Some of you may not be familiar with the terms “Rolling upgrade” or “Rolling restart". This is the action of upgrading or restarting a cluster without service interruption (alias zero downtime). In most cases, this is done node by node, but in fact it depends of the technology you’re managing and the number of active nodes in your cluster.
At Nousmotards we have several Java Spring Boot applications running. Restarting one application can take up to 1 min.
A few months ago, I already talked about offloading SSL with Nginx. I also wanted to try it with HAProxy which can be more interesting in some cases.
The good On HAProxy, the good thing is the simplicity to do it. First of all you need to have at least the version 1.5 of HAProxy so to get SSL support. Then you only need those lines to offload SSL:
frontend frontend-https bind :443 ssl crt /etc/haproxy/ssl/server-unified.
As I do not really have a good blog post for today, here are some things I wanted to talk in a summary. Some good projects have been updated like:
WordPress 4: no big updates at the first view but it’s nice to see how this project works and grow (videos, WYSIWYG,…) HAProxy 1.5: Offlloading SSL embedded, Full HTTP Keepalive, ACL enhancement… BIG update! MySecureShell 2.0: Using GnuTLS instead of OpenSSL, available in Debian upstream repositories, new doc… Nginx 1.
The Percona replication manager (PRM) is a framework using the Linux HA resource agent Pacemaker that manages replication and provides automatic failover. This covers the installation of the framework on a set of servers. The PRM framework is made of 4 components: Corosync, Pacemaker, the mysql resource agent and MySQL itself.
It’s easy to setup, better if you already know how to use Pacemaker and it works like a charm. In fact it setup a master and x slaves.
I recently played with HAProxy and discovered that I’ve never made a documentation regarding this fantastic software. So if you don’t know what HAProxy is, it’s a load balancer working on layer 7 and specialized for http protocol.
It is able to handle sticky sessions which is really powerful. More than that, it has a small footprint and can work under a high load traffic.
Here is my documentation.
Last week, I had a webinar regarding Galera Cluster given by Severalnines. As I found it very interesting, I’m sharing it with you :
DRBD is an old stable solution to create block disk replications over the network. However in some case, we need to deal with failures for several reasons (hardware, network issues…).
I’ve updated my current documentation on this point and hope this will help some of you. You can find the update here.