clickhouse

Ansible Clickhouse

This role is in charge of configuring & installing a clickhouse cluster with N shard and M replicas. apt and yum molecule tested on top of Vagrant.

The cluster is built based on ansible inventory groups aka inventory patterns, so following groups are mandatory to run the cluster:

clickhouse: contains all clickhouse inventory hosts. These hosts must set their hostname like: ch01-shard01-replica01 with regex like: ^ch\\d{2}-shard\\d{2}-replica\\d{2}
zookeeper: contains all zookeeper inventory hosts.

Take a look into defaults to check how shards and replicas are being calculated in function of hostnames:

Note that {{ inventory_hostname }} is the DNS or IP address while {{ ansible_hostname }} is the hostname of the machine.

In order to check if the hostname ({{ ansible_hostname}}) is properly set, check in the host machine with following command:

hostname

Requirements

Vagrant with virtual box (see https://www.vagrantup.com/intro)
ansible => 2.10
python3
pip3
Molecule (see https://molecule.readthedocs.io/en/latest/installation.html)

In order to install molecule use python3.


python3 -m pip install --user "molecule"
python3 -m pip install --user "molecule-vagrant"

This solution is based on the ability of resolve hostname in a private dns server. So, assuming vagrant does not provide any dns solution, the following software is installed in prepare.yml, providing vagrant infrastructure with internal dns resolver


---
- name: Prepare
  hosts: all
  tasks:
    - name: Install epel-release
      yum:
        name: epel-release
        state: present
  
    - name: Install nss-mdns
      yum:
        name: nss-mdns
        state: present

    - name: Stop service cron on debian, if running
      systemd:
        name: avahi-daemon
        state: started

with nss-mdns and avahi, vagrant is able to resolve dns just like .local

Notice that DNS resolution supposes to be more sophisticated in a real world use case.

In order to properly configure hostname, take a look into de the following link:

https://www.vagrantup.com/docs/networking/basic_usage#setting-hostname

Architecture

Clickhouse-cluster is built on top of hostname, so ensure hostname is properly set like

^ch\\d{2}-shard\\d{2}-replica\\d{2}

where chX-shardY-replicaZ is the key providing a way to easily discovering which replica belongs what shard based on regex.

Notice that the ansible_hostname is not the same than inventory_hostname:

ansible_hostname: It is the name in the OS: do hostname in the host machine
inventory_hostname: It is the URL (IP or DNS) of the host and must be resolved by other cluster replicas. IP address or DN are both valid.

{{ ansible_hostname }} will be used for discovering replicas while {{ inventory_hostname}} for enabling communications.

An example of inventory could be:

[clickhouse]
<URL-ch01-shard01-replica01> ansible_host=<ip>
<URL-ch01-shard01-replica02> ansible_host=<ip>
...
<URL-ch01-shard02-replica01> ansible_host=<ip>
<URL-ch01-shard02-replica02> ansible_host=<ip>
...
<URL-chX-shardY-replicaZ> ansible_host=<ip>

[zookeeper]
<URL-zookeeper01> ansible_host=<ip>
...
<URL-zookeeperN> ansible_host=<ip>

where URL (inventory_hostname) can be the IP or the DNS that will be resolved in runtime.

Notice that in order to build the cluster, clickhouse group and zookeeper group are mandatory

Design

Download: From yandex rpm repository. Downgrade is allowed by clickhouse_allow_downgrade property.
Config: Ensure clickhouse group & user. Ensure paths and config files. Discover replicas and shards based on regex.
Install: Download and install by yum
Users: Dynamic list to manage users. Password management is not implemented
RBAC: TO BE IMPLEMENTED

Role Default Variables

Check variables in defaults

Cluster Definition

Use these variable to set up the main definition. Notice that cluster configuration relies on hostname and with that, clickhouse_replica_name and clickhouse_shard_name take relevance while clickhouse_hostname_regex is the regex of the hostname definition: ^ch\\d{2}-(shard\\d{2})-replica\\d{2} See vars for more info.


# clickhouse cluster definition
clickhouse_version: "20.8.7.15"
clickhouse_allow_downgrade: false
clickhouse_cluster_name: "mycluster"

clickhouse_service_name: "clickhouse-server"
clickhouse_service_status: "started"

# user/group
clickhouse_group: "clickhouse"
clickhouse_user: "clickhouse"

Clickhouse Installation Support


# yum support 
clickhouse_yum_repo: "https://repo.clickhouse.tech/rpm/stable/x86_64/"
clickhouse_yum_repo_key: "https://repo.clickhouse.tech//CLICKHOUSE-KEY.GPG"
clickhouse_yum_package:
  - "clickhouse-client-{{ clickhouse_version }}"
  - "clickhouse-common-static-{{ clickhouse_version }}"
  - "clickhouse-server-{{ clickhouse_version }}"

# apt support
clickhouse_apt_repo: "deb https://repo.clickhouse.tech/deb/stable/ main/"
clickhouse_apt_repo_keyserver: "keyserver.ubuntu.com"
clickhouse_apt_repo_key: "E0C56BD4"
clickhouse_apt_package:
  - "clickhouse-client={{ clickhouse_version }}"
  - "clickhouse-common-static={{ clickhouse_version }}"
  - "clickhouse-server={{ clickhouse_version }}"

# path
clickhouse_path_config: "/etc/clickhouse-server"
clickhouse_path_config_d: "{{ clickhouse_path_config }}/config.d"
clickhouse_path_log: "/var/log/clickhouse-server"
clickhouse_path_data: "/var/lib/clickhouse"

Clickhouse Configuration

With those variables, main configuration is set up.


clickhouse_config:
  max_connections: 2048
  keep_alive_timeout: 3
  max_concurrent_queries: 100
  uncompressed_cache_size: 8589934592
  mark_cache_size: 5368709120
  builtin_dictionaries_reload_interval: 3600
  max_session_timeout: 3600
  default_session_timeout: 60
  mlock_status: false
  merge_tree_config: []

Networking

These variable are related to networking configuration


clickhouse_http_port: 8123
clickhouse_https_port: 8443
clickhouse_tcp_port: 9000
clickhouse_tcp_secure_port: 9440
clickhouse_interserver_http: 9009

# see vars for clickhouse_listen_host_default
clickhouse_listen_host: "{{ clickhouse_listen_host_default + clickhouse_listen_host_custom }}"

Notice that clickhouse_listen_host must allow ch members listening

Users

Use these vars to customize users. In order to remove a user use clickhouse config attributes: https://clickhouse.tech/docs/en/operations/configuration-files/


# ch users: https://clickhouse.tech/docs/en/operations/configuration-files/
clickhouse_users_list:
  - { user_name: "default",
      profile: "default",
      networks: ["::/1"],
      quota: "default" }

Zookeeper

Zookeeper host list is based on inventory groups pattern


# zookeeper is not at all mandatory. If zookeeper is not installed, 
# replication must be accomplished by the client side
clickhouse_zookeeper_list: "{{ groups['zookeeper'] }}"
clickhouse_zookeeper_port: "2181"

Role Vars Variables

They are variables that are greatest than defaults and inventory groups vars. They can only be overridden by some higher precedence, but normally they are not.

Notice that cluster configuration relies on hostname and with that, clickhouse_replica_name and clickhouse_shard_name take relevance while clickhouse_hostname_regex is the regex of the hostname definition: ^ch\\d{2}-(shard\\d{2})-replica\\d{2} See vars for more info.

Check variables in vars


---
# regex to discover shards and replicas
clickhouse_hostname_regex: "^ch\\d{2}-(shard\\d{2})-(replica\\d{2})"
#discover based on regex
clickhouse_shard_name: "{{ ansible_hostname | regex_search(clickhouse_hostname_regex, '\\1') | first }}"
clickhouse_replica_name: "{{ ansible_hostname | regex_search(clickhouse_hostname_regex, '\\2') | first }}"

# shard list are calculated by replica list. The must accomplish with regex: see clickhouse_hostname_regex in vars/main.yml
clickhouse_shard_list: "{{ clickhouse_replica_list | map('extract', hostvars, 'ansible_hostname') | map('regex_search', clickhouse_hostname_regex, '\\1') | unique | map ('first') }}"
# replica list are all host of a group
clickhouse_replica_list: "{{ groups['clickhouse'] }}"

clickhouse_listen_host_default:
  - "{{ inventory_hostname }}"
  - "127.0.0.1"
  - "::1"

They should NEVER be overridden as they are the core of how the rol discover and relate replicas with shards.

Role Tags

Following tags are supported in this role:

ch:configure: For executing only config tasks.
ch:install: For download and install the software.
ch:service: For managing systemctl service status

Dependencies

Clickhouse depends on Zookeeper to achieve consistency

Example Playbook

Including an example of how to use the rol


    - hosts: my_clickhouse_group
      tasks:
        - include_role:
            name: javigs82.clickhouse
          vars:
            clickhouse_cluster_name: "e-commerce"
            clickhouse_replica_list: "{{ groups['my_clickhouse_group'] }}"
            clickhouse_zookeeper_list: "{{ groups['my_zookeeper_group'] }}"