Merge pull request ClickHouse#62425 from Avogar/fix-partitioned-write

Respect settings truncate_on_insert/create_new_file_on_insert in s3/hdfs/azure engines during partitioned write

Author: Avogar

src/Storages/HDFS/StorageHDFS.cpp

@@ -871,6 +871,40 @@ class HDFSSink : public SinkToStorage
     bool cancelled = false;
 };
 
+namespace
+{
+    std::optional<String> checkAndGetNewFileOnInsertIfNeeded(const ContextPtr & context, const String & uri, size_t sequence_number)
+    {
+        const auto [path_from_uri, uri_without_path] = getPathFromUriAndUriWithoutPath(uri);
+
+        HDFSBuilderWrapper builder = createHDFSBuilder(uri_without_path + "/", context->getGlobalContext()->getConfigRef());
+        HDFSFSPtr fs = createHDFSFS(builder.get());
+
+        if (context->getSettingsRef().hdfs_truncate_on_insert || hdfsExists(fs.get(), path_from_uri.c_str()))
+            return std::nullopt;
+
+        if (context->getSettingsRef().hdfs_create_new_file_on_insert)
+        {
+            auto pos = uri.find_first_of('.', uri.find_last_of('/'));
+            String new_uri;
+            do
+            {
+                new_uri = uri.substr(0, pos) + "." + std::to_string(sequence_number) + (pos == std::string::npos ? "" : uri.substr(pos));
+                ++sequence_number;
+            }
+            while (!hdfsExists(fs.get(), new_uri.c_str()));
+
+            return new_uri;
+        }
+
+            throw Exception(
+                ErrorCodes::BAD_ARGUMENTS,
+                "File with path {} already exists. If you want to overwrite it, enable setting hdfs_truncate_on_insert, "
+                "if you want to create new file on each insert, enable setting hdfs_create_new_file_on_insert",
+                path_from_uri);
+    }
+}
+
 class PartitionedHDFSSink : public PartitionedSink
 {
 public:
@@ -894,6 +928,8 @@ class PartitionedHDFSSink : public PartitionedSink
     {
         auto path = PartitionedSink::replaceWildcards(uri, partition_id);
         PartitionedSink::validatePartitionKey(path, true);
+        if (auto new_path = checkAndGetNewFileOnInsertIfNeeded(context, path, 1))
+            path = *new_path;
         return std::make_shared<HDFSSink>(path, format, sample_block, context, compression_method);
     }
 
@@ -1056,7 +1092,7 @@ void ReadFromHDFS::initializePipeline(QueryPipelineBuilder & pipeline, const Bui
 
 SinkToStoragePtr StorageHDFS::write(const ASTPtr & query, const StorageMetadataPtr & metadata_snapshot, ContextPtr context_, bool /*async_insert*/)
 {
-    String current_uri = uris.back();
+    String current_uri = uris.front();
 
     bool has_wildcards = current_uri.find(PartitionedSink::PARTITION_ID_WILDCARD) != String::npos;
     const auto * insert_query = dynamic_cast<const ASTInsertQuery *>(query.get());
@@ -1078,34 +1114,10 @@ SinkToStoragePtr StorageHDFS::write(const ASTPtr & query, const StorageMetadataP
         if (is_path_with_globs)
             throw Exception(ErrorCodes::DATABASE_ACCESS_DENIED, "URI '{}' contains globs, so the table is in readonly mode", uris.back());
 
-        const auto [path_from_uri, uri_without_path] = getPathFromUriAndUriWithoutPath(current_uri);
-
-        HDFSBuilderWrapper builder = createHDFSBuilder(uri_without_path + "/", context_->getGlobalContext()->getConfigRef());
-        HDFSFSPtr fs = createHDFSFS(builder.get());
-
-        bool truncate_on_insert = context_->getSettingsRef().hdfs_truncate_on_insert;
-        if (!truncate_on_insert && !hdfsExists(fs.get(), path_from_uri.c_str()))
+        if (auto new_uri = checkAndGetNewFileOnInsertIfNeeded(context_, uris.front(), uris.size()))
         {
-            if (context_->getSettingsRef().hdfs_create_new_file_on_insert)
-            {
-                auto pos = uris[0].find_first_of('.', uris[0].find_last_of('/'));
-                size_t index = uris.size();
-                String new_uri;
-                do
-                {
-                    new_uri = uris[0].substr(0, pos) + "." + std::to_string(index) + (pos == std::string::npos ? "" : uris[0].substr(pos));
-                    ++index;
-                }
-                while (!hdfsExists(fs.get(), new_uri.c_str()));
-                uris.push_back(new_uri);
-                current_uri = new_uri;
-            }
-            else
-                throw Exception(
-                    ErrorCodes::BAD_ARGUMENTS,
-                    "File with path {} already exists. If you want to overwrite it, enable setting hdfs_truncate_on_insert, "
-                    "if you want to create new file on each insert, enable setting hdfs_create_new_file_on_insert",
-                    path_from_uri);
+            uris.push_back(*new_uri);
+            current_uri = *new_uri;
         }
 
         return std::make_shared<HDFSSink>(current_uri,

src/Storages/StorageAzureBlob.cpp

@@ -634,6 +634,36 @@ class StorageAzureBlobSink : public SinkToStorage
     std::mutex cancel_mutex;
 };
 
+namespace
+{
+    std::optional<String> checkAndGetNewFileOnInsertIfNeeded(const ContextPtr & context, AzureObjectStorage * object_storage, const String & path, size_t sequence_number)
+    {
+        if (context->getSettingsRef().azure_truncate_on_insert || !object_storage->exists(StoredObject(path)))
+            return std::nullopt;
+
+        if (context->getSettingsRef().azure_create_new_file_on_insert)
+        {
+            auto pos = path.find_first_of('.');
+            String new_path;
+            do
+            {
+            new_path = path.substr(0, pos) + "." + std::to_string(sequence_number) + (pos == std::string::npos ? "" : path.substr(pos));
+                ++sequence_number;
+            }
+            while (object_storage->exists(StoredObject(new_path)));
+
+            return new_path;
+        }
+
+        throw Exception(
+            ErrorCodes::BAD_ARGUMENTS,
+            "Object with key {} already exists. "
+            "If you want to overwrite it, enable setting azure_truncate_on_insert, if you "
+            "want to create a new file on each insert, enable setting azure_create_new_file_on_insert",
+            path);
+    }
+}
+
 class PartitionedStorageAzureBlobSink : public PartitionedSink, WithContext
 {
 public:
@@ -660,6 +690,8 @@ class PartitionedStorageAzureBlobSink : public PartitionedSink, WithContext
     {
         auto partition_key = replaceWildcards(blob, partition_id);
         validateKey(partition_key);
+        if (auto new_path = checkAndGetNewFileOnInsertIfNeeded(getContext(), object_storage, partition_key, 1))
+            partition_key = *new_path;
 
         return std::make_shared<StorageAzureBlobSink>(
             format,
@@ -837,8 +869,9 @@ void ReadFromAzureBlob::initializePipeline(QueryPipelineBuilder & pipeline, cons
 
 SinkToStoragePtr StorageAzureBlob::write(const ASTPtr & query, const StorageMetadataPtr & metadata_snapshot, ContextPtr local_context, bool /*async_insert*/)
 {
+    auto path = configuration.blobs_paths.front();
     auto sample_block = metadata_snapshot->getSampleBlock();
-    auto chosen_compression_method = chooseCompressionMethod(configuration.blobs_paths.back(), configuration.compression_method);
+    auto chosen_compression_method = chooseCompressionMethod(path, configuration.compression_method);
     auto insert_query = std::dynamic_pointer_cast<ASTInsertQuery>(query);
 
     auto partition_by_ast = insert_query ? (insert_query->partition_by ? insert_query->partition_by : partition_by) : nullptr;
@@ -854,44 +887,18 @@ SinkToStoragePtr StorageAzureBlob::write(const ASTPtr & query, const StorageMeta
             format_settings,
             chosen_compression_method,
             object_storage.get(),
-            configuration.blobs_paths.back());
+            path);
     }
     else
     {
         if (configuration.withGlobs())
             throw Exception(ErrorCodes::DATABASE_ACCESS_DENIED,
                             "AzureBlobStorage key '{}' contains globs, so the table is in readonly mode", configuration.blob_path);
 
-        bool truncate_in_insert = local_context->getSettingsRef().azure_truncate_on_insert;
-
-        if (!truncate_in_insert && object_storage->exists(StoredObject(configuration.blob_path)))
+        if (auto new_path = checkAndGetNewFileOnInsertIfNeeded(local_context, object_storage.get(), path, configuration.blobs_paths.size()))
         {
-
-            if (local_context->getSettingsRef().azure_create_new_file_on_insert)
-            {
-                size_t index = configuration.blobs_paths.size();
-                const auto & first_key = configuration.blobs_paths[0];
-                auto pos = first_key.find_first_of('.');
-                String new_key;
-
-                do
-                {
-                    new_key = first_key.substr(0, pos) + "." + std::to_string(index) + (pos == std::string::npos ? "" : first_key.substr(pos));
-                    ++index;
-                }
-                while (object_storage->exists(StoredObject(new_key)));
-
-                configuration.blobs_paths.push_back(new_key);
-            }
-            else
-            {
-                throw Exception(
-                    ErrorCodes::BAD_ARGUMENTS,
-                    "Object in bucket {} with key {} already exists. "
-                    "If you want to overwrite it, enable setting azure_truncate_on_insert, if you "
-                    "want to create a new file on each insert, enable setting azure_create_new_file_on_insert",
-                    configuration.container, configuration.blobs_paths.back());
-            }
+            configuration.blobs_paths.push_back(*new_path);
+            path = *new_path;
         }
 
         return std::make_shared<StorageAzureBlobSink>(
@@ -901,7 +908,7 @@ SinkToStoragePtr StorageAzureBlob::write(const ASTPtr & query, const StorageMeta
             format_settings,
             chosen_compression_method,
             object_storage.get(),
-            configuration.blobs_paths.back());
+            path);
     }
 }
 

src/Storages/StorageFile.cpp

@@ -1968,22 +1968,22 @@ SinkToStoragePtr StorageFile::write(
                                 "Table '{}' is in readonly mode because of globs in filepath",
                                 getStorageID().getNameForLogs());
 
-            path = paths.back();
+            path = paths.front();
             fs::create_directories(fs::path(path).parent_path());
 
             std::error_code error_code;
             if (!context->getSettingsRef().engine_file_truncate_on_insert && !is_path_with_globs
                 && !FormatFactory::instance().checkIfFormatSupportAppend(format_name, context, format_settings)
-                && fs::file_size(paths.back(), error_code) != 0 && !error_code)
+                && fs::file_size(path, error_code) != 0 && !error_code)
             {
                 if (context->getSettingsRef().engine_file_allow_create_multiple_files)
                 {
-                    auto pos = paths[0].find_first_of('.', paths[0].find_last_of('/'));
+                    auto pos = path.find_first_of('.', path.find_last_of('/'));
                     size_t index = paths.size();
                     String new_path;
                     do
                     {
-                        new_path = paths[0].substr(0, pos) + "." + std::to_string(index) + (pos == std::string::npos ? "" : paths[0].substr(pos));
+                        new_path = path.substr(0, pos) + "." + std::to_string(index) + (pos == std::string::npos ? "" : path.substr(pos));
                         ++index;
                     }
                     while (fs::exists(new_path));

src/Storages/StorageS3.cpp

@@ -955,6 +955,36 @@ class StorageS3Sink : public SinkToStorage
     std::mutex cancel_mutex;
 };
 
+namespace
+{
+    std::optional<String> checkAndGetNewFileOnInsertIfNeeded(const ContextPtr & context, const StorageS3::Configuration & configuration, const String & key, size_t sequence_number)
+    {
+        if (context->getSettingsRef().s3_truncate_on_insert || !S3::objectExists(*configuration.client, configuration.url.bucket, key, configuration.url.version_id, configuration.request_settings))
+            return std::nullopt;
+
+        if (context->getSettingsRef().s3_create_new_file_on_insert)
+        {
+            auto pos = key.find_first_of('.');
+            String new_key;
+            do
+            {
+                new_key = key.substr(0, pos) + "." + std::to_string(sequence_number) + (pos == std::string::npos ? "" : key.substr(pos));
+                ++sequence_number;
+            }
+            while (S3::objectExists(*configuration.client, configuration.url.bucket, new_key, configuration.url.version_id, configuration.request_settings));
+
+            return new_key;
+        }
+
+        throw Exception(
+            ErrorCodes::BAD_ARGUMENTS,
+            "Object in bucket {} with key {} already exists. "
+            "If you want to overwrite it, enable setting s3_truncate_on_insert, if you "
+            "want to create a new file on each insert, enable setting s3_create_new_file_on_insert",
+            configuration.url.bucket, key);
+    }
+}
+
 
 class PartitionedStorageS3Sink : public PartitionedSink, WithContext
 {
@@ -988,6 +1018,9 @@ class PartitionedStorageS3Sink : public PartitionedSink, WithContext
         auto partition_key = replaceWildcards(key, partition_id);
         validateKey(partition_key);
 
+        if (auto new_key = checkAndGetNewFileOnInsertIfNeeded(getContext(), configuration, partition_key, /* sequence_number */1))
+            partition_key = *new_key;
+
         return std::make_shared<StorageS3Sink>(
             format,
             sample_block,
@@ -1248,6 +1281,7 @@ void ReadFromStorageS3Step::initializePipeline(QueryPipelineBuilder & pipeline,
 SinkToStoragePtr StorageS3::write(const ASTPtr & query, const StorageMetadataPtr & metadata_snapshot, ContextPtr local_context, bool /*async_insert*/)
 {
     auto query_configuration = updateConfigurationAndGetCopy(local_context);
+    auto key = query_configuration.keys.front();
 
     auto sample_block = metadata_snapshot->getSampleBlock();
     auto chosen_compression_method = chooseCompressionMethod(query_configuration.keys.back(), query_configuration.compression_method);
@@ -1267,43 +1301,19 @@ SinkToStoragePtr StorageS3::write(const ASTPtr & query, const StorageMetadataPtr
             chosen_compression_method,
             query_configuration,
             query_configuration.url.bucket,
-            query_configuration.keys.back());
+            key);
     }
     else
     {
         if (query_configuration.withGlobs())
             throw Exception(ErrorCodes::DATABASE_ACCESS_DENIED,
                             "S3 key '{}' contains globs, so the table is in readonly mode", query_configuration.url.key);
 
-        bool truncate_in_insert = local_context->getSettingsRef().s3_truncate_on_insert;
-
-        if (!truncate_in_insert && S3::objectExists(*query_configuration.client, query_configuration.url.bucket, query_configuration.keys.back(), query_configuration.url.version_id, query_configuration.request_settings))
+        if (auto new_key = checkAndGetNewFileOnInsertIfNeeded(local_context, configuration, query_configuration.keys.front(), query_configuration.keys.size()))
         {
-            if (local_context->getSettingsRef().s3_create_new_file_on_insert)
-            {
-                size_t index = query_configuration.keys.size();
-                const auto & first_key = query_configuration.keys[0];
-                auto pos = first_key.find_first_of('.');
-                String new_key;
-                do
-                {
-                    new_key = first_key.substr(0, pos) + "." + std::to_string(index) + (pos == std::string::npos ? "" : first_key.substr(pos));
-                    ++index;
-                }
-                while (S3::objectExists(*query_configuration.client, query_configuration.url.bucket, new_key, query_configuration.url.version_id, query_configuration.request_settings));
-
-                query_configuration.keys.push_back(new_key);
-                configuration.keys.push_back(new_key);
-            }
-            else
-            {
-                throw Exception(
-                    ErrorCodes::BAD_ARGUMENTS,
-                    "Object in bucket {} with key {} already exists. "
-                    "If you want to overwrite it, enable setting s3_truncate_on_insert, if you "
-                    "want to create a new file on each insert, enable setting s3_create_new_file_on_insert",
-                    query_configuration.url.bucket, query_configuration.keys.back());
-            }
+            query_configuration.keys.push_back(*new_key);
+            configuration.keys.push_back(*new_key);
+            key = *new_key;
         }
 
         return std::make_shared<StorageS3Sink>(
@@ -1314,7 +1324,7 @@ SinkToStoragePtr StorageS3::write(const ASTPtr & query, const StorageMetadataPtr
             chosen_compression_method,
             query_configuration,
             query_configuration.url.bucket,
-            query_configuration.keys.back());
+            key);
     }
 }