Other engines

MergeTree is the workhorse, but ClickHouse ships dozens of other table engines. This page is a catalog of the non-MergeTree engines under src/Storages/.

Distributed and Merge

• StorageDistributed (StorageDistributed.cpp, ~90 KB) — a stateless query router. It holds no data; it forwards SELECT/INSERT to shards in a cluster. Reads emit RemoteSource processors that issue the rewritten sub-query to each shard. Writes either go through the same path (sync) or are queued in a local "sharded" directory and flushed asynchronously. See Distributed queries.

• StorageMerge (StorageMerge.cpp) — virtual UNION over a set of tables matching a regex. Useful for transparent table sharding.

In-memory and pseudo engines

• StorageMemory — rows in RAM. Used for ephemeral tables and as the default engine for the _local database in clickhouse-local.
• StorageBuffer (StorageBuffer.cpp) — write-through buffer in front of another table. Aggregates inserts in memory and flushes them when thresholds are hit.
• StorageNull — /dev/null for inserts; reads always return zero rows. Used to discard data while keeping triggers (e.g. MATERIALIZED VIEW TO Null()).
• StorageDummy — placeholder used during analyzer rewrites.
• StorageDictionary (StorageDictionary.cpp) — table view over a configured Dictionary. See Dictionaries.
• StorageJoin (StorageJoin.cpp) — a pre-built hash table queryable as joinGet(...) and reusable across queries.
• StorageSet (StorageSet.cpp) — a pre-built set for IN-operator reuse.
• StorageView / StorageMaterializedView — non-materialized and incrementally materialized views. See Materialized views.
• StorageWindowView (src/Storages/WindowView/) — streaming windowed aggregation.
• StorageGenerateRandom — produces synthetic rows from a schema. Used in tests and demos.
• StorageInput — an injection point for clickhouse-local / external pipelines.
• StorageLoop — repeats a sub-query infinitely (useful for stream-like demos).
• StorageExecutable (StorageExecutable.cpp) — runs an external script and reads its stdout.
• StorageFuzzJSON / StorageFuzzQuery — fuzzers as table engines.

Log family

Log, TinyLog, StripeLog (StorageLog.cpp, StorageStripeLog.cpp) are simple append-only column-stripe formats from before MergeTree. They have no indices and limited concurrency, but they are extremely simple and fast for tiny tables.

Object storage

The directory src/Storages/ObjectStorage/ is the modern object-storage layer. It abstracts S3, Azure, HDFS, and a "local fake" object store behind one set of engines:

• S3, AzureBlobStorage, HDFS — read/write tables backed by a bucket with one of the supported file formats (Parquet, ORC, JSON, CSV, …).
• Iceberg, DeltaLake, Hudi — read tables in those open table formats.
• *Cluster variants distribute reads across cluster nodes for parallelism.

The reading code lives under src/Storages/ObjectStorage/StorageObjectStorage.cpp, partitioning under src/Storages/HivePartitioningUtils.cpp. The actual transport is IObjectStorage from src/Disks/ObjectStorages/. See Object storage tables.

Streaming engines

Pull-based ingestion engines that hold a consumer thread and stream rows into materialized views:

• Kafka (src/Storages/Kafka/) — librdkafka-based consumer.
• RabbitMQ (src/Storages/RabbitMQ/) — AMQP-CPP-based consumer.
• NATS (src/Storages/NATS/) — cnats-based consumer.
• FileLog (src/Storages/FileLog/) — tail a directory of log files.
• S3Queue, AzureQueue (src/Storages/ObjectStorageQueue/) — newest-file ingestion from a bucket with at-least-once semantics tracked in Keeper.

The shared sink-side code is in src/Storages/MessageQueueSink.cpp and src/Storages/StreamingStorageRegistry.cpp. Each engine pairs with a producer (via IMessageProducer) for INSERT support.

External database engines

Engines that proxy to another database. Reads are translated; writes are forwarded.

• StorageMySQL, StoragePostgreSQL, StorageMongoDB, StorageRedis, StorageSQLite — first-class connectors.
• StorageXDBC (StorageXDBC.cpp) — generic ODBC/JDBC bridge that talks to clickhouse-odbc-bridge / clickhouse-library-bridge via src/BridgeHelper/.

Predicate pushdown for these engines lives in src/Storages/transformQueryForExternalDatabase.cpp and the analyzer-aware variant.

Time series and key-value

• StorageTimeSeries (StorageTimeSeries.cpp) — Prometheus-style time series (a tag-aware schema with hash-based grouping).
• StoragePrometheusQuery — proxy to a Prometheus instance.
• StorageKeeperMap (StorageKeeperMap.cpp, ~70 KB) — a key-value table whose data lives directly in Keeper. Useful for small dictionaries with strong consistency.
• src/Storages/RocksDB/ — embedded RocksDB-backed KV table.

Hive

src/Storages/Hive/ provides a reader for tables described in a Hive Metastore. Now largely superseded by Iceberg/DeltaLake engines but still in use.

ArrowFlight

StorageArrowFlight (StorageArrowFlight.cpp) exposes a table over Arrow Flight RPC. The corresponding handler is src/Server/ArrowFlight*.

Smaller helpers

• IStorageCluster — the base for *Cluster variants of object-storage engines.
• StorageProxy.h / StorageTableProxy.h / StorageTableFunction.h — proxy storages used during query rewriting.
• StorageLoop.cpp — loops a sub-query for streaming demos.
• StorageFromMergeTreeDataPart.cpp, StorageFromMergeTreeProjection.cpp — tiny "engines" used to read a single part or projection inside the optimizer.

See also

• MergeTree and Replicated MergeTree
• Distributed queries
• Object storage tables
• Materialized views
• Dictionaries