Skip to content

Latest commit

 

History

History
309 lines (265 loc) · 11.1 KB

advanced.md

File metadata and controls

309 lines (265 loc) · 11.1 KB
Raw

NOTE: Description below is experimental and might not work as expected or not work at all, DO NOT USE IN PRODUCTION, but report any issues you've encountered, please:)

Advanced usage

Advanced usage includes ability to build cluster of multiple nodes that scale nicely and load balancing.

Good news: all your usual services like are already prepared for this, just need to be configured slightly differently.

For instance, MariaDB instance is not a regular MariaDB server, but rather MariaDB Galera cluster with just one node, so you can scale it to multiple nodes at any time.

The same about PhpMyAdmin, Nginx, PHP-FPM and SSH images - they are all ready to work in cluster, so you don't have to change your code at all, just modify declarative docker-compose.yml file.

TODO: backup/restore/upgrade details needs to be updated here Upgrade procedure remains the same as before, backup and restore depends on your persistent storage.

Setup that works and setup that doesn't

Setup described below is what should ideally work, but doesn't work quite yet (see docker/docker#31157 and linked issues for details).

The main issue is persistent storage. This repository contains Ceph image that can be used as persistent storage by mounting CephFS into target images, but Docker doesn't currently allow to mount anything within containers deployed using Docker Swarm Mode, so we can create Ceph cluster, but can't use it when needed.

In order to overcome this and use built-in clusterization support you'll need to do 2 things:

  • have existing persistent storage and mount volume /data (also /etc/ssh for SSH image) in target services
  • remove from docker-compose.yml following lines, since they are provided by previous step and not working yet:
    environment:
      CEPHFS_MOUNT: 1
# In order to access FUSE
    devices:
      - /dev/fuse:/dev/fuse
# In order to mount CEPHFS
    cap_add:
      - SYS_ADMIN
  • ignore Ceph and Consul images altogether if you don't need them for other purposes

Zero-configuration

All the features provided work with zero-configuration out of the box, all you need is declarative docker-compose.yml file.

IP addresses, configuration files and other stuff will be done for you completely automatically with reasonable defaults (which you can, of course, change at any time).

As you scale your services - new nodes will join existing cluster, removed nodes will leave cluster and failed instances will do their best to re-join existing cluster as soon as they start again.

Everything mentioned above makes building scalable infrastructure a breathe!

If you think some of the features are not configured as good as they can - feel free to open an issue or submit a pull request.

Ceph

One of the first major issue that needs to be solved when creating cluster is storage. While Docker supports volumes and their drivers, they need a lot of manual work to configure them, would be much better to have something more portable.

We solve storage problem with Ceph, more precisely, with Ceph FS.

In order for this to work, we create Ceph cluster and then mount necessary directories with ceph-fuse so that multiple containers on different nodes of the cluster will have access to the same files.

For this purpose we'll use nazarpc/webserver:ceph-v1 image. It can work in different modes, which is why we specify different commands:

# docker-compose.yml
version: '3.1'
services:
...
  ceph-mon:
    image: nazarpc/webserver:ceph-v1
    command: mon
    restart: always

  ceph-osd:
    image: nazarpc/webserver:ceph-v1
    command: osd
    restart: always

  ceph-mds:
    image: nazarpc/webserver:ceph-v1
    command: mds
    restart: always

Currently it is not possible to specify scaling in YAML file, so we'll likely need to scale ceph nodes like following:

docker-compose scale ceph-mon=3 ceph-osd=3 ceph-mds=3

Main environment variables supported:

  • CONSUL_SERVICE - OPTIONAL (defaults to consul), name of Consul service in docker-compose.yml declaration, needed for storing configuration
  • CEPH_MON_SERVICE OPTIONAL (defaults to ceph-mon), name of Ceph monitor service in docker-compose.yml declaration, needed in order to find Ceph monitor nodes in cluster

NOTE: Any features supported by upstream ceph/daemon image are inherently supported by this image.

Consul

Consul is an integral piece of the cluster, since it is actually required for Ceph to store configuration

For this purpose we'll use nazarpc/webserver:consul-v1 image.

# docker-compose.yml
version: '3.1'
services:
...
  consul:
    image: nazarpc/webserver:consul-v1
    restart: always

Currently it is not possible to specify scaling in YAML file (see docker/compose#1661 and docker/compose#2496), so we'll likely need to scale Consul nodes (to at least MIN_SERVERS) like following:

docker-compose scale consul=3

Main environment variables supported:

  • CONSUL_SERVICE - OPTIONAL (defaults to consul), name of Consul service in docker-compose.yml declaration (name of service itself actually), needed to find other Consul nodes in cluster
  • MIN_SERVERS - OPTIONAL (defaults to 3), Consul cluster will wait till MIN_SERVERS started before forming cluster and starting leader election

HAProxy

Load Balancing is also an integral part of cluster robustness and performance. HAProxy might be used to hide behind multiple nodes under single entry point and distribute incoming requests across those nodes. HAProxy is only useful for TCP connections.

For this purpose we'll use nazarpc/webserver:haproxy-v1 image, for instance:

# docker-compose.yml
version: '3.1'
services:
...
  mariadb-haproxy:
    image: nazarpc/webserver:haproxy-v1
    restart: always
    environment:
      SERVICE_NAME: mariadb
      SERVICE_PORTS: 3306

Main environment variables supported:

  • SERVICE_NAME - REQUIRED, name of service for load balancing, for multiple services create multiple HAProxy services, they're lightweight, so you can create as many of them as needed
  • SERVICE_PORTS - REQUIRED, coma and/or space-separated list of ports that HAProxy will listen, HAProxy works as transparent proxy, so input port is the same as output port

In example above all requests to the host mariadb-haproxy and port 3306 will be sent over to one of running mariadb instances. You can also scale mariadb-haproxy instance if needed.

MariaDB

In order to scale MariaDB we'll need to build MariaDB Galera cluster. nazarpc/webserver:mariadb-v1 image is in fact already a cluster with single node - so, it it ready to scale at any time with Master-Master replication mode.

# docker-compose.yml
version: '3.1'
services:
...
  mariadb:
    image: nazarpc/webserver:mariadb-v1
    restart: always
    environment:
      CEPHFS_MOUNT: 1
# In order to access FUSE
    devices:
      - /dev/fuse:/dev/fuse
# In order to mount CEPHFS
    cap_add:
      - SYS_ADMIN

Main environment variables supported:

  • SERVICE_NAME - OPTIONAL (defaults to mariadb), name of MariaDB service in docker-compose.yml declaration, needed in order to find other MariaDB nodes in cluster
  • CEPH_MON_SERVICE OPTIONAL (defaults to ceph-mon), name of Ceph monitor service in docker-compose.yml declaration, needed in order to find Ceph monitor nodes in cluster
  • CEPHFS_MOUNT OPTIONAL (defaults to 0), when set to 1, container will try to mount CephFS on start

If you scale mariadb instance - new nodes will join existing cluster and start to replicate:

docker-compose scale mariadb=3

Nginx, PHP-FPM and SSH

They are very similar to MariaDB in terms of configuration, but don't create clusters by themselves.

All you need to specify is that CephFS should be mounted:

# docker-compose.yml
version: '3.1'
services:
...
  nginx:
    image: nazarpc/webserver:nginx-v1
    restart: always
    environment:
      CEPHFS_MOUNT: 1
# In order to access FUSE
    devices:
      - /dev/fuse:/dev/fuse
# In order to mount CEPHFS
    cap_add:
      - SYS_ADMIN

  php:
    image: nazarpc/webserver:php-fpm-v1
    restart: always
    environment:
      CEPHFS_MOUNT: 1
# In order to access FUSE
    devices:
      - /dev/fuse:/dev/fuse
# In order to mount CEPHFS
    cap_add:
      - SYS_ADMIN

  ssh:
    image: nazarpc/webserver:ssh-v1
    restart: always
    environment:
      CEPHFS_MOUNT: 1
# In order to access FUSE
    devices:
      - /dev/fuse:/dev/fuse
# In order to mount CEPHFS
    cap_add:
      - SYS_ADMIN

Main environment variables supported (common to all 3 images):

  • CEPH_MON_SERVICE OPTIONAL (defaults to ceph-mon), name of Ceph monitor service in docker-compose.yml declaration, needed in order to find Ceph monitor nodes in cluster
  • CEPHFS_MOUNT OPTIONAL (defaults to 0), when set to 1, container will try to mount CephFS on start

Combined example

# docker-compose.yml
version: '3.1'
services:
  consul:
    image: nazarpc/webserver:consul-v1
    restart: always

  ceph-mon:
    image: nazarpc/webserver:ceph-v1
    command: mon
    restart: always

  ceph-osd:
    image: nazarpc/webserver:ceph-v1
    command: osd
    restart: always

  ceph-mds:
    image: nazarpc/webserver:ceph-v1
    command: mds
    restart: always

  mariadb-haproxy:
    image: nazarpc/webserver:haproxy-v1
    restart: always
    environment:
      SERVICE_NAME: mariadb
      SERVICE_PORTS: 3306

  mariadb:
    image: nazarpc/webserver:mariadb-v1
    restart: always
    environment:
      CEPHFS_MOUNT: 1
# In order to access FUSE
    devices:
      - /dev/fuse:/dev/fuse
# In order to mount CEPHFS
    cap_add:
      - SYS_ADMIN

  nginx:
    image: nazarpc/webserver:nginx-v1
    restart: always
    environment:
      CEPHFS_MOUNT: 1
# In order to access FUSE
    devices:
      - /dev/fuse:/dev/fuse
# In order to mount CEPHFS
    cap_add:
      - SYS_ADMIN

  php:
    image: nazarpc/webserver:php-fpm-v1
    restart: always
    environment:
      CEPHFS_MOUNT: 1
# In order to access FUSE
    devices:
      - /dev/fuse:/dev/fuse
# In order to mount CEPHFS
    cap_add:
      - SYS_ADMIN

  phpmyadmin:
    image: nazarpc/webserver:phpmyadmin-v1
    restart: always
    environment:
      MYSQL_HOST: mariadb
  
  ssh:
    image: nazarpc/webserver:ssh-v1
    restart: always
    environment:
      CEPHFS_MOUNT: 1
# In order to access FUSE
    devices:
      - /dev/fuse:/dev/fuse
# In order to mount CEPHFS
    cap_add:
      - SYS_ADMIN
docker-compose up -d
docker-compose scale consul=3 ceph-mon=3 ceph-osd=3 ceph-mds=3 mariadb-haproxy=2 mariadb=3 nginx=2 php=3

After this we'll have:

  • 3 Consul nodes in cluster
  • Ceph cluster with 3 monitors, 3 OSDs and 3 metadata servers
  • 2 HAProxy instances for MariaDB load balancing
  • 3 MariaDB Galera nodes in cluster with Master-Master replication mode
  • 2 Nginx instances
  • 3 PHP-FPM instances
  • 1 PhpMyAdmin instance to work with MariaDB Galera cluster in GUI mode
  • 1 SSH server to access files and configs

TODO: Extend example with Docker Swarm Mode