Skip to main content

· 3 min read

Question

How do I set up ClickHouse with a Docker image to connect to Microsoft SQL Server?

Answer

Notes on this example

  • Uses the ClickHouse Docker Ubuntu image
  • Uses the FreeTDS Driver
  • Uses MSSQL Server 2012R2
  • Windows hostname for this example is MARSDB2.marsnet2.local at IP: 192.168.1.133 (update with your hostname and/or IP)
  • MSSQL Instance name MARSDB2
  • MSSQL Login and datbase users are sql_user

Example setup in MSSQL for testing

Database and table created in MSSQL:

Screenshot 2024-01-01 at 8 25 50 PM

MSSQL Login User, sql_user:

Screenshot 2024-01-01 at 8 27 11 PM

Database membership roles for sql_user:

Screenshot 2024-01-01 at 8 27 35 PM

Database User with Login:

Screenshot 2024-01-01 at 8 35 34 PM

Configuring ClickHouse with ODBC

Create a working directory:

mkdir ch-odbc-mssql
cd ch-odbc-mssql

Create an odbc.ini file:

vim odbc.ini

Add the following entries to update the name of the DSN and IP:

[marsdb2_mssql]
Driver = FreeTDS
Server = 192.168.1.133

Create an odbcinst.ini file:

vim odbcinst.ini

Add the following entries (trace is optional but helps with debugging):

[ODBC]
Trace = Yes
TraceFile = /tmp/odbc.log

[FreeTDS]
Description = FreeTDS
Driver = /usr/lib/aarch64-linux-gnu/odbc/libtdsodbc.so
Setup = /usr/lib/x86_64-linux-gnu/odbc/libtdsS.so
UsageCount = 1

Configure a Dockerfile to download the image and add the TDS and required ODBC libraries

Create the Dockerfile:

vim Dockerfile

Add the contents of the Dockerfile:

FROM clickhouse/clickhouse-server:23.10

# Install the ODBC driver

RUN apt-get update && apt-get install -y --no-install-recommends unixodbc \
&& apt-get install -y freetds-bin freetds-common freetds-dev libct4 libsybdb5 \
&& apt-get install tdsodbc

Build the new docker image:

docker build . -t marsnet/clickhouse-odbc:23.10

Create a docker-compose.yml file:

vim docker-compose.yml

Add the following contents to the YAML:

version: '3.7'
services:
clickhouse:
image: marsnet/clickhouse-odbc:23.10
container_name: clickhouse-odbc
hostname: clickhouse-host
ports:
- "9000:9000"
- "8123:8123"
- "9009:9009"
volumes:
- ./odbc.ini:/etc/odbc.ini
- ./odbcinst.ini:/etc/odbcinst.ini
restart: always
ulimits:
memlock:
soft: -1
hard: -1
nofile:
soft: 262144
hard: 262144
deploy:
resources:
limits:
memory: 4g

Start the container:

docker compose up --detach

After you start the container, you should see something like this:

ch-odbc-mssql % docker compose up --detach
[+] Running 1/1
✔ Container clickhouse-odbc Started

Check to ensure the container is running:

ch-odbc-mssql % docker ps
CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
87a400b803ce marsnet/clickhouse-odbc:23.10 "/entrypoint.sh" 57 minutes ago Up About a minute 0.0.0.0:8123->8123/tcp, 0.0.0.0:9000->9000/tcp, 0.0.0.0:9009->9009/tcp clickhouse-odbc

Test ODBC connection

Login with the ClickHouse client:

./clickhouse client

Test the SELECT using the odbc table function to the remote MSSQL Database table:

clickhouse-host :) SELECT * from odbc('DSN=marsdb2_mssql;port=1433;Uid=sql_user;Pwd=ClickHouse123;Database=db1', 'table1');

SELECT *
FROM odbc('DSN=marsdb2_mssql;port=1433;Uid=sql_user;Pwd=ClickHouse123;Database=db1', 'table1')

Query id: 23494da2-6e12-4ade-95fa-372a0420cac1

┌─id─┬─column1─┐
│ 1 │ abc │
│ 2 │ def │
│ 3 │ ghi │
└────┴─────────┘

3 rows in set. Elapsed: 0.188 sec.

You can also create a remote table using the odbc table engine:

CREATE TABLE table1_odbc_mssql
(
`id` Int32,
`column1` String
)
ENGINE = ODBC('DSN=marsdb2_mssql;port=1433;Uid=sql_user;Pwd=ClickHouse123;Database=db1', 'dbo', 'table1')

Use a SELECT query to test the new remote table:

clickhouse-host :) select * from table1_odbc_mssql;

SELECT *
FROM table1_odbc_mssql

Query id: 94724368-485d-4364-ae58-a435a225c37d

┌─id─┬─column1─┐
│ 1 │ abc │
│ 2 │ def │
│ 3 │ ghi │
└────┴─────────┘

3 rows in set. Elapsed: 0.218 sec.

For more information, please see:

· 2 min read

Question

How do I work with JSON message using a source or landing table to extract with a Materialized View?
How do I work with JSON without the experimental JSON Object?

Answer

A common pattern to work with JSON data is to send the data to a landing table and use JSONExtract functions to pull the data onto a new table using a Materialized View trigger. This is normally done in the following flow and pattern:

source data --> MergeTree table --> Materialized View (with base table) --> application/client

The landing table should have a raw string field where you would store the raw json. It should also have one to two other fields that can be used for management of that table so that it could be partitioned and trimmed as the data ages.

*some integrations can add fields to the original data for example if using the ClickHouse Kafka Connector Sink.

Simplified example below:

  • create the example database
create database db1;
  • create a landing table where your raw json will be inserted:
create table db1.table2_json_raw
(
id Int32,
timestamp DateTime,
raw String
)
engine = MergeTree()
order by timestamp;
  • create the base table for the materialized view
create table db1.table2_json_mv_base
(
id Int32,
timestamp DateTime,
raw_string String,
custId Int8,
custName String
)
engine = MergeTree()
order by timestamp;
  • create the materialized view to the base table
create materialized view db1.table2_json_mv to db1.table2_json_mv_base
AS SELECT
id,
timestamp,
raw as raw_string,
simpleJSONExtractRaw(raw, 'customerId') as custId,
simpleJSONExtractRaw(raw, 'customerName') as custName
FROM
db1.table2_json_raw;
  • insert some sample rows
 insert into db1.table2_json_raw
values
(1, '2024-05-16 00:00:00', '{"customerId":1, "customerName":"ABC"}'),
(2, '2024-05-16 00:00:01', '{"customerId":2, "customerName":"XYZ"}');
  • view the results from the extraction and the materialized view that would be used in the queries
clickhouse-cloud :) select * from db1.table2_json_mv;

SELECT *
FROM db1.table2_json_mv

Query id: 12655fd3-567a-4dfb-9ef7-abc4b11ad044

┌─id─┬───────────timestamp─┬─raw_string─────────────────────────────┬─custId─┬─custName─┐
│ 1 │ 2024-05-16 00:00:00 │ {"customerId":1, "customerName":"ABC"} │ 1 │ "ABC" │
│ 2 │ 2024-05-16 00:00:01 │ {"customerId":2, "customerName":"XYZ"} │ 2 │ "XYZ" │
└────┴─────────────────────┴────────────────────────────────────────┴────────┴──────────┘

Additional Reference links:
Materialized Views: https://clickhouse.com/docs/en/guides/developer/cascading-materialized-views
Working with JSON: https://clickhouse.com/docs/en/integrations/data-formats/json#other-approaches
JSON functions: https://clickhouse.com/docs/en/sql-reference/functions/json-functions

· 6 min read

This example shows how AD users that belong to different AD security groups can be given role access in ClickHouse. It also shows how a user may be added to multiple AD user groups so they can have access provided by multiple roles.

In this environment, we have the following:

  • A Windows Active Directory domain: marsnet2.local
  • A ClickHouse Cluster, cluster_1S_3R with 3 nodes on a cluster configuration of 1 shard, 3 replicas
  • 3 AD users
AD UserDescription
clickhouse_ad_adminClickHouse Admin user
clickhouse_db1_userUser with access to db1.table1
clickhouse_db2_userUser with access to db2.table1
ch_db1_db2_userUser with access to both db1.table1 and db2.table1
  • 3 AD security groups
AD GroupDescription
clickhouse_ad_adminsClickHouse Admins group
clickhouse_ad_db1_usersGroup to map with access to db1.table1
clickhouse_ad_db2_usersGroup to map with access to db2.table1
  • Example AD Environment and UO structure:

Example_AD_Env_and_UO_structure

  • Example AD Security Group Configuration:

Example_AD_Group_clickhouse_ad_db1_users

  • Example AD User Configuration:

Example_AD_user_clickhouse_db1_user

  1. In Windows AD Users and Groups, add each user to their respective group(s), they will be mapped to the ClickHouse roles (example in the next step).
AD Security GroupClickHouse Role
clickhouse_ad_adminclickhouse_ad_admins
clickhouse_db1_userclickhouse_ad_db1_users
clickhouse_db2_userclickhouse_ad_db2_users
ch_db1_db2_userclickhouse_ad_db1_users and clickhouse_ad_db2_users
  • Example user group membership:

Example_AD_user_to_group

  1. In ClickHouse config.xml, add the ldap_servers configuration to each ClickHouse node.
<ldap_servers>
<marsnet2_ad>
<host>marsdc1.marsnet2.local</host>
<port>389</port>
<bind_dn>{user_name}@marsnet2.local</bind_dn>
<user_dn_detection>
<base_dn>OU=Users,OU=ClickHouse,DC=marsnet2,DC=local</base_dn>
<search_filter>(&amp;(objectClass=user)(sAMAccountName={user_name}))</search_filter>
</user_dn_detection>
<enable_tls>no</enable_tls>
</marsnet2_ad>
</ldap_servers>
xml tagDescriptionExample Value
ldap_serversTag used to define the ldap servers that will be used by ClickHouseNA
marsnet_adThis tag is arbitrary and is just a label to use to identify the server in <user_directories> sectionNA
hostFQDN or IP Address of Active Directory server or domainmarsdc1.marsnet2.local
portActive Directory Port, usually 389 for non-ssl or 636 for SSL389
bind_dnWhich user will be used to create the bind to AD, it can be a dedicated user if regular users are not allowed to{user_name}@marsnet2.local
user_dn_detectionSettings on how ClickHouse will find the AD usersNA
base_dnAD OU path to start the search for the usersOU=Users,OU=ClickHouse,DC=marsnet2,DC=local
search_filterldap search filter to find the AD user(&(objectClass=user)(sAMAccountName={user_name}))

Refer to documentation for full set of options: https://clickhouse.com/docs/en/operations/external-authenticators/ldap#ldap-server-definition

  1. In ClickHouse config.xml, add the <user_directories> configuration with <ldap> entries to each ClickHouse node.
<user_directories>
<users_xml>
<path>users.xml</path>
</users_xml>
<local_directory>
<path>/var/lib/clickhouse/access/</path>
</local_directory>
<ldap>
<server>marsnet2_ad</server>
<role_mapping>
<base_dn>OU=Groups,OU=ClickHouse,DC=marsnet2,DC=local</base_dn>
<search_filter>(&amp;(objectClass=group)(member={user_dn}))</search_filter>
<attribute>CN</attribute>
<scope>subtree</scope>
<prefix>clickhouse_</prefix>
</role_mapping>
</ldap>
</user_directories>
xml tagDescriptionExample Value
user_directoriesDefines which authenticators will be usedNA
ldapThis contains the settings for the ldap servers, in this AD that will be usedNA
serverThis is the tag that was define in the <ldap_servers> sectionmarsnet2_ad
role_mappingdefinition on how the users authenticated will be mapped between AD groups and ClickHouse rolesNA
base_dnAD path that the system will use to start search for AD groupsOU=Groups,OU=ClickHouse,DC=marsnet2,DC=local
search_filterldap search filter to find the AD groups(&(objectClass=group)(member={user_dn}))
attributeWhich AD attribute field should be used to identify the userCN
scopeWhich levels in the base DN the system should search for the groupssubtree
prefixPrefix for the names of the groups in AD, this prefix will be removed to find the roles in ClickHouseclickhouse_

Refer to documentation for full set of options: https://clickhouse.com/docs/en/operations/external-authenticators/ldap#ldap-external-user-directory

note::: Since the AD security groups were prefixed in the example - i.e. clickhouse_ad_db1_users- when the system retrieves them, the prefix will be removed and the system will look for a ClickHouse role called ad_db1_users to map to clickhouse_ad_db1_users. :::

  1. Create example databases.
create database db1 on cluster 'cluster_1S_3R';
create database db2 on cluster 'cluster_1S_3R';
  1. Create example tables.
create table db1.table1 on cluster 'cluster_1S_3R'
(
id Int32,
column1 String
)
engine = MergeTree()
order by id;

create table db2.table1 on cluster 'cluster_1S_3R
(
id Int32,
column1 String
)
engine = MergeTree()
order by id;
  1. Insert sample data.
insert into db1.table1
values
(1, 'a');

insert into db2.table1
values
(2, 'b');
  1. Create ClickHouse Roles.
create role ad_admins on cluster 'cluster_1S_3R';
create role ad_db1_users on cluster 'cluster_1S_3R';
create role ad_db2_users on cluster 'cluster_1S_3R';
  1. Grant the privileges to the roles.
GRANT SHOW, SELECT, INSERT, ALTER, CREATE, DROP, UNDROP TABLE, TRUNCATE, OPTIMIZE, BACKUP, KILL QUERY, KILL TRANSACTION, MOVE PARTITION BETWEEN SHARDS, ACCESS MANAGEMENT, SYSTEM, dictGet, displaySecretsInShowAndSelect, INTROSPECTION, SOURCES, CLUSTER ON *.* on cluster 'cluster_1S_3R' TO ad_admins WITH GRANT OPTION;

GRANT SELECT ON db1.table1 on cluster 'cluster_1S_3R' TO ad_db1_users;

GRANT SELECT ON db2.table1 on cluster 'cluster_1S_3R' TO ad_db2_users;
  1. Test access for restricted db1 user. For example:
root@chnode1:/etc/clickhouse-server# clickhouse-client --user clickhouse_db1_user --password MyPassword123  --secure --port 9440 --host chnode1.marsnet.local
ClickHouse client version 24.1.3.31 (official build).
Connecting to chnode1.marsnet.local:9440 as user clickhouse_db1_user.
Connected to ClickHouse server version 24.1.3.


clickhouse :) select * from db1.table1;

SELECT *
FROM db1.table1

Query id: b04b92d6-5b8b-40a2-a92a-f06f15774930

┌─id─┬─column1─┐
│ 1 │ a │
└────┴─────────┘

1 row in set. Elapsed: 0.004 sec.

clickhouse :) select * from db2.table1;

SELECT *
FROM db2.table1

Query id: 7f7eaa44-7b47-4184-807a-6968a56057ad


Elapsed: 0.115 sec.

Received exception from server (version 24.1.3):
Code: 497. DB::Exception: Received from chnode1.marsnet.local:9440. DB::Exception: clickhouse_db1_user: Not enough privileges. To execute this query, it's necessary to have the grant SELECT(id, column1) ON db2.table1. (ACCESS_DENIED)
  1. Test access for the user that has access to both databases, db1 and db2. For example:
root@chnode1:/etc/clickhouse-server# clickhouse-client --user ch_db1_db2_user --password MyPassword123  --secure --port 9440 --host chnode1.marsnet.local
ClickHouse client version 24.1.3.31 (official build).
Connecting to chnode1.marsnet.local:9440 as user ch_db1_db2_user.
Connected to ClickHouse server version 24.1.3.

clickhouse :) select * from db1.table1;

SELECT *
FROM db1.table1

Query id: 23084744-08c2-48bd-8635-a23438812026

┌─id─┬─column1─┐
│ 1 │ a │
└────┴─────────┘

1 row in set. Elapsed: 0.005 sec.

clickhouse :) select * from db2.table1;

SELECT *
FROM db2.table1

Query id: f9954ec4-d8d9-4b5a-9f68-a7aa79a1bb4a

┌─id─┬─column1─┐
│ 2 │ b │
└────┴─────────┘

1 row in set. Elapsed: 0.004 sec.
  1. Test access for the Admin user. For example:
root@chnode1:/etc/clickhouse-server# clickhouse-client --user clickhouse_ad_admin --password MyPassword123  --secure --port 9440 --host chnode1.marsnet.local
ClickHouse client version 24.1.3.31 (official build).
Connecting to chnode1.marsnet.local:9440 as user clickhouse_ad_admin.
Connected to ClickHouse server version 24.1.3.

clickhouse :) create table db1.table2 on cluster 'cluster_1S_3R'
(
id Int32,
column1 String
)
engine = MergeTree()
order by id;

CREATE TABLE db1.table2 ON CLUSTER cluster_1S_3R
(
`id` Int32,
`column1` String
)
ENGINE = MergeTree
ORDER BY id

Query id: 6041fd32-4294-44bd-b442-3fdd41333e6f

┌─host──────────────────┬─port─┬─status─┬─error─┬─num_hosts_remaining─┬─num_hosts_active─┐
│ chnode1.marsnet.local │ 9440 │ 0 │ │ 2 │ 2 │
└───────────────────────┴──────┴────────┴───────┴─────────────────────┴──────────────────┘
┌─host──────────────────┬─port─┬─status─┬─error─┬─num_hosts_remaining─┬─num_hosts_active─┐
│ chnode2.marsnet.local │ 9440 │ 0 │ │ 1 │ 1 │
└───────────────────────┴──────┴────────┴───────┴─────────────────────┴──────────────────┘
┌─host──────────────────┬─port─┬─status─┬─error─┬─num_hosts_remaining─┬─num_hosts_active─┐
│ chnode3.marsnet.local │ 9440 │ 0 │ │ 0 │ 0 │
└───────────────────────┴──────┴────────┴───────┴─────────────────────┴──────────────────┘

· 6 min read

Overview: This article walks through the process of sending data from a Kafka topic to a ClickHouse table. We’ll use the Wiki recent changes feed, which provides a stream of events that represent changes made to various Wikimedia properties. The steps include:

  1. How to setup Kafka on Ubuntu
  2. Ingest a stream of data into a Kakfa topic
  3. Create a ClickHouse table that subscribes to the topic

1. Setup Kafka on Ubuntu

  1. Create an Ubuntu ec2 instance and SSH on to it:
ssh -i ~/training.pem [email protected]
  1. Install Kafka (based on the instructions here: https://www.linode.com/docs/guides/how-to-install-apache-kafka-on-ubuntu/):
sudo apt update
sudo apt install openjdk-11-jdk

mkdir /home/ubuntu/kafka
cd /home/ubuntu/kafka/

wget https://downloads.apache.org/kafka/3.7.0/kafka_2.13-3.7.0.tgz

tar -zxvf kafka_2.13-3.7.0.tgz
  1. Start ZooKeeper:
cd kafka_2.13-3.7.0
bin/zookeeper-server-start.sh config/zookeeper.properties
  1. Open a new console and launch Kafka:
ssh -i ~/training.pem [email protected]
cd kafka/kafka_2.13-3.7.0/
bin/kafka-server-start.sh config/server.properties
  1. Open a third console and create a topic named wikimedia:
ssh -i ~/training.pem [email protected]
cd kafka/kafka_2.13-3.7.0/

bin/kafka-topics.sh --create --topic wikimedia --bootstrap-server localhost:9092
  1. You can verify it was created successfully by:
bin/kafka-topics.sh --list --bootstrap-server localhost:9092

2. Ingest the Wikimedia Stream into Kafka

  1. We need some utilities first:
sudo apt-get install librdkafka-dev libyajl-dev
sudo apt-get install kafkacat
  1. The data is sent to Kafka using a clever curl command that grabs the latest Wikimedia events, parses out the JSON data and sends that to the Kafka topic:
curl -N https://stream.wikimedia.org/v2/stream/recentchange  | awk '/^data: /{gsub(/^data: /, ""); print}' | kafkacat -P -b localhost:9092 -t wikimedia
  1. You can "describe" the topic:
bin/kafka-topics.sh --describe --topic wikimedia --bootstrap-server localhost:9092
  1. Let's verify everything is working by consuming some events:
bin/kafka-console-consumer.sh --topic wikimedia --from-beginning --bootstrap-server localhost:9092
  1. Hit Ctrl+c to kill the previous command.

3. Ingest the Data into ClickHouse

  1. Here is what the incoming data looks like:
{
"$schema": "/mediawiki/recentchange/1.0.0",
"meta": {
"uri": "https://www.wikidata.org/wiki/Q45791749",
"request_id": "f64cfb17-04ba-4d09-8935-38ec6f0001c2",
"id": "9d7d2b5a-b79b-45ea-b72c-69c3b69ae931",
"dt": "2024-04-18T13:21:21Z",
"domain": "www.wikidata.org",
"stream": "mediawiki.recentchange",
"topic": "eqiad.mediawiki.recentchange",
"partition": 0,
"offset": 5032636513
},
"id": 2196113017,
"type": "edit",
"namespace": 0,
"title": "Q45791749",
"title_url": "https://www.wikidata.org/wiki/Q45791749",
"comment": "/* wbsetqualifier-add:1| */ [[Property:P1545]]: 20, Modify PubMed ID: 7292984 citation data from NCBI, Europe PMC and CrossRef",
"timestamp": 1713446481,
"user": "Cewbot",
"bot": true,
"notify_url": "https://www.wikidata.org/w/index.php?diff=2131981357&oldid=2131981341&rcid=2196113017",
"minor": false,
"patrolled": true,
"length": {
"old": 75618,
"new": 75896
},
"revision": {
"old": 2131981341,
"new": 2131981357
},
"server_url": "https://www.wikidata.org",
"server_name": "www.wikidata.org",
"server_script_path": "/w",
"wiki": "wikidatawiki",
"parsedcomment": "<span dir=\"auto\"><span class=\"autocomment\">Added qualifier: </span></span> <a href=\"/wiki/Property:P1545\" title=\"series ordinal | position of an item in its parent series (most frequently a 1-based index), generally to be used as a qualifier (different from &quot;rank&quot; defined as a class, and from &quot;ranking&quot; defined as a property for evaluating a quality).\"><span class=\"wb-itemlink\"><span class=\"wb-itemlink-label\" lang=\"en\" dir=\"ltr\">series ordinal</span> <span class=\"wb-itemlink-id\">(P1545)</span></span></a>: 20, Modify PubMed ID: 7292984 citation data from NCBI, Europe PMC and CrossRef"
}
  1. We will need the Kafka table engine to pull the data from the Kafka topic:
CREATE OR REPLACE TABLE wikiQueue
(
`id` UInt32,
`type` String,
`title` String,
`title_url` String,
`comment` String,
`timestamp` UInt64,
`user` String,
`bot` Bool,
`server_url` String,
`server_name` String,
`wiki` String,
`meta` Tuple(uri String, id String, stream String, topic String, domain String)
)
ENGINE = Kafka(
'ec2.compute.amazonaws.com:9092',
'wikimedia',
'consumer-group-wiki',
'JSONEachRow'
);
  1. For some reason the Kafka table engine seems to take the public ec2 URL and convert it to the private DNS name, so I had to add that to my local /etc/hosts file:
52.14.154.92  ip.us-east-2.compute.internal
  1. You can read from a Kafka table, you just have to enable a setting:
SELECT *
FROM wikiQueue
LIMIT 20
FORMAT Vertical
SETTINGS stream_like_engine_allow_direct_select = 1;

The rows should come back nicely parsed based on the columns defined in the wikiQueue table:

id:          2473996741
type: edit
title: File:Père-Lachaise - Division 6 - Cassereau 05.jpg
title_url: https://commons.wikimedia.org/wiki/File:P%C3%A8re-Lachaise_-_Division_6_-_Cassereau_05.jpg
comment: /* wbcreateclaim-create:1| */ [[d:Special:EntityPage/P921]]: [[d:Special:EntityPage/Q112327116]], [[:toollabs:quickstatements/#/batch/228454|batch #228454]]
timestamp: 1713457283
user: Ameisenigel
bot: false
server_url: https://commons.wikimedia.org
server_name: commons.wikimedia.org
wiki: commonswiki
meta: ('https://commons.wikimedia.org/wiki/File:P%C3%A8re-Lachaise_-_Division_6_-_Cassereau_05.jpg','01a832e2-24c5-4ccb-bd93-8e2c0e429418','mediawiki.recentchange','eqiad.mediawiki.recentchange','commons.wikimedia.org')
  1. We need a MergeTree table to store these incoming events:
CREATE TABLE rawEvents (
id UInt64,
type LowCardinality(String),
comment String,
timestamp DateTime64(3, 'UTC'),
title_url String,
topic LowCardinality(String),
user String
)
ENGINE = MergeTree
ORDER BY (type, timestamp);
  1. Let's define a materialized view that gets triggered when an insert occurs on the Kafka table and sends the data to our rawEvents table:
CREATE MATERIALIZED VIEW rawEvents_mv TO rawEvents
AS
SELECT
id,
type,
comment,
toDateTime(timestamp) AS timestamp,
title_url,
tupleElement(meta, 'topic') AS topic,
user
FROM wikiQueue
WHERE title_url <> '';
  1. You should start seeing data going into rawEvents almost immediately:
SELECT count()
FROM rawEvents;
  1. Let's view some of the rows:
SELECT *
FROM rawEvents
LIMIT 5
FORMAT Vertical
Row 1:
──────
id: 124842852
type: 142
comment: Pere prlpz commented on "Plantilles Enciclopèdia Catalana" (Diria que no cal fer res als articles. Es pot actualitzar els enllaços que es facin servir a les referències (tot i que l'antic encara ha...)
timestamp: 2024-04-18 16:22:29.000
title_url: https://ca.wikipedia.org/wiki/Tema:Wu36d6vfsiuu4jsi
topic: eqiad.mediawiki.recentchange
user: Pere prlpz

Row 2:
──────
id: 2473996748
type: categorize
comment: [[:File:Ruïne van een poortgebouw, RP-T-1976-29-6(R).jpg]] removed from category
timestamp: 2024-04-18 16:21:20.000
title_url: https://commons.wikimedia.org/wiki/Category:Pieter_Moninckx
topic: eqiad.mediawiki.recentchange
user: Warburg1866

Row 3:
──────
id: 311828596
type: categorize
comment: [[:Cujo (película)]] añadida a la categoría
timestamp: 2024-04-18 16:21:21.000
title_url: https://es.wikipedia.org/wiki/Categor%C3%ADa:Pel%C3%ADculas_basadas_en_obras_de_Stephen_King
topic: eqiad.mediawiki.recentchange
user: Beta15

Row 4:
──────
id: 311828597
type: categorize
comment: [[:Cujo (película)]] eliminada de la categoría
timestamp: 2024-04-18 16:21:21.000
title_url: https://es.wikipedia.org/wiki/Categor%C3%ADa:Trabajos_basados_en_obras_de_Stephen_King
topic: eqiad.mediawiki.recentchange
user: Beta15

Row 5:
──────
id: 48494536
type: categorize
comment: [[:braiteremmo]] ajoutée à la catégorie
timestamp: 2024-04-18 16:21:21.000
title_url: https://fr.wiktionary.org/wiki/Cat%C3%A9gorie:Wiktionnaire:Exemples_manquants_en_italien
topic: eqiad.mediawiki.recentchange
user: Àncilu bot
  1. Let's see what types of events are coming in:
SELECT
type,
count()
FROM rawEvents
GROUP BY type
   ┌─type───────┬─count()─┐
1. │ 142 │ 1 │
2. │ new │ 1003 │
3. │ categorize │ 12228 │
4. │ log │ 1799 │
5. │ edit │ 17142 │
└────────────┴─────────┘

Let's define a materialized view chained to our current materialized view. We will keep track of some aggregated stats per minute:

CREATE TABLE byMinute
(
`dateTime` DateTime64(3, 'UTC') NOT NULL,
`users` AggregateFunction(uniq, String),
`pages` AggregateFunction(uniq, String),
`updates` AggregateFunction(sum, UInt32)
)
ENGINE = AggregatingMergeTree
ORDER BY dateTime;

CREATE MATERIALIZED VIEW byMinute_mv TO byMinute
AS SELECT
toStartOfMinute(timestamp) AS dateTime,
uniqState(user) AS users,
uniqState(title_url) AS pages,
sumState(toUInt32(1)) AS updates
FROM rawEvents
GROUP BY dateTime;
  1. We will need -Merge functions to view the results:
SELECT
dateTime AS dateTime,
uniqMerge(users) AS users,
uniqMerge(pages) AS pages,
sumMerge(updates) AS updates
FROM byMinute
GROUP BY dateTime
ORDER BY dateTime DESC
LIMIT 10;

· One min read

In order to build and contribute to ClickHouse, you must use LLVM and Clang.

These are the commands to build the latest version of LLVM and Clang on Linux:

git clone [email protected]:llvm/llvm-project.git
mkdir llvm-build
cd llvm-build
cmake -GNinja -DCMAKE_BUILD_TYPE:STRING=Release -DLLVM_ENABLE_PROJECTS=all -DLLVM_TARGETS_TO_BUILD=all ../llvm-project/llvm
time ninja
sudo ninja install

· 3 min read

This is useful when there are tables that have similar naming conventions and similar columns but are not replicated. An example is searching the system database for entries in the query log tables.

The query_log table is not replicated, and only queries that are executed on a specific node get logged. Data may also roll to a different table For example, data may be inserted into query_log_0, query_log_1, etc. Since one node may roll at a different time than others, it is useful to try to find the data we're looking for in tables that are not exactly named the same.

In essence, we need to do something like this, but in ClickHouse syntax:

SELECT column1, column2 FROM my_db.my_table_*

For this, we can use the clusterAllReplicas() to search all the nodes and the merge() table function to be able to use a regex pattern to search the multiple tables.

The following example shows how to query all tables with the prefix query_log:

clickhouse-cloud :) SELECT 
`event_time`,
`query_id`,
`query`,
`type`
FROM
clusterAllReplicas(default,merge('system', '^query_log*'))
WHERE
query ilike '%db1.table1%' and event_time > now() - toIntervalMinute(5);

SELECT
event_time,
query_id,
query,
type
FROM clusterAllReplicas(default, merge('system', '^query_log*'))
WHERE (query ILIKE '%db1.table1%') AND (event_time > (now() - toIntervalMinute(5)))

Query id: de95c13e-5759-436e-90d9-a12c1327889e

┌──────────event_time─┬─query_id─────────────────────────────┬─query──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┬─type────────┐
│ 2024-02-08 00:15:20 │ d1dd0d6a-4337-4e58-bdd1-c2c827b6dfe2 │ /* ddl_entry=query-0000000428 */ CREATE TABLE db1.table1 UUID '781f25db-3cd1-47c6-a76e-701945a67485' (`id` Int32, `string_column` String) ENGINE = ReplicatedMergeTree ORDER BY id │ QueryStart │
│ 2024-02-08 00:15:20 │ d1dd0d6a-4337-4e58-bdd1-c2c827b6dfe2 │ /* ddl_entry=query-0000000428 */ CREATE TABLE db1.table1 UUID '781f25db-3cd1-47c6-a76e-701945a67485' (`id` Int32, `string_column` String) ENGINE = ReplicatedMergeTree ORDER BY id │ QueryFinish │
└─────────────────────┴──────────────────────────────────────┴────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┴─────────────┘
┌──────────event_time─┬─query_id─────────────────────────────┬─query──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┬─type────────┐
│ 2024-02-08 00:15:20 │ f0ca43b2-544e-4b94-a21d-0f05e777fa96 │ /* ddl_entry=query-0000000428 */ CREATE TABLE db1.table1 UUID '781f25db-3cd1-47c6-a76e-701945a67485' (`id` Int32, `string_column` String) ENGINE = ReplicatedMergeTree ORDER BY id │ QueryStart │
│ 2024-02-08 00:15:20 │ f0ca43b2-544e-4b94-a21d-0f05e777fa96 │ /* ddl_entry=query-0000000428 */ CREATE TABLE db1.table1 UUID '781f25db-3cd1-47c6-a76e-701945a67485' (`id` Int32, `string_column` String) ENGINE = ReplicatedMergeTree ORDER BY id │ QueryFinish │
└─────────────────────┴──────────────────────────────────────┴────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┴─────────────┘
┌──────────event_time─┬─query_id─────────────────────────────┬─query──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┬─type────────┐
│ 2024-02-08 00:15:20 │ 5cc0a508-7f64-460b-a5be-949ef1d1f2ca │ /* ddl_entry=query-0000000428 */ CREATE TABLE db1.table1 UUID '781f25db-3cd1-47c6-a76e-701945a67485' (`id` Int32, `string_column` String) ENGINE = ReplicatedMergeTree ORDER BY id │ QueryStart │
│ 2024-02-08 00:15:20 │ 5cc0a508-7f64-460b-a5be-949ef1d1f2ca │ /* ddl_entry=query-0000000428 */ CREATE TABLE db1.table1 UUID '781f25db-3cd1-47c6-a76e-701945a67485' (`id` Int32, `string_column` String) ENGINE = ReplicatedMergeTree ORDER BY id │ QueryFinish │
│ 2024-02-08 00:15:20 │ d1e01cb0-a27c-44b2-829c-90fb2596c9c0 │ create table db1.table1
(
id Int32,
string_column String
)
engine = MergeTree
order by id │ QueryStart │
│ 2024-02-08 00:15:20 │ d1e01cb0-a27c-44b2-829c-90fb2596c9c0 │ create table db1.table1
(
id Int32,
string_column String
)
engine = MergeTree
order by id │ QueryFinish │
│ 2024-02-08 00:15:27 │ 6c2c6c3f-173e-464f-bfa0-643089ca085e │ insert into db1.table1
values
│ QueryStart │
│ 2024-02-08 00:15:27 │ 6c2c6c3f-173e-464f-bfa0-643089ca085e │ insert into db1.table1
values
│ QueryFinish │
└─────────────────────┴──────────────────────────────────────┴────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┴─────────────┘

10 rows in set. Elapsed: 0.046 sec. Processed 317.27 thousand rows, 33.57 MB (6.89 million rows/s., 729.43 MB/s.)
Peak memory usage: 67.04 MiB.

Note that the columns you select must exist on each of the tables being queried or you may encounter an error such as:

Received exception from server (version 24.0.2):
Code: 47. DB::Exception: Received from abc123.us-west-2.aws.clickhouse.cloud:9440. DB::Exception: Missing columns: 'hostname' while processing query: 'WITH 'query_log_0' AS _table

Alternatively, you can use the EXCEPT clause to exclude any columns that may not be present on different tables.

example:

SELECT * EXCEPT (used_privileges, missing_privileges)
FROM clusterAllReplicas(default, merge('system', 'query_log[\\_]*'))
WHERE (query_id = '31a93b8e-1149-4edd-a33d-f03f47a676cc') AND (event_time = '2024-10-21 12:49:04')

· 2 min read

Dictionaries created in ClickHouse Cloud may experience inconsistency during the initial creation phase. This means that you may not see any data in the dictionary right after creation. However, after several retries, the creation query may land on different replicas, and data will be visible.

This sometimes occurs because the dictionary was created before the part reached the server. As an example:

2024-01-25 13:38:25.615837 - CREATE DICTIONARY received
2024-01-25 13:38:25.626468 - CREATE DICTIONARY finished
2024-01-25 13:38:25.733008 - Part all_0_0_0 downloaded

As you can see, the part only arrived after the dictionary was created. This can be a bigger problem if you are using LIFETIME(MIN 0 MAX 0) because this means that dictionary will never be refreshed automatically. Therefore, the dictionary will remain empty until the command RELOAD DICTIONARIES is executed.

The solution to this issue is to use a SELECT query instead of specifying a source table when creating the dictionary and enabling the setting select_sequential_consistency=1.

Instead of specifying a source table:

SOURCE(CLICKHOUSE(
table 'test.temp_title_table_1706189903924'
user default password 'PASSWORD'))

Use a SELECT query with select_sequential_consistency=1:

SOURCE(CLICKHOUSE(QUERY
'SELECT songTitle, mappedTitle
FROM test.temp_title_table_1706189903924
SETTINGS select_sequential_consistency=1' USER default PASSWORD ''))

Why does this issue occur?

When you insert data and then create or reload a dictionary, the DDL may reach a replica before the data (or new data) does. This leads to the dictionaries being inconsistent between replicas. Then, depending on which replica receives the query, you may get different results.

Note that the same thing happens when you insert and immediately after read from a table. If you read from a replica that hasn't replicated the data yet, you won't see the newly inserted data. When you need sequential consistency, at the cost of performance (which is why it's generally not recommended to use) you can enable select_sequential_consistency.

The case of dictionaries is a bit trickier since dictionaries don't use the settings from the query, but the settings from the server. As a result, when loading data into the dictionary, even if you SET select_sequential_consistency=1 data may load inconsistently across replicas. Specifying select_sequential_consistency=1 in the dictionary source query allows the dictionary to adhere to this setting even if it's not globally enabled as a server setting.

· 3 min read

Question

How can I backup a specific partition in ClickHouse?

Answer

See the below example, this uses the S3(Minio) disk configuration listed in our docker compose examples page.

Note

This does NOT apply to ClickHouse Cloud

Create a table:

ch_minio_s3 :) CREATE TABLE my_table
(
`event_time` DateTime,
`field_foo` String,
`field_bar` String,
`number` UInt256
)
ENGINE = MergeTree
PARTITION BY number % 2
ORDER BY tuple()

CREATE TABLE my_table
(
`event_time` DateTime,
`field_foo` String,
`field_bar` String,
`number` UInt256
)
ENGINE = MergeTree
PARTITION BY number % 2
ORDER BY tuple()

Query id: a1a54a5a-eac0-477c-b847-b40acaa62780

Ok.

0 rows in set. Elapsed: 0.016 sec.

Add some data that will fill both partitions equally:

ch_minio_s3 :) INSERT INTO my_table SELECT
toDateTime(now() + number) AS event_time,
randomPrintableASCII(10) AS field_foo,
randomPrintableASCII(20) AS field_bar,
number
FROM numbers(1000000)

INSERT INTO my_table SELECT
toDateTime(now() + number) AS event_time,
randomPrintableASCII(10) AS field_foo,
randomPrintableASCII(20) AS field_bar,
number
FROM numbers(1000000)

Query id: bf6ef803-5747-4ea1-ad00-a17967e349b6

Ok.

0 rows in set. Elapsed: 0.282 sec. Processed 1.00 million rows, 8.00 MB (3.55 million rows/s., 28.39 MB/s.)

verify data:

ch_minio_s3 :) SELECT
_partition_id AS partition_id,
cityHash64(sum(number)) AS hash,
count() AS count
FROM my_table
GROUP BY partition_id

SELECT
_partition_id AS partition_id,
cityHash64(sum(number)) AS hash,
count() AS count
FROM my_table
GROUP BY partition_id

Query id: d8febfb0-5339-4f97-aefa-ef0003128526

┌─partition_id─┬─cityHash64(sum(number))─┬──count─┐
015460940821314360342500000
111827822647069388611500000
└──────────────┴─────────────────────────┴────────┘

2 rows in set. Elapsed: 0.025 sec. Processed 1.00 million rows, 32.00 MB (39.97 million rows/s., 1.28 GB/s.)

backup partition with id 1 to configured s3 disk:

ch_minio_s3 :) BACKUP TABLE my_table PARTITION 1 TO Disk('s3','backups/');

BACKUP TABLE my_table PARTITION 1 TO Disk('s3', 'backups/')

Query id: 810f6144-e282-42e2-99d0-9a80c75a927d

┌─id───────────────────────────────────┬─status─────────┐
4d1da197-c4c9-4b6e-966c-76202eadbd53 │ BACKUP_CREATED │
└──────────────────────────────────────┴────────────────┘

1 row in set. Elapsed: 0.095 sec.

Drop the table:

ch_minio_s3 :) DROP TABLE my_table

DROP TABLE my_table

Query id: c3456044-4689-406e-82ac-8d08b8b618fe

Ok.

0 rows in set. Elapsed: 0.007 sec.

restore just partition with id 1 from backup:

ch_minio_s3 :) RESTORE TABLE my_table PARTITION 1 FROM Disk('s3','backups/');

RESTORE TABLE my_table PARTITION 1 FROM Disk('s3', 'backups/')

Query id: ea306c73-83c5-479f-9c0c-391594facc69

┌─id───────────────────────────────────┬─status───┐
│ ec6841a8-0607-465e-bc4d-d446f960d40a │ RESTORED │
└──────────────────────────────────────┴──────────┘

1 row in set. Elapsed: 0.065 sec.

validate the restored data:

ch_minio_s3 :) SELECT
_partition_id AS partition_id,
cityHash64(sum(number)) AS hash,
count() AS count
FROM my_table
GROUP BY partition_id

SELECT
_partition_id AS partition_id,
cityHash64(sum(number)) AS hash,
count() AS count
FROM my_table
GROUP BY partition_id

Query id: a916176d-6a6e-47fc-ba7d-79bb33b152d8

┌─partition_id─┬─────────────────hash─┬──count─┐
111827822647069388611500000
└──────────────┴──────────────────────┴────────┘

1 row in set. Elapsed: 0.012 sec. Processed 500.00 thousand rows, 16.00 MB (41.00 million rows/s., 1.31 GB/s.)

· 2 min read

Question

I'm writing data into ClickHouse cloud and need to be able ,when reading data, to guarantee that I'm getting the latest complete information.

Answer

Talking to same node

If you are using native protocol, or a session to do your write/read, you should then be connected to the same replica: in this scenario you're reading directly from the node where you're writing, then your read will always be consistent.

Talking to a random node

If you can't guarantee you're talking to the same node (for example talking to the node via HTTPS calls which get shuffled via the a load load balancer), you can either:

A)

  1. write your data
  2. connect to a new replica
  3. run SYSTEM SYNC REPLICA db.table_name LIGHTWEIGHT
  4. read the latest data

See SYSTEM commands reference

OR

B) read anytime with sequential consistency

SELECT 
...
SETTINGS select_sequential_consistency = 1

note when using ClickHouse Cloud and its default SharedMergeTree Engine, using insert_quorum_parallel is not required (it's a given)

Using SYSTEM SYNC REPLICAS or select_sequential_consistency will increase the load on ClickHouse Keeper and might have slower performance depending on the load on the service.

The recommended approach is to do the writes/read using the same session or the native protocol (sticky connection).

· 2 min read

If you install ClickHouse using brew, you may encounter an error from MacOS. By default, MacOS will not run applications or tools created by a developer who cannot be verified. When attempting to run any clickhouse command, you may see this error:

MacOS showing a developer verification error.

To get around this verification error, you need to remove the app from MacOS' quarintine bin either by finding the appropriate setting in your System Settings window, or using the terminal.

System settings process

The easiest way to remove the clickhouse executable from the quarintine bin is to:

  1. Open System settings.

  2. Navigate to Privacy & Security:

  3. Scroll to the bottom of the window to find a message saying _"clickhouse-macos-aarch64" was blocked from use because it is not from an identified developer".

  4. Click Allow Anyway.

  5. Enter your MacOS user password.

You should now be able to run clickhouse commands in your terminal.

Terminal process

You can perform this process using the command-line:

First find out where Homebrew installed the clickhouse executable:

which clickhouse

This should output something like:

/opt/homebrew/bin/clickhouse

Remove clickhouse from the quarantine bin by running xattr -d com.apple.quarantine following by the path from the previous command:

xattr -d com.apple.quarantine /opt/homebrew/bin/clickhouse

You should now be able to run the clickhouse executable:

clickhouse

This should output something like:

Use one of the following commands:
clickhouse local [args]
clickhouse client [args]
clickhouse benchmark [args]
...