SERVER-37199 Yield locks of transactions in secondary application.r4.1.6

8 files changed, 222 insertions, 11 deletions

diff --git a/src/mongo/db/concurrency/lock_manager.cpp b/src/mongo/db/concurrency/lock_manager.cpp
index 1b8f5f644a5..1a5520494bf 100644
--- a/src/mongo/db/concurrency/lock_manager.cpp
+++ b/src/mongo/db/concurrency/lock_manager.cpp
@@ -1033,6 +1033,7 @@ void LockRequest::initNew(Locker* locker, LockGrantNotification* notify) {
     partitioned = false;
     mode = MODE_NONE;
     convertMode = MODE_NONE;
+    unlockPending = 0;
 }
 
 
diff --git a/src/mongo/db/concurrency/lock_state.cpp b/src/mongo/db/concurrency/lock_state.cpp
index e47fd47990f..421b201af00 100644
--- a/src/mongo/db/concurrency/lock_state.cpp
+++ b/src/mongo/db/concurrency/lock_state.cpp
@@ -423,6 +423,40 @@ void LockerImpl::endWriteUnitOfWork() {
     }
 }
 
+bool LockerImpl::releaseWriteUnitOfWork(LockSnapshot* stateOut) {
+    // Only the global WUOW can be released.
+    invariant(_wuowNestingLevel == 1);
+    --_wuowNestingLevel;
+    invariant(!isGlobalLockedRecursively());
+
+    // All locks should be pending to unlock.
+    invariant(_requests.size() == _numResourcesToUnlockAtEndUnitOfWork);
+    for (auto it = _requests.begin(); it; it.next()) {
+        // No converted lock so we don't need to unlock more than once.
+        invariant(it->unlockPending == 1);
+    }
+    _numResourcesToUnlockAtEndUnitOfWork = 0;
+
+    return saveLockStateAndUnlock(stateOut);
+}
+
+void LockerImpl::restoreWriteUnitOfWork(OperationContext* opCtx,
+                                        const LockSnapshot& stateToRestore) {
+    if (stateToRestore.globalMode != MODE_NONE) {
+        restoreLockState(opCtx, stateToRestore);
+    }
+
+    invariant(_numResourcesToUnlockAtEndUnitOfWork == 0);
+    for (auto it = _requests.begin(); it; it.next()) {
+        invariant(_shouldDelayUnlock(it.key(), (it->mode)));
+        invariant(it->unlockPending == 0);
+        it->unlockPending++;
+    }
+    _numResourcesToUnlockAtEndUnitOfWork = static_cast<unsigned>(_requests.size());
+
+    beginWriteUnitOfWork();
+}
+
 LockResult LockerImpl::lock(OperationContext* opCtx,
                             ResourceId resId,
                             LockMode mode,
diff --git a/src/mongo/db/concurrency/lock_state.h b/src/mongo/db/concurrency/lock_state.h
index 599cb95bd2d..ab21b58bb3a 100644
--- a/src/mongo/db/concurrency/lock_state.h
+++ b/src/mongo/db/concurrency/lock_state.h
@@ -149,8 +149,8 @@ public:
     virtual LockResult lockRSTLBegin(OperationContext* opCtx);
     virtual LockResult lockRSTLComplete(OperationContext* opCtx, Date_t deadline);
 
-    virtual void beginWriteUnitOfWork();
-    virtual void endWriteUnitOfWork();
+    virtual void beginWriteUnitOfWork() override;
+    virtual void endWriteUnitOfWork() override;
 
     virtual bool inAWriteUnitOfWork() const {
         return _wuowNestingLevel > 0;
@@ -194,6 +194,10 @@ public:
         restoreLockState(nullptr, stateToRestore);
     }
 
+    bool releaseWriteUnitOfWork(LockSnapshot* stateOut) override;
+    void restoreWriteUnitOfWork(OperationContext* opCtx,
+                                const LockSnapshot& stateToRestore) override;
+
     virtual void releaseTicket();
     virtual void reacquireTicket(OperationContext* opCtx);
 
diff --git a/src/mongo/db/concurrency/lock_state_test.cpp b/src/mongo/db/concurrency/lock_state_test.cpp
index a0094d3d485..b731f556057 100644
--- a/src/mongo/db/concurrency/lock_state_test.cpp
+++ b/src/mongo/db/concurrency/lock_state_test.cpp
@@ -254,6 +254,138 @@ TEST(LockerImpl, saveAndRestoreDBAndCollection) {
     ASSERT(locker.unlockGlobal());
 }
 
+TEST(LockerImpl, releaseWriteUnitOfWork) {
+    Locker::LockSnapshot lockInfo;
+
+    LockerImpl locker;
+
+    const ResourceId resIdDatabase(RESOURCE_DATABASE, "TestDB"_sd);
+    const ResourceId resIdCollection(RESOURCE_COLLECTION, "TestDB.collection"_sd);
+
+    locker.beginWriteUnitOfWork();
+    // Lock some stuff.
+    locker.lockGlobal(MODE_IX);
+    ASSERT_EQUALS(LOCK_OK, locker.lock(resIdDatabase, MODE_IX));
+    ASSERT_EQUALS(LOCK_OK, locker.lock(resIdCollection, MODE_X));
+    // Unlock them so that they will be pending to unlock.
+    ASSERT_FALSE(locker.unlock(resIdCollection));
+    ASSERT_FALSE(locker.unlock(resIdDatabase));
+    ASSERT_FALSE(locker.unlockGlobal());
+
+    ASSERT(locker.releaseWriteUnitOfWork(&lockInfo));
+
+    // Things shouldn't be locked anymore.
+    ASSERT_EQUALS(MODE_NONE, locker.getLockMode(resIdDatabase));
+    ASSERT_EQUALS(MODE_NONE, locker.getLockMode(resIdCollection));
+    ASSERT_FALSE(locker.isLocked());
+
+    // Destructor should succeed since the locker's state should be empty.
+}
+
+TEST(LockerImpl, restoreWriteUnitOfWork) {
+    Locker::LockSnapshot lockInfo;
+
+    LockerImpl locker;
+
+    const ResourceId resIdDatabase(RESOURCE_DATABASE, "TestDB"_sd);
+    const ResourceId resIdCollection(RESOURCE_COLLECTION, "TestDB.collection"_sd);
+
+    locker.beginWriteUnitOfWork();
+    // Lock some stuff.
+    locker.lockGlobal(MODE_IX);
+    ASSERT_EQUALS(LOCK_OK, locker.lock(resIdDatabase, MODE_IX));
+    ASSERT_EQUALS(LOCK_OK, locker.lock(resIdCollection, MODE_X));
+    // Unlock them so that they will be pending to unlock.
+    ASSERT_FALSE(locker.unlock(resIdCollection));
+    ASSERT_FALSE(locker.unlock(resIdDatabase));
+    ASSERT_FALSE(locker.unlockGlobal());
+
+    ASSERT(locker.releaseWriteUnitOfWork(&lockInfo));
+
+    // Things shouldn't be locked anymore.
+    ASSERT_EQUALS(MODE_NONE, locker.getLockMode(resIdDatabase));
+    ASSERT_EQUALS(MODE_NONE, locker.getLockMode(resIdCollection));
+    ASSERT_FALSE(locker.isLocked());
+
+    // Restore lock state.
+    locker.restoreWriteUnitOfWork(nullptr, lockInfo);
+
+    // Make sure things were re-locked.
+    ASSERT_EQUALS(MODE_IX, locker.getLockMode(resIdDatabase));
+    ASSERT_EQUALS(MODE_X, locker.getLockMode(resIdCollection));
+    ASSERT(locker.isLocked());
+
+    locker.endWriteUnitOfWork();
+
+    ASSERT_EQUALS(MODE_NONE, locker.getLockMode(resIdDatabase));
+    ASSERT_EQUALS(MODE_NONE, locker.getLockMode(resIdCollection));
+    ASSERT_FALSE(locker.isLocked());
+}
+
+TEST(LockerImpl, releaseAndRestoreReadOnlyWriteUnitOfWork) {
+    Locker::LockSnapshot lockInfo;
+
+    LockerImpl locker;
+
+    const ResourceId resIdDatabase(RESOURCE_DATABASE, "TestDB"_sd);
+    const ResourceId resIdCollection(RESOURCE_COLLECTION, "TestDB.collection"_sd);
+
+    // Snapshot transactions delay shared locks as well.
+    locker.setSharedLocksShouldTwoPhaseLock(true);
+
+    locker.beginWriteUnitOfWork();
+    // Lock some stuff in IS mode.
+    locker.lockGlobal(MODE_IS);
+    ASSERT_EQUALS(LOCK_OK, locker.lock(resIdDatabase, MODE_IS));
+    ASSERT_EQUALS(LOCK_OK, locker.lock(resIdCollection, MODE_IS));
+    // Unlock them.
+    ASSERT_FALSE(locker.unlock(resIdCollection));
+    ASSERT_FALSE(locker.unlock(resIdDatabase));
+    ASSERT_FALSE(locker.unlockGlobal());
+    ASSERT_EQ(3u, locker.numResourcesToUnlockAtEndUnitOfWorkForTest());
+
+    // Things shouldn't be locked anymore.
+    ASSERT_TRUE(locker.releaseWriteUnitOfWork(&lockInfo));
+
+    ASSERT_EQUALS(MODE_NONE, locker.getLockMode(resIdDatabase));
+    ASSERT_EQUALS(MODE_NONE, locker.getLockMode(resIdCollection));
+    ASSERT_FALSE(locker.isLocked());
+
+    // Restore lock state.
+    locker.restoreWriteUnitOfWork(nullptr, lockInfo);
+
+    ASSERT_EQUALS(MODE_IS, locker.getLockMode(resIdDatabase));
+    ASSERT_EQUALS(MODE_IS, locker.getLockMode(resIdCollection));
+    ASSERT_TRUE(locker.isLocked());
+
+    locker.endWriteUnitOfWork();
+
+    ASSERT_EQUALS(MODE_NONE, locker.getLockMode(resIdDatabase));
+    ASSERT_EQUALS(MODE_NONE, locker.getLockMode(resIdCollection));
+    ASSERT_FALSE(locker.isLocked());
+}
+
+TEST(LockerImpl, releaseAndRestoreEmptyWriteUnitOfWork) {
+    Locker::LockSnapshot lockInfo;
+    LockerImpl locker;
+
+    // Snapshot transactions delay shared locks as well.
+    locker.setSharedLocksShouldTwoPhaseLock(true);
+
+    locker.beginWriteUnitOfWork();
+
+    // Nothing to yield.
+    ASSERT_FALSE(locker.releaseWriteUnitOfWork(&lockInfo));
+    ASSERT_FALSE(locker.isLocked());
+
+    // Restore lock state.
+    locker.restoreWriteUnitOfWork(nullptr, lockInfo);
+    ASSERT_FALSE(locker.isLocked());
+
+    locker.endWriteUnitOfWork();
+    ASSERT_FALSE(locker.isLocked());
+}
+
 TEST(LockerImpl, DefaultLocker) {
     const ResourceId resId(RESOURCE_DATABASE, "TestDB"_sd);
 
diff --git a/src/mongo/db/concurrency/locker.h b/src/mongo/db/concurrency/locker.h
index d1ea2830a91..ee4b7118b5d 100644
--- a/src/mongo/db/concurrency/locker.h
+++ b/src/mongo/db/concurrency/locker.h
@@ -206,8 +206,8 @@ public:
 
     /**
      * beginWriteUnitOfWork/endWriteUnitOfWork are called at the start and end of WriteUnitOfWorks.
-     * They can be used to implement two-phase locking. Each call to begin should be matched with an
-     * eventual call to end.
+     * They can be used to implement two-phase locking. Each call to begin or restore should be
+     * matched with an eventual call to end or release.
      *
      * endWriteUnitOfWork, if not called in a nested WUOW, will release all two-phase locking held
      * lock resources.
@@ -375,6 +375,15 @@ public:
     virtual void restoreLockState(const LockSnapshot& stateToRestore) = 0;
 
     /**
+     * releaseWriteUnitOfWork opts out two-phase locking and yield the locks after a WUOW
+     * has been released. restoreWriteUnitOfWork reaquires the locks and resume the two-phase
+     * locking behavior of WUOW.
+     */
+    virtual bool releaseWriteUnitOfWork(LockSnapshot* stateOut) = 0;
+    virtual void restoreWriteUnitOfWork(OperationContext* opCtx,
+                                        const LockSnapshot& stateToRestore) = 0;
+
+    /**
      * Releases the ticket associated with the Locker. This allows locks to be held without
      * contributing to reader/writer throttling.
      */
diff --git a/src/mongo/db/concurrency/locker_noop.h b/src/mongo/db/concurrency/locker_noop.h
index 37abca2c99e..332e410ff91 100644
--- a/src/mongo/db/concurrency/locker_noop.h
+++ b/src/mongo/db/concurrency/locker_noop.h
@@ -111,9 +111,9 @@ public:
         MONGO_UNREACHABLE;
     }
 
-    virtual void beginWriteUnitOfWork() {}
+    virtual void beginWriteUnitOfWork() override {}
 
-    virtual void endWriteUnitOfWork() {}
+    virtual void endWriteUnitOfWork() override {}
 
     virtual bool inAWriteUnitOfWork() const {
         return false;
@@ -187,6 +187,15 @@ public:
         MONGO_UNREACHABLE;
     }
 
+    bool releaseWriteUnitOfWork(LockSnapshot* stateOut) override {
+        MONGO_UNREACHABLE;
+    }
+
+    void restoreWriteUnitOfWork(OperationContext* opCtx,
+                                const LockSnapshot& stateToRestore) override {
+        MONGO_UNREACHABLE;
+    };
+
     virtual void releaseTicket() {
         MONGO_UNREACHABLE;
     }
diff --git a/src/mongo/db/transaction_participant.cpp b/src/mongo/db/transaction_participant.cpp
index 05947daa982..cc9726c29b5 100644
--- a/src/mongo/db/transaction_participant.cpp
+++ b/src/mongo/db/transaction_participant.cpp
@@ -529,6 +529,12 @@ TransactionParticipant::TxnResources::TxnResources(OperationContext* opCtx, bool
     }
     _locker->unsetThreadId();
 
+    // On secondaries, we yield the locks for transactions.
+    if (!opCtx->writesAreReplicated()) {
+        _lockSnapshot = std::make_unique<Locker::LockSnapshot>();
+        _locker->releaseWriteUnitOfWork(_lockSnapshot.get());
+    }
+
     // This thread must still respect the transaction lock timeout, since it can prevent the
     // transaction from making progress.
     auto maxTransactionLockMillis = maxTransactionLockRequestTimeoutMillis.load();
@@ -553,25 +559,35 @@ TransactionParticipant::TxnResources::~TxnResources() {
         // when starting a new transaction before completing an old one.  So we should
         // be at WUOW nesting level 1 (only the top level WriteUnitOfWork).
         _recoveryUnit->abortUnitOfWork();
-        _locker->endWriteUnitOfWork();
+        // If locks are not yielded, release them.
+        if (!_lockSnapshot) {
+            _locker->endWriteUnitOfWork();
+        }
         invariant(!_locker->inAWriteUnitOfWork());
     }
 }
 
 void TransactionParticipant::TxnResources::release(OperationContext* opCtx) {
     // Perform operations that can fail the release before marking the TxnResources as released.
+
+    // Restore locks if they are yielded.
+    if (_lockSnapshot) {
+        invariant(!_locker->isLocked());
+        // opCtx is passed in to enable the restoration to be interrupted.
+        _locker->restoreWriteUnitOfWork(opCtx, *_lockSnapshot);
+        _lockSnapshot.reset(nullptr);
+    }
     _locker->reacquireTicket(opCtx);
 
     invariant(!_released);
     _released = true;
 
-    // We intentionally do not capture the return value of swapLockState(), which is just an empty
-    // locker. At the end of the operation, if the transaction is not complete, we will stash the
-    // operation context's locker and replace it with a new empty locker.
-
     // It is necessary to lock the client to change the Locker on the OperationContext.
     stdx::lock_guard<Client> lk(*opCtx->getClient());
     invariant(opCtx->lockState()->getClientState() == Locker::ClientState::kInactive);
+    // We intentionally do not capture the return value of swapLockState(), which is just an empty
+    // locker. At the end of the operation, if the transaction is not complete, we will stash the
+    // operation context's locker and replace it with a new empty locker.
     opCtx->swapLockState(std::move(_locker));
     opCtx->lockState()->updateThreadIdToCurrentThread();
 
@@ -701,6 +717,11 @@ void TransactionParticipant::unstashTransactionResources(OperationContext* opCtx
                                                  : SpeculativeTransactionOpTime::kLastApplied);
         }
 
+        // All locks of transactions must be acquired inside the global WUOW so that we can
+        // yield and restore all locks on state transition. Otherwise, we'd have to remember
+        // which locks are managed by WUOW.
+        invariant(!opCtx->lockState()->isLocked());
+
         // Stashed transaction resources do not exist for this in-progress multi-document
         // transaction. Set up the transaction resources on the opCtx.
         opCtx->setWriteUnitOfWork(std::make_unique<WriteUnitOfWork>(opCtx));
diff --git a/src/mongo/db/transaction_participant.h b/src/mongo/db/transaction_participant.h
index e7c69a69c76..85b0b0e3611 100644
--- a/src/mongo/db/transaction_participant.h
+++ b/src/mongo/db/transaction_participant.h
@@ -113,6 +113,7 @@ public:
     private:
         bool _released = false;
         std::unique_ptr<Locker> _locker;
+        std::unique_ptr<Locker::LockSnapshot> _lockSnapshot;
         std::unique_ptr<RecoveryUnit> _recoveryUnit;
         repl::ReadConcernArgs _readConcernArgs;
         WriteUnitOfWork::RecoveryUnitState _ruState;