() const {
-    return Timestamp(_oplogNeededForRollback.load());
+StatusWith<Timestamp> WiredTigerKVEngine::getOplogNeededForRollback() const {
+    // Get the current stable timestamp and use it throughout this function, ignoring updates from
+    // another thread.
+    auto stableTimestamp = _stableTimestamp.load();
+
+    // Only one thread can set or execute this callback.
+    stdx::lock_guard<stdx::mutex> lk(_oldestActiveTransactionTimestampCallbackMutex);
+    boost::optional<Timestamp> oldestActiveTransactionTimestamp;
+    if (_oldestActiveTransactionTimestampCallback) {
+        auto status = _oldestActiveTransactionTimestampCallback(Timestamp(stableTimestamp));
+        if (status.isOK()) {
+            oldestActiveTransactionTimestamp.swap(status.getValue());
+        } else {
+            LOG(1) << "getting oldest active transaction timestamp: " << status.getStatus();
+            return status.getStatus();
+        }
+    }
+
+    if (oldestActiveTransactionTimestamp) {
+        return std::min(oldestActiveTransactionTimestamp.value(), Timestamp(stableTimestamp));
+    } else {
+        return Timestamp(stableTimestamp);
+    }
 }
 
 boost::optional<Timestamp> WiredTigerKVEngine::getOplogNeededForCrashRecovery() const {
@@ -1853,21 +1882,37 @@ boost::optional<Timestamp> WiredTigerKVEngine::getOplogNeededForCrashRecovery()
 }
 
 Timestamp WiredTigerKVEngine::getPinnedOplog() const {
-    stdx::lock_guard<stdx::mutex> lock(_oplogPinnedByBackupMutex);
-    if (!storageGlobalParams.allowOplogTruncation) {
-        // If oplog truncation is not allowed, then return the min timestamp so that no history is
-        // ever allowed to be deleted.
-        return Timestamp::min();
-    }
-    if (_oplogPinnedByBackup) {
-        // All the oplog since `_oplogPinnedByBackup` should remain intact during the backup.
-        return _oplogPinnedByBackup.get();
+    {
+        stdx::lock_guard<stdx::mutex> lock(_oplogPinnedByBackupMutex);
+        if (!storageGlobalParams.allowOplogTruncation) {
+            // If oplog truncation is not allowed, then return the min timestamp so that no history
+            // is
+            // ever allowed to be deleted.
+            return Timestamp::min();
+        }
+        if (_oplogPinnedByBackup) {
+            // All the oplog since `_oplogPinnedByBackup` should remain intact during the backup.
+            return _oplogPinnedByBackup.get();
+        }
     }
+
+    auto oplogNeededForCrashRecovery = getOplogNeededForCrashRecovery();
     if (!_keepDataHistory) {
         // We use rollbackViaRefetch, so we only need to pin oplog for crash recovery.
-        return getOplogNeededForCrashRecovery().value_or(Timestamp::max());
+        return oplogNeededForCrashRecovery.value_or(Timestamp::max());
+    }
+
+    if (oplogNeededForCrashRecovery) {
+        return oplogNeededForCrashRecovery.value();
     }
-    return getOplogNeededForCrashRecovery().value_or(getOplogNeededForRollback());
+
+    auto status = getOplogNeededForRollback();
+    if (status.isOK()) {
+        return status.getValue();
+    }
+
+    // If getOplogNeededForRollback fails, don't truncate any oplog right now.
+    return Timestamp::min();
 }
 
 bool WiredTigerKVEngine::supportsReadConcernSnapshot() const {
diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.h b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.h
index 7c5328476e5..1b1c9af1648 100644
--- a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.h
+++ b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.h
@@ -39,6 +39,7 @@
 #include "mongo/bson/ordering.h"
 #include "mongo/bson/timestamp.h"
 #include "mongo/db/storage/kv/kv_engine.h"
+#include "mongo/db/storage/storage_engine.h"
 #include "mongo/db/storage/wiredtiger/wiredtiger_oplog_manager.h"
 #include "mongo/db/storage/wiredtiger/wiredtiger_session_cache.h"
 #include "mongo/db/storage/wiredtiger/wiredtiger_util.h"
@@ -94,6 +95,9 @@ public:
         return _ephemeral;
     }
 
+    void setOldestActiveTransactionTimestampCallback(
+        StorageEngine::OldestActiveTransactionTimestampCallback callback) override;
+
     RecoveryUnit* newRecoveryUnit() override;
 
     Status createRecordStore(OperationContext* opCtx,
@@ -193,9 +197,7 @@ public:
 
     void setJournalListener(JournalListener* jl) final;
 
-    void setStableTimestamp(Timestamp stableTimestamp,
-                            boost::optional<Timestamp> maximumTruncationTimestamp,
-                            bool force) override;
+    void setStableTimestamp(Timestamp stableTimestamp, bool force) override;
 
     void setInitialDataTimestamp(Timestamp initialDataTimestamp) override;
 
@@ -321,10 +323,11 @@ public:
 
     /**
      * Returns the minimum possible Timestamp value in the oplog that replication may need for
-     * recovery in the event of a rollback. This value gets updated on every `setStableTimestamp`
-     * call.
+     * recovery in the event of a rollback. This value depends on the timestamp passed to
+     * `setStableTimestamp` and on the set of active MongoDB transactions. Returns an error if it
+     * times out querying the active transctions.
      */
-    Timestamp getOplogNeededForRollback() const;
+    StatusWith<Timestamp> getOplogNeededForRollback() const;
 
     /**
      * Returns the minimum possible Timestamp value in the oplog that replication may need for
@@ -405,6 +408,10 @@ private:
 
     std::uint64_t _getCheckpointTimestamp() const;
 
+    mutable stdx::mutex _oldestActiveTransactionTimestampCallbackMutex;
+    StorageEngine::OldestActiveTransactionTimestampCallback
+        _oldestActiveTransactionTimestampCallback;
+
     WT_CONNECTION* _conn;
     WiredTigerFileVersion _fileVersion;
     WiredTigerEventHandler _eventHandler;
@@ -458,7 +465,6 @@ private:
     // Tracks the stable and oldest timestamps we've set on the storage engine.
     AtomicWord<std::uint64_t> _oldestTimestamp;
     AtomicWord<std::uint64_t> _stableTimestamp;
-    AtomicWord<std::uint64_t> _oplogNeededForRollback{Timestamp::min().asULL()};
 
     // Timestamp of data at startup. Used internally to advise checkpointing and recovery to a
     // timestamp. Provided by replication layer because WT does not persist timestamps.
diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine_test.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine_test.cpp
index a9b6f3a5d83..9229c4df0f8 100644
--- a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine_test.cpp
+++ b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine_test.cpp
@@ -240,13 +240,30 @@ TEST_F(WiredTigerKVEngineTest, TestOplogTruncation) {
     _engine->setInitialDataTimestamp(Timestamp(1, 1));
     wiredTigerGlobalOptions.checkpointDelaySecs = 1;
 
+    // Simulate the callback that queries config.transactions for the oldest active transaction.
+    boost::optional<Timestamp> oldestActiveTxnTimestamp;
+    AtomicWord<bool> callbackShouldFail{false};
+    auto callback = [&](Timestamp stableTimestamp) {
+        using ResultType = StorageEngine::OldestActiveTransactionTimestampResult;
+        if (callbackShouldFail.load()) {
+            return ResultType(ErrorCodes::ExceededTimeLimit, "timeout");
+        }
+
+        return ResultType(oldestActiveTxnTimestamp);
+    };
+
+    _engine->setOldestActiveTransactionTimestampCallback(callback);
+
     // A method that will poll the WiredTigerKVEngine until it sees the amount of oplog necessary
     // for crash recovery exceeds the input.
     auto assertPinnedMovesSoon = [this](Timestamp newPinned) {
         // If the current oplog needed for rollback does not exceed the requested pinned out, we
         // cannot expect the CheckpointThread to eventually publish a sufficient crash recovery
         // value.
-        ASSERT_TRUE(_engine->getOplogNeededForRollback() >= newPinned);
+        auto needed = _engine->getOplogNeededForRollback();
+        if (needed.isOK()) {
+            ASSERT_TRUE(needed.getValue() >= newPinned);
+        }
 
         // Do 100 iterations that sleep for 100 milliseconds between polls. This will wait for up
         // to 10 seconds to observe an asynchronous update that iterates once per second.
@@ -264,17 +281,31 @@ TEST_F(WiredTigerKVEngineTest, TestOplogTruncation) {
         FAIL("");
     };
 
-    _engine->setStableTimestamp(Timestamp(10, 1), boost::none, false);
+    oldestActiveTxnTimestamp = boost::none;
+    _engine->setStableTimestamp(Timestamp(10, 1), false);
     assertPinnedMovesSoon(Timestamp(10, 1));
 
-    _engine->setStableTimestamp(Timestamp(20, 1), Timestamp(15, 1), false);
+    oldestActiveTxnTimestamp = Timestamp(15, 1);
+    _engine->setStableTimestamp(Timestamp(20, 1), false);
     assertPinnedMovesSoon(Timestamp(15, 1));
 
-    _engine->setStableTimestamp(Timestamp(30, 1), Timestamp(19, 1), false);
+    oldestActiveTxnTimestamp = Timestamp(19, 1);
+    _engine->setStableTimestamp(Timestamp(30, 1), false);
     assertPinnedMovesSoon(Timestamp(19, 1));
 
-    _engine->setStableTimestamp(Timestamp(30, 1), boost::none, false);
+    oldestActiveTxnTimestamp = boost::none;
+    _engine->setStableTimestamp(Timestamp(30, 1), false);
     assertPinnedMovesSoon(Timestamp(30, 1));
+
+    callbackShouldFail.store(true);
+    ASSERT_NOT_OK(_engine->getOplogNeededForRollback());
+    _engine->setStableTimestamp(Timestamp(40, 1), false);
+    // Await a new checkpoint. Oplog needed for rollback does not advance.
+    sleepmillis(1100);
+    ASSERT_EQ(_engine->getOplogNeededForCrashRecovery().get(), Timestamp(30, 1));
+    _engine->setStableTimestamp(Timestamp(30, 1), false);
+    callbackShouldFail.store(false);
+    assertPinnedMovesSoon(Timestamp(40, 1));
 }
 
 std::unique_ptr<KVHarnessHelper> makeHelper() {
diff --git a/src/mongo/db/transaction_participant.cpp b/src/mongo/db/transaction_participant.cpp
index 7e5d3334b3d..16cec7c5e7a 100644
--- a/src/mongo/db/transaction_participant.cpp
+++ b/src/mongo/db/transaction_participant.cpp
@@ -36,6 +36,7 @@
 
 #include "mongo/db/transaction_participant.h"
 
+#include "mongo/db/catalog/database_holder.h"
 #include "mongo/db/catalog/index_catalog.h"
 #include "mongo/db/catalog_raii.h"
 #include "mongo/db/commands/test_commands_enabled.h"
@@ -332,19 +333,63 @@ void TransactionParticipant::performNoopWrite(OperationContext* opCtx, StringDat
     }
 }
 
-boost::optional<Timestamp> TransactionParticipant::getOldestActiveTimestamp(
-    OperationContext* opCtx) {
-    DBDirectClient client(opCtx);
-    Query q(BSON(SessionTxnRecord::kStateFieldName << "prepared"));
-    q.sort(SessionTxnRecord::kStartOpTimeFieldName.toString());
-    auto result = client.findOne(NamespaceString::kSessionTransactionsTableNamespace.ns(), q);
-    if (result.isEmpty()) {
-        return boost::none;
-    }
+StorageEngine::OldestActiveTransactionTimestampResult
+TransactionParticipant::getOldestActiveTimestamp(Timestamp stableTimestamp) {
+    // Read from config.transactions at the stable timestamp for the oldest active transaction
+    // timestamp. Use a short timeout: another thread might have the global lock e.g. to shut down
+    // the server, and it both blocks this thread from querying config.transactions and waits for
+    // this thread to terminate.
+    auto client = getGlobalServiceContext()->makeClient("OldestActiveTxnTimestamp");
+    AlternativeClientRegion acr(client);
+
+    try {
+        auto opCtx = cc().makeOperationContext();
+        auto nss = NamespaceString::kSessionTransactionsTableNamespace;
+        auto deadline = Date_t::now() + Milliseconds(100);
+        Lock::DBLock dbLock(opCtx.get(), nss.db(), MODE_IS, deadline);
+        Lock::CollectionLock collLock(opCtx.get()->lockState(), nss.toString(), MODE_IS, deadline);
+
+        auto databaseHolder = DatabaseHolder::get(opCtx.get());
+        auto db = databaseHolder->getDb(opCtx.get(), nss.db());
+        if (!db) {
+            // There is no config database, so there cannot be any active transactions.
+            return boost::none;
+        }
+
+        auto collection = db->getCollection(opCtx.get(), nss);
+        if (!collection) {
+            return boost::none;
+        }
 
-    auto txnRecord =
-        SessionTxnRecord::parse(IDLParserErrorContext("parse oldest active txn record"), result);
-    return txnRecord.getStartOpTime()->getTimestamp();
+        if (!stableTimestamp.isNull()) {
+            opCtx->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kProvided,
+                                                          stableTimestamp);
+        }
+
+        // Scan. We guess that occasional scans are cheaper than the write overhead of an index.
+        boost::optional<Timestamp> oldestTxnTimestamp;
+        auto cursor = collection->getCursor(opCtx.get());
+        while (auto record = cursor->next()) {
+            auto doc = record.get().data.toBson();
+            auto txnRecord = SessionTxnRecord::parse(
+                IDLParserErrorContext("parse oldest active txn record"), doc);
+            if (txnRecord.getState() != DurableTxnStateEnum::kPrepared) {
+                continue;
+            }
+
+            // A prepared transaction must have a start timestamp.
+            // TODO(SERVER-40013): Handle entries with state "prepared" and no "startTimestamp".
+            invariant(txnRecord.getStartOpTime());
+            auto ts = txnRecord.getStartOpTime()->getTimestamp();
+            if (!oldestTxnTimestamp || ts < oldestTxnTimestamp.value()) {
+                oldestTxnTimestamp = ts;
+            }
+        }
+
+        return oldestTxnTimestamp;
+    } catch (const DBException&) {
+        return exceptionToStatus();
+    }
 }
 
 const LogicalSessionId& TransactionParticipant::Observer::_sessionId() const {
diff --git a/src/mongo/db/transaction_participant.h b/src/mongo/db/transaction_participant.h
index 6d05089b284..2cbb0aaaf7d 100644
--- a/src/mongo/db/transaction_participant.h
+++ b/src/mongo/db/transaction_participant.h
@@ -48,6 +48,7 @@
 #include "mongo/db/session_txn_record_gen.h"
 #include "mongo/db/single_transaction_stats.h"
 #include "mongo/db/storage/recovery_unit.h"
+#include "mongo/db/storage/storage_engine.h"
 #include "mongo/db/transaction_metrics_observer.h"
 #include "mongo/stdx/unordered_map.h"
 #include "mongo/util/assert_util.h"
@@ -774,10 +775,12 @@ public:
 
 
     /**
-     * Returns the timestamp of the oldest oplog entry written across all open transactions.
-     * Returns boost::none if there are no active transactions.
+     * Returns the timestamp of the oldest oplog entry written across all open transactions, at the
+     * time of the stable timestamp. Returns boost::none if there are no active transactions, or an
+     * error if it fails.
      */
-    static boost::optional<Timestamp> getOldestActiveTimestamp(OperationContext* opCtx);
+    static StorageEngine::OldestActiveTransactionTimestampResult getOldestActiveTimestamp(
+        Timestamp stableTimestamp);
 
     /**
      * Append a no-op to the oplog, for cases where we haven't written in this unit of work but
diff --git a/src/mongo/db/transaction_participant_test.cpp b/src/mongo/db/transaction_participant_test.cpp
index 3a20184e9dd..97070fb2fc6 100644
--- a/src/mongo/db/transaction_participant_test.cpp
+++ b/src/mongo/db/transaction_participant_test.cpp
@@ -29,6 +29,8 @@
 
 #include "mongo/platform/basic.h"
 
+#include <boost/optional/optional_io.hpp>
+
 #include "mongo/db/client.h"
 #include "mongo/db/db_raii.h"
 #include "mongo/db/dbdirectclient.h"
@@ -3820,5 +3822,92 @@ TEST_F(TxnParticipantTest, ResponseMetadataHasReadOnlyFalseIfAborted) {
     ASSERT_FALSE(txnParticipant.getResponseMetadata().getReadOnly());
 }
 
+TEST_F(TxnParticipantTest, OldestActiveTransactionTimestamp) {
+    auto nss = NamespaceString::kSessionTransactionsTableNamespace;
+
+    auto insertTxnRecord = [&](unsigned i) {
+        Timestamp ts(1, i);
+        SessionTxnRecord record;
+        record.setStartOpTime(repl::OpTime(ts, 0));
+        record.setState(DurableTxnStateEnum::kPrepared);
+        record.setSessionId(makeLogicalSessionIdForTest());
+        record.setTxnNum(1);
+        record.setLastWriteOpTime(repl::OpTime(ts, 0));
+        record.setLastWriteDate(Date_t::now());
+
+        AutoGetOrCreateDb autoDb(opCtx(), nss.db(), MODE_X);
+        WriteUnitOfWork wuow(opCtx());
+        auto coll = autoDb.getDb()->getCollection(opCtx(), nss.ns());
+        ASSERT(coll);
+        OpDebug* const nullOpDebug = nullptr;
+        ASSERT_OK(
+            coll->insertDocument(opCtx(), InsertStatement(record.toBSON()), nullOpDebug, false));
+        wuow.commit();
+    };
+
+    auto deleteTxnRecord = [&](unsigned i) {
+        Timestamp ts(1, i);
+        AutoGetOrCreateDb autoDb(opCtx(), nss.db(), MODE_X);
+        WriteUnitOfWork wuow(opCtx());
+        auto coll = autoDb.getDb()->getCollection(opCtx(), nss.ns());
+        ASSERT(coll);
+        auto cursor = coll->getCursor(opCtx());
+        while (auto record = cursor->next()) {
+            auto bson = record.get().data.toBson();
+            if (bson["state"].String() != "prepared"_sd) {
+                continue;
+            }
+
+            if (bson["startOpTime"]["ts"].timestamp() == ts) {
+                coll->deleteDocument(opCtx(), kUninitializedStmtId, record->id, nullptr);
+                wuow.commit();
+                return;
+            }
+        }
+        FAIL(mongoutils::str::stream() << "No prepared transaction with start timestamp (1, " << i
+                                       << ")");
+    };
+
+    auto oldestActiveTransactionTS = [&]() {
+        return TransactionParticipant::getOldestActiveTimestamp(Timestamp()).getValue();
+    };
+
+    auto assertOldestActiveTS = [&](boost::optional<unsigned> i) {
+        if (i.has_value()) {
+            ASSERT_EQ(Timestamp(1, i.value()), oldestActiveTransactionTS());
+        } else {
+            ASSERT_EQ(boost::none, oldestActiveTransactionTS());
+        }
+    };
+
+    assertOldestActiveTS(boost::none);
+    insertTxnRecord(1);
+    assertOldestActiveTS(1);
+    insertTxnRecord(2);
+    assertOldestActiveTS(1);
+    deleteTxnRecord(1);
+    assertOldestActiveTS(2);
+    deleteTxnRecord(2);
+    assertOldestActiveTS(boost::none);
+
+    // Add a newer transaction, then an older one, to test that order doesn't matter.
+    insertTxnRecord(4);
+    insertTxnRecord(3);
+    assertOldestActiveTS(3);
+    deleteTxnRecord(4);
+    assertOldestActiveTS(3);
+    deleteTxnRecord(3);
+    assertOldestActiveTS(boost::none);
+};
+
+TEST_F(TxnParticipantTest, OldestActiveTransactionTimestampTimeout) {
+    // Block getOldestActiveTimestamp() by locking the config database.
+    auto nss = NamespaceString::kSessionTransactionsTableNamespace;
+    AutoGetOrCreateDb autoDb(opCtx(), nss.db(), MODE_X);
+    auto statusWith = TransactionParticipant::getOldestActiveTimestamp(Timestamp());
+    ASSERT_FALSE(statusWith.isOK());
+    ASSERT_TRUE(ErrorCodes::isInterruption(statusWith.getStatus().code()));
+};
+
 }  // namespace
 }  // namespace mongo
diff --git a/src/mongo/dbtests/storage_timestamp_tests.cpp b/src/mongo/dbtests/storage_timestamp_tests.cpp
index 2a326ced824..6a45c1d0223 100644
--- a/src/mongo/dbtests/storage_timestamp_tests.cpp
+++ b/src/mongo/dbtests/storage_timestamp_tests.cpp
@@ -432,8 +432,8 @@ public:
 
     void assertOldestActiveTxnTimestampEquals(const boost::optional<Timestamp>& ts,
                                               const Timestamp& atTs) {
-        OneOffRead oor(_opCtx, atTs);
-        ASSERT_EQ(TransactionParticipant::getOldestActiveTimestamp(_opCtx), ts);
+        auto oldest = TransactionParticipant::getOldestActiveTimestamp(atTs);
+        ASSERT_EQ(oldest, ts);
     }
 
     void assertHasStartOpTime() {