Skip to main content

HDFS

This engine provides integration with the Apache Hadoop ecosystem by allowing to manage data on HDFS via ClickHouse. This engine is similar to the File and URL engines, but provides Hadoop-specific features.

Usage​

ENGINE = HDFS(URI, format)

Engine Parameters

  • URI - whole file URI in HDFS. The path part of URI may contain globs. In this case the table would be readonly.
  • format - specifies one of the available file formats. To perform SELECT queries, the format must be supported for input, and to perform INSERT queries – for output. The available formats are listed in the Formats section.

Example:

1. Set up the hdfs_engine_table table:

CREATE TABLE hdfs_engine_table (name String, value UInt32) ENGINE=HDFS('hdfs://hdfs1:9000/other_storage', 'TSV')

2. Fill file:

INSERT INTO hdfs_engine_table VALUES ('one', 1), ('two', 2), ('three', 3)

3. Query the data:

SELECT * FROM hdfs_engine_table LIMIT 2
β”Œβ”€name─┬─value─┐
β”‚ one β”‚ 1 β”‚
β”‚ two β”‚ 2 β”‚
β””β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”˜

Implementation Details​

  • Reads and writes can be parallel.
  • Zero-copy replication is supported.
  • Not supported:
    • ALTER and SELECT...SAMPLE operations.
    • Indexes.

Globs in path

Multiple path components can have globs. For being processed file should exists and matches to the whole path pattern. Listing of files determines during SELECT (not at CREATE moment).

  • * β€” Substitutes any number of any characters except / including empty string.
  • ? β€” Substitutes any single character.
  • {some_string,another_string,yet_another_one} β€” Substitutes any of strings 'some_string', 'another_string', 'yet_another_one'.
  • {N..M} β€” Substitutes any number in range from N to M including both borders.

Constructions with {} are similar to the remote table function.

Example

  1. Suppose we have several files in TSV format with the following URIs on HDFS:

    • 'hdfs://hdfs1:9000/some_dir/some_file_1'
    • 'hdfs://hdfs1:9000/some_dir/some_file_2'
    • 'hdfs://hdfs1:9000/some_dir/some_file_3'
    • 'hdfs://hdfs1:9000/another_dir/some_file_1'
    • 'hdfs://hdfs1:9000/another_dir/some_file_2'
    • 'hdfs://hdfs1:9000/another_dir/some_file_3'
  2. There are several ways to make a table consisting of all six files:

CREATE TABLE table_with_range (name String, value UInt32) ENGINE = HDFS('hdfs://hdfs1:9000/{some,another}_dir/some_file_{1..3}', 'TSV')

Another way:

CREATE TABLE table_with_question_mark (name String, value UInt32) ENGINE = HDFS('hdfs://hdfs1:9000/{some,another}_dir/some_file_?', 'TSV')

Table consists of all the files in both directories (all files should satisfy format and schema described in query):

CREATE TABLE table_with_asterisk (name String, value UInt32) ENGINE = HDFS('hdfs://hdfs1:9000/{some,another}_dir/*', 'TSV')
warning

If the listing of files contains number ranges with leading zeros, use the construction with braces for each digit separately or use ?.

Example

Create table with files named file000, file001, … , file999:

CREATE TABLE big_table (name String, value UInt32) ENGINE = HDFS('hdfs://hdfs1:9000/big_dir/file{0..9}{0..9}{0..9}', 'CSV')

Configuration​

Similar to GraphiteMergeTree, the HDFS engine supports extended configuration using the ClickHouse config file. There are two configuration keys that you can use: global (hdfs) and user-level (hdfs_*). The global configuration is applied first, and then the user-level configuration is applied (if it exists).

  <!-- Global configuration options for HDFS engine type -->
<hdfs>
<hadoop_kerberos_keytab>/tmp/keytab/clickhouse.keytab</hadoop_kerberos_keytab>
<hadoop_kerberos_principal>[email protected]</hadoop_kerberos_principal>
<hadoop_security_authentication>kerberos</hadoop_security_authentication>
</hdfs>

<!-- Configuration specific for user "root" -->
<hdfs_root>
<hadoop_kerberos_principal>[email protected]</hadoop_kerberos_principal>
</hdfs_root>

Configuration Options​

Supported by libhdfs3​

parameterdefault value
rpc_client_connect_tcpnodelaytrue
dfs_client_read_shortcircuittrue
output_replace-datanode-on-failuretrue
input_notretry-another-nodefalse
input_localread_mappedfiletrue
dfs_client_use_legacy_blockreader_localfalse
rpc_client_ping_interval10 * 1000
rpc_client_connect_timeout600 * 1000
rpc_client_read_timeout3600 * 1000
rpc_client_write_timeout3600 * 1000
rpc_client_socekt_linger_timeout-1
rpc_client_connect_retry10
rpc_client_timeout3600 * 1000
dfs_default_replica3
input_connect_timeout600 * 1000
input_read_timeout3600 * 1000
input_write_timeout3600 * 1000
input_localread_default_buffersize1 1024 1024
dfs_prefetchsize10
input_read_getblockinfo_retry3
input_localread_blockinfo_cachesize1000
input_read_max_retry60
output_default_chunksize512
output_default_packetsize64 * 1024
output_default_write_retry10
output_connect_timeout600 * 1000
output_read_timeout3600 * 1000
output_write_timeout3600 * 1000
output_close_timeout3600 * 1000
output_packetpool_size1024
output_heeartbeat_interval10 * 1000
dfs_client_failover_max_attempts15
dfs_client_read_shortcircuit_streams_cache_size256
dfs_client_socketcache_expiryMsec3000
dfs_client_socketcache_capacity16
dfs_default_blocksize64 1024 1024
dfs_default_uri"hdfs://localhost:9000"
hadoop_security_authentication"simple"
hadoop_security_kerberos_ticket_cache_path""
dfs_client_log_severity"INFO"
dfs_domain_socket_path""

HDFS Configuration Reference might explain some parameters.

ClickHouse extras​

parameterdefault value
hadoop_kerberos_keytab""
hadoop_kerberos_principal""
hadoop_kerberos_kinit_commandkinit
libhdfs3_conf""

Limitations​

  • hadoop_security_kerberos_ticket_cache_path and libhdfs3_conf can be global only, not user specific

Kerberos support​

If the hadoop_security_authentication parameter has the value kerberos, ClickHouse authenticates via Kerberos. Parameters are here and hadoop_security_kerberos_ticket_cache_path may be of help. Note that due to libhdfs3 limitations only old-fashioned approach is supported, datanode communications are not secured by SASL (HADOOP_SECURE_DN_USER is a reliable indicator of such security approach). Use tests/integration/test_storage_kerberized_hdfs/hdfs_configs/bootstrap.sh for reference.

If hadoop_kerberos_keytab, hadoop_kerberos_principal or hadoop_kerberos_kinit_command is specified, kinit will be invoked. hadoop_kerberos_keytab and hadoop_kerberos_principal are mandatory in this case. kinit tool and krb5 configuration files are required.

HDFS Namenode HA support​

libhdfs3 support HDFS namenode HA.

  • Copy hdfs-site.xml from an HDFS node to /etc/clickhouse-server/.
  • Add following piece to ClickHouse config file:
  <hdfs>
<libhdfs3_conf>/etc/clickhouse-server/hdfs-site.xml</libhdfs3_conf>
</hdfs>
  • Then use dfs.nameservices tag value of hdfs-site.xml as the namenode address in the HDFS URI. For example, replace hdfs://[email protected]:8020/abc/ with hdfs://[email protected]_nameservice/abc/.

Virtual Columns​

  • _path β€” Path to the file.
  • _file β€” Name of the file.

See Also

Original article