High Availability using centminmod

Post at Load Balancing | Centmin Mod Community asked same question and is pretty much still the same answer.

Google has alot of guides for haproxy and glusterfs/nfs and nginx as a load balancer !
Centmin Mod is provide as is, so short of scripted related bugs or issues, any further optimisation to the web stack components - nginx, php-fpm, mariadb mysql, csf firewall etc or web app specific configurations are left to the Centmin Mod user to deal with. So I do not provide any free support for such.

However, Centmin Mod users are free to help each other out and ask questions or give answers on this community forum. My hopes are that this community forum evolves so that more veteran long time Centmin Mod users help new Centmin Mod users out

Centmin Mod's Nginx has all the nginx modules it requires for you to configure nginx reverse proxy, nginx based load balancing and proxy caching (with exception of the shared file system you'd need i.e. NFS, GlusterFS and any mysql database cluster/replication). You just have to configure it yourself

For nginx reverse proxy & load balancing read the documentation, tutorial and guides below

• NGINX Reverse Proxy | NGINX
• Module ngx_http_proxy_module
• Setting Up simple Proxy server
• How to Set Up Reverse Proxy with Nginx | MaxCDN One
• Understanding Nginx HTTP Proxying, Load Balancing, Buffering, and Caching | DigitalOcean
• Module ngx_http_upstream_module
• Module ngx_stream_upstream_module
• NGINX Load Balancing - HTTP and TCP Load Balancer
• Using nginx as HTTP load balancer
• Load Balancing with NGINX and NGINX Plus, Part 2 | NGINX
• NGINX Boosts Python Performance: Monitor & Load Balance

 

For real high availability you need more than 2 servers, a typical setup minimum setup would be load balanced web servers with load balancer failover and mysql load balanced backends either using mysql master/slave replication or mysql/mariadb galera cluster and then distributed storage via something like GlusterFS with/without NFS or lsyncd/csync and finally a dedicated memcached/redis server(s) for any caching used.

web end

• 3x web server backends
• 2x web load balancers with failover (keepalived/pacemaker etc) talking to those 3x web server backends

distributed storage

• GlusterFS with/without NFS or lsyncd/csync or even a mix for specific types of files/directories with 3-5 servers

mysql end using mysql master/slave replication

• 2x mysql load balancers talking to the mysql master/slave replication with failover (keepalived/pacemaker etc)
• 1x mysql master
• 2x mysql slave

or mysql end using mysql/mariadb galera cluster

• 2x mysql load balancers talking to the mysql master/slave replication with failover
• 3x mariadb/mysql galera cluster

Then for memcached/redis need memcache/redis server

• 1x redis/memcached server
• or best to choose redis in redis master/slave/sentinel based high availability setup which consists of 1x redis master + 1 redis slave + 3x redis sentinel with quorum of 2. Sentinels handle redis master failure/failover to redis slave

Then

• For web load balancers either nginx or haproxy
• For mysql load balancers either haproxy, proxysql or mariadb maxscale

So that is a total of between 14 to 20 servers you'd need. If you want real high availability, you'd also want a 2nd set of same servers in standby configuration too or as a staging set, so you can test HA config changes on the 2nd test set before live production deployment.

You'd need to use dedicated servers for this not VPS unless you have VPS which are not on same VPS host node and have private secure LAN network communication between all VPS servers with at least 1Gbps network connectivity.

 

at say US$60-150 per dedicated server, would be between US$1,000 to US$3,000 per month per set (with 2nd staging/test set 2x costs) but can be 2x to 10x times that depending on disk space as you'd want at least 2x disk raid 1 or 4x disk raid 10 for each server.

all of that though excludes costs of respective server backups done locally + remotely

also you'd need the know-how knowledge of how to install, configure, manage and troubleshoot all the moving pieces of software and setups otherwise any downtime or issues due to your lack of knowledge would effectively negate/cancel out any benefits of such a high availability configuration.

 

Yeah that's part of why Centmin Mod tries to squeeze as much out of each software component performance wise so cumulatively it allows you to run a single server or simplistic server setup (web + db server) much longer before you need to look into HA or multi-server setup when it comes to performance (high availability aside though) or use less servers in a multi-server setup.

So

• if Centmin Mod Nginx server can handle and perform 50% faster than standard distro Nginx versions, that means instead of 15x web servers, you'd only need 10x web servers in a load balanced web server setup. Thus saving you the cost of 5x web servers
• if Centmin Mod PHP-FPM server with PHP 7+ with Profile Guided Optimizations can perform 25% faster than standard PHP-FPM distro binaries, that means instead of 15x PHP-FPM servers, you'd only need 12x PHP-FPM servers. Thus saving you the cost of 3x PHP-FPM servers

Yeah HA only makes sense when the benefits outweigh the additional man hour and financial costs involved. Basically, if you make more $$$$ than the cost of implementing such a HA setup

And yes if you run additional servers like Elasticsearch or Sphinx, then those are extra servers to add to the mix.