From 6b3e341703b781bb1ff7b1263406a0f1d28dd77c Mon Sep 17 00:00:00 2001
From: Gregory Noma <gregory.noma@gmail.com>
Date: Tue, 25 Aug 2020 16:07:00 -0400
Subject: SERVER-50365 Use short WT transaction rollback timeout in the
 multi-document transaction expirer thread

---
 ...iodic_runner_job_abort_expired_transactions.cpp |  5 +++++
 src/mongo/db/storage/recovery_unit.h               | 10 +++++++++
 .../wiredtiger/wiredtiger_recovery_unit.cpp        | 25 +++++++++++++---------
 src/mongo/db/transaction_participant.cpp           |  7 ++++++
 src/mongo/db/transaction_participant.h             |  2 ++
 5 files changed, 39 insertions(+), 10 deletions(-)

(limited to 'src/mongo')

diff --git a/src/mongo/db/periodic_runner_job_abort_expired_transactions.cpp b/src/mongo/db/periodic_runner_job_abort_expired_transactions.cpp
index 4ab8aed0dec..8038e7bbcb5 100644
--- a/src/mongo/db/periodic_runner_job_abort_expired_transactions.cpp
+++ b/src/mongo/db/periodic_runner_job_abort_expired_transactions.cpp
@@ -102,6 +102,11 @@ void PeriodicThreadToAbortExpiredTransactions::_init(ServiceContext* serviceCont
             // non-transaction, exclusive lock taking operation blocked
             // behind an active transaction's intent lock.
             opCtx->lockState()->setMaxLockTimeout(Milliseconds(0));
+
+            // This thread needs storage rollback to complete timely, so instruct the storage
+            // engine to not do any extra eviction for this thread, if supported.
+            opCtx->recoveryUnit()->setNoEvictionAfterRollback();
+
             try {
                 killAllExpiredTransactions(opCtx.get());
             } catch (ExceptionForCat<ErrorCategory::CancelationError>& ex) {
diff --git a/src/mongo/db/storage/recovery_unit.h b/src/mongo/db/storage/recovery_unit.h
index 6242a69d777..5c8be96b528 100644
--- a/src/mongo/db/storage/recovery_unit.h
+++ b/src/mongo/db/storage/recovery_unit.h
@@ -633,6 +633,14 @@ public:
         _mustBeTimestamped = true;
     }
 
+    void setNoEvictionAfterRollback() {
+        _noEvictionAfterRollback = true;
+    }
+
+    bool getNoEvictionAfterRollback() const {
+        return _noEvictionAfterRollback;
+    }
+
 protected:
     RecoveryUnit();
 
@@ -673,6 +681,8 @@ protected:
 
     bool _mustBeTimestamped = false;
 
+    bool _noEvictionAfterRollback = false;
+
 private:
     // Sets the snapshot associated with this RecoveryUnit to a new globally unique id number.
     void assignNextSnapshotId();
diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit.cpp
index 8eba439563c..1167fd673f3 100644
--- a/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit.cpp
+++ b/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit.cpp
@@ -374,17 +374,22 @@ void WiredTigerRecoveryUnit::_txnClose(bool commit) {
         }
 
         wtRet = s->commit_transaction(s, conf.str().c_str());
-        LOGV2_DEBUG(22412,
-                    3,
-                    "WT commit_transaction for snapshot id {snapshotId}",
-                    "snapshotId"_attr = getSnapshotId().toNumber());
+
+        LOGV2_DEBUG(
+            22412, 3, "WT commit_transaction", "snapshotId"_attr = getSnapshotId().toNumber());
     } else {
-        wtRet = s->rollback_transaction(s, nullptr);
-        invariant(!wtRet);
-        LOGV2_DEBUG(22413,
-                    3,
-                    "WT rollback_transaction for snapshot id {snapshotId}",
-                    "snapshotId"_attr = getSnapshotId().toNumber());
+        StringBuilder config;
+        if (_noEvictionAfterRollback) {
+            // The only point at which rollback_transaction() can time out is in the bonus-eviction
+            // phase. If the timeout expires here, the function will stop the eviction and return
+            // success. It cannot return an error due to timeout.
+            config << "operation_timeout_ms=1,";
+        }
+
+        wtRet = s->rollback_transaction(s, config.str().c_str());
+
+        LOGV2_DEBUG(
+            22413, 3, "WT rollback_transaction", "snapshotId"_attr = getSnapshotId().toNumber());
     }
 
     if (_isTimestamped) {
diff --git a/src/mongo/db/transaction_participant.cpp b/src/mongo/db/transaction_participant.cpp
index f4e650b47f3..d3772edb7db 100644
--- a/src/mongo/db/transaction_participant.cpp
+++ b/src/mongo/db/transaction_participant.cpp
@@ -830,6 +830,10 @@ void TransactionParticipant::TxnResources::release(OperationContext* opCtx) {
     readConcernArgs = _readConcernArgs;
 }
 
+void TransactionParticipant::TxnResources::setNoEvictionAfterRollback() {
+    _recoveryUnit->setNoEvictionAfterRollback();
+}
+
 TransactionParticipant::SideTransactionBlock::SideTransactionBlock(OperationContext* opCtx)
     : _opCtx(opCtx) {
     // Do nothing if we are already in a SideTransactionBlock. We can tell we are already in a
@@ -1662,6 +1666,9 @@ void TransactionParticipant::Participant::_abortTransactionOnSession(OperationCo
                                                      : TransactionState::kAbortedWithoutPrepare;
 
     stdx::lock_guard<Client> lk(*opCtx->getClient());
+    if (o().txnResourceStash && opCtx->recoveryUnit()->getNoEvictionAfterRollback()) {
+        o(lk).txnResourceStash->setNoEvictionAfterRollback();
+    }
     _resetTransactionState(lk, nextState);
 }
 
diff --git a/src/mongo/db/transaction_participant.h b/src/mongo/db/transaction_participant.h
index 8fad2122368..1c3164a51fc 100644
--- a/src/mongo/db/transaction_participant.h
+++ b/src/mongo/db/transaction_participant.h
@@ -221,6 +221,8 @@ public:
             return _readConcernArgs;
         }
 
+        void setNoEvictionAfterRollback();
+
     private:
         bool _released = false;
         std::unique_ptr<Locker> _locker;
-- 
cgit v1.2.1