Поправил по комментариям разработчика.
9.0 KiB
toc_priority | toc_title |
---|---|
68 | External Disks for Storing Data |
External Disks for Storing Data
Data, processed in ClickHouse, is usually stored in the local file system — on the same machine with the ClickHouse server. That requires large-capacity disks, which can be expensive enough. To avoid that you can store the data remotely — on Amazon S3 disks or in the Hadoop Distributed File System (HDFS).
To work with data stored on Amazon S3
disks use S3 table engine, and to work with data in the Hadoop Distributed File System — HDFS table engine.
To load data from a web server with static files use a disk with type web.
Zero-copy Replication
ClickHouse supports zero-copy replication for S3
and HDFS
disks, which means that if the data is stored remotely on several machines and needs to be synchronized, then only the metadata is replicated (paths to the data parts), but not the data itself.
Configuring HDFS
MergeTree and Log family table engines can store data to HDFS using a disk with type HDFS
.
Configuration markup:
<yandex>
<storage_configuration>
<disks>
<hdfs>
<type>hdfs</type>
<endpoint>hdfs://hdfs1:9000/clickhouse/</endpoint>
</hdfs>
</disks>
<policies>
<hdfs>
<volumes>
<main>
<disk>hdfs</disk>
</main>
</volumes>
</hdfs>
</policies>
</storage_configuration>
<merge_tree>
<min_bytes_for_wide_part>0</min_bytes_for_wide_part>
</merge_tree>
</yandex>
Required parameters:
endpoint
— HDFS endpoint URL inpath
format. Endpoint URL should contain a root path to store data.
Optional parameters:
min_bytes_for_seek
— The minimal number of bytes to use seek operation instead of sequential read. Default value:1 Mb
.
Using Virtual File System for Data Encryption
You can encrypt the data stored on S3, or HDFS external disks, or on a local disk. To turn on the encryption mode, in the configuration file you must define a disk with the type encrypted
and choose a disk on which the data will be saved. An encrypted
disk ciphers all written files on the fly, and when you read files from an encrypted
disk it deciphers them automatically. So you can work with an encrypted
disk like with a normal one.
Example of disk configuration:
<disks>
<disk1>
<type>local</type>
<path>/path1/</path>
</disk1>
<disk2>
<type>encrypted</type>
<disk>disk1</disk>
<path>path2/</path>
<key>_16_ascii_chars_</key>
</disk2>
</disks>
For example, when ClickHouse writes data from some table to a file store/all_1_1_0/data.bin
to disk1
, then in fact this file will be written to the physical disk along the path /path1/store/all_1_1_0/data.bin
.
When writing the same file to disk2
, it will actually be written to the physical disk at the path /path1/path2/store/all_1_1_0/data.bin
in encrypted mode.
Required parameters:
type
—encrypted
. Otherwise the encrypted disk is not created.disk
— Type of disk for data storage.key
— The key for encryption and decryption. Type: Uint64. You can usekey_hex
parameter to encrypt in hexadecimal form. You can specify multiple keys using theid
attribute (see example above).
Optional parameters:
path
— Path to the location on the disk where the data will be saved. If not specified, the data will be saved in the root directory.current_key_id
— The key used for encryption. All the specified keys can be used for decryption, and you can always switch to another key while maintaining access to previously encrypted data.algorithm
— Algorithm for encryption. Possible values:AES_128_CTR
,AES_192_CTR
orAES_256_CTR
. Default value:AES_128_CTR
. The key length depends on the algorithm:AES_128_CTR
— 16 bytes,AES_192_CTR
— 24 bytes,AES_256_CTR
— 32 bytes.
Example of disk configuration:
<yandex>
<storage_configuration>
<disks>
<disk_s3>
<type>s3</type>
<endpoint>...
</disk_s3>
<disk_s3_encrypted>
<type>encrypted</type>
<disk>disk_s3</disk>
<algorithm>AES_128_CTR</algorithm>
<key_hex id="0">00112233445566778899aabbccddeeff</key_hex>
<key_hex id="1">ffeeddccbbaa99887766554433221100</key_hex>
<current_key_id>1</current_key_id>
</disk_s3_encrypted>
</disks>
</storage_configuration>
</yandex>
Storing Data on Web Server
There is a tool clickhouse-static-files-uploader
, which prepares a data directory for a given table (SELECT data_paths FROM system.tables WHERE name = 'table_name'
). For each table you need, you get a directory of files. These files can be uploaded to, for example, a web server with static files. After this preparation, you can load this table into any ClickHouse server via DiskWeb
.
The following types of queries are not supported: CREATE TABLE, ALTER TABLE, RENAME TABLE, DETACH TABLE and TRUNCATE TABLE.
Web server storage is supported only for the MergeTree and Log engine families. To access the data stored on a web
disk, use the storage_policy setting when executing the query. For example, ATTACH TABLE table_web UUID '{}' (id Int32) ENGINE = MergeTree() ORDER BY id SETTINGS storage_policy = 'web'
.
A ready test case. You need to add this configuration to config:
<yandex>
<storage_configuration>
<disks>
<web>
<type>web</type>
<endpoint>https://clickhouse-datasets.s3.yandex.net/disk-with-static-files-tests/test-hits/</endpoint>
</web>
</disks>
<policies>
<web>
<volumes>
<main>
<disk>web</disk>
</main>
</volumes>
</web>
</policies>
</storage_configuration>
</yandex>
And then execute this query (https://github.com/ClickHouse/ClickHouse/blob/master/src/Disks/DiskWebServer.h#L17).
Required parameters:
type
—web
. Otherwise the disk is not created.endpoint
— The endpoint URL inpath
format. Endpoint URL must contain a root path to store data, where they were uploaded.
Optional parameters:
min_bytes_for_seek
— The minimal number of bytes to use seek operation instead of sequential read. Default value:1
Mb.remote_fs_read_backoff_threashold
— The maximum wait time when trying to read data for remote disk. Default value:10000
seconds.remote_fs_read_backoff_max_tries
— The maximum number of attempts to read with backoff. Default value:5
.
If query fails with an exception DB:Exception Unreachable URL
, then may help to adjust settings: http_connection_timeout, http_receive_timeout, keep_alive_timeout.
To get files for upload run:
clickhouse static-files-disk-uploader --metadata-path <path> --output-dir <dir>
(--metadata-path
can be found in query SELECT data_paths FROM system.tables WHERE name = 'table_name'
).
When loading files by endpoint
they must be loaded into <endpoint>/store/
path, but config must contain only endpoint
.
If URL is not reachable on disk load when the server is starting up tables, then all errors are caught. If in this case there were errors, tables can be reloaded (become visible) via DETACH TABLE table_name
-> ATTACH TABLE table_name
. If metadata was successfully loaded at server startup, then tables are available straight away.
Use http_max_single_read_retries setting to limit the maximum number of retries during a single HTTP read.