Skip to main content

EmbeddedRocksDB Engine

This engine allows integrating ClickHouse with rocksdb.

Creating a Table​

CREATE TABLE [IF NOT EXISTS] [db.]table_name [ON CLUSTER cluster]
(
    name1 [type1] [DEFAULT|MATERIALIZED|ALIAS expr1],
    name2 [type2] [DEFAULT|MATERIALIZED|ALIAS expr2],
    ...
) ENGINE = EmbeddedRocksDB([ttl, rocksdb_dir, read_only]) PRIMARY KEY(primary_key_name)
[ SETTINGS name=value, ... ]

Engine parameters:

  • ttl - time to live for values. TTL is accepted in seconds. If TTL is 0, regular RocksDB instance is used (without TTL).
  • rocksdb_dir - path to the directory of an existed RocksDB or the destination path of the created RocksDB. Open the table with the specified rocksdb_dir.
  • read_only - when read_only is set to true, read-only mode is used. For storage with TTL, compaction will not be triggered (neither manual nor automatic), so no expired entries are removed.
  • primary_key_name – any column name in the column list.
  • primary key must be specified, it supports only one column in the primary key. The primary key will be serialized in binary as a rocksdb key.
  • columns other than the primary key will be serialized in binary as rocksdb value in corresponding order.
  • queries with key equals or in filtering will be optimized to multi keys lookup from rocksdb.

Engine settings:

  • optimize_for_bulk_insert – Table is optimized for bulk insertions (insert pipeline will create SST files and import to rocksdb database instead of writing to memtables); default value: 1.
  • bulk_insert_block_size - Minimum size of SST files (in term of rows) created by bulk insertion; default value: 1048449.

Example:

CREATE TABLE test
(
    `key` String,
    `v1` UInt32,
    `v2` String,
    `v3` Float32
)
ENGINE = EmbeddedRocksDB
PRIMARY KEY key

Metrics​

There is also system.rocksdb table, that expose rocksdb statistics:

SELECT
    name,
    value
FROM system.rocksdb

┌─name──────────────────────┬─value─┐
│ no.file.opens             │     1 │
│ number.block.decompressed │     1 │
└───────────────────────────┴───────┘

Configuration​

You can also change any rocksdb options using config:

<rocksdb>
    <options>
        <max_background_jobs>8</max_background_jobs>
    </options>
    <column_family_options>
        <num_levels>2</num_levels>
    </column_family_options>
    <tables>
        <table>
            <name>TABLE</name>
            <options>
                <max_background_jobs>8</max_background_jobs>
            </options>
            <column_family_options>
                <num_levels>2</num_levels>
            </column_family_options>
        </table>
    </tables>
</rocksdb>

By default trivial approximate count optimization is turned off, which might affect the performance count() queries. To enable this optimization set up optimize_trivial_approximate_count_query = 1. Also, this setting affects system.tables for EmbeddedRocksDB engine, turn on the settings to see approximate values for total_rows and total_bytes.

Supported operations​

Inserts​

When new rows are inserted into EmbeddedRocksDB, if the key already exists, the value will be updated, otherwise a new key is created.

Example:

INSERT INTO test VALUES ('some key', 1, 'value', 3.2);

Deletes​

Rows can be deleted using DELETE query or TRUNCATE.

DELETE FROM test WHERE key LIKE 'some%' AND v1 > 1;
ALTER TABLE test DELETE WHERE key LIKE 'some%' AND v1 > 1;
TRUNCATE TABLE test;

Updates​

Values can be updated using the ALTER TABLE query. The primary key cannot be updated.

ALTER TABLE test UPDATE v1 = v1 * 10 + 2 WHERE key LIKE 'some%' AND v3 > 3.1;

Joins​

A special direct join with EmbeddedRocksDB tables is supported. This direct join avoids forming a hash table in memory and accesses the data directly from the EmbeddedRocksDB.

With large joins you may see much lower memory usage with direct joins because the hash table is not created.

To enable direct joins:

SET join_algorithm = 'direct, hash'
Tip

When the join_algorithm is set to direct, hash, direct joins will be used when possible, and hash otherwise.

Example​

Create and populate an EmbeddedRocksDB table:​
CREATE TABLE rdb
(
    `key` UInt32,
    `value` Array(UInt32),
    `value2` String
)
ENGINE = EmbeddedRocksDB
PRIMARY KEY key
INSERT INTO rdb
    SELECT
        toUInt32(sipHash64(number) % 10) as key,
        [key, key+1] as value,
        ('val2' || toString(key)) as value2
    FROM numbers_mt(10);
Create and populate a table to join with table rdb:​
CREATE TABLE t2
(
    `k` UInt16
)
ENGINE = TinyLog
INSERT INTO t2 SELECT number AS k
FROM numbers_mt(10)
Set the join algorithm to direct:​
SET join_algorithm = 'direct'
An INNER JOIN:​
SELECT *
FROM
(
    SELECT k AS key
    FROM t2
) AS t2
INNER JOIN rdb ON rdb.key = t2.key
ORDER BY key ASC
┌─key─┬─rdb.key─┬─value──┬─value2─┐
│   0 │       0 │ [0,1]  │ val20  │
│   2 │       2 │ [2,3]  │ val22  │
│   3 │       3 │ [3,4]  │ val23  │
│   6 │       6 │ [6,7]  │ val26  │
│   7 │       7 │ [7,8]  │ val27  │
│   8 │       8 │ [8,9]  │ val28  │
│   9 │       9 │ [9,10] │ val29  │
└─────┴─────────┴────────┴────────┘

More information on Joins​