Merge pull request #30658 from ClickHouse/std_mutex_alter_lock

Change `alter_lock` from `RWLock` to `std::timed_mutex`

src/Interpreters/InterpreterAlterQuery.cpp

@@ -76,7 +76,8 @@ BlockIO InterpreterAlterQuery::executeToTable(const ASTAlterQuery & alter)
     StoragePtr table = DatabaseCatalog::instance().getTable(table_id, getContext());
     if (table->isStaticStorage())
         throw Exception(ErrorCodes::TABLE_IS_READ_ONLY, "Table is read-only");
-    auto alter_lock = table->lockForAlter(getContext()->getCurrentQueryId(), getContext()->getSettingsRef().lock_acquire_timeout);
+    auto table_lock = table->lockForShare(getContext()->getCurrentQueryId(), getContext()->getSettingsRef().lock_acquire_timeout);
+    auto alter_lock = table->lockForAlter(getContext()->getSettingsRef().lock_acquire_timeout);
     auto metadata_snapshot = table->getInMemoryMetadataPtr();
 
     /// Add default database to table identifiers that we can encounter in e.g. default expressions, mutation expression, etc.

src/Storages/IStorage.cpp

@@ -60,27 +60,31 @@ TableLockHolder IStorage::lockForShare(const String & query_id, const std::chron
     return result;
 }
 
-TableLockHolder IStorage::lockForAlter(const String & query_id, const std::chrono::milliseconds & acquire_timeout)
+IStorage::AlterLockHolder IStorage::lockForAlter(const std::chrono::milliseconds & acquire_timeout)
 {
-    TableLockHolder result = tryLockTimed(alter_lock, RWLockImpl::Write, query_id, acquire_timeout);
+    AlterLockHolder lock{alter_lock, std::defer_lock};
+
+    if (!lock.try_lock_for(acquire_timeout))
+        throw Exception(ErrorCodes::DEADLOCK_AVOIDED,
+                        "Locking attempt for ALTER on \"{}\" has timed out! ({} ms) "
+                        "Possible deadlock avoided. Client should retry.",
+                        getStorageID().getFullTableName(), std::to_string(acquire_timeout.count()));
 
     if (is_dropped)
         throw Exception("Table is dropped", ErrorCodes::TABLE_IS_DROPPED);
 
-    return result;
+    return lock;
 }
 
 
 TableExclusiveLockHolder IStorage::lockExclusively(const String & query_id, const std::chrono::milliseconds & acquire_timeout)
 {
     TableExclusiveLockHolder result;
-    result.alter_lock = tryLockTimed(alter_lock, RWLockImpl::Write, query_id, acquire_timeout);
+    result.drop_lock = tryLockTimed(drop_lock, RWLockImpl::Write, query_id, acquire_timeout);
 
     if (is_dropped)
         throw Exception("Table is dropped", ErrorCodes::TABLE_IS_DROPPED);
 
-    result.drop_lock = tryLockTimed(drop_lock, RWLockImpl::Write, query_id, acquire_timeout);
-
     return result;
 }
 
@@ -126,7 +130,7 @@ Pipe IStorage::alterPartition(
     throw Exception("Partition operations are not supported by storage " + getName(), ErrorCodes::NOT_IMPLEMENTED);
 }
 
-void IStorage::alter(const AlterCommands & params, ContextPtr context, TableLockHolder &)
+void IStorage::alter(const AlterCommands & params, ContextPtr context, AlterLockHolder &)
 {
     auto table_id = getStorageID();
     StorageInMemoryMetadata new_metadata = getInMemoryMetadata();

src/Storages/IStorage.h

@@ -238,7 +238,8 @@ public:
 
     /// Lock table for alter. This lock must be acuqired in ALTER queries to be
     /// sure, that we execute only one simultaneous alter. Doesn't affect share lock.
-    TableLockHolder lockForAlter(const String & query_id, const std::chrono::milliseconds & acquire_timeout);
+    using AlterLockHolder = std::unique_lock<std::timed_mutex>;
+    AlterLockHolder lockForAlter(const std::chrono::milliseconds & acquire_timeout);
 
     /// Lock table exclusively. This lock must be acquired if you want to be
     /// sure, that no other thread (SELECT, merge, ALTER, etc.) doing something
@@ -417,7 +418,7 @@ public:
     /** ALTER tables in the form of column changes that do not affect the change
       * to Storage or its parameters. Executes under alter lock (lockForAlter).
       */
-    virtual void alter(const AlterCommands & params, ContextPtr context, TableLockHolder & alter_lock_holder);
+    virtual void alter(const AlterCommands & params, ContextPtr context, AlterLockHolder & alter_lock_holder);
 
     /** Checks that alter commands can be applied to storage. For example, columns can be modified,
       * or primary key can be changes, etc.
@@ -596,12 +597,9 @@ public:
     virtual bool dropTableImmediately() { return false; }
 
 private:
-    /// Lock required for alter queries (lockForAlter). Always taken for write
-    /// (actually can be replaced with std::mutex, but for consistency we use
-    /// RWLock). Allows to execute only one simultaneous alter query. Also it
-    /// should be taken by DROP-like queries, to be sure, that all alters are
-    /// finished.
-    mutable RWLock alter_lock = RWLockImpl::create();
+    /// Lock required for alter queries (lockForAlter).
+    /// Allows to execute only one simultaneous alter query.
+    mutable std::timed_mutex alter_lock;
 
     /// Lock required for drop queries. Every thread that want to ensure, that
     /// table is not dropped have to table this lock for read (lockForShare).

src/Storages/MergeTree/MergeTreeData.cpp

@@ -2211,7 +2211,7 @@ MergeTreeData::MutableDataPartPtr MergeTreeData::createPart(
 
 void MergeTreeData::changeSettings(
         const ASTPtr & new_settings,
-        TableLockHolder & /* table_lock_holder */)
+        AlterLockHolder & /* table_lock_holder */)
 {
     if (new_settings)
     {

src/Storages/MergeTree/MergeTreeData.h

@@ -583,7 +583,7 @@ public:
     /// Change MergeTreeSettings
     void changeSettings(
         const ASTPtr & new_settings,
-        TableLockHolder & table_lock_holder);
+        AlterLockHolder & table_lock_holder);
 
     /// Should be called if part data is suspected to be corrupted.
     void reportBrokenPart(const String & name) const

src/Storages/StorageBuffer.cpp

@@ -1066,7 +1066,7 @@ std::optional<UInt64> StorageBuffer::totalBytes(const Settings & /*settings*/) c
     return total_writes.bytes;
 }
 
-void StorageBuffer::alter(const AlterCommands & params, ContextPtr local_context, TableLockHolder &)
+void StorageBuffer::alter(const AlterCommands & params, ContextPtr local_context, AlterLockHolder &)
 {
     auto table_id = getStorageID();
     checkAlterIsPossible(params, local_context);

src/Storages/StorageBuffer.h

@@ -107,7 +107,7 @@ public:
     void checkAlterIsPossible(const AlterCommands & commands, ContextPtr context) const override;
 
     /// The structure of the subordinate table is not checked and does not change.
-    void alter(const AlterCommands & params, ContextPtr context, TableLockHolder & table_lock_holder) override;
+    void alter(const AlterCommands & params, ContextPtr context, AlterLockHolder & table_lock_holder) override;
 
     std::optional<UInt64> totalRows(const Settings & settings) const override;
     std::optional<UInt64> totalBytes(const Settings & settings) const override;

src/Storages/StorageDistributed.cpp

@@ -785,7 +785,7 @@ void StorageDistributed::checkAlterIsPossible(const AlterCommands & commands, Co
     }
 }
 
-void StorageDistributed::alter(const AlterCommands & params, ContextPtr local_context, TableLockHolder &)
+void StorageDistributed::alter(const AlterCommands & params, ContextPtr local_context, AlterLockHolder &)
 {
     auto table_id = getStorageID();
 

src/Storages/StorageDistributed.h

@@ -94,7 +94,7 @@ public:
 
     /// in the sub-tables, you need to manually add and delete columns
     /// the structure of the sub-table is not checked
-    void alter(const AlterCommands & params, ContextPtr context, TableLockHolder & table_lock_holder) override;
+    void alter(const AlterCommands & params, ContextPtr context, AlterLockHolder & table_lock_holder) override;
 
     void startup() override;
     void shutdown() override;

src/Storages/StorageMaterializedView.cpp

@@ -274,7 +274,7 @@ bool StorageMaterializedView::optimize(
 void StorageMaterializedView::alter(
     const AlterCommands & params,
     ContextPtr local_context,
-    TableLockHolder &)
+    AlterLockHolder &)
 {
     auto table_id = getStorageID();
     StorageInMemoryMetadata new_metadata = getInMemoryMetadata();

src/Storages/StorageMaterializedView.h

@@ -50,7 +50,7 @@ public:
         const Names & deduplicate_by_columns,
         ContextPtr context) override;
 
-    void alter(const AlterCommands & params, ContextPtr context, TableLockHolder & table_lock_holder) override;
+    void alter(const AlterCommands & params, ContextPtr context, AlterLockHolder & table_lock_holder) override;
 
     void checkMutationIsPossible(const MutationCommands & commands, const Settings & settings) const override;
 

src/Storages/StorageMerge.cpp

@@ -651,7 +651,7 @@ void StorageMerge::checkAlterIsPossible(const AlterCommands & commands, ContextP
 }
 
 void StorageMerge::alter(
-    const AlterCommands & params, ContextPtr local_context, TableLockHolder &)
+    const AlterCommands & params, ContextPtr local_context, AlterLockHolder &)
 {
     auto table_id = getStorageID();
 

src/Storages/StorageMerge.h

@@ -43,7 +43,7 @@ public:
 
     /// you need to add and remove columns in the sub-tables manually
     /// the structure of sub-tables is not checked
-    void alter(const AlterCommands & params, ContextPtr context, TableLockHolder & table_lock_holder) override;
+    void alter(const AlterCommands & params, ContextPtr context, AlterLockHolder & table_lock_holder) override;
 
     bool mayBenefitFromIndexForIn(
         const ASTPtr & left_in_operand, ContextPtr query_context, const StorageMetadataPtr & metadata_snapshot) const override;

src/Storages/StorageMergeTree.cpp

@@ -271,7 +271,7 @@ void StorageMergeTree::truncate(const ASTPtr &, const StorageMetadataPtr &, Cont
 void StorageMergeTree::alter(
     const AlterCommands & commands,
     ContextPtr local_context,
-    TableLockHolder & table_lock_holder)
+    AlterLockHolder & table_lock_holder)
 {
     auto table_id = getStorageID();
     auto old_storage_settings = getSettings();

src/Storages/StorageMergeTree.h

@@ -86,7 +86,7 @@ public:
     void drop() override;
     void truncate(const ASTPtr &, const StorageMetadataPtr &, ContextPtr, TableExclusiveLockHolder &) override;
 
-    void alter(const AlterCommands & commands, ContextPtr context, TableLockHolder & table_lock_holder) override;
+    void alter(const AlterCommands & commands, ContextPtr context, AlterLockHolder & table_lock_holder) override;
 
     void checkTableCanBeDropped() const override;
 

src/Storages/StorageNull.cpp

@@ -64,7 +64,7 @@ void StorageNull::checkAlterIsPossible(const AlterCommands & commands, ContextPt
 }
 
 
-void StorageNull::alter(const AlterCommands & params, ContextPtr context, TableLockHolder &)
+void StorageNull::alter(const AlterCommands & params, ContextPtr context, AlterLockHolder &)
 {
     auto table_id = getStorageID();
 

src/Storages/StorageNull.h

@@ -43,7 +43,7 @@ public:
 
     void checkAlterIsPossible(const AlterCommands & commands, ContextPtr context) const override;
 
-    void alter(const AlterCommands & params, ContextPtr context, TableLockHolder & table_lock_holder) override;
+    void alter(const AlterCommands & params, ContextPtr context, AlterLockHolder & table_lock_holder) override;
 
     std::optional<UInt64> totalRows(const Settings &) const override
     {

src/Storages/StorageProxy.h

@@ -91,7 +91,7 @@ public:
         IStorage::renameInMemory(new_table_id);
     }
 
-    void alter(const AlterCommands & params, ContextPtr context, TableLockHolder & alter_lock_holder) override
+    void alter(const AlterCommands & params, ContextPtr context, AlterLockHolder & alter_lock_holder) override
     {
         getNested()->alter(params, context, alter_lock_holder);
         IStorage::setInMemoryMetadata(getNested()->getInMemoryMetadata());

src/Storages/StorageReplicatedMergeTree.cpp

@@ -4355,7 +4355,8 @@ bool StorageReplicatedMergeTree::executeMetadataAlter(const StorageReplicatedMer
     zookeeper->multi(requests);
 
     {
-        auto lock = lockForAlter(RWLockImpl::NO_QUERY, getSettings()->lock_acquire_timeout_for_background_operations);
+        auto table_lock_holder = lockForShare(RWLockImpl::NO_QUERY, getSettings()->lock_acquire_timeout_for_background_operations);
+        auto alter_lock_holder = lockForAlter(getSettings()->lock_acquire_timeout_for_background_operations);
         LOG_INFO(log, "Metadata changed in ZooKeeper. Applying changes locally.");
 
         auto metadata_diff = ReplicatedMergeTreeTableMetadata(*this, getInMemoryMetadataPtr()).checkAndFindDiff(metadata_from_entry);
@@ -4427,7 +4428,7 @@ PartitionBlockNumbersHolder StorageReplicatedMergeTree::allocateBlockNumbersInAf
 
 
 void StorageReplicatedMergeTree::alter(
-    const AlterCommands & commands, ContextPtr query_context, TableLockHolder & table_lock_holder)
+    const AlterCommands & commands, ContextPtr query_context, AlterLockHolder & table_lock_holder)
 {
     assertNotReadonly();
 
@@ -4632,7 +4633,7 @@ void StorageReplicatedMergeTree::alter(
         }
     }
 
-    table_lock_holder.reset();
+    table_lock_holder.unlock();
 
     LOG_DEBUG(log, "Updated shared metadata nodes in ZooKeeper. Waiting for replicas to apply changes.");
     waitForLogEntryToBeProcessedIfNecessary(*alter_entry, query_context, "Some replicas doesn't finish metadata alter: ");

src/Storages/StorageReplicatedMergeTree.h

@@ -129,7 +129,7 @@ public:
         const Names & deduplicate_by_columns,
         ContextPtr query_context) override;
 
-    void alter(const AlterCommands & commands, ContextPtr query_context, TableLockHolder & table_lock_holder) override;
+    void alter(const AlterCommands & commands, ContextPtr query_context, AlterLockHolder & table_lock_holder) override;
 
     void mutate(const MutationCommands & commands, ContextPtr context) override;
     void waitMutation(const String & znode_name, size_t mutations_sync) const;

src/Storages/TableLockHolder.h

@@ -16,8 +16,6 @@ struct TableExclusiveLockHolder
 private:
     friend class IStorage;
 
-    /// Order is important.
-    TableLockHolder alter_lock;
     TableLockHolder drop_lock;
 };